Citation |

- Permanent Link:
- http://ufdc.ufl.edu/AA00011802/00001
## Material Information- Title:
- Comparison of males and females on math item performance analysis of response patterns
- Creator:
- Feliciano, Sonia, 1938-
- Publication Date:
- 1986
- Language:
- English
- Physical Description:
- x, 117 leaves : ill. ; 28 cm.
## Subjects- Subjects / Keywords:
- Arithmetic ( jstor )
College mathematics ( jstor ) Educational research ( jstor ) Learning ( jstor ) Mathematical variables ( jstor ) Mathematics ( jstor ) Mathematics achievement ( jstor ) Mathematics education ( jstor ) Problem solving ( jstor ) Sex linked differences ( jstor ) Dissertations, Academic -- Foundations of Education -- UF Foundations of Education thesis Ph. D Mathematical ability ( lcsh ) Sex differences ( lcsh ) City of Madison ( local ) - Genre:
- bibliography ( marcgt )
non-fiction ( marcgt )
## Notes- Thesis:
- Thesis (Ph. D.)--University of Florida, 1986.
- Bibliography:
- Bibliography: leaves 107-115.
- General Note:
- Typescript.
- General Note:
- Vita.
- Statement of Responsibility:
- by Sonia Feliciano.
## Record Information- Source Institution:
- University of Florida
- Holding Location:
- University of Florida
- Rights Management:
- Copyright [name of dissertation author]. Permission granted to the University of Florida to digitize, archive and distribute this item for non-profit research and educational purposes. Any reuse of this item in excess of fair use or other copyright exemptions requires permission of the copyright holder.
- Resource Identifier:
- 029704760 ( ALEPH )
15167202 ( OCLC ) AEJ5569 ( NOTIS )
## UFDC Membership |

Downloads |

## This item has the following downloads:
E7GIIQLHZ_4RJU7A.xml
AA00011802_00001.pdf AA00011802_00001_0094.txt AA00011802_00001_0026.txt AA00011802_00001_0235.txt comparisonofmale00feli_0055.txt comparisonofmale00feli_0071.txt AA00011802_00001_0201.txt AA00011802_00001_0067.txt comparisonofmale00feli_0079.txt comparisonofmale00feli_0101.txt comparisonofmale00feli_0028.txt comparisonofmale00feli_0104.txt AA00011802_00001_0215.txt AA00011802_00001_0070.txt AA00011802_00001_0029.txt AA00011802_00001_0095.txt AA00011802_00001_0017.txt comparisonofmale00feli_0037.txt comparisonofmale00feli_0123.txt AA00011802_00001_0021.txt AA00011802_00001_0031.txt AA00011802_00001_0065.txt comparisonofmale00feli_0041.txt AA00011802_00001_0124.txt comparisonofmale00feli_0115.txt AA00011802_00001_0086.txt AA00011802_00001_0009.txt AA00011802_00001_0160.txt AA00011802_00001_0176.txt comparisonofmale00feli_0058.txt AA00011802_00001_0119.txt comparisonofmale00feli_0002.txt comparisonofmale00feli_0082.txt comparisonofmale00feli_0051.txt comparisonofmale00feli_0044.txt comparisonofmale00feli_0090.txt AA00011802_00001_0110.txt comparisonofmale00feli_0106.txt comparisonofmale00feli_0019.txt comparisonofmale00feli_0054.txt comparisonofmale00feli_0021.txt AA00011802_00001_0001.txt comparisonofmale00feli_0015.txt comparisonofmale00feli_0085.txt AA00011802_00001_0004.txt comparisonofmale00feli_0031.txt AA00011802_00001_0082.txt comparisonofmale00feli_0094.txt AA00011802_00001_0025.txt AA00011802_00001_0192.txt comparisonofmale00feli_0118.txt AA00011802_00001_0016.txt AA00011802_00001_0044.txt AA00011802_00001_0143.txt AA00011802_00001_0219.txt AA00011802_00001_0040.txt comparisonofmale00feli_0017.txt AA00011802_00001_0147.txt comparisonofmale00feli_0110.txt comparisonofmale00feli_0046.txt AA00011802_00001_0136.txt comparisonofmale00feli_0049.txt comparisonofmale00feli_0072.txt comparisonofmale00feli_0020.txt AA00011802_00001_0039.txt AA00011802_00001_0210.txt AA00011802_00001_0208.txt AA00011802_00001_0248.txt comparisonofmale00feli_0117.txt AA00011802_00001_0232.txt comparisonofmale00feli_0077.txt comparisonofmale00feli_0048.txt comparisonofmale00feli_0006.txt comparisonofmale00feli_0099.txt AA00011802_00001_0171.txt AA00011802_00001_0060.txt AA00011802_00001_0184.txt comparisonofmale00feli_0119.txt AA00011802_00001_0059.txt AA00011802_00001_0130.txt AA00011802_00001_0175.txt AA00011802_00001_0178.txt AA00011802_00001_0172.txt AA00011802_00001_0083.txt AA00011802_00001_0209.txt AA00011802_00001_0011.txt AA00011802_00001_pdf.txt comparisonofmale00feli_0024.txt AA00011802_00001_0223.txt AA00011802_00001_0108.txt AA00011802_00001_0157.txt AA00011802_00001_0179.txt comparisonofmale00feli_0066.txt AA00011802_00001_0084.txt AA00011802_00001_0211.txt AA00011802_00001_0071.txt comparisonofmale00feli_0063.txt AA00011802_00001_0154.txt AA00011802_00001_0194.txt comparisonofmale00feli_0014.txt AA00011802_00001_0227.txt AA00011802_00001_0015.txt AA00011802_00001_0080.txt comparisonofmale00feli_0061.txt AA00011802_00001_0239.txt AA00011802_00001_0141.txt AA00011802_00001_0023.txt AA00011802_00001_0207.txt AA00011802_00001_0096.txt AA00011802_00001_0164.txt AA00011802_00001_0234.txt comparisonofmale00feli_0053.txt AA00011802_00001_0035.txt AA00011802_00001_0010.txt AA00011802_00001_0032.txt AA00011802_00001_0002.txt comparisonofmale00feli_0089.txt AA00011802_00001_0052.txt AA00011802_00001_0203.txt AA00011802_00001_0206.txt comparisonofmale00feli_0000.txt AA00011802_00001_0243.txt AA00011802_00001_0006.txt AA00011802_00001_0100.txt comparisonofmale00feli_0039.txt comparisonofmale00feli_0010.txt AA00011802_00001_0198.txt AA00011802_00001_0214.txt AA00011802_00001_0180.txt AA00011802_00001_0079.txt comparisonofmale00feli_0105.txt AA00011802_00001_0107.txt AA00011802_00001_0205.txt comparisonofmale00feli_0120.txt AA00011802_00001_0224.txt AA00011802_00001_0074.txt comparisonofmale00feli_0036.txt AA00011802_00001_0241.txt AA00011802_00001_0151.txt comparisonofmale00feli_0033.txt AA00011802_00001_0181.txt comparisonofmale00feli_0025.txt AA00011802_00001_0173.txt AA00011802_00001_0077.txt comparisonofmale00feli_0009.txt comparisonofmale00feli_0097.txt comparisonofmale00feli_0109.txt comparisonofmale00feli_0030.txt AA00011802_00001_0113.txt AA00011802_00001_0185.txt AA00011802_00001_0245.txt AA00011802_00001_0042.txt AA00011802_00001_0118.txt comparisonofmale00feli_0005.txt comparisonofmale00feli_0043.txt AA00011802_00001_0078.txt AA00011802_00001_0104.txt AA00011802_00001_0246.txt AA00011802_00001_0222.txt comparisonofmale00feli_0075.txt AA00011802_00001_0048.txt AA00011802_00001_0068.txt AA00011802_00001_0144.txt AA00011802_00001_0097.txt AA00011802_00001_0088.txt AA00011802_00001_0076.txt AA00011802_00001_0255.txt comparisonofmale00feli_0040.txt comparisonofmale00feli_0073.txt AA00011802_00001_0166.txt AA00011802_00001_0046.txt AA00011802_00001_0132.txt AA00011802_00001_0140.txt AA00011802_00001_0091.txt AA00011802_00001_0225.txt comparisonofmale00feli_0121.txt AA00011802_00001_0024.txt AA00011802_00001_0202.txt AA00011802_00001_0155.txt AA00011802_00001_0055.txt AA00011802_00001_0122.txt AA00011802_00001_0226.txt AA00011802_00001_0034.txt AA00011802_00001_0218.txt comparisonofmale00feli_0128.txt comparisonofmale00feli_0070.txt AA00011802_00001_0090.txt AA00011802_00001_0251.txt AA00011802_00001_0186.txt AA00011802_00001_0145.txt comparisonofmale00feli_0084.txt AA00011802_00001_0075.txt AA00011802_00001_0216.txt AA00011802_00001_0102.txt AA00011802_00001_0258.txt AA00011802_00001_0087.txt AA00011802_00001_0148.txt comparisonofmale00feli_0102.txt AA00011802_00001_0187.txt comparisonofmale00feli_0124.txt comparisonofmale00feli_0026.txt AA00011802_00001_0114.txt comparisonofmale00feli_0095.txt AA00011802_00001_0162.txt AA00011802_00001_0169.txt comparisonofmale00feli_0107.txt AA00011802_00001_0053.txt AA00011802_00001_0238.txt AA00011802_00001_0242.txt AA00011802_00001_0182.txt comparisonofmale00feli_0050.txt AA00011802_00001_0204.txt AA00011802_00001_0195.txt comparisonofmale00feli_0088.txt AA00011802_00001_0064.txt AA00011802_00001_0117.txt AA00011802_00001_0135.txt AA00011802_00001_0045.txt comparisonofmale00feli_0069.txt AA00011802_00001_0231.txt AA00011802_00001_0253.txt AA00011802_00001_0250.txt comparisonofmale00feli_0067.txt AA00011802_00001_0163.txt comparisonofmale00feli_0076.txt AA00011802_00001_0098.txt AA00011802_00001_0105.txt AA00011802_00001_0043.txt comparisonofmale00feli_0004.txt AA00011802_00001_0240.txt AA00011802_00001_0054.txt comparisonofmale00feli_0042.txt AA00011802_00001_0128.txt AA00011802_00001_0212.txt AA00011802_00001_0193.txt AA00011802_00001_0146.txt AA00011802_00001_0188.txt comparisonofmale00feli_0086.txt comparisonofmale00feli_0023.txt comparisonofmale00feli_0022.txt AA00011802_00001_0049.txt comparisonofmale00feli_0065.txt AA00011802_00001_0170.txt comparisonofmale00feli_0122.txt AA00011802_00001_0103.txt AA00011802_00001_0237.txt AA00011802_00001_0099.txt AA00011802_00001_0190.txt comparisonofmale00feli_0056.txt AA00011802_00001_0012.txt AA00011802_00001_0038.txt AA00011802_00001_0093.txt comparisonofmale00feli_0057.txt comparisonofmale00feli_0093.txt AA00011802_00001_0022.txt AA00011802_00001_0116.txt AA00011802_00001_0092.txt comparisonofmale00feli_0127.txt AA00011802_00001_0159.txt AA00011802_00001_0061.txt AA00011802_00001_0057.txt AA00011802_00001_0072.txt AA00011802_00001_0126.txt AA00011802_00001_0062.txt AA00011802_00001_0030.txt AA00011802_00001_0230.txt comparisonofmale00feli_0112.txt AA00011802_00001_0134.txt AA00011802_00001_0005.txt AA00011802_00001_0161.txt AA00011802_00001_0233.txt comparisonofmale00feli_0113.txt AA00011802_00001_0156.txt AA00011802_00001_0158.txt comparisonofmale00feli_0038.txt AA00011802_00001_0213.txt AA00011802_00001_0137.txt AA00011802_00001_0153.txt comparisonofmale00feli_0013.txt AA00011802_00001_0196.txt comparisonofmale00feli_0027.txt AA00011802_00001_0221.txt AA00011802_00001_0111.txt AA00011802_00001_0041.txt AA00011802_00001_0236.txt AA00011802_00001_0174.txt AA00011802_00001_0191.txt comparisonofmale00feli_0100.txt AA00011802_00001_0257.txt AA00011802_00001_0020.txt AA00011802_00001_0199.txt AA00011802_00001_0127.txt AA00011802_00001_0120.txt comparisonofmale00feli_0078.txt comparisonofmale00feli_0016.txt comparisonofmale00feli_0116.txt comparisonofmale00feli_0018.txt AA00011802_00001_0028.txt comparisonofmale00feli_0080.txt AA00011802_00001_0003.txt comparisonofmale00feli_0092.txt AA00011802_00001_0066.txt E7GIIQLHZ_4RJU7A_xml.txt AA00011802_00001_0167.txt comparisonofmale00feli_0114.txt AA00011802_00001_0013.txt AA00011802_00001_0123.txt comparisonofmale00feli_0008.txt comparisonofmale00feli_0103.txt AA00011802_00001_0152.txt AA00011802_00001_0244.txt AA00011802_00001_0027.txt AA00011802_00001_0036.txt comparisonofmale00feli_0074.txt AA00011802_00001_0019.txt AA00011802_00001_0142.txt AA00011802_00001_0089.txt AA00011802_00001_0051.txt comparisonofmale00feli_0001.txt AA00011802_00001_0069.txt AA00011802_00001_0168.txt comparisonofmale00feli_0125.txt comparisonofmale00feli_0034.txt AA00011802_00001_0228.txt AA00011802_00001_0183.txt comparisonofmale00feli_0111.txt AA00011802_00001_0177.txt AA00011802_00001_0165.txt AA00011802_00001_0247.txt AA00011802_00001_0047.txt AA00011802_00001_0007.txt AA00011802_00001_0018.txt comparisonofmale00feli_0012.txt AA00011802_00001_0033.txt comparisonofmale00feli_0052.txt comparisonofmale00feli_0035.txt comparisonofmale00feli_0064.txt AA00011802_00001_0254.txt AA00011802_00001_0101.txt AA00011802_00001_0139.txt AA00011802_00001_0249.txt comparisonofmale00feli_0081.txt AA00011802_00001_0200.txt AA00011802_00001_0121.txt comparisonofmale00feli_0068.txt AA00011802_00001_0189.txt comparisonofmale00feli_0060.txt comparisonofmale00feli_0011.txt comparisonofmale00feli_0087.txt AA00011802_00001_0106.txt AA00011802_00001_0138.txt AA00011802_00001_0150.txt comparisonofmale00feli_0029.txt AA00011802_00001_0073.txt comparisonofmale00feli_0091.txt AA00011802_00001_0125.txt AA00011802_00001_0037.txt AA00011802_00001_0217.txt comparisonofmale00feli_0032.txt AA00011802_00001_0085.txt AA00011802_00001_0252.txt comparisonofmale00feli_0045.txt comparisonofmale00feli_0126.txt AA00011802_00001_0058.txt AA00011802_00001_0112.txt AA00011802_00001_0256.txt AA00011802_00001_0056.txt comparisonofmale00feli_0059.txt comparisonofmale00feli_0047.txt AA00011802_00001_0081.txt AA00011802_00001_0197.txt AA00011802_00001_0229.txt AA00011802_00001_0131.txt AA00011802_00001_0149.txt AA00011802_00001_0129.txt comparisonofmale00feli_0083.txt AA00011802_00001_0008.txt comparisonofmale00feli_0062.txt AA00011802_00001_0063.txt AA00011802_00001_0133.txt AA00011802_00001_0220.txt AA00011802_00001_0014.txt comparisonofmale00feli_0096.txt AA00011802_00001_0115.txt comparisonofmale00feli_0007.txt AA00011802_00001_0050.txt comparisonofmale00feli_0003.txt comparisonofmale00feli_0098.txt AA00011802_00001_0109.txt comparisonofmale00feli_0108.txt |

Full Text |

COMPARISON OF MALES AND FEMALES ON MATH ITEM PERFORMANCE: ANALYSIS OF RESPONSE PATTERNS By SONIA FELICIANO A DISSERTATION PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY UNIVERSITY OF FLORIDA 1986 ACKNOWLEDGMENTS I would like to acknowledge my gratitude to several people who have influenced my formal education and/or made this study possible. My special thanks to Dr. James Algina, Chairman of the doctoral committee who contributed to the development of my love for research and statistics. He has been insuperable as professor and valued friend; his help and guidance in the preparation and completion of this study were invaluable. I extend my thanks to Dr. Linda Crocker for her advice and help during my doctoral studies at University of Florida. Thanks also go to Dr. Michael Nunnery, member of the doctoral committee. To Dr. Wilson Guertin, who was a friend for me and my family, I extend my special thanks. Thanks are also extended to Dr. Amalia Charneco, past Undersecretary of Education of the Puerto Rico Department of Education, for her continuous support. To my sister Nilda Santaellar who typed the thesis, I give my sincere thanks. Special thanks go to my family and to those friends who provided encouragement throughout this critical period of my life. TABLE OF CONTENTS Page ACKNOWLEDGMENTS.... .......................... ii LIST OF TABLES .. ................ ................... v LIST OF FIGURES...................... ............... vii ABSTRACT....... ...................................... viii CHAPTER I INTRODUCTION.................................. 1 Purpose of the Study ............... ........ 6 Significance of the Study................... 6 Organization of the Study................... 8 II REVIEW OF THE LITERATURE ...................... 9 Sex-related Differences in Incorrect Response Patterns........................ 10 Sex-related Differences in Problem Solving............... ..... ........... .... 14 Cognitive and Affective Variables that Influence Sex Differences in Mathematics Learning and Achievement................ 23 Differences in Formal Mathematics Education....................... 24 Differences in Spatial Ability.......... 26 Differentiated Effect of Affective Variables................... ....... ... 30 Problem Solving Performance and Related Variables.... ..................... ......... 37 Computational Skills and Problem Solving Performance. .......................... 38 Reading and Problem Solving Perfor- mance........................... .... .. 44 Attitudes Toward Problem Solving and Problem Solving Performance........... 50 iii III METHOD..... ... ...... ...... ..... ............... 54 The Sample ................ ................... 54 The Instrument............. ................ 55 Analysis of the Data....................... 57 Analysis of Sex by Option by Year Cross Classifications................ 57 Comparison of Males and Females in Problem Solving Performance........... 66 IV RESULTS........... ..... ....... ...... .. ........ 68 Introduction ............................. 68 Sex-related Differences in the Selection of Incorrect Responses.......... 68 Sex-related Differences in Problem Solving Performance.................... 81 Summary.................................... 93 V DISCUSSION..................................... 101 Summary and Interpretation of the Results ................................... 101 Implications of the Findings and Suggestions for Further Research.......... 103 Sex-related Differences in Incorrect Responses................ 104 Sex-related Differences in Problem Solving Performance................... 105 REFERENCES.......................................... 107 BIOGRAPHICAL SKETCH................................ 116 LIST OF TABLES Table Page 3.1 Hypothetical Probabilities of Option Choice Conditional on Year and Sex.......... 58 3.2 Hypothetical Probabilities of Option Choice Conditional on Year and Sex and Arranged by Sex......... ...... ........... 58 3.3 Hypothetical Joint Probabilities of Yearr Option, and Sex............................. 64 3.4 Hypothetical Probabilities of Option Choice Conditional on Sex and Year.......... 66 4.1 Summary of Significant Tests of Model 3.2..... 71 4.2 Chi-square Values for the Comparison of 75 Models (3.2) and (3.3) with Actual Sample Sizes and with the Corresponding Number of Subjects Needed for Significant Results 75 4.3 Means, Standard Deviations, and t-Test for the Eight Mathematical Variables............ 82 4.4 ANCOVA First Year: Multiplication Covariate 85 4.5 ANCOVA First Year: Division Covariate....... 85 4.6 ANCOVA Second Year: Subtraction Covariate... 86 4.7 Reliability of the Covariates for Each of the Three Years of Test Administra- tion........................................ 91 4.8 ANCOVA First Year: Other Covariates......... 92 4.9 Adjusted Means on Problem Solving, by Covariate and Sex.......................... 94 4.10 ANCOVA Second Year: Other Covariates........ 95 Table Page 4.11 Adjusted Means on Problem Solving, by Covariate and Sex........................ 96 4.12 ANCOVA Third Year........................ 97 4.13 Adjusted Means on Problem Solving, by Covariate and Sex....................... 98 LIST OF FIGURES Figure Page 3.1 Number of Questions by Skill Area in a Basic Skills Test in Mathematics-6.......... 56 4.1 Sex by Multiplication Interaction............. 87 4.2 Sex by Division Interaction ................ 88 4.3 Sex by Subtraction Interaction................ 89 vii Abstract of Dissertation Presented to the Graduate School of the University of Florida in Partial Fulfillment of the Requirements for the Degree of Doctor of Philosophy COMPARISON OF MALES AND FEMALES ON MATH ITEM PERFORMANCE: ANALYSIS OF RESPONSE PATTERNS By Sonia Feliciano August, 1986 Chairman: James Algina Major Department: Foundations of Education The first objective of this study was to investigate sex differences in the selection of incorrect responses on a mathematics multiple-choice test, and to determine whether these differences were consistent over three consecutive administrations of the test. A second objective was to compare male and female performance in problem solving after controlling for computational skills. The responses of all 6th grade students from the public schools in Puerto Rico who took the Basic Skills Test in Mathematics-6 ("Prueba de Destrezas B&sicas en Matem&ticas-6") during three academic years were used in the analyses relevant to the first objective. viii Log-linear models were used in the analysis of incorrect responses. The results of the analyses showed that for 100 of the 111 items of the test, males and females selected different incorrect options, and this pattern of responses was consistently found during the three years of test administration. However, for the vast majority of the 100 items the male-female differences were relatively small, considering the fact that the number of subjects needed to obtain statistical significance was very large. The responses of approximately 1,000 randomly selected students per academic year were analyzed in the comparison of male and female performance in problem solving. Females outperformed males in problem solving and in six of the seven computational variables. Males showed superiority in equivalence in all the three years, but statistical significance was obtained in only one of the years. Analysis of covariance (ANCOVA) was used in the comparison of male and female performance in problem solving after controlling for computational skills. Seven analyses of covariance tests were conducted, one for each of the covariates. Estimated true scores for observed scores were used in the analyses. The results tend to show that for examinees with similar levels of computational skills, sex-related differences in problem solving performance do not exist. Females retained their superiority in problem solving when equivalence (in all three years) and subtraction (in one year) were the controlling variables. The question of whether male-female differences in problem solving depend on computational skills was answered, partially, in the affirmative. CHAPTER I INTRODUCTION Sex differences in mathematics learning and achievement have been the subject of intensive research. Research done before 1974 has shown that male performance on mathematical achievement tests is superior to female performance by the time they reach upper elementary or junior high school (Fennema, 1976, p. 2). The literature strongly suggests that at the elementary level females outperform males in computation and males excel in mathematical reasoning (Glennon & Callahan, 1968; Jarvis, 1964; Maccoby, 1966). Since 1974, research findings have been less consistent. Fennema (1974), after reviewing 36 studies, found that during secondary school or earlier, sex-related differences in mathematics achievement are not so evident, but that when differences are founds they favor males in high level cognitive tasks (problem solving) and females in low level cognitive tasks (computation). As a result of a further review of the literature, Fennema (1977) concluded that at the elementary level, sex-related differences do not exist at all cognitive levels, from computation to problem solving. Many variables, cognitive affective, and educational, have been investigated since 1974 in relation to sex differences in mathematics learning and achievement. Fennema and Sherman (1977) investigated the effect of differential formal mathematics education. After controlling for the number of years of exposure to the study of mathematics, they found sex differences in only two of the four schools under study. However, in those schools where boys scored higher than girls, differences were also found in their attitudes toward mathematics. Hilton and Berglund (1974) found significant sex differences after controlling for the number of mathematics courses taken, and attributed them to sex differences in interests. "As the boys' interests in science increase related to the girls', their achievement in mathematics increases relative to that of the girls" (p. 234). Wise, Steel, and McDonald (1979) reanalyzed test data collected in a longitudinal study of 400,000 high school students (Project Talent). They found that when the effect of the number of high school mathematics courses was not controlled, no sex differences emerged for 9th graders, but that gains made by boys during the next three years were more than twice that of the girls. These differences between the sexes disappeared when the number of mathematics courses taken was controlled. Results of the 1978 Women and Mathematics National Survey, Survey I, indicated no significant sex differences for 8th grade students on measures of problem solving or algebra. However, females outperformed males in computation and spatial visualization. For the 12th grade students, statistically significant sex differences favoring males were found in problem solving, but not in algebra, computation, or spatial visualization. For males and females who had enrolled in courses beyond general mathematics and who had taken or were enrolled in courses such as pre-calculus, calculus, or geometry, differences in problem solving or spatial visualization did not exist. Sex differences favoring males were found on a total score obtained summing across the computation, problem solving, and algebra subtests (Armstrong, 1979). The mathematics data collected in the second survey by the 1978 National Assessment of Educational Progress showed significant sex differences for both 13- and 17-year-old students. The 13-year-old females outperformed males in the computational subtest and males outscored females by 1 1/2 percentage points in problem solving (statistically significant). No statistically significant differences were found in algebra. No sex differences were found for the 17-year-old group either in the computation subtest or in the algebra subtest. Males surpassed females in problem solving. A reanalysis of the data from the 17-year-old group confirmed male superiority in problem solving after controlling for mathematics preparation. Males who were enrolled or had completed algebra II outperformed the females in computation and problem solving but not in algebra. Males who studied beyond algebra II outscored females on all three subtests: computation, algebra, and problem solving (Armstrong, 1979). Carpenter, Lindquist, Mathews, and Silver (1984) analyzed the results of the Third National Assessment of Educational Progress (NAEP), and compared them with the First and Second Surveys. Between 1978 and 1982, the differences between the average performance of males and females remained stable at each age level. At ages 9 and 13, the overall performance of males and females was not significantly different. At age 17r males scored higher than females by about 3 percentage points. When course background was held constant, achievement differences still existed at age 17. For each category of course background, male achievement exceeded female achievement. Consistent with previous assessments, sex differences in problem solving in favor of males were found for the 17-year-old sample. At ages 9 and 13, no large differences were found between the sexes within any level of course background. Marshall (1981, 1984) investigated sex differences in mathematics performance. She found that males and females excel each other in solving different types of problems. Females were better on items of computation and males were more successful on word-story problem items (problem solving). She also found that females successful performance in the problem solving items was more dependent on their successful performance in the computation items. Males did not need, as much as females, to succeed in the computation items in order to answer correctly the problem solving items. Although the general findings seem to support sex differences in mathematics learning and achievement, the research done does not consistently support superiority for either sex. Most of the research has been concerned with how the sexes differ on subtests or total test scores in mathematics. Moreover, the great majority of the studies deal with correct responses. Sex differences in incorrect responses at the item level have not been fully researched. Only two studies dealing with sex differences in incorrect responses at the item level were found in the research literature (Marshall, 1981, 1983). Marshall investigated whether boys and girls made similar errors in computation and story problems. She analyzed boys' and girls' answers to six mathematics items and found that the sexes made different errors, possibly reflecting different problem solving strategies. Her original findings were supported when she studied the same problem using a large number of items three years later. Purpose of the Study The purpose of this study was twofold. The first was to investigate sex-related differences in the selection of incorrect responses in a multiple-choice achievement test in mathematics. For each test item, the research questions were as follows: 1. Is there a difference in the proportion of males and females choosing each incorrect option? 2. Is the same pattern of differences found in data obtained in three different administrations of the test? The second objective was to investigate sex-related differences in test scores in mathematics problem solving. The following questions were studied: 1. Do males and females differ in problem solving performance? 2. Do sex differences in problem solving persist after controlling for computational skills, and does the differential success of males and females on problem solving items depend on their success on the computation items? Significance of the Study Item response patterns are very useful techniques in the assessment of mathematics learning and achievement. Total test scores can be very misleading in the assessment of student performance and provide no diagnostic information about the nature and seriousness of student errors (Harnisch, 1983). Item response patterns are valuable for the identification of large group differences, including district-to-district, school-to-school, and classroom-to-classroom variations on different subsets of items. The response patterns can provide diagnostic information about the type of understanding the student has on various mathematics topics (e.g.r problem solving). Marshall (1981, 1984) has used the item response pattern technique and her findings indicate sex differences in mathematics performance at the item level. Females outperformed males in computation and males outscored females in problem solving. Also the success of girls in the problem solving items was dependent upon their success in the computation items; for boys, success in the problem solving items did not depend as much on their computational performance. Marshall (1981, 1983) has also reported that males and females differ in the selection of incorrect responses, reflecting differences in reasoning abilities. In Puerto Rico, a high percentage of children promoted to the 7th grade in the public schools does not master the basic skills in mathematics. If 6th grade male and female children can be diagnosed as having different problem solving abilities, as Marshall found with California children, teachers may need to provide tailor-made mathematics instruction for each sex, in order to ensure equal access to formal education and enhance mathematics achievement. Since there are no investigations reported in sex differences in item response patterns in Puerto Rico, research is needed. Organization of the Study A review of the literature on sex differences in mathematics performance is reported in Chapter II. The research methodology is presented in Chapter III. Research questions, sample, instrument, and data analysis are discussed in that chapter. Chapter IV is an exposition of the results of the study. Chapter V contains a summary and interpretation of the results of the study and the implications of the findings together with suggestions for further research. CHAPTER II REVIEW OF THE LITERATURE Sex differences in mathematics learning and achievement have been a subject of concern for educators and psychologists. Many studies found in the literature support the existence of these differences. Boys show superiority in higher level cognitive tasks (problem solving or mathematical reasoning) in the upper elementary years and in the early high school years (Fennemar 1974; Maccoby & Jacklin, 1974). Almost all the research carried out has dealt with analysis of total correct scores in mathematics aptitude and achievement tests or scores in subtests. The literature related to sex differences in incorrect responses, the main subject of the present study, is surprisingly sparse. For the most part, the studies have investigated the differences between the sexes in mathematics learning and achievement and the underlying variables causing the differences. Cognitive and affective variables have been the matter at issue in the establishment of sex differences. Although research in mathematics problem solving, the secondary subject of this investigation, is extensive, most of the studies consider sex differences incidental to the major study findings. The available literature offers very little research directly related to the problem of sex differences in this area. The review of the literature has been divided in four sections. The first section consists of a detailed summary of the available research on sex differences in incorrect responses. The second section deals with sex-related differences in problem solving performance. These sections are directly related to the objectives of the study. The third section is more peripheral, and contains a discussion of the more prevalent issues about the influence of cognitive and affective variables on sex differences in mathematics learning and achievement. The fourth is a summary of the research dealing with variables considered as influential to mathematics problem solving performance. Sex-related Differences in Incorrect Response Patterns Research findings tend to suggest that boys and girls may be approaching problem solving differently (e.g., Fennema and Sherman, 1978; Marshall, 1981, 1983; Meyer, 1978, among others). Marshall (1981) investigated whether 6th grade boys and girls approach mathematical problem solving with different strategies. Her specific interest was whether the sexes made the same errors. She analyzed the responses of 9,000 boys and 9,000 girls to 6 selected items, 2 computation items, and 1 story problem item from each of 2 of the 16 forms of the Survey of Basic Skills test administered during the academic year 1978-79. The Survey is a 30-item achievement test administered every year to all 6th grade children in California through the California Assessment Program. There are 16 forms of the test, to which approximately 9,000 boys and 9,000 girls respond each year. Of the 160 mathematics items contained in the 16 forms of the test, 32 are on measurement and graphing, 28 on number concepts, 28 on whole number arithmetics, 20 on fraction arithmetics, 20 on decimal arithmetics, 20 on geometry, and 12 on probability and statistics. The item analysis performed on the 1978-79 data showed that boys and girls tended to select different incorrect responses. In the first computation item (Form 1 of the test) both sexes reflected similar mistakes in carrying, but in different columns. In the second computation item, both sexes ignored the decimal points and selected the same incorrect response. However, more girls than boys chose this response. In the first computation item (Form 2 of the test) the incorrect choice of both sexes was option cr but the second most frequently selected option was a for boys and b for girls. In the second computation item of this test form, no sex differences were found in response patterns. Approximately 45% of each sex selected option c. The next popular choice for both sexes was option d, selected by approximately 35% of both boys and girls. On the story problem of Form 1, males and females responded alike. Their most popular incorrect response choice was option a for both males and females. The second most popular incorrect choice was option c for both sexes. Response to the story problem in Form 2 showed sex differences in response choice. Including the correct option, 33% of the girls selected option a, 20% chose option c, and 20% option d. For males approximately 25% selected option a and the same percent selected option d. Marshall concluded that although the analysis of incorrect responses does not explain why boys and girls differ in their responses, the analysis shows that boys and girls approach problems in different ways and these varying strategies can be useful in identifying how the sexes differ in reasoning abilities. Two years later, Marshall (1983) analyzed the responses of approximately 300,000 boys and girls to mathematics items contained in the 16 test forms of the Survey of Basic Skills during the years 1977, 1978, and 1979. She used log-linear models (explained in Chapter III) to investigate sex related differences in the selection of incorrect responses, and the consistency of such differences over three years of administration of the test. Based on her findings that sex differences were found in 80% of the items, Marshall classified the students' errors according to Radatz' (1979) five-category error classification. The categories are language (errors in semantics), spatial visualization, mastery, association, and use of irrelevant rules. It was found that girls' errors are more likely to be due to the misuse of spatial information, the use of irrelevant rules, or the choice of an incorrect operation. Girls also make relatively more errors of negative transfer and keyword association. Boys seem more likely than girls to make errors of perseverance and formula interference. Both sexes make language-related errors, but the errors are not the same. Available research is not extensive enough to make definite judgments about the sex-related differences observed in incorrect responses. Clearly more research is needed. Sex-related Differences in Problem Solving It has already been acknowledged that the subject of problem solving has been extensively researched. However, as early as 1969, Kilpatrick criticized the fact that the study of problem solving has not been systematic; some researchers have studied the characteristics of the problem while others have given their attention to the characteristics of the problem solvers. Moreover, differences in the tests used to measure problem solving performance also constitute an obstacle when trying to compare the results of the studies carried out. In order to avoid this pitfall and provide a basis for comparison, the studies reviewed in this section, dealing with sex-related differences in problem solving have been divided in two groups. The first comprises those studies that used the Romberg-Wearne Problem Solving Test. The second contains other relevant studies in which problem solving performance has been measured by means of other instruments. The Romberg-Wearne Problem Solving Test merits special mention because it was the first attempt "to develop a test to overcome the inadequacies of total test scores in explaining the reasons why some students are successful problem solvers and others are not" (Whitaker, 1976, pp. 9, 10). The test is composed of 23 items designed to yield 3 scores: a comprehension score, an application score, and a problem solving score. The comprehension question ascertains whether a child understands the information given in the item stem. The application question assesses the child's mastery of a prerequisite concept or skill of the problem solving question. The problem solving question poses a question whose solution is not immediately available, that is, a situation which does not lend itself to the immediate application of a rule or algorithm. The application and problem solving parts of the test may refer to a common unit of information (the item stem) but the questions are independent in that the response to the application question is not used to respond to the problem solving question. Meyer (1978) Whitaker (1976), Pennema and Sherman (1978), and Sherman (1979) have used the Romberg-Wearne test in their studies. Meyer (1978) investigated whether males and females differ in problem solving performance and examined their prerequisite computational skills and mathematical concepts for the problem solving questions. A sample of 179 students from the 4th grade were administered 19 "reference tests" for intellectual abilities and the Romberg-Wearne test. The analysis showed that males and females were not significantly different in the comprehension, application, and problem solving questions of the test. The sexes differed in only 2 of the 19 reference tests, Spatial Relations and Picture Group Name-Selection. A factor analysis, however, showed differences in the number and composition of the factors. For females, a general mathematics factor was determined by mathematics computation, comprehension, application, and problem solving. For males, the comprehension and application parts determined one factor; problem solving with two other reference tests (Gestalt and Omelet) determined another factor. Meyer concluded that comprehension of the data and mastery of the prerequisite mathematical concepts did not guarantee successful problem solving either for males or for females. Problem solving scores for both sexes were about one third their scores in comprehension and one fourth their scores in application. She also concluded that the sexes may have approached the problem solving questions differently. The methods used by females for solving problem situations may have paralleled their approach to the application parts. Males may have used established rules and algorithms for the application parts, but may have used more of a Gestalt approach to the problem solving situation. Whitaker (1976) investigated the relationship between the mathematical problem performance of 4th grade children and their attitudes toward problem solving, their teachers' attitude toward mathematical problem solving, and related sex and program-type differences. Although his main objective was to construct an attitude scale to measure attitudes toward problem solving, his study is important because his findings support Meyer's regarding the lack of significant sex-related differences in problem solving performance. Performance in the problem solving questions, for both males and females, was much lower than performance in the application questions, and much lower than performance in the comprehension questions. In fact, the mean score for each part of the item, for both males and females, was almost identical to the mean scores obtained by males and females in Meyer's study. Whitaker noted that each application item is more difficult than its preceding comprehension item, and that each problem solving item is more difficult than its preceding application item. No significant sex-related differences were found for any of the three parts of the item (comprehension, application, or problem solving). Fennema and Sherman (1978) investigated sex-related differences in mathematics achievement and cognitive and affective variables related to such differences. They administered the Romberg-Wearne Problem Solving Test to a representative sample of 1320 students (grades 6-8) from Madison, Wisconsin, predominantly middle-class, but including great diversity in SES. The sample consisted of students who had taken a similar number of mathematics courses and were in the top 85% of the class in mathematics achievement. They were tested in 1976. Four high school districts were included. In only one of the high school districts were sex-related differences in application and problem solving founds in favor of males. They concluded that when relevant factors are controlled, sex-related differences in favor of males do not appear often, and when they dor they are not large. Sherman (1980) investigated the causes of the emerging sex-related differences in mathematics performance, in favor of males, during adolescence (grades 8-11). She wanted to know if these differences emerge as a function of sex-related differences in spatial visualization and sociocultural influences that consider math as a male domain. In grade 8, she used the Romberg-Wearne Test and, in grade 11, a mathematical problem solving test derived from the French Kit of Tests. The analysis showed that for girls, problem solving performance remained stable across the years. Mean problem solving performance for boys, however, was higher in grade 11 than in grade 8. No sex-related differences were found in grade 8, but boys outperformed girls in grade 11, where the Stafford test was used. Sherman found that for both sexes problem solving performance in grade 8 was the best predictor of problem solving performance in grade 11. Spatial visualization was a stronger predictor for girls than for boys. Mathematics as a male domain was a good predictor for girls only; the less a girl stereotyped mathematics as a male domain in grade 8, the higher her problem solving score in grade 11. Attitudes toward success in mathematics in grade 8 was a more positive predictor of problem solving performance for boys than for girls; the more positive the attitudes toward success in mathematics in grade 8, the higher their performance in problem solving in grade 11. None of these four studies, all of which used the Romberg-Wearne Mathematics Problem Solving Test, show statistically significant sex-related differences in problem solving performance. In later studies other tests were used to measure this variable (Kaufman, 1984; Marshall, 1981, 1984). Kaufman (1983/1984) investigated if sex differences in problem solving, favoring males, exist in the 5th and 6th grades and if these differences were more pronounced in mathematically gifted students than in students of average mathematical ability. The Iowa Test of Basic Skills and a mathematics problem-solving test were administered to 504 subjects. Males in the average group as well as males in the gifted group outperformed females, but only the gifted group showed statistically significant differences. As a result of her investigations, Marshall (1981) concluded that sex-related differences in mathematics performance may be the result of comparing the sexes on total test scores. If the test contains more computation items than problem solving items, girls will perform better than boys, but if the test contains more problem solving items than computation ones, boys will outperform girls. With this in mind, Marshall investigated sex-related differences in computation and problem solving by analyzing the responses of approximately 18,000 students from grade 6 who had been administered the Survey of Basic Skills Test: Grade 6, during the academic year 1978-79. Two of the 16 test forms of the Survey were used to assess skills such as concepts of whole numbers, fractions, and decimals. These skills were tested both as simple computations and as story problems (problem solving). Two computation items and one story problem item were selected because they were particularly related; both computation items required skills needed in solving the corresponding story problem. It was assumed that correct solution of the computation item correlates with solving the story problem because the story problem requires a similar computation. Marshall found that girls were better in computation and boys were better in problem solving. She also found that boys were much more likely than girls to answer the story problem item correctly after giving incorrect responses to both computation items. Apparently, mastery of the skills required by the computation items is more important for girls than for boys. If girls cannot solve the computation items, they have little chance of solving the related story problem item. For girls, the probability of success in the story problem item after giving successful answers to both computation items is almost 2 1/2 times the probability of success after giving incorrect responses to both computation items. For boys, the probability of success in the story problem item after successful responses to the computation items is about 1 3/4 times the probability of success on the story problem item after incorrect responses to the computation items. Three years later, Marshall (1984) analyzed more in depth these phenomena of sex-related differences. Her interest was twofold. First, she wanted to know if there were differences in the rate of success for boys and girls in solving computation and story problem items. Second, she examined additional factors that interact with sex to influence mathematics performance, such as reading achievement, socio-economic status (SES), primary language, and chronological age. Two questions were raised: Do the probabilibities of successful solving of computation and story problem items increase with reading score? Are these probabilities different for the two sexes? Approximately 270,000 students from the 6th grade were administered the Survey of Basic Skills of the California Assessment Program, during the years 1977, 1978, and 1979. Responses were analyzed using log-linear models. Successful solving of computation items was positively associated with successful solving of story problems. Girls were more successful in computation than boys, and boys were more successful than girls in solving story problems. This finding supports reports from the National Assessment of Educational Progress (NAEP) (Armstrong, 1979). To investigate the effects of reading, SES, language and chronological ager only those test forms containing 2 computation items and 2 story problems were considered for analysis; 32 items from 8 test forms were included in the analysis. The results of these analyses showed that at every level of reading score, 6th grade children were more successful in computation than in story problems. Although the differences were not larger at every reading score boys consistently had higher probabilities of success in story problems than did girls, and girls consistently showed higher probabilities of success in computation than boys. Also, as the reading score increased, the difference between the probability of success in story problems and the probability of success in computation grew larger. This difference grew larger for girls than for boys. Although SES was a major factor in solving computation and story problem items successfully, the effect was similar for each sex. Sex-related differences by primary language or chronological age were not large. This research carried out by Marshall with elementary grade children supports previous research findings that males are better than females in mathematics problem solving (a higher order skill) and females are better than their counterpart males in computation (a lower level skill). Marshall's research also brought out a different aspect of this question: the notion that girls find it more necessary than boys to succeed in the computation items in order to successfully solve the story problem items. Cognitive and Affective Factors That Influence Sex Differences in Mathematics Learning and Achievement The research reviewed in the literature does not provide evidence of any unique variable that could serve as an explanation for the observed sex-related differences in mathematics learning and achievement. However, some issues have been discussed, among which the most prevalent are that sex-related differences in mathematics learning and achievement are a result of differences in formal education; that sex-related differences in mathematics learning and achievement arise from sex differences in spatial visualization; and that sex-related differences result from a differentiated effect of affective variables on the mathematics performance of males and females. Differences in Formal Mathematics Education (Differential Coursework Hypothesis) The basis for the differential coursework hypothesis is the fact that sex-related differences in mathematical learning and achievement show up when comparing groups which are not equal in previous mathematics learning. Atter the 8th grader boys tend to select mathematics courses more otten than girls. Therefore, girls show lower achievement scores in mathematics tests because their mathematics experience is not as strong as the boys' (Fennemar 1975; Fennema & Sherman, 1977; Sherman, 1979). Fennema and Sherman's study (1977) lends additional support to the feasibility of viewing sex differences in mathematics learning and achievement as reflecting something other than a difference in mathematics aptitude. After controlling for previous study of mathematics, they found significant sex differences in mathematics achievement in only two of the four schools under study, making the attribution to sex per se less likely. Controlling for the number of space visualization-related courses, the sex-related differences which originally emerged in spatial visualization scores became non-significant. In the two schools where sex differences in mathematics achievement were found, differences between the sexes were also found in their attitudes toward mathematics. Researchers like Backman (1972), who analyzed data from Project Talent, and Allen and Chambers (1977) have also hypothesized that sex-related differences in mathematics achievement may be related to different curricula followed by males and females. Allen and Chambers attributed male superiority in mathematics problem solving to differences in the number of mathematics courses taken in high school. This issue has been seriously questioned by Astin (1974) Fox (1975ar 1975b), and Benbow and Stanley (1980), among others. Astin and Fox have reported large differences in favor of males among gifted students taking the Scholastic Achievement Test. These differences occur as early as grade 7, when there are no sex differences in the number of courses taken. Benbow and Stanley (1980) compared mathematically precocious boys and girls in the 7th grader with similar mathematics background, and found sizeable sex-related differences favoring boys in mathematical reasoning ability. Five years later, they conducted a follow-up study which showed that boys maintained their superiority in mathematics ability during high school. While Fox attributed sex-related differences in mathematical achievement to differential exposure to mathematical games and activities outside school, Benbow and Stanley suggested that sex-related differences in mathematics performance stem from superior mathematical ability in males, not from differences in mathematics formal education. The differential coursework hypothesis is not totally convincing and, as reported before, it has been challenged by researchers such as Benbow and Stanley (1980). However, Pallas and Alexander (1983) have questioned the generalizaoility of Benbow and Stanley's findings based on the fact that they used highly precocious learners. The differential coursework hypothesis can be accepted only as a partial explanation of differences in mathematics performance found between the sexes. Differences in Spatial Ability The basic premise in this issue is that males and females differ in spatial visualization and this explains differential mathematics learning and achievement. Until recently, sex differences in spatial ability in favor ot males were believed to be a fact and were thought by some to be related to sex differences in mathematical achievement. Research findings in this area have been inconsistent. In 1966, Maccoby stated that "by early school years, boys consistently do better (than girls) on spatial tasks and this difference continues through the high school and college years" (p.26). In 1972, Maccoby and Jacklin said that the differences in spatial ability between the sexes "remain minimal and inconsistent until approximately the ages of 10 or 11, when the superiority of boys becomes consistent in a wide range of populations and tests" (p.41). In 1974, after a comprehensive literature search, Maccoby and Jacklin concluded that sex differences in spatial visualization become more pronounced between upper elementary years and the last year of high school, the years when sex-related differences in mathematics achievement favoring boys emerge. Guay and McDaniel (1977) supported in part Maccoby and Jacklin's 1974 findings. They found that among elementary school children, males had greater high level spatial ability than females, but that males and females were equal in low level spatial ability. This finding is inconsistent with that portion of Maccoby and Jacklin's review that suggests that sex differences become evident only during early adolescence. Cohen and Wilkie (1979) however, stated that in tests measuring distinct spatial tasks, males perform better than females in early adolescence and throughout their life span. Most studies carried out after 1974 have failed to support these sex differences in spatial abilities (Armstrong, 1979; Connor, Serbin, & Schackman, 1977; Fennema & Sherman, 1978; Sherman, 1979). Fennema and Sherman (1978) and Sherman (1979) have explored sex-related differences in mathematical achievement and cognitive and affective variables related to these differences. In a study involving students from grades 6, 7, and 8, from four school districts, Fennema and Sherman found that spatial visualization and problem solving were highly correlated for both sexes (.59 and .60). Even in the school district where sex differences were found in problem solving, no significant sex-related differences were found in spatial visualization. When Sherman (1980) compared groups of males and females in two different grades, 8 and 11, she found no sex-related differences in problem solving or in spatial visualization in grade 8. In grade 11, however, although the sexes differed in their problem solving performance, no sex-related differences were found in spatial vizualization. Even though spatial visualization in grade 8 was the second best predictor of problem solving performance in grade 11, sex differences in grade 11 were not a result of spatial visualization since no differences were found in that skill. In spite of the fact that no sex differences were found in spatial abilities, it is evident that males and females may use them in a different way. Meyer (1978), with an elementary grade sample, and Fennema and Tartree (1983) with an intermediate level sampler found that the influence of spatial visualization on solving mathematics problems is subtler and that males and females use their spatial skills differently in solving word story problems (problems that measure problem solving ability or reasoning). Fennema and Tartree (1983) carried out a three-year longitudinal study which showed that girls and boys with equivalent spatial visualization skills did not solve the same number of items, nor did they use the same processes in solving problems. The results also suggested that a low level of spatial visualization skills was a more debilitating factor for girls than for boys in problem solving performance. Landau (1984) also investigated the relationship between spatial visualization and mathematics achievement. She studied the performance of middle school children in mathematical problems of varying difficulty, and the extent to which a diagramatic representation is likely to facilitate solution. She found that spatial ability was strongly correlated to mathematical problem solving and that the effect of spatial ability was more influential for females. Females made more use of diagrams in the solution of problems, reducing the advantage of males over females in problem solving performance. The issue of sex-related differences in spatial visualization ability as an explanation for sex differences observed in mathematics achievement is less convincing and the findings more contradictory than in the issue of sex differences in formal education. Besides these cognitive issues, other issues, mostly affective in nature have also been studied in trying to explain the origin of these sex differences in mathematics achievement and learning. The studies dealing with these affective variables are reviewed in the next section. Differentiated Effect of Affective Variables Researchers have attempted to explain the effect of sex differences in internal beliefs, interests, and attitudes (affective variables) on mathematics learning and achievement. A brief statement of each explanation precedes the summary of studies conducted that support the explanation. Confidence as lerners of mathematics. Females, more than males, lack confidence in their ability to learn mathematics and this affects their achievement in mathematics and their election of more advanced mathematics courses. Maccoby and Jacklin (1974) reported that self- confidence in terms of grade expectancy and success in particular tasks was found to be consistently lower in women than in men. In 1978, Fennema and Sherman reported that in their study involving students from grades 6 through 12, boys showed a higher level of confidence in mathematics at each grade level. These differences between the sexes occurred in most instances even when no sex-related differences in mathematics achievement were found. The correlation between confidence in mathematics performance and mathematics achievement in this study was higher than for any other affective variable investigated. Sherman reported a similar finding in 1980; in males, the most important factor related to continuation in theoretical mathematics courses was confidence in learning mathematics. This variable weighed more than any of the cognitive variables: mathematics achievement, spatial visualization, general ability, and verbal skill. In the case of females, among the affective variables, confidence in learning mathematics was found to be second in importance to perceived usefulness of mathematics. Probert (1983) supported these findings with college students. A variable that needs discussion within the context of sex differences in confidence as learners of mathematics is causal attribution. Causal attribution models attempt to classify those factors to which one attributes success or failure. The model proposed by Weiner (1974) categorizes four dimensions of attribution ot success and failure: stable and internal, unstable and internal, stable and external, and unstable and external. For example, if one attributes success to an internal, stable attribute, such as ability, then one is confident of being successful in the future and will continue to strive in that area. If one attributes success to an external factor such as a teacher, or to an unstable one, such as effort, then one will not be as confident or success in the future and will cease to strive. Failure attribution patterns work this way: if failure is attributed to unstable causes, such as effort, failure can be avoided in the future and the tendency will be to persist in the task. However, if failure is attributed to a stable cause, such as ability, the belief that one cannot avoid failure will remain. Studies reported by Bar-Tal and Frieze (1977) suggest that males and females tend to exhibit different attributional patterns of success and failure. Males tend to attribute their success to internal causes and their failures to external or unstable ones. Females show a different pattern; they tend to attribute success to external or unstable causes and failures to internal ones. The pattern of attributions, success attributed externally and failure attributed internally, has become hypothesized to show a strong effect on mathematics achievement in females. Kloosterman's (1985) study supported these findings. According to Kloosterman, attributional variables appear to be more important achievement mediators for females than for males, as measured by mathematics word problems. More research is needed in this area. Mathematics as a male domain. Mathematics is an activity more closely related to the male sex domain than to the female sex domain (Eccles et al., 1983). Thus, the mathematical achievement or boys is higher than that of girls. According to John Ernest (1976) in his study Mathematics and Sex, mathematics is a sexist discipline. He attributed sex-related differences in mathematical achievement to the creation by society of sexual stereotypes and attitudes, restrictions, and constraints that promote the idea of the superiority of boys in mathematics. Ernest reported that boys, girls, and teachers, all believe that boys are superior in mathematics, at least by the time students reach adolescence. Bem and Bem (1970) agree and argue that an American woman is trained to "know her place" in society because or the pervasive sex-role concept which results in differential expectations and socialization practices. Plank and Plank (1954) were more specific. They discussed two hypotheses related to this view: the differential cultural reinforcement hypothesis and the masculine identification hypothesis. The differential cultural reinforcement hypothesis states that society in general perceives mathematics as a male domain, giving females less encouragement for excelling in it. The masculine identification hypothesis establishes that achievement and interest in mathematics result from identification with the masculine role. A study related to the differential cultural reinforcement hypothesis is that of Dwyer (1974). Dwyer examined the relationship between sex role standards (the extent to which an individual considers certain activities appropriate to males or females) and achievement in reading and arithmetic. Students from grades 2, 4, 6, 8, 10, and 12 participated in this study. She found that sex role standards contributed significant variance to reading and arithmetic achievement test scores and that the effect was stronger for males than for females. This led to her conclusion that sex-related differences in reading and arithmetic are more a function of the child's perception of these areas as sex-appropriate or sex-inappropriate than of the child's biological sex, individual preference for masculine and feminine sex roles, or liking or disliking reading or mathematics. In a study which agrees with the masculine identification hypothesis, Milton (1957) found that individuals who had received strong masculine orientation performed better in problem solving than individuals who received less masculine orientation. Elton and Rose (1967) found that women with high mathematical aptitude and average verbal aptitude scored higher on the masculinity scale of the Omnibus Personality Inventory (OPI) than those with average scores on both tasks. It is not until adolescence that sex differences in the perception of mathematics as a male domain are found (Fennemar 1976; Stein, 1971; Stein & Smithless, 1969; Verbeker 1983). In a study with 2nd, 6th and 12th graders, Stein and Smithless (1969) found that students' perceptions of spatial, mechanical, and arithmetic skills as masculine became more defined as these students got older. Fennema (1976) considers that the influence each sex exerts upon the other on all aspects of behavior is stronger during adolescence. Since during these years males stereotype mathematics as a male domain, they send this message to females who, in turn, tend to be influenced in their willingness to study or not to study mathematics. Before that stage, girls consider arithmetic feminine, while boys consider it appropriate for both sexes (Bobber 1971). Usefulness of mathematics. Females perceive mathema- tics as less useful to them than males do, and this perception occurs at a very young age. As a results females exert less effort than males to learn or elect to take advanced mathematics courses. Many studies reported before 1976 found that the perception of the usefulness of mathematics for one's future differs for males and females, and is related to course taking plans and behavior (Fox, 1977). If females do not perceive mathematics as useful for their future, they show less interest in the subject than counterpart males. These differences in interest are what Hilton and Berglund (1974) suggest to account for sex-related differences in mathematics achievement. Although the perception of the usefulness of mathematics is still an important predictor of course taking for girls, there is a growing similarity between males and females regarding the usefulness of mathematics (Armstrong & Pricer 1982; Fennema & Sherman, 1977; Moller, 1982/1983). Armstrong and Price investigated the relative influence of selected factors in sex-related differences in mathematics participation. Both males and females selected usefulness of mathematics as the most important factor in deciding whether or not to take more mathematics in high school. Moller's study revealed that both males and females based mathematics course-taking decisions on career usefulness. A Fennema and Sherman (1977) study showed only slight differences between males and females in their feelings about the usefulness of mathematics. In her study of this variable among college students, Probert (1983/1984) did not find any sex-related differences either. These have been the main affective variables researched in attempting to explain the underlying causes of sex-related differences observed in mathematics learning and achievement. In spite ot the great diversity of studies dealing with both cognitive and affective variables, there are no clear-cut findings to render unequivocal support to a particular variable as accounting for these sex-related differences. However, everything seems to point to the fact that affective, rather than cognitive variables play a more significant role in the sex-related differences observed in mathematics performance and learning. In most of the studies dealing with affective variables, findings consistently show that these factors influence mathematics performance in females more than in males. In at least one area confidence as learners of mathematics, Sherman (1980) found that this variable influenced course election more than all the cognitive variables previously discussed. The case for the societal influences on sex roles and expectations to account for the differences in mathematics learning is also supported in one way or another in the studies reported in the literature. Problem Solving Performance and Related Variables Problem solving has been perhaps the most extensively researched area in mathematics education. Published reviews by Kilpatrick (1969) Riedesel (1969), and Suydam and Weaver (1970-1975) attest to this. Much of the research done has focused on identifying the determinants of problem difficulty and the problem features that influence the solution process. At presents no set of variables has been clearly established as a determinant of problem difficulty. Several researchers have investigated the effect or reading and computation on problem solving performance. Others have studied the effect of student attitudes toward problem solving in problem solving learning and achievement. Typically, correlational methods have been used to investigate these questions. Computational Skills and Problem Solving Performance One of the first researchers to study the effect of computation and reading on problem solving performance was Hansen (1944). He investigated the relationship of arithmetical factors, mental factors, and reading factors to achievement in problem solving. Sixth grade students were administered tests in problem solving and categorized as superior achievers (best problem solvers) and inferior archievers (poorest problem solvers). The two groups were compared in selected factors believed to be related to success in arithmetic problem solving: arithmetical, mental and reading factors. After controlling for mental and chronological age, the superior achievers in problem solving surpassed the inferior achievers in mental and arithmetical factors. The superior group did better in only two of the six items under the reading factors: general language ability and the reading of graphs, charts, and tables. The findings suggest that reading factors are not as important as arithmetic and mental factors in problem solving performance. However, these findings should be taken cautiously, as the content of the Gates tests (used to measure reading) is literary and does not include mathematical material. Chase (1960) studied 15 variables in an effort to find out which ones have significant influence on the ability to solve verbal mathematics problems. Only computation, reading to note details, and fundamental knowledge were primarily related to problem solving. Computation accounted for 20.4% of the 32% variance directly associated with problem solving. Chase concluded that a pupil's ability in the mechanics of computation, comprehension of the principles that underline the number systems, and the extent to which important items of information are noticed when reading, are good predictors of the student's ability in solving verbal problems. Balow (1964) investigated the importance of reading ability and computation ability in problem solving performance. He objected to the approaches used by other researchers who in their analyses dichotomized research subjects as "poor" or "good" students, and who ignored the recognize effect of intelligence on reading and on mathematics achievement. Balow administered the Stanford Achievement Test (subtests of reading, arithmetic, and reasoning) and the California Short-Form test of mental ability to a group of 1,400 children from the 6th grade. All levels of achievement were included in the analysis. Analysis of variance and covariance were used and compared. He confirmed the findings of other researchers to the effect that there is a direct relationship between I.Q. and reading ability and between I.Q. and computational skills. The results of the analysis of variance revealed that increases in computation ability were associated with higher achievement in problem solving. A relationship between reading ability and problem solving was also found, but it was not as strong. Significant differences in problem solving performance associated with computational ability were found when intelligence was controlled. Balow concluded that computation is a much more important factor in problem solving than reading ability, and that when I.Q. is taken into consideration, the degree of the relationship between reading and problem solving ability becomes less pronounced. Intelligence tends to confound the relationship between these two variables. Knifong and Holtan (1976, 1977) attempted to investigate the types of difficulties children have in solving word problems. They administered the word problem section of the Metropolitan Achievement Test to 35 children from the 6th grade. Errors were classified in two categories. Category I included clerical and computational errors. Category II included other types of errors, such as average and area errors, use of wrong operation, no response, and erred responses offering no clues. It the student's work indicated the correct procedure and yet the problem was missed because of a computational or clerical errors it was assumed that the problem was read and understood. An analysis of frequencies showed that clerical errors were responsible for 3% of the problems incorrectly solved, computational errors accounted for 49%, and other errors for 48% of the erred problems. Knifong and Holtan concluded that "improved computational skills could have eliminated nearly half or the word problem errors" (p. 111). These computational errors were made in a context where other skills such as reading, interpretation of the problem, and integration of these skills necessary for the solution of word problems, might interact. However, Knifong and Holtan state that their findings neither confirm nor deny that improvement or reading skills will lead to improvement in problem solving. They conclude that "it is difficult to attribute major importance to reading as a source of failure" (p. ll). In a later analysis, looking for evidence of poor reading abilities affecting children's success in word problems, Knifong and Holtan (1977) interviewed the children whose errors fell under the category or "other errors." Students were asked to read each problem aloud and answer these questions: What kind of situation does the problem describe? What does the problem ask you to find? How would you work the problem? Ninety five percent of the students read the problem correctly; 98% explained the kind of situation the problem described in a correct manner; 92% correctly answered what the problem was asking them to find, and 36% correctly answered the question of how to work the problem. The fact that a large percent of the students whose errors were classified as "other errors" (in which reading skills might have been a factor) correctly stated how to work the problem, is strong evidence of their ability to read and interpret the problems correctly. The errors made by this group of students had a distinct origin, unrelated to reading ability. Zalewski (1974) investigated the relative contribution of verbal intelligence, reading comprehension, vocabulary, interpretation of graphs and tables, mathematical concepts, number sentence selection, and computation to successful mathematical word problem solution, and the relationship of the dependent variable to the eight independent variables. She worked with a group of 4th grade children who were administered the subtests of the Iowa Test of Basic Skills (ITBS) and the Weschler Intelligence Scale for Children (WISC). Multiple regression analysis was performed. A correlation of .769 was found between word problem solving and the eight independent variables. Correlations between word problem solving and the independent variables ranged from .363 (verbal intelligence) to .674 (mathematical concepts). Correlations between the independent variables ranged from .369 (verbal intelligence and computation) to .749 (reading comprehension and vocabulary). Mathematical concepts, computation, and number sentence selection were almost as effective as all eight independent variables in predicting achievement in mathematical word problem solving. Mathematical concepts, computations number sentence selection, and reading comprehension accounted for 58% of the variance, whereas all eight predictors accounted for 59% of the variance. The two best predictors were mathematical concepts and computations which accounted for 54% variance. Other variables accounted for about 40% of the variance. The author recommends that the findings of this study be interpreted cautiously because the correlation between the eight independent variables was high, and, according to Zalewski, in a study of this nature where the interest is primarily in the influence of several variables on one dependent variable, a low correlation between the independent variables is required. (p. 2804) In a more recent investigations Exedisis (1983) studied the contribution of reading ability, vocabulary, mathematical concepts, computation, sex, and race on problem-solving performance. The Iowa Test of Basic Skills was administered to a group of 6th, 7th, and 8th grade anglo and black Chicago male and female adolescents. Problem solving was highly correlated to an understanding of basic mathematical concepts, somewhat correlated to race, and weakly correlated to computational and vocabulary skills, sex, and reading ability. Although the findings of these studies show a relationship between computational skills and problem solving achievement, this relationship is not strong enough to be considered the most determinant factor in problem solving achievement, as some of the researchers have been careful to point out. In spite ot the dismissal of reading as a determinant factor in problem solving achievement by some of these same researchers, more recent studies in this area have led others to hold different views. Reading and Problem Solving Performance Martin (1964) studied the contribution of reading comprehension, computation, abstract verbal reasoning, and arithmetic concepts to arithmetic problem solving performance. Fourth and 8th grade students were administered the Iowa tests of Basic Skills and the Lorge-Thorndlike intelligence test (verbal). He found that in the 4th grade the correlations between problem solving and abstract verbal reasoning, reading comprehension, arithmetic concepts, and computation were .61, .64, .66, and .60 respectively, and .56, .68, .69, and .63 in the 8th grade. When computation was held constant, the correlation between problem solving and reading was .52 in grade 4 and .54 in grade 8. When reading was held constant the correlation between problem solving and computation was .43 in grade 4 and .42 in grade 8. Creswell (1982) worked with a sample of anglo and black adolescents from Chicago. Each subject was administered the California Achievement test. Multiple regression was used to analyze the data. The analysis showed that reading is more important than computation in predicting student performance in problem solving. Reading accounted for 49.5% of the variance; computation accounted for 14.6% of the variance. Ballew and Cunningham (1982) worked with 6th grade students in an attempt to find what proportion of students have as their main source of difficulty with word problems each of the following factors: a) computation skills, b) interpretation of the problem, c) reading and, d) integrating these skills in the solution of problems. They also wanted to know if a student can be efficiently diagnosed as having one of the four categories as his/her main difficulty with mathematics word problems. Their study is important because it represents an attempt to demonstrate that multiple factors can interact in the correct solution of a mathematics word problem. They constructed three graded tests from a basal mathematics series for grades 3 through 8. For test 1i the problems were set in pure computational form (the effects or reading, interpretation, as well as the necessity for integration were removed in an effort to measure the computational skills required by the word problems). For test 2, the effects of reading and computation were removed by reading the problems to the students and by giving scores based on whether or not the students set them up properly, in an attempt to measure problem interpretation alone. For test 3, the effect of computation was removed. The test yielded two scores--one by grading the students on whether or not they set up the problems properly and another by grading on the basis of the correct answer. The tests were administered to all 244 students from the 6th grade in two different schools. A diagnostic profile was obtained for each of the 217 students for which complete data were available: a computational score, a problem-interpretation score, a reading score, and a reading-problem solving score. They assumed that if the reading-problem interpretation score was lower (one or more levels lower) than the problem-interpretation score, the difficulty was due to reading ability. If the score of the lowest of the three areas (computation, problem interpretation, and reading-problem interpretation score) was the same as the reading-problem solving score, the student's area of greatest immediate need was either computation, problem interpretation, or reading. If the reading-problem solving score was lower than the lowest of the other three scores, the student's area of greatest immediate need was integration. Analysis of the data revealed that for 19% of the students, problem interpretation was their major difficulty; for 26% of the students, integration (total problem solving) was their greatest immediate need; for another 26%, computation was the major weakness; and for 29%, reading was their greatest immediate need. Seventy five percent of the students demonstrated clear strength in computation, 21% in problem interpretation, and 4% in reading-problem interpretation. An analysis across all students (including those without complete data) showed that 26% of the subjects could not work word problems at a level as high as that at which they could computer interpret problems, and read and interpret problems, when those areas were measured separately. This led them to conclude that knowing the skills or the components of solving word problems is not sufficient for success, since the components must be integrated into a whole process (mastery learning of the components cannot assure mastery of the process). Their analysis also led them to conclude that, in the case of 6th graders, inability to read problems is a major obstacle in solving word problems. Only 12% ot the subjects could read and set up problems correctly at a higher level than they could computer while 60% could compute correctly at a higher level than they could read and set up problems; 44% could set up problems better when they heard them read than when they read the problem themselves. Only 13% could set up problems better when they read them than when they heard them read. Muth (1984) investigated the role of reading and computational skills in the solution of word problems. A group of 200 students from the 6th grade were administered a test of basic skills and a mathematics word problem test. The word problem test consisted of 15 sample items supplied by the National Assessment of Educational Progress. The items were adapted to include some extraneous information and complex syntactic structure. Four versions of the test were constructed by combining two versions of problem information (absence vs. presence of extraneous information) with two versions of syntactic structure (simple vs. complex syntax). Task performance was measured by means of the number of problems answered correctly, number of problems set up correctly, and amount of time spent taking the test. Reading ability and computational ability were both positively correlated with number of correct answers and with number of problems correctly set up, and negatively correlated with test-taking time. Presence of extraneous information was negatively correlated with correct answers and correct set ups and positively correlated with test-taking time. Syntactic complexity was not significantly correlated with any of the performance measures. Results of a multiple regression analysis showed that reading accounted for 46% of the variance in total correct answers and computation accounted for 8%. Reading ability and computational ability uniquely accounted for 14% and 8% of the variance in the number of correct answers, respectively. Extraneous information added significantly to the variance explained in the number of correct answers, but syntactic structure did not. Reading ability accounted for 5% of the variance in test-taking time, but computation did not add significantly to the variance explained by reading. Muth concluded that reading and computation both contribute significantly to success in solving arithmetic word problems, but that reading plays a more significant role than does computation. The studies reviewed in this section show a positive relationship between reading and problem solving performance, but in the case of Ballew and Cunningham (1982) this relationship is not viewed singly but rather as one among the interacting factors that produce successful problem solving. The third variable reviewed is the effect of student attitudes toward problem solving on problem solving performance. Many researchers have tried to demonstrate that this variable is a determinant factor in problem solving achievement. Attitudes Toward Problem Solving and Problem Solving Performance Research studies support the existence of positive and rather stable relationships between student attitudes and achievement in mathematics. Aiken (1970) has suggested that an individual's attitude toward one aspect of the discipline (mathematics), such as problem solving, may be entirely different from his/her attitude toward another phase of the discipline, such as computation. Research, however, has been directed to the use of single, global measures of attitudes toward mathematics rather than to the investigation of attitudes toward a particular phase of the discipline. The studies described below are only part of the few investigations which have examined the relationship between student attitudes and performance in the area of problem solving. Carey (1958) constructed a scale to measure attitudes toward problem solving. Her interest was in general problem solving rather than in mathematical problem solving. Her work constitutes the first attempt to construct a measure of attitudes toward problem solving. The scale was used with a group of college students, and she found, among other things, that problem solving performance is positively related to problem solving attitudes and that, in the case of females, positive modification of attitudes toward problem solving brings a significant gain in problem solving performance. Lindgren, Silvar Faraco, and DaRocha (1964) adapted Carey's scale of attitudes toward problem solving and applied it to a group of 4th grade Brazilian children. Students also answered an arithmetic achievement test, a general intelligence test, and a socioeconomic (SE) scale. A low but significant positive correlation was found between arithmetic achievement and attitudes toward problem solving. A near zero correlation was found between attitudes toward problem solving and intelligence. Since problem solving is one aspect of the discipline of mathematics, this correlation between attitudes and arithmetic achievement can lead to a conclusion or a strong correlation between attitudes toward problem solving and problem solving performance. Whitaker (1976) constructed a student attitude scale to measure some aspects of 4th grade student attitudes toward mathematic problem solving. He included statements reflecting children's beliefs about the nature of various types of mathematical problems, the nature of the problem solving process, the desirability or persevering when solving a problem, and the value of generating several ideas for solving a problem. He correlated student attitudes toward problem solving with their scores in a mathematical test which yielded a comprehension score, an application score, and a problem solving score. He found a significant positive relationship between problem solving performance and student attitude scores on the subscale which measured reactions to such things as problem solving techniques or problem situations, or to the frustration or anxiety experienced when confronted with problem solving situations. In another part of this study, Whitaker investigated the relationship between the attitudes of 4th grade teachers toward problem solving and their students' performance in problem solving. A very weak and nonsignificant negative correlation was found between the teacher's attitudes toward problem solving and student performance. The studies reviewed have confirmed the relationship between problem solving performance and attitudes toward problem solving (Carey, 1958; Lindgren et al., 1964; Whitaker, 1976). However, the results reported in the studies that investigated the relationship between problem solving performance and computation and between reading and problem solving fail to be consistent in their conclusions. Hansen (1944), Chase (1960) Balow (1964), Knifong and Holtan (1976, 1977), and Zalewski (1974) concluded that computation is more strongly related to problem solving than is reading. Martin (1964), Creswell (1982), Ballew and Cunningham (1982), and Muth (1984), concluded that reading ability and mathematical problem solving show a stronger relationship than computation and problem solving. Exedisis's (1983) findings led to the conclusion that the effect or reading and computation in problem solving performance is unimportant. CHAPTER III METHOD The first objective of this study was to investigate sex differences in the selection of incorrect responses in a mathematics multiple-choice achievement test, and to determine whether these differences were consistent over three consecutive administrations of the test. The second objective was to investigate whether males and females differ in problem solving performance, if these differences persist after accounting for computational skills, and if the male-female differences depend on the level of computational skills. This chapter contains descriptions of the sample, the test instrument, and the statistical analysis used in achieving the above mentioned objectives. The Sample To achieve the first objective of the study, all the students who took the Basic Skills Test in Mathematics-6 ("Prueba de Destrezas Basicas en Matem&ticas-6") in each or the three years were included in the study. To achieve the second objective of the study, approximately 1,000 students were selected randomly for each year. For the first year, 1,002 were selected (492 boys and 510 girls); for the second year, 1,013 students were selected (504 boys and 509 girls); and, for the third years 1,013 students were selected (509 boys and 504 girls). The student population in Puerto Rico includes children from the urban and rural zones and comprises children from low and middle socioeconomic levels. Findings can be generalized only to this population. The Instrument The Basic Skills Test in Mathematics-6 is a criterion-referenced test used in the Department ot Education of Puerto Rico as part of the annual assessment program. The test measures academic achievement in operations, mathematical concepts, and story problems. It has a reported split half reliability or .95. The test was designed specifically for Puerto Rico. Its contents and the procedures followed for its development were formulated and reviewed by educators from the mathematics department of the Department of Education of Puerto Rico, in coordination with the Evaluation Center ot the Department of Education and mathematics teachers from the school districts. The emphasis placed on each skill area is depicted in Figure 3.1. Fig. 3.1 Number of Questions by Skill Area in the Basic Skills Test in Mathematics 6 Analysis of the Data Analysis of Sex by Option by Year Cross Classifications Log-linear models were used to analyze the sex by option by year cross classifications for each item. Two topics are addressed in this section, the hypotheses tested using log-linear models and a comparison of the hypotheses tested in this study with those tested by Marshall (1981, 1983). The object of the analysis was to test two hypotheses: 1. The proportion of males and the proportion of females choosing each incorrect option does not vary from year to year. Note that this hypothesis is stated in the null form. 2. Assuming that the first hypothesis is correct, the proportion of males who choose each incorrect option is different from the proportion of females who choose each incorrect option. This hypothesis is stated in the alternate form. In Table 3.1 a hypothetical cross classification of sex, option, and year is presented. Hypothesis 1 is true for this three dimensional contingency table. In Table 3.2 the three dimensional contingency table is rearranged to show the year by option contingency table for each gender. Table 3.1 Hypothetical Probabilities of Option Choice Conditional on Year and Sex Option Year Sex 1 2 3 First M .7 .2 .1 F .5 .3 .2 Second M .7 .2 .1 F .5 .3 .2 Third M .7 .2 .1 F .5 .3 .2 Table 3.2 Hypothetical Probabilities of Option Choice Conditional on Year and Sex and Arranged by Sex Option Sex Year 1 2 3 M First .7 .2 .1 Second .7 .2 .1 Third .7 .2 .1 F First .5 .3 .2 Second .5 .3 .2 Third .5 .3 .2 Inspection of the year by option contingency tables shows that year and option are independent for each gender. Thus, hypothesis 1 is equivalent to the hypothesis that year and option are independent conditional upon gender. Hypothesis 2 is also true for Table 3.1 and Table 3.2. Therefore, when hypothesis 1 is correct, hypothesis 2 is equivalent to the hypothesis that sex and option choice are dependent. In his discussion of the analysis of three dimensional contingency tables, Fienberg (1980) presents the following saturated model for the data: log mijk = Ui+Uj+Uk+Uij+Uik+Ujk+Uijk. (3.1) In this model, mijk is the expected value of the frequency in cell ijk of the three dimensional table. The model states that all three classification factors for a three dimensional contingency table are mutually dependent. In the present research i is the year index, j is the option index, and k is the sex index. Fienberg shows that deleting the terms Uij and Uijk yields a model in which year and option are independent conditional upon sex. This model is log mijk = Ui+Uj+Uk+Uik+Ujk. (3.2) Fienberg also shows that deleting the Ujk term from (3.2) to obtain log mijk= Ui+Uj+Uk+Uik. (3.3) yields a model that specifies that option is independent of sex. Based on Fienberg's presentation, an appropriate analysis for testing the hypotheses is 1. Conduct a likelihood ratio test of the adequacy of model (3.2). If this test is nonsignificant, then the data support the adequacy of the model, and the hypothesis that conditional on gender, options and year are independent. Because model (3.1) is a saturated model, testing the adequacy of model (3.2) is the same as comparing the adequacies of models (3.1) and (3.2). 2. Conduct a likelihood ratio test comparing .the adequacies of models (3.2) and (3.3). If this test is significant, then model (3.2) fits the data better than model (3.3) and the data support the hypothesis that the choice of incorrect option is dependent on sex. To summarized if the first test is nonsignificant and the second test is significant, then the choice of option is dependent on sex, and the pattern of dependency is the same for all three years. A problem arises in interpreting this analysis with the data used in this research. Over the three years, there were responses available from 135,340 students. Even if only 5% of the students answered an item incorrectly, the responses of 7,767 students would be used in analyzing this item. On the other hand, if 90% answered an item incorrectly, the responses of 121,806 students would be used in analyzing the data. As a result of the large sample size the tests described above are likely to be very powerful. In step 1 of the analysis, then, even a very small change from year to year in the proportion of males or females who choose an option is apt to be detected, and the results will indicate that hypothesis 1 is not supported. For step 2, even a very small dependence of option choice on sex is likely to be detected and hypothesis 2 is likely to be supported. In brief, the problem caused by the large sample size is that practically insignificant differences may yield statistical significance. Fortunately, the form of the test statistic used in the likelihood ratio test suggests a reasonable solution to this problem. The test statistic is G2 = 2 ZFo loge (Fo/Fe). (3.4) Here the summation is over all the cells in the contingency table Fo refers to the observed frequency in a cell, and Fe refers to the estimated expected frequency in a cell. Denoting the observed proportion in a cell by Po and the estimated expected proportion in a cell by Per the test statistic can be written G2 = 2 N TPo loge (Po/Pe)- (3.5) where N is the total number of subjects. This form of the test statistic suggests the following strategy. For any significant G2, using Po and Pe calculated from the total data set available for an item, calculate the minimum N required for G2 to be significant. If the minimum N is very larger this suggests that the statistically significant result is not practically significant since it can only be detected in very large samples. Of course, the question remains as to what can be considered a minimum large N. Although there is room for argument, it seems reasonable to claim that if an average of 1000 subjects per year is required to show significance, then the result is not likely to be practically significant. On the basis of this reasoning, it was proposed to ignore all significant results that would be nonsignificant if there were less than 3000 subjects available. In addition, all log-linear model tests were conducted using a .01 level of significance. Since this research is based on Marshall (1981, 1983), it is important to compare the method of analysis used in this study to the one used by Marshall. Marshall also used a two-step analysis. In the first step of her analysis she deleted the Uijk term from (3.1) and tested the adequacy of the model, log mij k = Ui+Uj+Uk+Uij+Uik+Ujk. (3.6) Following this, she deleted the Ujk term to obtain log mijk = Ui+Uj+Uk+Uij+Uik. (3.7) and compared these two models using a likelihood ratio test. If the first test was nonsignificant and the second significant, Marshall claimed that option choice was dependent on sex and the pattern of dependency was the same from year to year. This is the same claim that this study sought to establish. However, the approach used here was to present evidence that model (3.2) fits the data while Marshall tried to show that model (3.6) fits the data. The major difference between the two approaches concerns the operationalization of the concept that the 64 gender-option dependency is the same from year to year. In this study, a three dimensional table was considered to exhibit the same year to year pattern of gender-option dependency if, conditioned on gender, the same proportion of students chose each incorrect response over the three year period. This seems to be a straightforward and natural way to operationalize the concept. To illustrate how Marshall operationalized the concept in question, a hypothetical set of probabilities was constructed conforming to the pattern specified by Marshall. This is displayed in Table 3.3. Table 3.3 Hypothetical Joint Probabilities of Year, Option, and Sex Option Year Sex 1 2 3 ------------------------ First M .120 .045 .015 F .060 .060 .030 I------------ I Second M .144 .022 .019 F .072 .030 .039 I----------------------~ Third M .132 .040 .016 F .066 .054 .039 Note: Probabilities reported are truncated to three places. One important characteristic of this table involves two-by-two subtables of sex cross classified with two of the options for each year. For example, consider the three two-by-two tables obtained by cross classifying sex and option choices one and two for each year. These tables are indicated by dotted lines on Table 3.3. For this table, the ratio of the odds of a male choosing option one to the odds of a female choosing option one is the same (within truncating error) for each year. For example, this odds ratio for the first year is (.120/.045) / (.060/.060) = 2.67. Within the error caused by reporting truncated probabilities the odds ratio for years two and three is the same as that for year one. The equality of these odds ratios is Marshall's operationalization stage of the year- to-year gender-option dependency. To show that the odds ratio can be constant over years, but that the probabilities of option choice conditional on sex and year can change from year to year, for both males and females, the probabilities in Table 3.3 were converted to the probabilities of option choice conditional upon sex and year. These conditional probabilities are reported in Table 3.4. Unlike the probabilities in Table 3.2, those in Table 3.4 change from year to year for both males and females. Table 3.4 Hypothetical Probabilities of Option Choice Conditional on Sex and Year Option Year Sex 1 2 3 First M .666 .250 .083 F .400 .400 .200 Second M .774 .120 .104 F .510 .212 .276 Third M .698 .214 .087 F .431 .352 .215 Note: Probabilities reported are truncated to three places. Which procedure is more appropriate to test the claim that the pattern of male and female option choice remains the same from year to year? It seems more reasonable to test for a pattern like that in Table 3.2 than to test for a pattern like that in Table 3.4, and, consequently, this was the strategy adopted in this study. Comparison of Males and Females in Problem Solving Performance One object of the study was to compare the performance of males and females on problem solving. Two questions were addressed. First, do males and females differ in problem solving performance? Second, do these differences persist when computational skill is controlled for, and do these differences depend on the level of computational skill? Seven analyses of covariance (ANCOVA) tests were conducted, one with each of addition, subtraction, multiplication, division, addition of fractions, decimals in subtraction, and equivalence as covariates. A well known problem that arises in the use of ANCOVA is that an unreliable covariate can cause spurious differences between the sex groups. To solve this problem, Porter (1967/1968) proposed the use of estimated true scores for observed scores. Porter (1967/1968) conducted a simulation that gave empirical support to the adequacy of this strategy. Hunter and Cohen (1974) have provided theoretical support for this strategy. CHAPTER IV RESULTS Introduction The data gathered from 6th grade students from the public schools in Puerto Rico who took the Basic Skills Test in Mathematics-6 during three consecutive years were analyzed in this study. The first objective was to investigate whether boys and girls differ in the selection of incorrect responses, and if the pattern of differences was consistent throughout the three years in which the test was administered. The second objective was to investigate whether males and females differ in problem solving performance, if these differences persist after accounting for computational skills, and if the male-female differences depend on the level of computational skills. The results are discussed in two sections. Study findings in the area of sex differences in the selection of incorrect responses are discussed in the first section. The second section is devoted to an exposition of the findings in the area of sex-related differences in problem solving. Sex-Related Differences in the Selection of Incorrect Responses As indicated in Chapter III, there are two models of interest. The first model indicates that year and option are independent, conditional upon sex, log mijk = Ui+Uj+Uk+Uik+Ujk. (3.2) Substantively, this model implies that the pattern of male and female option choices is consistent over the three years of test administration. The second model indicates that option is independent of sex, log mijk = Ui+Uj+Uk+Uik- (3.3) This model implies that the pattern of option choice is the same for males and females. In order to determine if males and females differed in the selection of incorrect responses and if these differences were stable across the three years of test administration, a two-step test was performed. First, a likelihood ratio test of the adequacy of model (3.2) was conducted for each of the 111 items of the test. Under this test a model fits the data when the X2 value obtained is nonsignificant at a specified alpha level. For 50 (45%) of the items, model (3.2) fitted the data adequately. For these items the pattern of male and female option choices was consistent for the three years. The X2 values for the 61 items for which the likelihood ratio tests were significant at the .01 level are reported in Table 4.1. Also reported in this table are the actual sample sizes and the minimum sample sizes necessary for the likelihood ratio tests to be significant. Of the 61 items, 59 had minimum sample sizes greater than 3,000. Thus, although for the three years both males and females samples had inconsistent option choices on these 61 items, on 59 of the items the inconsistency was relatively minor. Consequently, these 59 items were included in step 2 along with the initial 50 items. In step 2, a likelihood ratio test was performed comparing the adequacy of the models in (3.2) and (3.3). The X2 values associated with models (3.2) and (3.3) for each of the 109 items subjected to step 2 are reported in columns b and c of Table 4.2. Also reported in Table 4.2 is the difference between the two X2 values (see column d). This latter figure is the test statistic for comparing the adequacies of models (3.2) and (3.3). Significant X2 statistics are indicated by asterisks. The X2 statistics were significant for 100 of the items indicating a male-female difference in option choice for these items. In column e of Table 4.2 the actual sample sizes are reported. In column f, the minimum sample sizes necessary for significance are reported. For those 100 items which had significant X2 statistics reported in column d, 94 (94%) had minimum sample sizes greater than 3,000. TABLE 4.1 Summary of Significant Tests of Model 3.2 Item Xe for Actual Sample Min. Sample Number Model 3.2* Size Size for Significance (a) (b) (c) (d) 8 13 23 25 28 29 30 31 34 36 37 42 45 48 49 52 54 56 57 46.15 25.74 29.25 35.90 46.00 61.48 25.19 23.26 58.25 84.36 23.70 23.26 22.41 179.25 27.82 26.01 41.41 21.62 28.51 16,796 15,718 60,377 26,707 84,847 88,396 87,473 32,870 82,257 86,137 37,068 59,140 63,098 97,593 16,766 39,858 74,502 44,107 53,075 3,352 5,624 19,011 6,852 16,988 13,242 31,982 13,015 13,006 9,404 14,405 23,417 25,932 5,014 5,550 14,114 16,570 18,789 17,146 TABLE 4.1 Continued Item Xe for Actual Sample Number Model 3.2* Size (a) (hi ( 50.96 66.43 21.39 55.03 158.36 88.01 56.28 50.51 34.28 48.01 75.29 44.39 14,425.97 24.61 22.75 99.68 79.38 64.56 23.98 --- Min. Sample Size for Significance (d) 7,560 8,387 15,342 8,485 3,028 7,660 6,943 14,352 13,935 12,128 7,647 14,011 54** 19,969 34,077 6,485 7,693 10,957 18,700 41,829 60,494 35,632 50,699 52,068 73,196 42,429 78,708 51,862 63,223 62,511 67,528 84,368 53,358 84,175 70,183 66,304 76,806 48,690 TABLE 4.1 Continued Item XZ for Actual Sample Min. Sample Number Model 3.2* Size Size for Significance (a) (b) (c) (d) 80 75.01 55,277 6,787 81 95.51 57,710 5,565 82 51.99 71,315 12,633 83 20.12 81,021 37,088 84 38.62 78,959 18,830 85 1240.78 41,180 306** 86 102.85 39,903 3,573 87 66.75 48,194 6,650 88 63.01 38,742 5,663 89 43.26 39,648 8,441 90 85.53 53,557 5,767 91 82.53 62,954 7,025 93 78.39 80,723 9,484 94 70.15 87,241 11,454 95 40.59 83,969 19,053 96 29.31 74,287 23,343 98 29.87 69,052 21,291 101 23.77 83,782 32,462 102 61.06 75,708 11,419 104 40.47 72,517 16,503 TABLE 4.1 Continued Item XZ for Actual Sample Min. Sample Number Model 3.2* Size Size for Significance (a) (b) (c) (d) 105 28.66 60,712 19,510 108 26.56 86,261 29,912 111 30.71 86,190 25,849 * p<.Ol ** Minimum sample size less than 3,000 TABLE 4.2 Chi-square Values for the Comparison of Models (3.2) and (3.3) with Actual Sample Sizes and with the Corresponding Number of Subjects Needed for Significant Results Item Xk for X2 for X1 for Actual Num- Model Model Difference Sample ber 3.2 3.3 Size lal 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 (b) 16.11 17.54 20.06 7.54 14.75 11.86 4.60 46.15 10.13 9.54 5.89 12.58 25.74 11.19 16.63 14.51 17.90 (c) 30.94 103.22 47.51 21.87 70.40 23.13 35.73 329.98 109.99 65.29 19.06 198.46 62.37 80.50 17.33 84.29 44.20 (_d) 14.83* 85.68* 27.45* 14.33* 55.65* 11.27* 31.13* 283.83* 99.86* 55.75* 13.17* 185.88* 36.63* 69.31* 0.70 69.78* 26.30* (e) 7,427 11,018 11,497 6,016 7,706 13,046 8,832 16,796 23,075 35,060 24,908 40,423 15,718 30,238 42,889 35,114 43,962 Minimum Sample Size for Signif. (f) 10,061 2,583** 8,733 8,414 2,782** 23,256 5,700 1,189** 4,642 12,634 37,996 4,369 8,621 8,765 1,230,914 10,109 33,582 TABLE 4.2. Continued Item Xe for X for X2 for Actual Minimum Num- Model Model Difference Sample Sample ber 3.2 3.3 Size Size for Signif. (a) (b) (c) (d) (e) (f) 18 11.71 15.68 3.97 52,340 264,864 19 9.37 18.77 9.40* 41,363 88,402 20 12.37 13.46 1.09 55,707 1,026,746 21 11.51 19.03 7.52 49,240 131,547 22 9.63 83.74 74.11* 51,724 14,022 23 29.25 153.74 124.49* 60,377 9,744 24 10.35 42.93 32.58* 68,426 42,194 25 35.90 147.40 111.50* 26,707 4,812 26 18.84 47.30 28.46* 56,717 40,037 27 8.78 222.87 214.09* 78,353 7,353 28 46.00 242.78 196.78* 84,847 8,662 29 61.48 170.22 108.74* 88,396 16,331 30 25.19 53.84 28.65* 87,473 61,338 31 23.26 47.42 24.16* 32,870 27,333 32 15.51 41.69 26.18* 48,777 37,430 33 7.57 48.84 41.27* 39,470 19,214 34 58.25 324.67 266.42* 82,257 6,203 35 6.91 225.66 218.75* 88,069 8,088 TABLE 4.2 Continued Item X2 for a for Xk for Actual Minimum Num- Model Model Difference Sample Sample ber 3.2 3.3 Size Size for Signif. (a) ( (c) (d) (e) (f) 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 84.36 23.70 13.64 13.35 6.67 12.65 23.26 16.29 3.55 22.41 6.56 7.82 179.25 27.82 14.23 16.37 26.01 6.69 41.41 134.56 76.85 84.24 757.60 701.39 603.42 189.11 21.65 115.84 55.90 16.30 76.15 452.64 59.75 240.27 224.54 96.14 27.49 95.41 50.20* 53.15* 70.60* 744.25* 694.72* 590.77* 165.85* 5.36 112.29* 33.49* 9.74* 68.33* 273.39* 31.93* 226.04* 208.17* 70.13* 20.80* 54.00* mE 86,137 37,068 38,870 61,057 56,438 57,145 59,140 33,712 67,192 13,098 56,727 56,555 97,593 16,766 25,410 24,715 39,858 44,694 74,502 34,472 14,011 11,0618 1,648** 1,630** 1,943** 7,154 126,357 12,021 37,851 117,007 16,628 7,172 10,549 2,258** 2,385** 11,418 43,168 27,718 TABLE 4.2 Continued Item Xz for Xa for Xl for Actual Minimum Num- Model Model Difference Sample Sample ber 3.2 3.3 Size Size for Signif. (a) (b) (c) (d) (e) (f) 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 72 73 90.87 70.14 35.78 51.69 73.68 148.39 394.54 300.95 278.34 246.48 94.11 113.88 180.60 225.46 251.66 271.35 208.90 37.80 163.01 140.26* 71.34* 48.52* 7.27 34.20* 22.72* 81.96* 373.15* 245.92* 119.98* 158.47* 37.83* 63.37* 146.32* 177.45* 176.37* 226.96* 191.49* 13.19* 19.53 21.62 28.51 17.49 50.96 66.43 21.39 55.03 158.36 88.01 56.28 50.51 34.28 48.01 75.29 44.39 17.41 24.61 11,770 18,263 146,668 25,873 36,987 14,828 1,918** 4,142 8,719 9,279 22,532 24,953 7,121 7,158 7,121 5,977 8,831 81,271 41,796 44,107 53,075 44,044 41,829 60,494 35,632 50,699 52,068 73,196 42,429 78,708 51,862 63,223 62,511 67,528 84,175 53,358 74 22.75 84,175 12,057 TABLE 4.2 Continued Item X1 for X1 for Xa for Actual Num- Model Model Difference Sample ber 3.2 3.3 Size (aL 75 76 77 78 79 80 81 82 83 84 86 87 88 89 90 91 92 93 94 (b) 8.69 99.68 79.38 64.56 23.98 75.01 95.51 51.99 20.12 38.62 102.85 66.75 63.01 43.26 85.53 82.53 16.36 78.39 70-15 (c) 237.07 167.72 432.36 127.04 72.59 128.13 106.82 108.12 95.18 69.05 433.47 430.40 126.27 289.14 405.04 85.48 93.12 229.02 137.27 (d) 228.38* 68.04* 352.98* 62.48* 48.61* 53.12* 11.31* 56.13* 75.06* 30.43* 330.62* 363.65* 63.26* 245.88* 319.51* 2.95 76.76* 150.63* 67.12* (e) 67,783 70,183 66,304 76,806 48,690 55,277 57,710 71,315 81,021 78,959 39,903 48,194 38,742 39,648 53,557 62,954 88,231 80,723 87,241 Minimum Sample Size for Signif. (f) 5,963 20,723 3,774 24,696 20,123 20,906 102,511 25,525 21,685 52,129 2,425** 2,662** 12,304 3,240 3,368 428,727 23,092 10,766 26,113 TABLE 4.2 Continued Item X1 for Xa for X for Actual Minimum Num- Model Model Difference Sample Sample ber 3.2 3.3 Size Size for Signif. (a) (b) (c) (d) (e) (f) 68.41 212.34 126.96 41.77 9.94 41.48 315.62 437.09 7.97 283.94 48.59 1,682.71 99.20 79.33 18.33 49.11 477.44 27.82* 183.03* 118.36* 11.90* 0.50 30.13* 291.85* 376.03* 1.85 243.47* 19.93* 1,669.34* 86.07* 52.77* 12.60* 31.31* 446.73* 83,969 74,287 50,028 69,052 72,647 60,931 83,782 75,708 50,663 72,517 60,712 71,946 72,482 86,261 76,800 70,980 86,190 60,638 8,154 8,492 116,576 2,918,956 40,627 5,767 4,045 550,173 5,984 61,199 866** 16,918 32,840 112,453 45,544 3,876 * P<.01 ** Minimum sample size less than 3,000 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 40.59 29.31 8.60 29.87 9.44 11.35 23.77 61.06 6.12 40.47 28.66 13.47 13.13 26.56 5.73 17.80 30.71 Sex-related Differences in Problem Solving Performance In this part, results for the two questions related to the second objective of the study are presented. These questions serve as the framework for the presentation. Each question is stated, followed by the results pertaining to that question. Question 1: Do males and females differ in problem solving performance? The responses of 492 males and 510 females who took the Puerto Rico Basic Skills Test-6 during the spring of the first year were analyzed in this study. Also, data from 504 males and 509 females tested in the second year and from 509 males and 504 females tested in the third year were included in the analysis. The mean performance scores and the standard deviations for each of the eight variables are presented in Table 4.3. Results of t-tests are also presented in this table. Females outperformed males in problem solving, a finding consistently present in all the three years of test administration. Over the three-year period the mean differences favored females in all variables except equivalence. The sex-related differences in problem solving were significant (p<.01) for all three years. TABLE 4.3 Means, Standard Deviations, and t-Tests for the Eight Mathematical Subtests Year/ Males Females Subtest X SD X SD t First Problem Solving Addition Subtraction Multi pl i ca ti on Division Fracadd Decsub Equivalence Second Problem Solving Addition Subtraction Multiplication Division Fracadd 3.436 2.32 5.412 1.00 4.550 1.66 4.014 1.80 3.136 1.87 2.475 1.75 2.648 1.83 .77 .77 N = 492 3.632 5.470 4.843 4.148 3.242 2.565 2.28 .98 1.52 1.73 1.77 1.74 3.862 5.592 4.852 4.433 3.580 2.743 3.321 .73 2.45 .86 1.56 1.68 1.91 1.81 1.86 .76 2.82* 3.05* 1.22 3.81* 3.72* 2.32** 5.77* - .83 N = 510 4.015 5.603 4.923 4.550 3.632 2.903 2.36 .86 1.44 1.67 1.86 1.91 2.65* 2.29** .86 3.76* 3.42* 2.94* TABLE 4.3 Continued Year/ Males Females Subtest X SD X SD t Decsub 2.863 1.87 3.440 1.80 5.00* Equivalence .76 .76 .65 .75 -2.32** N = 504 N = 509 Third Problem Solving 3.927 2.49 4.341 2.44 2.67* Addition 5.536 .84 5.704 .64 3.59* Subtraction 4.836 1.52 4.958 1.46 1.30 Multiplication 4.168 1.74 4.541 1.68 3.47* Division 3.343 1.88 3.795 1.83 3.87* Fracadd 2.819 1.89 3.117 1.85 2.53** Decsub 3.021 1.93 3.448 1.85 3.60* Equivalance .830 .82 .800 .78 .60 N = 509 N = 504 Note: The number of items in the problem solving subtest was 9. In each computation subtest, the number of items was 6. An item was included in the computation subtest only if it measured a computation skill required to solve a problem solving item. * p <.01 ** p <.05 Consistent significant differences were also found for addition, multiplication, division, addition of fractions, and subtraction of decimals. For subtraction the difference was not statistically significant. Question 2: Do sex-related differences in problem solving persist when computational skills are controlled for, and is the male-female differences in problem solving dependent on level of computational skills? To address the question of dependence of male-female problem solving differences in computational skills for each year and computation subtest, the possibility of an interaction was investigated. For the first year, statistically significant interactions were found between sex and multiplication, F (1,998) = 8.59, p<.01; and sex and division, F (1,998) = 4.25, p<.01. A significant interaction was found between sex and subtraction in the second year, F (1,1009) = 6.39, p<.05. No significant interactions were found in the third year. Analysis of covariance summary tables are shown as Tables 4.4, 4.5, and 4.6. Also, the three interactions are depicted in Figures 4.1, 4.2, and 4.3. Each figure indicates that at lower levels of computational skills, males outperformed females in problem solving, with the reverse happening at higher levels of computational skills. TABLE 4.4 ANCOVA Summary Table: Multiplication Covariate First Year Source df SS MS F Multiplication (M) 1 1195.90 1195.90 265.66* Sex (S) 1 31.95 31.95 7.07* M x S 1 38.80 38.80 8.59* Error 998 4509.00 4.51 p <.01 TABLE 4.5 ANCOVA Summary Table: Divison Covariate First Year Source df SS MS F Division (D) 1 1195.90 1195.90 264.66* Sex (S) 1 13.94 13.94 3.22 D x S 1 18.39 18.39 4.25* Error 998 4317.00 4.32 * p<.01 TABLE 4.6 ANCOVA Summary Table: Subtraction Covariate Second Year Source df SS MS F Subtraction (S) 1 93.14 593.14 122.80* Sex (S) 1 18.58 18.58 3.85* SxS 1 30.85 30.85 6.39* Error 1009 4873.40 4.83 *p <. 5 87 9 8 7 6 . 0 I~ 5 o ,. 4 3 2 1 1 2 3 4 5 6 Multiplication Fig. 4.1 Sex by Multiplication Interaction. 88 9 8. 7 6 5 4 3 2 1 1 2 3 4 5 6 Division Fig. 4.2 Sex by Division Interaction 89 9 8. 7. 6 : 5 *r* O U) 1 4 1 2 3 4 5 6 Subtraction Fig. 4.3 Sex by Subtraction Interaction The significant interactions found between sex and multiplications and sex and division (first year), and sex and subtraction (second year), answered, in part, the question of whether male-female differences in problem solving performance depend on computational ability. However, the evidence is quite weak. Of 21 possible interactions, only three were significant. No variable exhibited a significant interaction for each of the three years. The analysis of covariance was also used to determine if sex-related differences exist after controlling for computational skills. Analyses were conducted for those variables that did not exhibit significant interactions with sex. As discussed in Chapter III, estimated true scores were used for observed scores to adjust for unreliability of the covariates (the computational subtests). Reliability coefficients calculated for each covariate are shown in Table 4.7. Summaries of the analyses of covariance for the first year are reported in Table 4.8. The results show that females retained their superiority in problem solving performance when equivalence was the controlling variable in the analysis of covariancer the only variable in which males outperformed females (nonsignificant). When the controlling variables were addition, subtraction, addition of fractions, and subtraction with decimals, female |

Full Text |

84
Consistent significant differences were also found for addition/ multiplication/ division/ addition of fractions/ and subtraction of decimals. For subtraction the difference was not statistically significant. Question 2: Do sex-related differences in problem solving persist when computational skills are controlled for/ and is the male-female differences in problem solving dependent on level of computational skills? To address the question of dependence of male-female problem solving differences in computational skills for each year and computation subtest/ the possibility of an interaction was investigated. For the first year/ statistically significant interactions were found between sex and multiplication/ F (1/998) = 8.59/ p second year/ F (1/1009) = 6.39/ p<.05. No significant interactions were found in the third year. Analysis of covariance summary tables are shown as Tables 4.4/ 4.5/ and 4.6. Also/ the three interactions are depicted in Figures 4.1/ 4.2/ and 4.3. Each figure indicates that at lower levels of computational skills/ males outperformed females in problem solving/ with the reverse happening at higher levels of computational skills. 16 of the test. The sexes differed in only 2 of the 19 reference tests Spatial Relations and Picture Group Name-Selection. A factor analysis however showed differences in the number and composition of the factors. For females a general mathematics factor was determined by mathematics computation comprehension application and problem solving. For males the comprehension and application parts determined one factor; problem solving with two other reference tests (Gestalt and Omelet) determined another factor. Meyer concluded that comprehension of the data and mastery of the prerequisite mathematical concepts did not guarantee successful problem solving either for males or for females. Problem solving scores for both sexes were about one third their scores in comprehension and one fourth their scores in application. She also concluded that the sexes may have approached the problem solving questions differently. The methods used by females for solving problem situations may have paralleled their approach to the application parts. Males may have used established rules and algorithms for the application parts but may have used more of a Gestalt approach to the problem solving situation. Whitaker (1976) investigated the relationship between the mathematical problem performance of 4th grade children and their attitudes toward problem solving their teachers' 96 TABLE 4 .11 Adjusted Means on Problem Solving/ by Covariate and Sex Second Year Covariate Adjusted Means Males Females Addition 3 .69 3 .96 Multiplication 3 .80 3.85 Division 3.80 3.85 Fracadd 3.78 3 .87 Decsub 3.90 3 .75 Equivalence 3.46 44.18 62 Here the summation is over all the cells in the contingency table/ F0 refers to the observed frequency in a cell/ and Fe refers to the estimated expected frequency in a cell. Denoting the observed proportion in a cell by PG and the estimated expected proportion in a cell by Pe/ the test statistic can be written G2 = 2 N Â£^P0 1 ge (P0/Pe>- (3.5) where N is the total number of subjects. This form of the test statistic suggests the following strategy. For any significant G2/ using P0 and Pe calculated from the total data set available for an item/ calculate the minimum N required for G2 to be significant. If the minimum N is very large/ this suggests that the statistically significant result is not practically significant since it can only be detected in very large samples. Of course/ the question remains as to what can be considered a minimum large N. Although there is room for argument/ it seems reasonable to claim that if an average of 1000 subjects per year is required to show significance/ then the result is not likely to be practically significant. On the basis of this reasoning/ it was proposed to ignore all significant results that would be nonsignificant if there were less than 3000 subjects available. In addition/ all log-linear 57 Analysis of the Data Analysis of Sex bv Option by Year Cross Classifications Log-linear models were used to analyze the sex by option by year cross classifications for each item. Two topics are addressed in this section/ the hypotheses tested using log-linear models and a comparison of the hypotheses tested in this study with those tested by Marshall (1981/ 1983). The object of the analysis was to test two hypotheses: 1. The proportion of males and the proportion of females choosing each incorrect option does not vary from year to year. Note that this hypothesis is stated in the null form. 2. Assuming that the first hypothesis is correct/ the proportion of males who choose each incorrect option is different from the proportion of females who choose each incorrect option. This hypothesis is stated in the alternate form. In Table 3.1 a hypothetical cross classification of sexz option/ and year is presented. Hypothesis 1 is true for this three dimensional contingency table. In Table 3.2 the three dimensional contingency table is rearranged to show the year by option contingency table for each gender. 76 TABLE 4.2. Continued Item Num ber X2 for Model 3.2 X2 for Model 3.3 X2 for Difference Actual Sample Size Minimum Sample Size for Signif. (a) (b) (c) (d) (e) (f) 18 11.71 15.68 3.97 52/3 40 264/864 19 9.37 18.77 9.40* 41/363 88/402 20 12.37 13.46 1.09 55 /707 1/026 /7 46 21 11.51 19.03 7.52 49/240 131/547 22 9.63 83 .74 74.11* 51/7 24 14/022 23 29.25 153.74 124.49* 60/377 9/744 24 10.35 42.93 32.58* 68/426 42/194 25 35.90 147.40 111.50* 26 /707 4/812 26 18.84 47 .30 28.46* 56/717 40 /037 27 8.78 222.87 214.09* 78/353 7/353 28 46.00 242.78 196.78* 84/847 8/662 29 61.48 170.22 108.74* 88/396 16/331 30 25.19 53 .84 28.65* 87/473 61/338 31 23 .26 47.42 24.16* 32/870 27/333 32 15.51 41.69 26.18* 48/777 37/430 33 7.57 48.84 41.27* 39/470 19/214 34 58.25 324.67 266.42* 82 /257 6/203 35 6.91 225 .66 218.75* 88/069 8/088 60 Fienberg also shows that deleting the Ujterm from (3.2) to obtain log mijk= Ui+Uj+Ufc+Uik. (3.3) yields a model that specifies that option is independent of sex. Based on Fienberg's presentation/ an appropriate analysis for testing the hypotheses is 1. Conduct a likelihood ratio test of the adequacy of model (3.2). If this test is nonsignificant/ then the data support the adequacy of the modelr and the hypothesis that conditional on gender option/ and year are independent. Because model (3.1) is a saturated model/ testing the adequacy of model (3.2) is the same as comparing the adequacies of models (3.1) and (3.2). 2. Conduct a likelihood ratio test comparing the adequacies of models (3.2) and (3.3). If this test is significant/ then model (3.2) fits the data better than model (3.3) and the data support the hypothesis that the choice of incorrect option is dependent on sex. To summarize/ if the first test is nonsignificant and the second test is significant/ then the choice of option is dependent on sex and the pattern of dependency is the same for all three years. 19 solving performance in grade 11. Spatial visualization was a stronger predictor for girls than for boys. Mathematics as a male domain was a good predictor for girls only; the less a girl stereotyped mathematics as a male domain in grade 8# the higher her problem solving score in grade 11. Attitudes toward success in mathematics in grade 8 was a more positive predictor of problem solving performance for boys than for girls; the more positive the attitudes toward success in mathematics in grade 8/ the higher their performance in problem solving in grade 11. None of these four studies# all of which used the Romberg-Wearne Mathematics Problem Solving Test# show statistically significant sex-related differences in problem solving performance. In later studies other tests were used to measure this variable (Kaufman# 1984; Marshall# 1981# 1984). Kaufman (1983/1984) investigated if sex differences in problem solving# favoring males# exist in the 5th and 6th grades and if these differences were more pronounced in mathematically gifted students than in students of average mathematical ability. The Iowa Test of Basic Skills and a mathematics problem-solving test were administered to 504 subjects. Males in the average group as well as males in the gifted group outperformed females# but only the gifted group showed statistically significant differences. 85 TABLE 4.4 ANCOVA Summary Table: Multiplication Covariate First Year Source df SS MS F Multiplication (M) 1 1195.90 1195.90 265.66* Sex (S) 1 31.95 31.95 7.07* M x S 1 38.80 38.80 8.59* Error 998 4509.00 4.51 * p <;.oi TABLE 4.5 ANCOVA Summary Table: Divison Covariate First Year Source df SS MS F Division (D) 1 1195.90 1195.90 264.66* Sex (S) 1 13.94 13.94 3.22 D x S 1 18.39 18.39 4.25* Error 998 4317.00 4.32 * pC.Ol 7 errors (Harnisch# 1983). Item response patterns are valuable for the identification of large group differences# including district-to-district # school-to-school# and classroom-to-classroom variations on different subsets of items. The response patterns can provide diagnostic information about the type of understanding the student has on various mathematics topics (e.g.r problem solving). Marshall (1981# 1984) has used the item response pattern technique and her findings indicate sex differences in mathematics performance at the item level. Females outperformed males in computation and males outscored females in problem solving. Also# the success of girls in the problem solving items was dependent upon their success in the computation items; for boys# success in the problem solving items did not depend as much on their computational performance. Marshall (1981# 1983) has also reported that males and females differ in the selection of incorrect responses# reflecting differences in reasoning abilities. In Puerto Rico# a high percentage of children promoted to the 7th grade in the public schools does not master the basic skills in mathematics. If 6th grade male and female children can be diagnosed as having different problem solving abilities# as Marshall found with California children# teachers may need to provide tailor-made mathematics instruction for each sex# in order to ensure equal access to formal education and enhance mathematics 11 She analyzed the responses of 9*000 boys and 9*000 girls to 6 selected items* 2 computation items* and 1 story problem item from each of 2 of the 16 forms of the Survey of Basic Skills test administered during the academic year 1978-79. The Survey is a 30-item achievement test administered every year to all 6th grade children in California through the California Assessment Program. There are 16 forms of the test* to which approximately 9 *000 boys and 9 *000 girls respond each year. Of the 160 mathematics items contained in the 16 forms of the test* 32 are on measurement and graphing* 28 on number concepts* 28 on whole number arithmetics* 20 on fraction arithmetics* 20 on decimal arithmetics* 20 on geometry* and 12 on probability and statistics. The item analysis performed on the 1978-79 data showed that boys and girls tended to select different incorrect responses. In the first computation item (Form 1 of the test) both sexes reflected similar mistakes in carrying* but in different columns. In the second computation item* both sexes ignored the decimal points and selected the same incorrect response. However* more girls than boys chose this response. In the first computation item (Form 2 of the test) the incorrect choice of both sexes was option Â£* but the second most frequently selected option was a for boys and b for girls. In the second computation item of this 21 story problem item correctly after giving incorrect responses to both computation items. Apparently mastery of the skills required by the computation items is more important for girls than for boys. If girls cannot solve the computation items they have little chance of solving the related story problem item. For girls the probability of success in the story problem item after giving successful answers to both computation items is almost 2 1/2 times the probability of success after giving incorrect responses to both computation items. For boys the probability of success in the story problem item after successful responses to the computation items is about 1 3/4 times the probability of success on the story problem item after incorrect responses to the computation items. Three years later Marshall (1984) analyzed more in depth these phenomena of sex-related differences. Her interest was twofold. First she wanted to know if there were differences in the rate of success for boys and girls in solving computation and story problem items. Second she examined additional factors that interact with sex to influence mathematics performance such as reading achievement socio-economic status (SES) primary language and chronological age. Two questions were raised: Do the probabilibities of successful solving of computation and story problem items increase with reading score? Are these probabilities different for the two sexes? 55 1 /000 students were selected randomly for each year. For the first year / 1 / 00 2 were selected (492 boys and 510 girls); for the second year/ 1/013 students were selected (504 boys and 509 girls); and/ for the third yearz 1/013 students were selected (509 boys and 504 girls). The student population in Puerto Rico includes children from the urban and rural zones and comprises children from low and middle socioeconomic levels. Findings can be generalized only to this population. The Basic Skills Test in Mathematics-6 is a criterion-referenced test used in the Department ot Education of Puerto Rico as part of the annual assessment program. The test measures academic achievement in operations/ mathematical concepts/ and story problems. It has a reported split half reliability ot .95. The test was designed specifically for Puerto Rico. Its contents and the procedures followed for its development were formulated and reviewed by educators from the mathematics department of the Department of Education of Puerto Rico/ in coordination with the Evaluation Center ot the Department of Education and mathematics teachers from the school districts. The emphasis placed on each skill area is depicted in Figure 3.1. 114 Sherman/ J. (197 9) Predicting mathematics performance in high school girls and boys. Journal of Educational Psychology/ 71/ 242-249. Sherman/ J. (1980) Mathematics/ spatial visualization/ and related factors: Changes in girls and boys/ grades 8-11. Journal of Educational Psychology/ 22/ 476-482. Stein/ A. H. (1971). The effects of sex role standards for achievement and sex role preference on three determinants of achievement motiviation. Developmental Psychology/ 4.* 219-231. Stein/ A. H./ & Smithless/ J. (1969). Age and sex differences in children's sex role standards about achievement/ Developmental Psychology/ 1/ 252-259. Suydam/ M. N. (1971). Research on mathematics education (K-12) reported in 1970. Journal for Research in Mathematics Education/ Z, 257-298. Suydam/ M. N. (1972). Research on mathematics education (K-12) reported in 1971. Journal for Research in Mathematics Education/ 2/ 196-232. Suydam/ M. N. (197 3) Research on mathematics education (K-12) reported in 1972. J Mathematics. Education/ / 205-242. Suydam/ M. N. (1974). Research on mathematics education (K-12) recorded in 1973. Journal for Research in Mathematics Education/ 2/ 238-272. Suydam/ M. N./ & Weaver/ J. F. (1975). Research on mathematics education (K-12) recorded in 1974. Journal for Research in Mathematics Education/ Â£/ 253-282. Verbeke/ K. A. (1983). Sex-related differences in mathematically gifted secondary students: An investigation of selected cognitive/ affective/ and educational factors (Doctoral dissertation/ University of Maryland/ 1982) Dissertation Abstracts International/ 42/ 2267A-2268A. Weiner/ B. (1974). Achievement motivation and attribution theory. Morristown/ N J: General Learning Press. 76 TABLE 4.2. Continued Item Num ber X2 for Model 3.2 X2 for Model 3.3 X2 for Difference Actual Sample Size Minimum Sample Size for Signif. (a) tb) (c) . (d). (e) (f) 18 11.71 15.68 3.97 52/340 264,864 19 9.37 18.77 9.40* 41/363 88,402 20 12.37 13.46 1.09 55/707 1/026,746 21 11.51 19.03 7.52 49/240 131/547 22 9.63 83.74 74.11* 51/724 14,022 23 29.25 153.74 124.49* 60/377 9/744 24 10.35 42.93 32.58* 68/426 42/194 25 35.90 147.40 111.50* 26/707 4/812 26 18.84 47.30 28.46* 56/717 40 /037 27 8.78 222.87 214.09* 78,353 7/353 28 46.00 242.78 196.78* 84/847 8,662 29 61.48 170.22 108.74* 88/396 16/331 30 25.19 53.84 28.65* 87/473 61/338 31 23.26 47.42 24.16* 32/870 27,333 32 15.51 41.69 26.18* 48,777 37/430 33 7.57 48.84 41.27* 39,470 19/214 34 58.25 324.67 266.42* 82,257 6/203 35 6.91 225.66 218.75* 88,069 8,088 Log-linear models were used in the analysis of incorrect responses. The results of the analyses showed that for 100 of the 111 items of the test* males and females selected different incorrect optionsr and this pattern of responses was consistently found during the three years of test administration. However for the vast majority of the 100 items the male-female differences were relatively small considering the fact that the number of subjects needed to obtain statistical significance was very large. The responses of approximately 1000 randomly selected students per academic year were analyzed in the comparison of male and female performance in problem solving. Females outperformed males in problem solving and in six of the seven computational variables. Males showed superiority in equivalence in all the three years but statistical significance was obtained in only one of the years. Analysis of covariance (ANCOVA) was used in the comparison of male and female performance in problem solving after controlling for computational skills. Seven analyses of covariance tests were conducted one for each of the covariates. Estimated true scores for observed scores were used in the analyses. The results tend to show that for examinees with similar levels of computational ix 21 story problem item correctly after giving incorrect responses to both computation items. Apparently* mastery of the skills required by the computation items is more important for girls than for boys. If girls cannot solve the computation items* they have little chance of solving the related story problem item. For girls* the probability of success in the story problem item after giving successful answers to both computation items is almost 2 1/2 times the probability of success after giving incorrect responses to both computation items. For boys* the probability of success in the story problem item after successful responses to the computation items is about 1 3/4 times the probability of success on the story problem item after incorrect responses to the computation items. Three years later* Marshall (1984) analyzed more in depth these phenomena of sex-related differences. Her interest was twofold. First* she wanted to know if there were differences in the rate of success for boys and girls in solving computation and story problem items. Second* she examined additional factors that interact with sex to influence mathematics performance* such as reading achievement* socio-economic status (SES) primary language* and chronological age. Two questions were raised: Do the probabilibities of successful solving of computation and story problem items increase with reading score? Are these probabilities different for the two sexes? 37 dealing with both cognitive and affective variables/ there are no clear-cut findings to render unequivocal support to a particular variable as accounting for these sex-related differences. However/ everything seems to point to the fact that affective/ rather than cognitive variables play a more significant role in the sex-related differences observed in mathematics performance and learning. In most of the studies dealing with affective variables/ findings consistently show that these factors influence mathematics performance in females more than in males. In at least one area/ confidence as learners of mathematics/ Sherman (1980) found that this variable influenced course election more than all the cognitive variables previously discussed. The case for the societal influences on sex roles and expectations to account for the differences in mathematics learning is also supported in one way or another in the studies reported in the literature. Problem Solving Performance and Related Variables Problem solving has been perhaps the most extensively researched area in mathematics education. Published reviews by Kilpatrick (1969)/ Riedesel (1969)/ and Suydam and Weaver (1970-1975) attest to this. Much of the research done has focused on identifying the determinants of problem difficulty and the problem features that influence the solution process. 109 Creswell/ J. L. (1982/ February). Sex-related differences in problem solving in rural black; anglo and Chicago adolescents. Paper presented at the annual meeting of the Southwest Educational Research Association. Austin/ TX. (ERIC Document Reproduction Service No. ED 216 895) Dwyer/ C. A. (1974). Influence of children's sex role standards on reading and arithmentic achievement. Journal of Educational Psychology/ Â£Â£/ 811-816. Eccles/ J. ( 1983 ) Sex differences in mathematics participation. In N. Steinkamp & M. Maehr (Eds.)/ Women in Science (pp. 80-100). Greenwich/ CT: JAI Press. Elton/ C. F./ & Rose/ H. A. (1967). Traditional sex attitudes and discrepant ability measures in college women. jQucpai of CQhhseling Psycology* Hr 538-548. Ernest/ J. (1976). Mai.hem3tics and Sex. Santa Barbara: University of California. Exedisis/ R. H. (1983). An investigation of the relation- ship of reading comprehension/ vocabulary/ mathematical concepts/ and computation on problem solving among angloz black/ and Chicago male and female middle school adolescents (Doctoral Dissertation/ University of Houston/ 1982). ni_s.sgrtatieii_Abstracts International/ 41/ 2264A-2265A. Fennema/ E. (1974). Mathematics learning and the sexes: A review. Journal for Research in Mathematics -Educfltioji/ 5./ 126-139. Fennema/ E. (1975). Mathematics/ spatial ability and the sexes. In E. Fennema/ (Ed.)/ Mathematics: What r_esga-rch_savs about_sex differences (pp. 33-44) . Columbus: Ohio State University/ College of Education. (ERIC Document Reproduction Service No. ED 128-195) Fennema/ E. (1976) Influences of selected cognitive afcfcectiahd_educatpn,al_,variables on sex-related _djj.fergji.ces in.mathematics learning and studying. Madison: University of Wisconsin/ Department of Curriculum and Instruction. (ERIC Document Reproduction Service No. ED 142 906) 100 exception was that/ when equivalence was controlled/ sex-related differences in favor of females/ persisted. The question of whether male-female differences in problem solving depend on computational skills was answered partially/ in the affirmative. Interactions/ indicating the dependence/ were significant only for multiplication and division in the first year and for subtraction in the second year. COMPARISON OF MALES AND FEMALES ON MATH ITEM PERFORMANCE: ANALYSIS OF RESPONSE PATTERNS By SONIA FELICIANO A DISSERTATION PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY UNIVERSITY OF FLORIDA 1986 Ill METHOD 54 The Sample 54 The Instrument 55 Analysis of the Data 57 Analysis of Sex by Option by Year Cross Classifications 57 Comparison of Males and Females in Problem Solving Performance 66 IV RESULTS 68 Introduction 68 Sex-related Differences in the Selection of Incorrect Responses 68 Sex-related Differences in Problem Solving Performance 81 Summary 93 V DISCUSSION 101 Summary and Interpretation of the Results 101 Implications of the Findings and Suggestions for Further Research 103 Sex-related Differences in Incorrect Responses 104 Sex-related Differences in Problem Solving Performance 105 REFERENCES 107 BIOGRAPHICAL SKETCH 116 iv .UL 59 60 61 62 63 64 65 66 67 68 69 70 71 73 74 76 77 78 79 72 TABLE 4.1 Continued X1 for Model 3.2* Actual Sample Size Min. Sample Size for Significance (b) (c) (d) 50.96 41/829 7/560 66.43 60/494 8/387 21.39 35/632 15/342 55.03 50/699 8/485 158.36 52/068 3/028 88.01 73/196 7/660 56.28 42/429 6/943 50.51 78/708 14/352 34.28 51/862 13/935 48.01 63/223 12/128 75.29 62/511 7/647 44.39 67/528 14/011 14/425 .97 84/368 54** 24.61 53/358 19/969 22.75 84/175 34/077 99.68 70/183 6/485 79.38 66/304 7/693 64.56 76/806 10/957 23.98 48/690 18/700 15 The test is composed of 23 items designed to yield 3 scores: a comprehension score an application score and a problem solving score. The comprehension question ascertains whether a child understands the information given in the item stem. The application question assesses the child's mastery of a prerequisite concept or skill of the problem solving question. The problem solving question poses a question whose solution is not immediately available that is a situation which does not lend itself to the immediate application of a rule or algorithm. The application and problem solving parts of the test may refer to a common unit of information (the item stem) but the questions are independent in that the response to the application question is not used to respond to the problem solving question. Meyer (1978) Whitaker (1976) Fennema and Sherman (1978) and Sherman (1979) have used the Romberg-Wearne test in their studies. Meyer (1978) investigated whether males and females differ in problem solving performance and examined their prerequisite computational skills and mathematical concepts for the problem solving questions. A sample of 179 students from the 4th grade were administered 19 "reference tests" for intellectual abilities and the Romberg-Wearne test. The analysis showed that males and females were not significantly different in the comprehension application and problem solving questions 53 Tne studies reviewed have confirmed the relationship between problem solving performance and attitudes toward problem solving (Carey 1958; Lindgren et al. 1964; Whitaker 1976). However the results reportea in the studies that investigated the relationship between problem solving performance and computation and between reading and problem solving fail to be consistent in their conclusions. Hansen (1944) Chase (I960) Balow (1964) Kmfong and Holtan (1976 1977) and Zalewski (1974; concluded that computation is more strongly related to problem solving than is reading. Martin (1964) Creswell (1982) Ballew and Cunningham (1982) and Muth (1984) concluded that reading aoility and mathematical problem solving show a stronger relationsnip than computation and problem solving. Exedisis's (1983) findings led to the conclusion that the effect or reading and computation in problem solving performance is unimportant. 6 Purpose of the Study The purpose of this study was twofold. The first was to investigate sex-related differences in the selection of incorrect responses in a multiple-choice achievement test in mathematics. For each test item* the research questions were as follows: 1. Is there a difference in the proportion of males and females choosing each incorrect option? 2. Is the same pattern of differences found in data obtained in three different administrations of the test? The second objective was to investigate sex-related differences in test scores in mathematics problem solving. The following questions were studied: 1. Do males and females differ in problem solving performance? 2. Do sex differences in problem solving persist after controlling for computational skills* and does the differential success of males and females on problem solving items depend on their success on the computation items? significenge Qf the Study Item response patterns are very useful techniques in the assessment of mathematics learning and achievement. Total test scores can be very misleading in the assessment of student performance and provide no diagnostic information about the nature and seriousness of student 34 achievement and interest in mathematics result from identification with the masculine role. A study related to the differential cultural reinforcement hypothesis is that of Dwyer (1974). Dwyer examined the relationship between sex role standards (the extent to which an individual considers certain activities appropriate to males or females) and achievement in reading and arithmetic. Students from grades 2/ 4* 6/ 8# 10/ and 12 participated in this study. She found that sex role standards contributed significant variance to reading and arithmetic achievement test scores and that the effect was stronger for males than for females. This led to her conclusion that sex-related differences in reading and arithmetic are more a function of the child's perception of these areas as sex-appropriate or sex-inappropriate than of the child's biological sex# individual preference for masculine and feminine sex roles/ or liking or disliking reading or mathematics. In a study which agrees with the masculine identification hypothesis/ Milton (1957) found that individuals who had received strong masculine orientation performed better in problem solving than individuals who received less masculine orientation. Elton and Rose (1967) found that women with high mathematical aptitude and average verbal aptitude scored higher on the masculinity scale of the Omnibus Personality Inventory (OPI) than those with average scores on both tasks. 4 mathematics preparation. Males who were enrolled or had completed algebra II outperformed the females in computation and problem solving but not in algebra. Males who studied beyond algebra II outscored females on all three subtests: computation/ algebra/ and problem solving (Armstrong/ 1979) . Carpenter/ Lindquist/ Mathews/ and Silver (1984) analyzed the results of the Third National Assessment of Educational Progress (NAEP)/ and compared them with the First and Second Surveys. Between 1978 and 1982/ the differences between the average performance of males and females remained stable at each age level. At ages 9 and 13/ the overall performance of males and females was not significantly different. At age 17/ males scored higher than females by about 3 percentage points. When course background was held constant/ achievement differences still existed at age 17. For each category of course background/ male achievement exceeded female achievement. Consistent with previous assessments/ sex differences in problem solving in favor of males were found for the 17-year-old sample. At ages 9 and 13/ no large differences were found between the sexes within any level of course background. Marshall (1981/ 1984) investigated sex differences in mathematics performance. She found that males and females excel each other in solving different types of problems. 80 TABLE 4.2 Continued Item Num ber X2 for Model 3.2 X2 for Model 3.3 X2 for Difference Actual Sample Size Minimum Sample Size for Signif. (a) (b) (c) (d) (e) (f) 95 40 .59 68.41 27 .82* 83 /969 60/638 96 29.31 212.34 183.03* 74/287 8/154 97 8.60 126.96 118.36* 50/028 8,492 98 29.87 41.77 11.90* 69/052 116/576 99 9.44 9.94 0.50 72/647 2/918/956 100 11.35 41.48 30.13* 60/931 40 /627 101 23.77 315.62 291.85* 83 /782 5/767 102 61.06 437.09 376.03* 75/708 4/045 103 6.12 7.97 1.85 50/663 550/173 104 40 .47 283.94 243 .47* 72/517 5/984 105 28.66 48.59 19.93* 60/712 61/199 106 13 .47 1/682 .71 1/669.34* 71/946 866* 107 13.13 99.20 86.07* 72/482 16 ,918 108 26 .56 79.33 52.77* 86/261 32,840 109 5.73 18.33 12.60* 76/800 112/453 110 17.80 49.11 31.31* 70/980 45 ,544 111 30.71 477.44 446 .73* 86/190 3/876 pC.oi Minimum sample size less than 3/000 30 The issue of sex-related differences in spatial visualization ability as an explanation for sex differences observed in mathematics achievement is less convincing and the findings more contradictory than in the issue of sex differences in formal education. Besides these cognitive issues/ other issues/ mostly affective in nature/ have also been studied in trying to explain the origin of these sex differences in mathematics achievement and learning. The studies dealing with these affective variables are reviewed in the next section. Differentiated Effect of Affective Variables Researchers have attempted to explain the effect of sex differences in internal beliefs/ interests/ and attitudes (affective variables) on mathematics learning and achievement. A brief statement of each explanation precedes the summary of studies conducted that support the explanation. than males/ lack confidence in their ability to learn mathematics and this affects their achievement in mathematics and their election of more advanced mathematics courses. Maccoby and Jacklin (1974) reported that self- confidence in terms of grade expectancy and success in particular tasks was found to be consistently lower in women than in men. In 1978/ Fennema and Sherman reported 8 achievement. Since there are no investigations reported in sex differences in item response patterns in Puerto Rico* research is needed. Organization of the Study A review of the literature on sex differences in mathematics performance is reported in Chapter II. The research methodology is presented in Chapter III. Research questions sample instrument and data analysis are discussed in that chapter. Chapter IV is an exposition of the results of the study. Chapter V contains a summary and interpretation of the results of the study and the implications of the findings together with suggestions for further research. LIST OF FIGURES Figure Page 3.1 Number of Questions by Skill Area in a Basic Skills Test in Mathematics-6 56 4.1 Sex by Multiplication Interaction 87 4.2 Sex by Division Interaction 88 4.3 Sex by Subtraction Interaction 89 vii 29 females may use them in a different way. Meyer (1978) , with an elementary grade sample/ and Pennema and Tartree (1983) / with an intermediate level sampler found that the influence of spatial visualization on solving mathematics problems is subtler and that males and females use their spatial skills differently in solving word story problems (problems that measure problem solving ability or reasoning) Fennema and Tartree (1983) carried out a three-year longitudinal study which showed that girls and boys with equivalent spatial visualization skills did not solve the same number of items/ nor did they use the same processes in solving problems. The results also suggested that a low level of spatial visualization skills was a more debilitating factor for girls than for boys in problem solving performance. Landau (1984) also investigated the relationship between spatial visualization and mathematics achievement. She studied the performance of middle school children in mathematical problems of varying difficulty/ and the extent to which a diagramatic representation is likely to facilitate solution. She found that spatial ability was strongly correlated to mathematical problem solving and that the effect of spatial ability was more influential for females. Females made more use of diagrams in the solution of problems/ reducing the advantage of males over females in problem solving performance. lal 8 13 23 25 28 29 30 31 34 36 37 42 45 48 49 52 54 56 57 71 TABLE 4 .1 Summary of Significant Tests of Model 3.2 X1 for odel 3.2* Actual Sample Size Min. Sample Size for Significance (b) (c) (d) 46 .15 16,796 3,352 25.74 15,718 5,624 29.25 60,377 19,011 35.90 26 ,707 6,852 46 .00 84 ,847 16,988 61.48 88,396 13,242 25.19 87 ,473 31,982 23 .26 32,870 13 ,015 58.25 82 ,257 13 ,006 84.36 86 ,137 9,404 23 .70 37,068 14,405 23 .26 59,140 23 ,417 22.41 63,098 25,932 179.25 97 ,593 5,014 27.82 16,766 5,550 26.01 39,858 14,114 41.41 74,502 16,570 21.62 44,107 18,789 28.51 53 ,075 17 ,146 69 are independent/ conditional upon sex, log m i j k = Ui + Uj + Uk+Uik+Uj k* (3.2) Substantively/ this model implies that the pattern of male and female option choices is consistent over the three years of test administration. The second model indicates that option is independent of sex/ log mj.jk = u i+ u j + uk+ ui k (3.3) This model implies that the pattern of option choice is the same for males and females. In order to determine if males and females differed in the selection of incorrect responses and if these differences were stable across the three years of test administration/ a two-step test was performed. First/ a likelihood ratio test of the adequacy of model (3.2) was conducted for each of the 111 items of the test. Under this test a model fits the data when the value obtained is nonsignificant at a specified alpha level. For 50 (45%) of the items/ model (3.2) fitted the data adequately. For these items the pattern of male and female option choices was consistent for the three years. The values for the 61 items for which the likelihood ratio 109 Creswell/ J. L. (1982/ February). Sex-related differences i..n_problem solving in rural black/ anglo and Chicago adolescents. Paper presented at the annual meeting of the Southwest Educational Research Association. Austin/ TX. (ERIC Document Reproduction Service No. ED 216 895) Dwyer/ C. A. ( 197 4) Influence of children's sex role standards on reading and arithmentic achievement. Journal of Educational Psychology/ Â£Â£/ 811-816. Eccles/ J. ( 1 9 83 ) Sex differences in mathematics participation. In M. Steinkamp & M. Maehr (Eds.)/ Women in Science (pp. 80-100). Greenwich/ CT: JAI Press. Elton/ C. F. / & Rose/ H. A. ( 1967). Traditional sex attitudes and discrepant ability measures in college women. Journal of Counseling Psycology/ M* 538-548. Ernest/ J. (1976). Mathematics and Sex. Santa Barbara: University of California. Exedisisz R. H. (1983). An investigation of the relation ship of reading comprehension/ vocabulary/ mathematical concepts/ and computation on problem solving among anglo/ black/ and Chicago male and female middle school adolescents (Doctoral Dissertation/ University of Houston/ 1982). Dissertation,Abstracts International/ 42., 2264A-2265A. Fennema/ E. (1974). Mathematics learning and the sexes: A review. Journal for Research in Mathematics Education/ / 126-139. Fennema/ E. (1975). Mathematics/ spatial ability and the sexes. In E. Fennema/ (Ed.)/ Mathematics: What research savs about sex differences (pp. 33-44). Columbus: Ohio State University/ College of Education. (ERIC Document Reproduction Service No. ED 128-195) Fennema/ E. (1976) Influences of selected cognitive affective/ and educational variables on sex-related djff_exgnc.e.s^,n_mathematics learning and studying. Madison: University of Wisconsin/ Department of Curriculum and Instruction. (ERIC Document Reproduction Service No. ED 142 906) 93 superiority in problem solving lost statistical significance. Adjusted means, associated with the analyses, are reported in Table 4.9. Summaries of the analyses of covariance for the second year are reported in Table 4.10. The analyses show that females retained their superiority in problem solving after controlling for equivalence. When the controlling variables were addition, multiplication, division, addition of fractions, and subtraction with decimals females superiority in problem solving lost statistical significance. Adjusted means are reported in Table 4.11. The analyses of covariance for the third year are summarized in Table 4.12. The results of the analyses show that females maintained their superiority in problem solving when performance on either subtraction or equivalence was controlled. When addition, multiplication, division, addition of fractions, and subtraction of decimals were the controlling variables, female superiority in problem solving lost statistical significance. Adjusted means are reported in Table 4.13. Summary Sex-related differences in mathematics achievement were the subject of this study. The first objective was to investigate whether males and females differ in the selection of incorrect responses, and if the pattern of 32 attributes success to an internal, stable attribute, such as ability, then one is confident of being successful in the future and will continue to strive in that area. If one attributes success to an external factor such as a teacher, or to an unstable one, such as effort, then one will not be as confident of success in the future and will cease to strive. Failure attribution patterns work this way: if failure is attributed to unstable causes, such as effort, failure can be avoided in the future and the tendency will be to persist in the task. However, if failure is attributed to a stable cause, such as ability, the belief that one cannot avoid failure will remain. Studies reported by Bar-Tal and Frieze (1977) suggest that males and females tend to exhibit different attributional patterns of success and failure. Males tend to attribute their success to internal causes and their failures to external or unstable ones. Females show a different pattern; they tend to attribute success to external or unstable causes and failures to internal ones. The pattern of attributions, success attributed externally and failure attributed internally, has become hypothesized to show a strong effect on mathematics achievement in females. Kloosterman's (1985) study supported these findings. According to Kloosterman, attributional variables appear to be more important achievement mediators for females than for males, as measured by mathematics word problems. More research is needed in this area. 50 The studies reviewed in this section show a positive relationship between reading and problem solving performance* but in the case of Ballew and Cunningham (1982)* this relationship is not viewed singly but rather as one among the interacting factors that produce successful problem solving. The third variable reviewed is the effect of student attitudes toward problem solving on problem solving performance. Many researchers have tried to demonstrate that this variable is a determinant factor in problem solving achievement. Attitudes. Toward, Problem Solving and Problem Solving performance Research studies support the existence of positive and rather stable relationships between student attitudes and achievement in mathematics. Aiken (1970) has suggested that an individual's attitude toward one aspect of the discipline (mathematics) such as problem solving* may be entirely different from his/her attitude toward another phase of the discipline* such as computation. Research* however* has been directed to the use of single* global measures of attitudes toward mathematics rather than to the investigation of attitudes toward a particular phase of the discipline. The studies described below are only part of the few investigations which have examined the relationship between 45 He found that in the 4th grade the correlations between problem solving and abstract verbal reasoning reading comprehension/ arithmetic concepts/ and computation were .61/ .64/ .66/ and .60 respectively/ and .56/ .68/ .69/ and .63 in the 8th grade. When computation was held constant/ the correlation between problem solving and reading was .52 in grade 4 and .54 in grade 8. When reading was held constant the correlation between problem solving and computation was .43 in grade 4 and .42 in grade 8. Creswell (1982) worked with a sample of anglo and black adolescents from Chicago. Each subject was administered the California Achievement test. Multiple regression was used to analyze the data. The analysis showed that reading is more important than computation in predicting student performance in problem solving. Reading accounted for 49.5% of the variance; computation accounted for 14.6% of the variance. Ballew and Cunningham (1982) worked with 6th grade students in an attempt to find what proportion of students have as their main source of difficulty with word problems each of the following factors: a) computation skills/ b) interpretation of the problem/ c) reading and/ d) integrating these skills in the solution of problems. They also wanted to know if a student can be efficiently diagnosed as having one of the four categories as his/her main difficulty with mathematics word problems. 49 number of problems set up correctly* and amount of time spent taking the test. Reading ability and computational ability were both positively correlated with number of correct answers and with number of problems correctly set up* and negatively correlated with test-taking time. Presence of extraneous information was negatively correlated with correct answers and correct set ups and positively correlated with test-taking time. Syntactic complexity was not significantly correlated with any of the performance measures. Results of a multiple regression analysis showed that reading accounted for 46% of the variance in total correct answers and computation accounted for 8%. Reading ability and computational ability uniquely accounted for 14% and 8% of the variance in the number of correct answers* respectively. Extraneous information added significantly to the variance explained in the number of correct answers* but syntactic structure did not. Reading ability accounted for 5% of the variance in test-taking time* but computation did not add significantly to the variance explained by reading. Muth concluded that reading and computation both contribute significantly to success in solving arithmetic word problems* but that reading plays a more significant role than does computation. COMPARISON OF MALES AND FEMALES ON MATH ITEM PERFORMANCE: ANALYSIS OF RESPONSE PATTERNS By SONIA FELICIANO A DISSERTATION PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY UNIVERSITY OF FLORIDA 1986 36 they show less interest in the subject than counterpart males. These differences in interest are what Hilton and Berglund (1974) suggest to account for sex-related differences in mathematics achievement. Although the perception of the usefulness of mathematics is still an important predictor of course taking for girls* there is a growing similarity between males and females regarding the usefulness of mathematics (Armstrong & Price* 1982; Fennema & Sherman* 1977; Moller* 1982/1983). Armstrong and Price investigated the relative influence of selected factors in sex-related differences in mathematics participation. Both males and females selected usefulness of mathematics as the most important factor in deciaing whether or not to take more mathematics in high school. Moller's study revealed that both males and females based mathematics course-taking decisions on career usefulness. A Fennema and Sherman (1977) study showed only slight differences between males and females in their feelings about the usefulness of mathematics. In her study of this variable among college students* Probert (1983/1984) did not find any sex-related differences either. These have been the main affective variables researched in attempting to explain the underlying causes of sex-related differences observed in mathematics learning and achievement. In spite of the great diversity of studies 39 The findings suggest that reading factors are not as important as arithmetic and mental factors in problem solving performance. However these findings should be taken cautiously* as the content of the Gates tests (used to measure reading) is literary and does not include mathematical material. Chase (1960) studied 15 variables in an effort to find out which ones have significant influence on the aDility to solve verbal mathematics problems. Only computation* reading to note details* and fundamental knowledge were primarily related to problem solving. Computation accounted for 20.4% of the 32% variance directly associated with problem solving. Chase concluded that a pupil's ability in the mechanics of computation* comprehension of the principles that underline the number systems* and the extent to which important items of information are noticed when reading* are good predictors of the student's ability in solving verbal problems. Balow (1964) investigated the importance of reading ability and computation ability in problem solving performance. He objected to the approaches used by other researchers who in their analyses dichotomized research subjects as "poor" or "good" students* and who ignored the recognizea effect of intelligence on reading and on mathematics achievement Balow administered the Stanford 33 Mathematics as a male domain. Mathematics is an activity more closely related to the male sex domain than to the female sex domain (Eccles et al.r 1983). Thus* the mathematical achievement ot boys is higher than that of gi rls. According to John Ernest (1976) in his study Mathematics and Sex> mathematics is a sexist discipline. He attributed sex-related differences in mathematical achievement to the creation by society of sexual stereotypes and attitudes* restrictions* and constraints that promote the idea of the superiority of boys in mathematics. Ernest reported that boys* girls* and teachers* all believe that boys are superior in mathematics* at least by the time students reach adolescence. Bern and Bern (1970) agree and argue that an American woman is trained to "know her place" in society because ot the pervasive sex-role concept which results in differential expectations and socialization practices. Plank and Plank (195 4) were more specific. They discussed two hypotheses related to this view: the differential cultural reinforcement hypothesis and the masculine identification hypothesis. The differential cultural reinforcement hypothesis states that society in general perceives mathematics as a male domain* giving females less encouragement for excelling in it. The masculine identification hypothesis establishes that 80 81 82 83 84 85 86 87 88 89 90 91 93 94 95 96 98 101 102 104 73 TABLE 4.1 Continued x* for Model 3.2* Actual Sample Size Min. Sample Size for Significance lb) (c) (d) 75.01 55,277 6,787 95.51 57 *710 5,565 51.99 71/315 12/633 20.12 81/021 37/088 38.62 78/959 18,830 1240.78 41,180 306** 102.85 39/903 3/573 66.75 48,194 6,650 63.01 38/742 5,663 43.26 39,648 8,441 85.53 53/557 5/767 82.53 62,954 7,025 78.39 80/723 9,484 70.15 87,241 11,454 40.59 83,969 19,053 29.31 74,287 23/343 29.87 69/052 21,291 23.77 83,782 32,462 61.06 75/708 11,419 40.47 72,517 16/503 70 tests were significant at the .01 level are reported in Table 4.1. Also reported in this table are the actual sample sizes and the minimum sample sizes necessary for the likelihood ratio tests to be significant. Of the 61 items* 59 had minimum sample sizes greater than 3*000. Thus* although for the three years both males and females samples had inconsistent option choices on these 61 items* on 59 of the items the inconsistency was relatively minor. Consequently* these 59 items were included in step 2 along with the initial 50 items. In step 2* a likelihood ratio test was performed comparing the adequacy of the models in (3.2) and (3.3). The values associated with models (3.2) and (3.3) for each of the 109 items subjected to step 2 are reported in columns b and c of Table 4.2. Also reported in Table 4.2 is the difference between the two values (see column d). This latter figure is the test statistic for comparing the adequacies of models (3.2) and (3.3). Significant X^ statistics are indicated by asterisks. The X^ statistics were significant for 100 of the items indicating a male-female difference in option choice for these items. In column e of Table 4.2 the actual sample sizes are reported. In column f* the minimum sample sizes necessary for significance are reported. For those 100 items which had significant X^ statistics reported in column d* 94 (94%) had minimum sample sizes greater than 3*000. CHAPTER I INTRODUCTION Sex differences in mathematics learning and achievement have been the subject of intensive research. Research done before 1974 has shown that male performance on mathematical achievement tests is superior to female performance by the time they reach upper elementary or junior high school (Fennema# 1976# p. 2). The literature strongly suggests that at the elementary level females outperform males in computation and males excel in mathematical reasoning (Glennon & Callahan# 1968; Jarvis# 1964; Maccoby# 1966). Since 1974# research findings have been less consistent. Fennema (1974)# after reviewing 36 studies# found that during secondary school or earlier# sex-related differences in mathematics achievement are not so evident# but that when differences are found# they favor males in high level cognitive tasks (problem solving) and females in low level cognitive tasks (computation). As a result of a further review of the literature# Fennema (1977) concluded that at the elementary level# sex-related differences do not exist at all cognitive levels# from computation to problem solving. 1 COMPARISON OF MALES AND FEMALES ON MATH ITEM PERFORMANCE: ANALYSIS OF RESPONSE PATTERNS By SONIA FELICIANO A DISSERTATION PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY UNIVERSITY OF FLORIDA 1986 ACKNOWLEDGMENTS I would like to acknowledge my gratitude to several people who have influenced my formal education and/or made this study possible. My special thanks to Dr. James Alginar Chairman of the doctoral committee/ who contributed to the development of my love for research and statistics. He has been insuperable as professor and valued friend; his help and guidance in the preparation and completion of this study were invaluable. I extend my thanks to Dr. Linda Crocker for her advice and help during my doctoral studies at University of Florida. Thanks also go to Dr. Michael Nunnery/ member of the doctoral committee. To Dr. Wilson Guertin* who was a friend for me and my family/ I extend my special thanks. Thanks are also extended to Dr. Amalia Charneco/ past Undersecretary of Education of the Puerto Rico Department of Education/ for her continuous support. To my sister Nilda Santaella/ who typed the thesis/ I give my sincere thanks. Special thanks go to my family and to those friends who provided encouragement throughout this critical period of my life. ii TABLE OF CONTENTS Page ACKNOWLEDGMENTS Ãœ LIST OF TABLES V LIST OF FIGURES VÃœ ABSTRACT viii CHAPTER I INTRODUCTION 1 Purpose of the Study 6 Significance of the Study 6 Organization of the Study 8 II REVIEW OF THE LITERATURE 9 Sex-related Differences in Incorrect Response Patterns 10 Sex-related Differences in Problem Solving 14 Cognitive and Affective Variables that Influence Sex Differences in Mathematics Learning and Achievement 23 Differences in Formal Mathematics Education 24 Differences in Spatial Ability 26 Differentiated Effect of Affective Variables 30 Problem Solving Performance and Related Variables 37 Computational Skills and Problem Solving Performance 38 Reading and Problem Solving PerforÂ¬ mance 44 Attitudes Toward Problem Solving and Problem Solving Performance 50 iii IllMETHOD 54 The Sample 54 The Instrument 55 Analysis of the Data 57 Analysis of Sex by Option by Year Cross Classifications 57 Comparison of Males and Females in Problem Solving Performance 66 IV RESULTS 6 8 Introduction 68 Sex-related Differences in the Selection of Incorrect Responses 68 Sex-related Differences in Problem Solving Performance 81 Summary 93 V DISCUSSION 101 Summary and Interpretation of the Results 101 Implications of the Findings and Suggestions for Further Research 103 Sex-related Differences in Incorrect Responses 104 Sex-related Differences in Problem Solving Performance 105 REFERENCES 107 BIOGRAPHICAL SKETCH 116 IV LIST OF TABLES Table Page 3.1 Hypothetical Probabilities of Option Choice Conditional on Year and Sex 58 3.2 Hypothetical Probabilities of Option Choice Conditional on Year and Sex and Arranged by Sex 58 3.3 Hypothetical Joint Probabilities of Year/ Option/ and Sex 64 3.4 Hypothetical Probabilities of Option Choice Conditional on Sex and Year 66 4.1 Summary of Significant Tests of Model 3.2 71 4.2 Chi-square Values for the Comparison of 75 Models (3.2) and (3.3) with Actual Sample Sizes and with the Corresponding Number of Subjects Needed for Significant Results 75 4.3 Means/ Standard Deviations/ and t-Test for the Eight Mathematical Variables 82 4.4 ANCOVA - First Year: Multiplication Covariate 85 4.5 ANCOVA - First Year: Division Covariate 85 4.6 ANCOVA - Second Year: Subtraction Covariate... 86 4.7 Reliability of the Covariates for Each of the Three Years of Test AdministraÂ¬ tion 91 4.8 ANCOVA - First Year: Other Covariates 92 4.9 Adjusted Means on Problem Solving/ by Covariate and Sex 94 4.10 ANCOVA - Second Year: Other Covariates 95 v Table Page 4.11 Adjusted Means on Problem Solving* by Covariate and Sex 96 4.12 ANCOVA - Third Year 97 4.13 Adjusted Means on Problem Solving* by Covariate and Sex 98 vi LIST OF FIGURES Figure Page 3.1 Number of Questions by Skill Area in a Basic Skills Test in Mathematics-6 56 4.1 Sex by Multiplication Interaction 87 4.2 Sex by Division Interaction 88 4.3 Sex by Subtraction Interaction 89 vii Abstract of Dissertation Presented to the Graduate School of the University of Florida in Partial Fulfillment of the Requirements for the Degree of Doctor of Philosophy COMPARISON OF MALES AND FEMALES ON MATH ITEM PERFORMANCE: ANALYSIS OF RESPONSE PATTERNS By Sonia Feliciano August/ 1986 Chairman: James Algina Major Department: Foundations of Education The first objective of this study was to investigate sex differences in the selection of incorrect responses on a mathematics multiple-choice test, and to determine whether these differences were consistent over three consecutive administrations of the test. A second objective was to compare male and female performance in problem solving after controlling for computational skills. The responses of all 6th grade students from the public schools in Puerto Rico who took the Basic Skills Test in Mathematics-6 ("Prueba de Destrezas BÃ¡sicas en MatemÃ¡ticas-6") during three academic years were used in the analyses relevant to the first objective. VXll Log-linear models were used in the analysis of incorrect responses. The results of the analyses showed that for 100 of the 111 items of the testr males and females selected different incorrect options/ and this pattern of responses was consistently found during the three years of test administration. However/ for the vast majority of the 100 items the male-female differences were relatively small/ considering the fact that the number of subjects needed to obtain statistical significance was very large. The responses of approximately 1/000 randomly selected students per academic year were analyzed in the comparison of male and female performance in problem solving. Females outperformed males in problem solving and in six of the seven computational variables. Males showed superiority in equivalence in all the three years/ but statistical significance was obtained in only one of the years. Analysis of covariance (ANCOVA) was used in the comparison of male and female performance in problem solving after controlling for computational skills. Seven analyses of covariance tests were conducted/ one for each of the covariates. Estimated true scores for observed scores were used in the analyses. The results tend to show that for examinees with similar levels of computational IX skills/ sex-related differences in problem solving performance do not exist. Females retained their superiority in problem solving when equivalence (in all three years) and subtraction (in one year) were the controlling variables. The question of whether male-female differences in problem solving depend on computational skills was answered/ partially/ in the affirmative. x CHAPTER I INTRODUCTION Sex differences in mathematics learning and achievement have been the subject of intensive research. Research done before 1974 has shown that male performance on mathematical achievement tests is superior to female performance by the time they reach upper elementary or junior high school (Fennema/ 1976/ p. 2). The literature strongly suggests that at the elementary level females outperform males in computation and males excel in mathematical reasoning (Glennon & Callahan/ 1968; Jarvis/ 1964; Maccoby/ 1966). Since 1974/ research findings have been less consistent. Fennema (1974)/ after reviewing 36 studies/ found that during secondary school or earlier/ sex-related differences in mathematics achievement are not so evident/ but that when differences are found/ they favor males in high level cognitive tasks (problem solving) and females in low level cognitive tasks (computation) . As a result of a further review of the literature/ Fennema (1977) concluded that at the elementary level/ sex-related differences do not exist at all cognitive levels/ from computation to problem solving. 1 2 Many variables* cognitive* affective* and educational* have been investigated since 1974 in relation to sex differences in mathematics learning and achievement. Fennema and Sherman (1977) investigated the effect of differential formal mathematics education. After controlling for the number of years of exposure to the study of mathematics* they found sex differences in only two of the four schools under study. However* in those schools where boys scored higher than girls* differences were also found in their attitudes toward mathematics. Hilton and Berglund (1974) found significant sex differences after controlling for the number of mathematics courses taken* and attributed them to sex differences in interests. "As the boys' interests in science increase related to the girls'* their achievement in mathematics increases relative to that of the girls" (p. 234) . Wise* Steel* and McDonald (1979) reanalyzed test data collected in a longitudinal study of 400*000 high school students (Project Talent). They found that when the effect of the number of high school mathematics courses was not controlled* no sex differences emerged for 9th graders* but that gains made by boys during the next three years were more than twice that of the girls. These differences between the sexes disappeared when the number of mathematics courses taken was controlled. Results of the 1978 Women and Mathematics National Survey* Survey I* 3 indicated no significant sex differences for 8th grade students on measures of problem solving or algebra. However/ females outperformed males in computation and spatial visualization. For the 12th grade students/ statistically significant sex differences favoring males were found in problem solving/ but not in algebra/ computation/ or spatial visualization. For males and females who had enrolled in courses beyond general mathematics and who had taken or were enrolled in courses such as pre-calculus/ calculus/ or geometry/ differences in problem solving or spatial visualization did not exist. Sex differences favoring males were found on a total score obtained summing across the computation/ problem solving/ and algebra subtests (Armstrong/ 1979). The mathematics data collected in the second survey by the 1978 National Assessment of Educational Progress showed significant sex differences for both 13- and 17-year-old students. The 13-year-old females outperformed males in the computational subtest and males outscored females by 1 1/2 percentage points in problem solving (statistically significant) . No statistically significant differences were found in algebra. No sex differences were found for the 17-year-old group either in the computation subtest or in the algebra subtest. Males surpassed females in problem solving. A reanalysis of the data from the 17-year-old group confirmed male superiority in problem solving after controlling for 4 mathematics preparation. Males who were enrolled or had completed algebra II outperformed the females in computation and problem solving but not in algebra. Males who studied beyond algebra II outscored females on all three subtests: computation/ algebra/ and problem solving (Armstrong/ 1979) . Carpenter/ Lindquist/ Mathews/ and Silver (1984) analyzed the results of the Third National Assessment of Educational Progress (NAEP)/ and compared them with the First and Second Surveys. Between 1978 and 1982/ the differences between the average performance of males and females remained stable at each age level. At ages 9 and 13/ the overall performance of males and females was not significantly different. At age 17/ males scored higher than females by about 3 percentage points. When course background was held constant/ achievement differences still existed at age 17. For each category of course background/ male achievement exceeded female achievement. Consistent with previous assessments/ sex differences in problem solving in favor of males were found for the 17-year-old sample. At ages 9 and 13/ no large differences were found between the sexes within any level of course background. Marshall (1981/ 1984) investigated sex differences in mathematics performance. She found that males and females excel each other in solving different types of problems. 5 Females were better on items of computation and males were more successful on word-story problem items (problem solving). She also found that females successful performance in the problem solving items was more dependent on their successful performance in the computation items. Males did not need/ as much as females/ to succeed in the computation items in order to answer correctly the problem solving items. Although the general findings seem to support sex differences in mathematics learning and achievement/ the research done does not consistently support superiority for either sex. Most of the research has been concerned with how the sexes differ on subtests or total test scores in mathematics. Moreover/ the great majority of the studies deal with correct responses. Sex differences in incorrect responses at the item level have not been fully researched. Only two studies dealing with sex differences in incorrect responses at the item level were found in the research literature (Marshall/ 1981/ 1983). Marshall investigated whether boys and girls made similar errors in computation and story problems. She analyzed boys' and girls' answers to six mathematics items and found that the sexes made different errors/ possibly reflecting different problem solving strategies. Her original findings were supported when she studied the same problem using a large number of items three years later. 6 Purpose of the Study The purpose of this study was twofold. The first was to investigate sex-related differences in the selection of incorrect responses in a multiple-choice achievement test in mathematics. For each test item* the research questions were as follows: 1. Is there a difference in the proportion of males and females choosing each incorrect option? 2. Is the same pattern of differences found in data obtained in three different administrations of the test? The second objective was to investigate sex-related differences in test scores in mathematics problem solving. The following questions were studied: 1. Do males and females differ in problem solving perf ormance? 2. Do sex differences in problem solving persist after controlling for computational skillsf and does the differential success of males and females on problem solving items depend on their success on the computation items? Significance of the Study Item response patterns are very useful techniques in the assessment of mathematics learning and achievement. Total test scores can be very misleading in the assessment of student performance and provide no diagnostic information about the nature and seriousness of student 7 errors (Harnisch? 1983). Item response patterns are valuable for the identification of large group differences/ including d i st rict-to-district/ school-to-school/ and classroom-to-classroom variations on different subsets of items. The response patterns can provide diagnostic information about the type of understanding the student has on various mathematics topics (e.g. / problem solving). Marshall (1981/ 1984) has used the item response pattern technique and her findings indicate sex differences in mathematics performance at the item level. Females outperformed males in computation and males outscored females in problem solving. Also/ the success of girls in the problem solving items was dependent upon their success in the computation items; for boys/ success in the problem solving items did not depend as much on their computational performance. Marshall (1981/ 1983) has also reported that males and females differ in the selection of incorrect responses/ reflecting differences in reasoning abilities. In Puerto Rico/ a high percentage of children promoted to the 7th grade in the public schools does not master the basic skills in mathematics. If 6th grade male and female children can be diagnosed as having different problem solving abilities/ as Marshall found with California children/ teachers may need to provide tailor-made mathematics instruction for each sex/ in order to ensure equal access to formal education and enhance mathematics 8 achievement. Since there are no investigations reported in sex differences in item response patterns in Puerto Rico* research is needed. Organization of the â– Study A review of the literature on sex differences in mathematics performance is reported in Chapter II. The research methodology is presented in Chapter III. Research questionsÂ» sampleÂ» instrumentÂ» and data analysis are discussed in that chapter. Chapter IV is an exposition of the results of the study. Chapter V contains a summary and interpretation of the results of the study and the implications of the findings together with suggestions for further research. CHAPTER II REVIEW OF THE LITERATURE Sex differences in mathematics learning and achievement have been a subject of concern for educators and psychologists. Many studies found in the literature support the existence of these differences. Boys show superiority in higher level cognitive tasks (problem solving or mathematical reasoning) in the upper elementary years and in the early high school years (Fennema/ 1974; Maccoby & Jacklin/ 1974) . Almost all the research carried out has dealt with analysis of total correct scores in mathematics aptitude and achievement tests or scores in subtests. The literature related to sex differences in incorrect responses/ the main subject of the present study/ is surprisingly sparse. For the most part/ the studies have investigated the differences between the sexes in mathematics learning and achievement and the underlying variables causing the differences. Cognitive and affective variables have been the matter at issue in the establishment of sex differences. Although research in mathematics problem solving/ the secondary subject of this investigation/ is extensive/ most 9 10 of the studies consider sex differences incidental to the major study findings. The available literature offers very little research directly related to the problem of sex differences in this area. The review of the literature has been divided in four sections. The first section consists of a detailed summary of the available research on sex diferences in incorrect responses. The second section deals with sex-related differences in problem solving performance. These sections are directly related to the objectives of the study. The third section is more peripheralÂ» and contains a discussion of the more prevalent issues about the influence of cognitive and affective variables on sex differences in mathematics learning and achievement. The fourth is a summary of the research dealing with variables considered as influential to mathematics problem solving performance. Sex-related Differences in Incorrect Response Patterns Research findings tend to suggest that boys and girls may be approaching problem solving differently (e.g.Â» Fennema and ShermanÂ» 1978; MarshallÂ» 1981Â» 1983; MeyerÂ» 1978Â» among others). Marshall (1981) investigated whether 6th grade boys and girls approach mathematical problem solving with different strategies. Her specific interest was whether the sexes made the same errors. 11 She analyzed the responses of 9/000 boys and 9/000 girls to 6 selected items/ 2 computation items/ and 1 story problem item from each of 2 of the 16 forms of the Survey of Basic Skills test administered during the academic year 1978-79. The Survey is a 30-item achievement test administered every year to all 6th grade children in California through the California Assessment Program. There are 16 forms of the test/ to which approximately 9 /000 boys and 9 /000 girls respond each year. Of the 160 mathematics items contained in the 16 forms of the test/ 32 are on measurement and graphing/ 28 on number concepts/ 28 on whole number arithmetics/ 20 on fraction arithmetics/ 20 on decimal arithmetics/ 20 on geometry/ and 12 on probability and statistics. The item analysis performed on the 1978-79 data showed that boys and girls tended to select different incorrect responses. In the first computation item (Form 1 of the test) both sexes reflected similar mistakes in carrying/ but in different columns. In the second computation item/ both sexes ignored the decimal points and selected the same incorrect response. However/ more girls than boys chose this response. In the first computation item (Form 2 of the test) the incorrect choice of both sexes was option Â£/ but the second most frequently selected option was for boys and b for girls. In the second computation item of this 12 test form/ no sex differences were found in response patterns. Approximately 45% of each sex selected option c. The next popular choice for both sexes was option df selected by approximately 35% of both boys and girls. On the story problem of Form If males and females responded alike. Their most popular incorrect response choice was option a for both males and females. The second most popular incorrect choice was option c for both sexes. Response to the story problem in Form 2 showed sex differences in response choice. Including the correct option/ 33% of the girls selected option a./ 20% chose option C/ and 20% option d. For males/ approximately 25% selected option a and the same percent selected option d. Marshall concluded that although the analysis of incorrect responses does not explain why boys and girls differ in their responses/ the analysis shows that boys and girls approach problems in different ways and these varying strategies can be useful in indentifying how the sexes differ in reasoning abilities. Two years later/ Marshall (1983) analyzed the responses of approximately 300/000 boys and girls to mathematics items contained in the 16 test forms of the Survey of Basic Skills during the years 1977/ 1978/ and 1979. She used log-linear models (explained in Chapter 13 III) to investigate sex related differences in the selection of incorrect responses/ and the consistency of such differences over three years of administration of the test. Based on her findings that sex differences were found in 80% of the items/ Marshall classified the students' errors according to Radatz' (1979) five-category error classification. The categories are language (errors in semantics)/ spatial visualization/ mastery/ association/ and use of irrelevant rules. It was found that girls' errors are more likely to be due to the misuse of spatial information/ the use of irrelevant rules/ or the choice of an incorrect operation. Girls also make relatively more errors of negative transfer and keyword association. Boys seem more likely than girls to make errors of perseverance and formula interference. Both sexes make language-related errors/ but the errors are not the same. Available research is not extensive enough to make definite judgments about the sex-related differences observed in incorrect responses. Clearly more research is needed. 14 Sex-related Differences in Problem Solving It has already been acknowledged that the subject of problem solving has been extensively researched. However* as early as 1969* Kilpatrick criticized the fact that the study of problem solving has not been systematic; some researchers have studied the characteristics of the problem while others have given their attention to the characteristics of the problem solvers. Moreover* differences in the tests used to measure problem solving performance also constitute an obstacle when trying to compare the results of the studies carried out. In order to avoid this pitfall and provide a basis for comparison* the studies reviewed in this section* dealing with sex-related differences in problem solving* have been divided in two groups. The first comprises those studies that used the Romberg-Wearne Problem Solving Test. The second contains other relevant studies in which problem solving performance has been measured by means of other instruments. The Romberg-Wearne Problem Solving Test merits special mention because it was the first attempt "to develop a test to overcome the inadequacies of total test scores in explaining the reasons why some students are successful problem solvers and others are not" (Whitaker* 1976* pp. 9* 10) . 15 The test is composed of 23 items designed to yield 3 scores: a comprehension score/ an application score/ and a problem solving score. The comprehension question ascertains whether a child understands the information given in the item stem. The application question assesses the child's mastery of a prerequisite concept or skill of the problem solving question. The problem solving question poses a question whose solution is not immediately available/ that is/ a situation which does not lend itself to the immediate application of a rule or algorithm. The application and problem solving parts of the test may refer to a common unit of information (the item stem) but the questions are independent in that the response to the application question is not used to respond to the problem solving question. Meyer (1978)/ Whitaker (1976)/ Fennema and Sherman (1978)/ and Sherman (1979) have used the Romberg-Wearne test in their studies. Meyer (1978) investigated whether males and females differ in problem solving performance and examined their prerequisite computational skills and mathematical concepts for the problem solving questions. A sample of 179 students from the 4th grade were administered 19 "reference tests" for intellectual abilities and the Romberg-Wearne test. The analysis showed that males and females were not significantly different in the comprehension/ application/ and problem solving questions 16 of the test. The sexes differed in only 2 of the 19 reference testsÂ» Spatial Relations and Picture Group Name-Selection. A factor analysisÂ» howeverÂ» showed differences in the number and composition of the factors. For females/ a general mathematics factor was determined by mathematics computation/ comprehension/ application/ and problem solving. For males/ the comprehension and application parts determined one factor; problem solving with two other reference tests (Gestalt and Omelet) determined another factor. Meyer concluded that comprehension of the data and mastery of the prerequisite mathematical concepts did not guarantee successful problem solving either for males or for females. Problem solving scores for both sexes were about one third their scores in comprehension and one fourth their scores in application. She also concluded that the sexes may have approached the problem solving questions differently. The methods used by females for solving problem situations may have paralleled their approach to the application parts. Males may have used established rules and algorithms for the application parts/ but may have used more of a Gestalt approach to the problem solving situation. Whitaker (1976) investigated the relationship between the mathematical problem performance of 4th grade children and their attitudes toward problem solving/ their teachers' 17 attitude toward mathematical problem solving/ and related sex and program-type differences. Although his main objective was to construct an attitude scale to measure attitudes toward problem solving/ his study is important because his findings support Meyer's regarding the lack of significant sex-related differences in problem solving performance. Performance in the problem solving questions/ for both males and females/ was much lower than performance in the application questions/ and much lower than performance in the comprehension questions. In fact/ the mean score for each part of the item/ for both males and females/ was almost identical to the mean scores obtained by males and females in Meyer's study. Whitaker noted that each application item is more difficult than its preceding comprehension item/ and that each problem solving item is more difficult than its preceding application item. No significant sex-related differences were found for any of the three parts of the item (comprehension/ application/ or problem solving). Fennema and Sherman (1978) investigated sex-related differences in mathematics achievement and cognitive and affective variables related to such differences. They administered the Romberg-Wearne Problem Solving Test to a representative sample of 1320 students (grades 6-8) from Madison/ Wisconsin/ predominantly middle-class/ but including great diversity in SES. The sample consisted of 18 students who had taken a similar number of mathematics courses and were in the top 85% of the class in mathematics achievement. They were tested in 1976. Four high school districts were included. In only one of the high school districts were sex-related differences in application and problem solving found# in favor of males. They concluded that when relevant factors are controlled/ sex-related differences in favor of males do not appear often/ and when they do/ they are not large. Sherman (1980) investigated the causes of the emerging sex-related differences in mathematics performance/ in favor of males/ during adolescence (grades 8-11). She wanted to know if these differences emerge as a function of sex-related differences in spatial visualization and sociocultural influences that consider math as a male domain. In grade 8/ she used the Romberg-Wearne Test and/ in grade 11/ a mathematical problem solving test derived from the French Kit of Tests. The analysis showed that for girls/ problem solving performance remained stable across the years. Mean problem solving performance for boys/ however/ was higher in grade 11 than in grade 8. No sex-related differences were found in grade 8/ but boys outperformed girls in grade 11/ where the Stafford test was used. Sherman found that for both sexes problem solving performance in grade 8 was the best predictor of problem 19 solving performance in grade 11. Spatial visualization was a stronger predictor for girls than for boys. Mathematics as a male domain was a good predictor for girls only; the less a girl stereotyped mathematics as a male domain in grade 8/ the higher her problem solving score in grade 11. Attitudes toward success in mathematics in grade 8 was a more positive predictor of problem solving performance for boys than for girls; the more positive the attitudes toward success in mathematics in grade 8/ the higher their performance in problem solving in grade 11. None of these four studiesÂ» all of which used the Romberg-Wearne Mathematics Problem Solving TestÂ» show statistically significant sex-related differences in problem solving performance. In later studies other tests were used to measure this variable (KaufmanÂ» 1984; MarshallÂ» 1981Â» 1984) . Kaufman (1983/1984) investigated if sex differences in problem solvingÂ» favoring malesÂ» exist in the 5th and 6th grades and if these differences were more pronounced in mathematically gifted students than in students of average mathematical ability. The Iowa Test of Basic Skills and a mathematics problem-solving test were administered to 504 subjects. Males in the average group as well as males in the gifted group outperformed femalesÂ» but only the gifted group showed statistically significant differences. 20 As a result of her investigations/ Marshall (1981) concluded that sex-related differences in mathematics performance may be the result of comparing the sexes on total test scores. If the test contains more computation items than problem solving items/ girls will perform better than boys/ but if the test contains more problem solving items than computation ones/ boys will outperform girls. With this in mind/ Marshall investigated sex-related differences in computation and problem solving by analyzing the responses of approximately 18/000 students from grade 6 who had been administered the Survey of Basic Skills Test: Grade 6/ during the academic year 1978-79. Two of the 16 test forms of the Survey were used to assess skills such as concepts of whole numbers/ fractions/ and decimals. These skills were tested both as simple computations and as story problems (problem solving). Two computation items and one story problem item were selected because they were particularly related; both computation items required skills needed in solving the corresponding story problem. It was assumed that correct solution of the computation item correlates with solving the story problem because the story problem requires a similar computation. Marshall found that girls were better in computation and boys were better in problem solving. She also found that boys were much more likely than girls to answer the 21 story problem item correctly after giving incorrect responses to both computation items. Apparently* mastery of the skills required by the computation items is more important for girls than for boys. If girls cannot solve the computation items* they have little chance of solving the related story problem item. For girls* the probability of success in the story problem item after giving successful answers to both computation items is almost 2 1/2 times the probability of success after giving incorrect responses to both computation items. For boys* the probability of success in the story problem item after successful responses to the computation items is about 1 3/4 times the probability of success on the story problem item after incorrect responses to the computation items. Three years later* Marshall (1984) analyzed more in depth these phenomena of sex-related differences. Her interest was twofold. First* she wanted to know if there were differences in the rate of success for boys and girls in solving computation and story problem items. Second* she examined additional factors that interact with sex to influence mathematics performance* such as reading achievement* socio-economic status (SES) * primary language* and chronological age. Two questions were raised: Do the probabilibities of successful solving of computation and story problem items increase with reading score? Are these probabilities different for the two sexes? 22 Approximately 270/000 students from the 6th grade were administered the Survey of Basic Skills of the California Assessment Program/ during the years 1977/ 1978/ and 1979. Responses were analyzed using log-linear models. Successful solving of computation items was positively associated with successful solving of story problems. Girls were more successful in computation than boys/ and boys were more successful than girls in solving story problems. This finding supports reports from the National Assessment of Educational Progress (NAEP) (Armstrong/ 1979) . To investigate the effects of reading/ SES/ language/ and chronological age/ only those test forms containing 2 computation items and 2 story problems were considered for analysis; 32 items from 8 test forms were included in the analysis. The results of these analyses showed that at every level of reading score/ 6th grade children were more successful in computation than in story problems. Although the differences were not large/ at every reading score boys consistently had higher probabilities of success in story problems than did girls/ and girls consistently showed higher probabilities of success in computation than boys. Also/ as the reading score increased/ the difference between the probability of success in story problems and the probability of success in 23 computation grew larger. This difference grew larger for girls than for boys. Although SES was a major factor in solving computation and story problem items successfully/ the effect was similar for each sex. Sex-related differences by primary language or chronological age were not large. This research carried out by Marshall with elementary grade children supports previous research findings that males are better than females in mathematics problem solving (a higher order skill) and females are better than their counterpart males in computation (a lower level skill). Marshall's research also brought out a different aspect of this question: the notion that girls find it more necessary than boys to succeed in the computation items in order to successfully solve the story problem items. Cognitive and Affective Factors That Influence_Sex Differences in Mathematics Learning and Achievement The research reviewed in the literature does not provide evidence of any unique variable that could serve as an explanation for the observed sex-related differences in mathematics learning and achievement. However/ some issues have been discussed/ among which the most prevalent are that sex-related differences in mathematics learning and achievement are a result of differences in formal education; that sex-related differences in mathematics 24 learning and achievement arise from sex differences in spatial visualization; and that sex-related differences result from a differentiated effect of affective variables on the mathematics performance of males and females. Differences in Formal Mathematics Education (Differential Coursework Hypothesis) The basis for the differential coursework hypothesis is the fact that sex-related differences in mathematical learning and achievement show up when comparing groups which are not equal in previous mathematics learning. After the 8th grade/ boys tend to select mathematics courses more otten than girls. Therefore/ girls show lower achievement scores in mathematics tests because their mathematics experience is not as strong as the boys' (Fennema/ 1975; Fennema & Sherman/ 1977; Sherman/ 1979). Fennema and Sherman's study (1977) lends additional support to the feasibility of viewing sex differences in mathematics learning and achievement as reflecting something other than a difference in mathematics aptitude. After controlling for previous study of mathematics/ they found significant sex differences in mathematics achievement in only two of the four schools under study/ making the attribution to sex per se less likely. Controlling for the number of space visualization-related courses/ the sex-related differences which originally 25 emerged in spatial visualization scores became non-significant. In the two schools where sex differences in mathematics achievement were found/ differences between the sexes were also found in their attitudes toward mathematics. Researchers like Backman (1972) / who analyzed data from Project Talent/ and Allen and Chambers (1977) have also hypothesized that sex-related differences in mathematics achievement may be related to different curricula followed by males and females. Allen and Chambers attributed male superiority in mathematics problem solving to differences in the number of mathematics courses taken in high school. This issue has been seriously questioned by Astin (1974)/ Fox (1975a/ 1975b)/ and Benbow and Stanley (1980)/ among others. Astin and Fox have reported large differences in favor of males among gifted students taking the Scholastic Achievement Test. These differences occur as early as grade 7/ when there are no sex differences in the number of courses taken. Benbow and Stanley (1980) compared mathematically precocious boys and girls in the 7th grade/ with similar mathematics background/ and found sizeable sex-related differences favoring boys in mathematical reasoning ability. Five years later/ they conducted a follow-up study which showed that boys 26 maintained their superiority in mathematics ability during high school. While Fox attributed sex-related differences in mathematical achievement to differential exposure to mathematical games and activities outside school/ Benbow and Stanley suggested that sex-related differences in mathematics performance stem from superior mathematical ability in malesÂ» not from differences in mathematics formal education. The differential coursework hypothesis is not totally convincing and/ as reporteu berore/ it has been challenged by researchers such as Benbow and Stanley (1980). However/ Pallas and Alexander (1983) have questioneu the generalizaoility of Benbow and Stanley's findings based on the fact that they used highly precocious learners. The differential coursework hypothesis can be accepted only as a partial explanation of differences in mathematics performance found between the sexes. Differenees in Spatial Ability The basic premise in this issue is that males and females differ in spatial visualization and this explains differential mathematics learning and achievement. Until recently/ sex differences in spatial ability in favor of males were believed to be a fact and were thought by some to be related to sex differences in mathematical achievement. 27 Research findings in this area have been inconsistent. In 1966 / Maccoby stated that "by early school years/ boys consistently do better (than girls) on spatial tasks and this difference continues through the high school and college years" (p.26). In 1972/ Maccoby and Jacklin said that the differences in spatial ability between the sexes "remain minimal and inconsistent until approximately the ages of 10 or 11/ when the superiority of boys becomes consistent in a wide range of populations and tests" (p- 41) - In 1974 / after a comprehensive literature search/ Maccoby and Jacklin concluded that sex differences in spatial visualization become more pronounced between upper elementary years and the last year of high school/ the years when sex-related differences in mathematics achievement favoring boys emerge. Guay and McDaniel (1977) supported in part Maccoby and Jacklin's 1974 findings. They found that among elementary school children/ males had greater high level spatial ability than females/ but that males and females were equal in low level spatial ability. This finding is inconsistent with that portion of Maccoby and Jacklin's review that suggests that sex differences become evident only during early adolescence. Cohen and Wilkie (1979) however/ stated that in tests measuring distinct spatial tasks/ males perform better than females in early adolescence and throughout their life span. Most studies carried out after 28 1974 have failed to support these sex differences in spatial abilities (Armstrong/ 1979; Connor/ Serbin, & Schackmanr 1977; Fennema & Sherman/ 1978; Sherman/ 1979). Fennema and Sherman (1978) and Sherman (1979) have explored sex-related differences in mathematical achievement and cognitive and affective variables related to these differences. In a study involving students from grades 6/ 7/ and 8/ from four school districts/ Fennema and Sherman found that spatial visualization and problem solving were highly correlated for both sexes (.59 and .60). Even in the school district where sex differences were found in problem solving/ no significant sex-related differences were found in spatial visualization. When Sherman (1980) compared groups of males and females in two different grades/ 8 and 11/ she found no sex-related differences in problem solving or in spatial visualization in grade 8. In grade 11/ however/ although the sexes differed in their problem solving performance/ no sex-related differences were found in spatial vizualization. Even though spatial visualization in grade 8 was the second best predictor of problem solving performance in grade 11/ sex differences in grade 11 were not a result of spatial visualization since no differences were found in that skill. In spite of the fact that no sex differences were found in spatial abilities/ it is evident that males and 29 females may use them in a different way. Meyer (1978) Â» with an elementary grade sampleÂ» and Fennema and Tartree (1983) Â» with an intermediate level sampleÂ» found that the influence of spatial visualization on solving mathematics problems is subtleÂ» and that males and females use their spatial skills differently in solving word story problems (problems that measure problem solving ability or reasoning). Fennema and Tartree (1983) carried out a three-year longitudinal study which showed that girls and boys with equivalent spatial visualization skills did not solve the same number of itemsÂ» nor did they use the same processes in solving problems. The results also suggested that a low level of spatial visualization skills was a more debilitating factor for girls than for boys in problem solving performance. Landau (1984) also investigated the relationship between spatial visualization and mathematics achievement. She studied the performance of middle school children in mathematical problems of varying difficultyÂ» and the extent to which a diagramatic representation is likely to facilitate solution. She found that spatial ability was strongly correlated to mathematical problem solving and that the effect of spatial ability was more influential for females. Females made more use of diagrams in the solution of problemsÂ» reducing the advantage of males over females in problem solving performance. 30 The issue of sex-related differences in spatial visualization ability as an explanation for sex differences observed in mathematics achievement is less convincing and the findings more contradictory than in the issue of sex differences in formal education. Besides these cognitive issues/ other issues/ mostly affective in nature/ have also been studied in trying to explain the origin of these sex differences in mathematics achievement and learning. The studies dealing with these affective variables are reviewed in the next section. Differentiated Effect of Affective Variables Researchers have attempted to explain the effect of sex differences in internal beliefs/ interests/ and attitudes (affective variables) on mathematics learning and achievement. A brief statement of each explanation precedes the summary of studies conducted that support the explanation. ;. Females/ more than males/ lack confidence in their ability to learn mathematics and this affects their achievement in mathematics and their election of more advanced mathematics courses. Maccoby and Jacklin (1974) reported that self- confidence in terms of grade expectancy and success in particular tasks was found to be consistently lower in women than in men. In 1978/ Fennema and Sherman reported 31 that in their study involving students from grades 6 through 12/ boys showed a higher level of confidence in mathematics at each grade level. These differences between the sexes occurred in most instances even when no sex-related differences in mathematics achievement were found. The correlation between confidence in mathematics performance and mathematics achievement in this study was higher than for any other affective variable investigated. Sherman reported a similar finding in 1980; in males/ the most important factor related to continuation in theoretical mathematics courses was confidence in learning mathematics. This variable weighed more than any of the cognitive variables: mathematics achievement/ spatial visualization/ general ability/ and verbal skill. In the case of females/ among the affective variables/ confidence in learning mathematics was found to be second in importance to perceived usefulness of mathematics. Probert (1983) supported these findings with college students. A variable that needs discussion within the context of sex differences in confidence as learners of mathematics is causal attribution. Causal attribution models attempt to classify those factors to which one attributes success or failure. The model proposed by Weiner (1974) categorizes four dimensions of attribution of success and failure: stable and internal/ unstable and internal/ stable and external/ and unstable and external. For example/ if one 32 attributes success to an internal, stable attribute, such as ability, then one is confident of being successful in the future and will continue to strive in that area. If one attributes success to an external factor such as a teacher, or to an unstable one, such as effort, then one will not be as confident of success in the future and will cease to strive. Failure attribution patterns work this way: if failure is attributed to unstable causes, such as effort, failure can be avoided in the future and the tendency will be to persist in the task. However, if failure is attributed to a stable cause, such as ability, the belief that one cannot avoid failure will remain. Studies reported by Bar-Tal and Frieze (1977) suggest that males and females tend to exhibit different attributional patterns of success and failure. Males tend to attribute their success to internal causes and their failures to external or unstable ones. Females show a different pattern; they tend to attribute success to external or unstable causes and failures to internal ones. The pattern of attributions, success attributed externally and failure attributed internally, has become hypothesized to show a strong effect on mathematics achievement in females. Kloosterman's (1985) study supported these findings. According to Kloosterman, attributional variables appear to be more important achievement mediators for females than for males, as measured by mathematics word problems. More research is needed in this area. 33 Mathematics as a male domain. Mathematics is an activity more closely related to the male sex domain than to the female sex domain (Eccles et al.r 1983). Thus* the mathematical achievement ot boys is higher than that of gi rls. According to John Ernest (1976) * in his study Mathematics and Sex> mathematics is a sexist discipline. He attributed sex-related differences in mathematical achievement to the creation by society of sexual stereotypes and attitudes* restrictions* and constraints that promote the idea of the superiority of boys in mathematics. Ernest reported that boys* girls* and teachers* all believe that boys are superior in mathematics* at least by the time students reach adolescence. Bern and Bern (1970) agree and argue that an American woman is trained to "know her place" in society because ot the pervasive sex-role concept which results in differential expectations and socialization practices. Plank and Plank (195 4) were more specific. They discussed two hypotheses related to this view: the differential cultural reinforcement hypothesis and the masculine identification hypothesis. The differential cultural reinforcement hypothesis states that society in general perceives mathematics as a male domain* giving females less encouragement for excelling in it. The masculine identification hypothesis establishes that 34 achievement and interest in mathematics result from identification with the masculine role. A study related to the differential cultural reinforcement hypothesis is that of Dwyer (1974). Dwyer examined the relationship between sex role standards (the extent to which an individual considers certain activities appropriate to males or females) and achievement in reading and arithmetic. Students from grades 2Â» 4* 6Â» 8Â» 10Â» and 12 participated in this study. She found that sex role standards contributed significant variance to reading and arithmetic achievement test scores and that the effect was stronger for males than for females. This led to her conclusion that sex-related differences in reading and arithmetic are more a function of the child's perception of these areas as sex-appropriate or sex-inappropriate than of the child's biological sexÂ» individual preference for masculine and feminine sex rolesÂ» or liking or disliking reading or mathematics. In a study which agrees with the masculine identification hypothesisÂ» Milton (1957) found that individuals who had received strong masculine orientation performed better in problem solving than individuals who received less masculine orientation. Elton and Rose (1967) found that women with high mathematical aptitude and average verbal aptitude scored higher on the masculinity scale of the Omnibus Personality Inventory (OPI) than those with average scores on both tasks. 35 It is not until adolescence that sex differences in the perception of mathematics as a male domain are found (Fennemar 1976; Stein/ 1971; Stein & Smithless/ 1969; Verbeke/ 1983). In a study with 2nd/ 6th and 12th graders/ Stein and Smithless (1969) found that students' perceptions of spatial/ mechanical/ and arithmetic skills as masculine became more defined as these students got older. Fennema (1976) considers that the influence each sex exerts upon the other on all aspects of behavior is stronger during adolescence. Since during these years males stereotype mathematics as a male domain/ they send this message to females who/ in turn/ tend to be influenced in their willingness to study or not to study mathematics. Before that stage/ girls consider arithmetic feminine/ while boys consider it appropriate for both sexes (Bobbe/ 1971). Usefulness of mathematics. Females perceive mathemaÂ¬ tics as less useful to them than males do/ and this perception occurs at a very young age. As a result/ females exert less effort than males to learn or elect to take advanced mathematics courses. Many studies reported before 1976 found that the perception of the usefulness of mathematics for one's future differs for males and females/ and is related to course taking plans and behavior (Fox/ 1977). If females do not perceive mathematics as useful for their future/ 36 they show less interest in the subject than counterpart males. These differences in interest are what Hilton and Berglund (1974) suggest to account for sex-related differences in mathematics achievement. Although the perception of the usefulness of mathematics is still an important predictor of course taking for girls* there is a growing similarity between males and females regarding the usefulness of mathematics (Armstrong & Price* 1982; Fennema & Sherman* 1977; Moller* 1982/1983). Armstrong and Price investigated the relative influence of selected factors in sex-related differences in mathematics participation. Both males and females selected usefulness of mathematics as the most important factor in deciaing whether or not to take more mathematics in high school. Moller's study revealed that both males and females based mathematics course-taking decisions on career usefulness. A Fennema and Sherman (1977) study showed only slight differences between males and females in their feelings about the usefulness of mathematics. In her study of this variable among college students* Probert (1983/1984) did not find any sex-related differences either. These have been the main affective variables researched in attempting to explain the underlying causes of sex-related differences observed in mathematics learning and achievement. In spite of the great diversity of studies 37 dealing with both cognitive and affective variables/ there are no clear-cut findings to render unequivocal support to a particular variable as accounting for these sex-related differences. However/ everything seems to point to the fact that affective/ rather than cognitive variables play a more significant role in the sex-related differences observed in mathematics performance and learning. In most of the studies dealing with affective variables/ findings consistently show that these factors influence mathematics performance in females more than in males. In at least one area/ confidence as learners of mathematics/ Sherman (1980) found that this variable influenced course election more than all the cognitive variables previously discussed. The case for the societal influences on sex roles and expectations to account for the differences in mathematics learning is also supported in one way or another in the studies reported in the literature. Problem Solving Performance and Related Variables Problem solving has been perhaps the most extensively researched area in mathematics education. Published reviews by Kilpatrick (1969)/ Riedesel (1969)/ and Suydam and Weaver (1970-1975) attest to this. Much of the research done has focused on identifying the determinants of problem difficulty and the problem features that influence the solution process. 38 At presentÂ» no set of variables has been clearly established as a determinant of problem difficulty. Several researchers have investigated the effect of reading and computation on problem solving performance. Others have studied the effect of student attitudes toward problem solving in problem solving learning and achievement. Typically/ correlational methods have been used to investigate these questions. Computational Skills and Problem Solving Performance One of the first researchers to study the effect of computation and reading on problem solving performance was Hansen (1944). He investigated the relationship of arithmetical factors/ mental factors/ and reading factors to achievement in problem solving. Sixth grade students were administered tests in problem solving and categorized as superior achievers (best problem solvers) and inferior archievers (poorest problem solvers). The two groups were compared in selected factors believed to be related to success in arithmetic problem solving: arithmetical/ mental and reading factors. After controlling for mental and chronological age/ the superior achievers in problem solving surpassed the inferior achievers in mental and arithmetical factors. The superior group did better in only two of the six items under the reading factors: general language ability and the reading of graphs/ charts/ and tables. 39 The findings suggest that reading factors are not as important as arithmetic and mental factors in problem solving performance. HoweverÂ» these findings should be taken cautiously* as the content of the Gates tests (used to measure reading) is literary and does not include mathematical material. Chase (1960) studied 15 variables in an effort to find out which ones have significant influence on the aDility to solve verbal mathematics problems. Only computation* reading to note details* and fundamental knowledge were primarily related to problem solving. Computation accounted for 20.4% of the 32% variance directly associated with problem solving. Chase concluded that a pupil's ability in the mechanics of computation* comprehension of the principles that underline the number systems* and the extent to which important items of information are noticed when reading* are good predictors of the student's ability in solving verbal problems. Balow (1964) investigated the importance of reading ability and computation ability in problem solving performance. He objected to the approaches used by other researchers who in their analyses dichotomized research subjects as "poor" or "good" students* and who ignored the recognizea effect of intelligence on reading and on mathematics achievement Balow administered the Stanford 40 Achievement Test (subtests of reading/ arithmetic/ and reasoning) and the California Short-Form test of mental ability to a group of 1/400 children from the 6th grade. All levels of achievement were included in the analysis. Analysis of variance and covariance were used and compared. He confirmed the findings of other researchers to the effect that there is a direct relationship between I.Q. and reading ability/ and between I.Q. and computational skills. The results of the analysis of variance revealed that increases in computation ability were associated with higher achievement in problem solving. A relationship between reading ability and problem solving was also found/ but it was not as strong. Significant differences in problem solving performance associated with computational ability were found when intelligence was controlled. Balow concluded that computation is a much more important factor in problem solving than reading ability/ and that when I.Q. is taken into consideration/ the degree of the relationship between reading and problem solving ability becomes less pronounced. Intelligence tends to confound the relationship between these two variables. Knifong and Holtan (1976/ 1977) attempted to investigate the types of difficulties children have in solving word problems. They administered the word problem section of the Metropolitan Achievement Test to 35 children from the 6th grade. Errors were classified in two 41 categories. Category I included clerical and computational errors. Category II included other types of errorsÂ» such as average and area errors/ use of wrong operation/ no response/ and erred responses offering no clues. It the student's work indicated the correct procedure and yet the problem was missed because of a computational or clerical error/ it was assumed that the problem was read and understood. An analysis of frequencies showed that clerical errors were responsible for 3% of the problems incorrectly solved/ computational errors accounted for 49%/ and other errors for 48% of the erred problems. Knifong and Holtan concluded that "improved computational skills could have eliminated nearly half ot the word problem errors" (p. Ill). These computational errors were made in a context where other skills such as reading/ interpretation of the problem/ and integration of these skills necessary for the solution of word problems/ might interact. However/ Knifong and Holtan state that their findings neither confirm nor deny that improvement ot reading skills will lead to improvement in problem solving. They conclude that "it is difficult to attribute major importance to reading as a source of failure" (p. 111). In a later analysis/ looking for evidence of poor reading abilities affecting children's success in word problems/ Knifong and Holtan (1977) interviewed the 42 children whose errors fell under the category or "other errors." Students were asked to read each problem aloud and answer these questions: What kind of situation does the problem describe? What does the problem ask you to find? How would you work the problem? Ninety five percent of the students read the problem correctly; 98% explained the kind of situation the problem described in a correct manner; 92% correctly answered what the problem was asking them to find* and 36% correctly answered the question of how to work the problem. The fact that a large percent of the students whose errors were classified as "other errors" (in which reading skills might have been a factor) correctly stated how to work the problem* is strong evidence of their ability to read and interpret the problems correctly. The errors made by this group of students had a distinct origin* unrelated to reading ability. Zalewski (1974) investigated the relative contribution of verbal intelligence* reading comprehension* vocabulary* interpretation of graphs and tables* mathematical concepts* number sentence selection* and computation to successful mathematical word problem solution* and the relationship of the dependent variable to the eight independent variables. She worked with a group of 4th grade children who were administered the subtests of the Iowa Test of Basic Skills (ITBS) and the Weschler Intelligence Scale for 43 Children (WISC). Multiple regression analysis was performed. A correlation of .769 was found between word problem solving and the eight independent variables. Correlations between word problem solving and the independent variables ranged from .363 (verbal intelligence) to .674 (mathematical concepts). Correlations between the independent variables ranged from .369 (verbal intelligence and computation) to .749 (reading comprehension and vocabulary). Mathematical concepts/ computation/ and number sentence selection were almost as effective as all eight independent variables in predicting achievement in mathematical word problem solving. Mathematical concepts/ computation/ number sentence selection/ and reading comprehension accounted for 58% of the variance/ whereas all eight predictors accounted for 59% of the variance. The two best predictors were mathematical concepts and computation/ which accounted for 54% variance. Other variables accounted for about 40% of the variance. The author recommends that the findings of this study be interpreted cautiously because the correlation between the eight independent variables was high/ and/ according to Zal ewski / in a study of this nature where the interest is primarily in the influence of several variables on one dependent variable/ a low correlation between the independent variables is required. (p. 2804) 44 In a more recent investigation? Exedisis (1983) studied the contribution of reading ability, vocabulary, mathematical concepts, computation, sex, and race on problem-solving performance. The Iowa Test of Basic Skills was administered to a group of 6th, 7th, and 8th grade anglo and black Chicago male and female adolescents. Problem solving was highly correlated to an understanding of basic mathematical concepts, somewhat correlated to race, and weakly correlated to computational and vocabulary skills, sex, and reading ability. Although the findings of these studies show a relationship between computational skills and problem solving achievement, this relationship is not strong enough to be considered the most determinant factor in problem solving achievement, as some of the researchers have been careful to point out. In spite of the dismissal of reading as a determinant factor in problem solving achievement by some of these same researchers, more recent studies in this area have led others to hold different views. Headin_g_and_J)r obiem_ Solving Performance Martin (1964) studied the contribution of reading comprehension, computation, abstract verbal reasoning, and arithmetic concepts to arithmetic problem solving performance. Fourth and 8th grade students were administered the Iowa tests of Basic Skills and the Lorge-Thorndlike intelligence test (verbal). 45 He found that in the 4th grade the correlations between problem solving and abstract verbal reasoningÂ» reading comprehension/ arithmetic concepts/ and computation were .61/ .64/ .66/ and .60 respectively/ and .56/ .68/ .69/ and .63 in the 8th grade. When computation was held constant/ the correlation between problem solving and reading was .52 in grade 4 and .54 in grade 8. When reading was held constant the correlation between problem solving and computation was .43 in grade 4 and .42 in grade 8. Creswell (1982) worked with a sample of anglo and black adolescents from Chicago. Each subject was administered the California Achievement test. Multiple regression was used to analyze the data. The analysis showed that reading is more important than computation in predicting student performance in problem solving. Reading accounted for 49.5% of the variance; computation accounted for 14.6% of the variance. Ballew and Cunningham (1982) worked with 6th grade students in an attempt to find what proportion of students have as their main source of difficulty with word problems each of the following factors: a) computation skills/ b) interpretation of the problem/ c) reading and/ d) integrating these skills in the solution of problems. They also wanted to know if a student can be efficiently diagnosed as having one of the four categories as his/her main difficulty with mathematics word problems. 46 Their study is important because it represents an attempt to demonstrate that multiple factors can interact in the correct solution of a mathematics word problem. They constructed three graded tests from a basal mathematics series for grades 3 through 8. For test 1 / the problems were set in pure computational form (the effects or reading/ interpretation/ as well as the necessity for integration were removed in an effort to measure the computational skills required by the word problems). For test 2/ the effects of reading and computation were removed by reading the problems to the students and by giving scores based on whether or not the students set them up properly/ in an attempt to measure problem interpretation alone. For test 3/ the effect of computation was removed. The test yielded two scoresâ€”one by grading the students on whether or not they set up the problems properly and another by grading on the basis of the correct answer. The tests were administered to all 244 students from the 6th grade in two different schools. A diagnostic profile was obtained for each of the 217 students for which complete data were available: a computational score/ a probiem-interpretation score/ a reading score/ and a reading-problem solving score. They assumed that if the reading-prob1em interpretation score was lower (one or more levels lower) 47 than the problem-interpretation score? the difficulty was due to reading ability. If the score of the lowest of the three areas (computation? problem interpretation? and reading-problem interpretation score) was the same as the reading-problem solving score? the student's area of greatest immediate need was either computation? problem interpretation? or reading. If the reading-problem solving score was lower than the lowest of the other three scores? the student's area of greatest immediate need was integration. Analysis of the data revealed that for 19% of the students? problem interpretation was their major difficulty; for 26% of the students? integration (total problem solving) was their greatest immediate need; for another 26%? computation was the major weakness; and for 29%? reading was their greatest immediate need. Seventy five percent of the students demonstrated clear strength in computation? 21% in problem interpretation? and 4% in reading-problem interpretation. An analysis across all students (including those without complete data) showed that 26% of the subjects could not work word problems at a level as high as that at which they could compute? interpret problems? and read and interpret problems? when those areas were measured separately. This led them to conclude that knowing the skills or the components of solving word problems is not sufficient for 48 success* since the components must be integrated into a whole process (mastery learning of the components cannot assure mastery of the process). Their analysis also led them to conclude that* in the case of 6th graders* inability to read problems is a major obstacle in solving word problems. Only 12% of the subjects could read and set up problems correctly at a higher level than they could compute* while 60% could compute correctly at a higher level than they could read and set up problems; 44% could set up problems better when they heard them read than when they read the problem themselves. Only 13% could set up problems better when they read them than when they heard them read. Muth ( 1984 ) investigated the role of reading and computational skills in the solution of word problems. A group of 200 students from the 6th grade were administered a test of basic skills and a mathematics word problem test. The word problem test consisted of 15 sample items supplied by the National Assessment of Educational Progress. The items were adapted to include some extraneous information and complex syntactic structure. Four versions of the test were constructed by combining two versions of problem information (absence vs. presence of extraneous information) with two versions of syntactic structure (simple vs. complex syntax) . Task performance was measured by means of the number of problems answered correctly* 49 number of problems set up correctly/ and amount of time spent taking the test. Reading ability and computational ability were both positively correlated with number of correct answers and with number of problems correctly set up/ and negatively correlated with test-taking time. Presence of extraneous information was negatively correlated with correct answers and correct set ups and positively correlated with test-taking time. Syntactic complexity was not significantly correlated with any of the performance measures. Results of a multiple regression analysis showed that reading accounted for 46% of the variance in total correct answers and computation accounted for 8%. Reading ability and computational ability uniquely accounted for 14% and 8% of the variance in the number of correct answers/ respectively. Extraneous information added significantly to the variance explained in the number of correct answers/ but syntactic structure did not. Reading ability accounted for 5% of the variance in test-taking time/ but computation did not add significantly to the variance explained by reading. Muth concluded that reading and computation both contribute significantly to success in solving arithmetic word problems/ but that reading plays a more significant role than does computation. 50 The studies reviewed in this section show a positive relationship between reading and problem solving performance/ but in the case of Ballew and Cunningham (1982)/ this relationship is not viewed singly but rather as one among the interacting factors that produce successful problem solving. The third variable reviewed is the effect of student attitudes toward problem solving on problem solving performance. Many researchers have tried to demonstrate that this variable is a determinant factor in problem solving achievement. Attitudes Toward Problem Solving and Problem Solving Performance Research studies support the existence of positive and rather stable relationships between student attitudes and achievement in mathematics. Aiken (1970) has suggested that an individual's attitude toward one aspect of the discipline (mathematics) / such as problem solving/ may be entirely different from his/her attitude toward another phase of the discipline/ such as computation. Research/ however/ has been directed to the use of single/ global measures of attitudes toward mathematics rather than to the investigation of attitudes toward a particular phase of the discipline. The studies described below are only part of the few investigations which have examined the relationship between 51 student attitudes and performance in the area of problem solving. Carey (1958) constructed a scale to measure attitudes toward problem solving. Her interest was in general problem solving rather than in mathematical problem solving. Her work constitutes the first attempt to construct a measure of attitudes toward problem solving. The scale was used with a group of college students/ and she foundf among other things/ that problem solving performance is positively related to problem solving attitudes and that/ in the case of females/ positive modification of attitudes toward problem solving brings a significant gain in problem solving performance. Lindgren/ Silva/ Faracoz and DaRocha (1964) adapted Carey's scale of attitudes toward problem solving and applied it to a group of 4th grade Brazilian children. Students also answered an arithmetic achievement test/ a general intelligence test/ and a socioeconomic (SE) scale. A low but significant positive correlation was found between arithmetic achievement and attitudes toward problem solving. A near zero correlation was found between attitudes toward problem solving and intelligence. Since problem solving is one aspect of the discipline of mathematics/ this correlation between attitudes and arithmetic achievement can lead to a conclusion or a strong 52 correlation between attitudes toward problem solving and problem solving performance. Whitaker (1976) constructed a student attitude scale to measure some aspects of 4th grade student attitudes toward mathematic problem solving. He included statements reflecting children's beliefs about the nature of various types of mathematical problems/ the nature of the problem solving process/ the desirability or persevering when solving a problem/ and the value of generating several ideas for solving a problem. He correlated student attitudes toward problem solving with their scores in a mathematical test which yielded a comprehension score/ an application score/ and a problem solving score. He found a significant positive relationship between problem solving performance and student attitude scores on the subscale which measured reactions to such things as problem solving techniques or problem situations/ or to the frustation or anxiety experienced when confronted with problem solving situations. In another part of this study/ Whitaker investigated the relationship between the attitudes of 4th grade teachers toward problem solving and their students' performance in problem solving. A very weak and nonsignificant negative correlation was found between the teacher's attitudes toward problem solving and student performance. 53 Tne studies reviewed have confirmed the relationship between problem solving performance and attitudes toward problem solving (CareyÂ» 1958; Lindgren et al.Â» 1964; WhitakerÂ» 1976). HoweverÂ» the results reportea in the studies that investigated the relationship between problem solving performance and computation and between reading and problem solving fail to be consistent in their conclusions. Hansen (1944)Â» Chase (I960)Â» Balow (1964)Â» Kmfong and Holtan (1976Â» 1977)Â» and Zalewski (1974; concluded that computation is more strongly related to problem solving than is reading. Martin (1964)Â» Creswell (1982)Â» Ballew and Cunningham (1982)Â» and Muth (1984)Â» concluded that reading aoility and mathematical problem solving show a stronger relationsnip than computation and problem solving. Exedisis's (1983) findings led to the conclusion that the effect or reading and computation in problem solving performance is unimportant. CHAPTER III METHOD The first objective of this study was to investigate sex differences in the selection of incorrect responses in a mathematics multiple-choice achievement test, and to determine whether these differences were consistent over three consecutive administrations of the test. The second objective was to investigate whether males and females differ in problem solving performance* if these differences persist after accounting for computational skills* and if the male-female differences depend on the level of computational skills. This chapter contains descriptions of the sample* the test instrument* and the statistical analysis used in achieving the above mentioned objectives. The Sample To achieve the first objective of the study* all the students who took the Basic Skills Test in Mathematics-6 ("Prueba de Destrezas BÃ¡sicas en MatemÃ¡ticas-6") in each or the three years were included in the study. To achieve the second objective of the study* approximately 54 55 1/000 students were selected randomly for each year. For the first yearr 1/002 were selected (492 boys and 510 girls); for the second year/ 1/013 students were selected (504 boys and 509 girls); and/ for the third year/ 1/013 students were selected (509 boys and 504 girls). The student population in Puerto Rico includes children from the urban and rural zones and comprises children from low and middle socioeconomic levels. Findings can be generalized only to this population. The Instrument The Basic Skills Test in Mathematics-6 is a criter ion-referenced test used in the Department of Education of Puerto Rico as part of the annual assessment program. The test measures academic achievement in operations/ mathematical concepts/ and story problems. It has a reported split half reliability or .95. The test was designed specifically for Puerto Rico. Its contents and the procedures followed for its development were formulated and reviewed by educators from the mathematics department of the Department of Education of Puerto Rico/ in coordination with the Evaluation Center of the Department of Education and mathematics teachers from the school districts. The emphasis placed on each skill area is depicted in Figure 3.1. MATH CONCEPTS Fractions j 2 items) j / Numbers j \ (18 items ) \ / Fig. 3.1 Number of Questions by Skill Area in the Basic Skills Test in Mathematics - 6 57 Analysis of the Data Analysis of Sex bv Option by Year Cross Classifications Log-linear models were used to analyze the sex by option by year cross classifications for each item. Two topics are addressed in this section/ the hypotheses tested using log-linear models and a comparison of the hypotheses tested in this study with those tested by Marshall (1981/ 1983) . The object of the analysis was to test two hypotheses: 1. The proportion of males and the proportion of females choosing each incorrect option does not vary from year to year. Note that this hypothesis is stated in the null form. 2. Assuming that the first hypothesis is correct/ the proportion of males who choose each incorrect option is different from the proportion of females who choose each incorrect option. This hypothesis is stated in the alternate form. In Table 3.1 a hypothetical cross classification of sex/ option/ and year is presented. Hypothesis 1 is true for this three dimensional contingency table. In Table 3.2 the three dimensional contingency table is rearranged to show the year by option contingency table for each gender. 58 Table 3.1 Hypothetical Probabilities of Option Choice Conditional on Year and Sex Year Sex 1 Option 2 3 First M .7 .2 .1 F .5 .3 .2 Second M .7 .2 .1 F .5 .3 .2 Third M .7 .2 .1 F .5 .3 .2 Table 3.2 Hypothetical Conditional on Probabilities of Year and Sex and Option Choice Arranged by Sex Sex Year 1 Option 2 3 M First .7 .2 .1 Second .7 .2 .1 Third .7 .2 .1 F First .5 .3 .2 Second .5 .3 .2 Third .5 .3 .2 59 Inspection of the year by option contingency tables shows that year and option are independent for each gender. Thus/ hypothesis 1 is equivalent to the hypothesis that year and option are independent conditional upon gender. Hypothesis 2 is also true for Table 3.1 and Table 3.2. Therefore/ when hypothesis 1 is correct/ hypothesis 2 is equivalent to the hypothesis that sex and option choice are dependent. In his discussion of the analysis of three dimensional contingency tables/ Fienberg (1980) presents the following saturated model for the data: log m i j k = Ã¼ i + Uj + Uk+Uij + Uik+Ujk+U i j k. (3.1) In this model/ mijk is the expected value of the frequency in cell ijk of the three dimensional table. The model states that all three classification factors for a three dimensional contingency table are mutually dependent. In the present research i is the year index/ j is the option index/ and k is the sex index. Fienberg shows that deleting the terms Uij and Uijk yields a model in which year and option are independent conditional upon sex. This model is log mij k Ui + Uj + Uk+Uik+Uj k. (3.2) 60 Fienberg also shows that deleting the Ujterm from (3.2) to obtain log mij k = Ui + Uj+Uk + Uik. (3.3) yields a model that specifies that option is independent of sex. Based on Fienberg1 s presentationÂ» an appropriate analysis for testing the hypotheses is 1. Conduct a likelihood ratio test of the adequacy of model (3.2). If this test is nonsignificantÂ» then the data support the adequacy of the modelÂ» and the hypothesis that conditional on genderÂ» optionÂ» and year are independent. Because model (3.1) is a saturated modelÂ» testing the adequacy of model (3.2) is the same as comparing the adequacies of models (3.1) and (3.2). 2. Conduct a likelihood ratio test comparing the adequacies of models (3.2) and (3.3). If this test is significantÂ» then model (3.2) fits the data better than model (3.3) and the data support the hypothesis that the choice of incorrect option is dependent on sex. To summarizeÂ» if the first test is nonsignificant and the second test is significantÂ» then the choice of option is dependent on sexÂ» and the pattern of dependency is the same for all three years. 61 A problem arises in interpreting this analysis with the data used in this research. Over the three years? there were responses available from 135/340 students. Even if only 5% of the students answered an item incorrectly/ the responses of 7/767 students would be used in analyzing this item. On the other hand/ if 90% answered an item incorrectly/ the responses of 121/806 students would be used in analyzing the data. As a result of the large sample size/ the tests described above are likely to be very powerful. In step 1 of the analysis/ then/ even a very small change from year to year in the proportion of males or females who choose an option is apt to be detected/ and the results will indicate that hypothesis 1 is not supported. For step 2/ even a very small dependence of option choice on sex is likely to be detected and hypothesis 2 is likely to be supported. In brief/ the problem caused by the large sample size is that practically insignificant differences may yield statistical significance. Fortunately/ the form of the test statistic used in the likelihood ratio test suggests a reasonable solution to this problem. The test statistic is - 2 ^ F0 lÂ°9e (F0/Fe). (3.4) 62 Here the summation is over all the cells in the contingency table/ F0 refers to the observed frequency in a cell/ and Fe refers to the estimated expected frequency in a cell. Denoting the observed proportion in a cell by PG and the estimated expected proportion in a cell by Pe/ the test statistic can be written G2 = 2 N Â£^P0 1 Â° ge (Po/Pe5- (3.5) where N is the total number of subjects. This form of the test statistic suggests the following strategy. For any significant G2/ using P0 and Pe calculated from the total data set available for an item/ calculate the minimum N required for G2 to be significant. If the minimum N is very large/ this suggests that the statistically significant result is not practically significant since it can only be detected in very large samples. Of course/ the question remains as to what can be considered a minimum large N. Although there is room for argument/ it seems reasonable to claim that if an average of 1000 subjects per year is required to show significance/ then the result is not likely to be practically significant. On the basis of this reasoning/ it was proposed to ignore all significant results that would be nonsignificant if there were less than 3000 subjects available. In addition/ all log-linear 63 model tests were conducted using a .01 level of significance. Since this research is based on Marshall (1981/ 1983)/ it is important to compare the method of analysis used in this study to the one used by Marshall. Marshall also used a two-step analysis. In the first step of her analysis she deleted the Uijk term from (3.1) and tested the adequacy of the model/ 1 Â° g m j. j k = Ã¼ i + U j + o k+ U i j + Ui k+ Ãœ j k â€¢ (3.6) Following this/ she deleted the Ujk term to obtain log m Â¿ j k = Ui + Uj + Uk+Uij + Uik. (3.7) and compared these two models using a likelihood ratio test. If the first test was nonsignificant and the second significant/ Marshall claimed that option choice was dependent on sex and the pattern of dependency was the same from year to year. This is the same claim that this study sought to establish. However/ the approach used here was to present evidence that model (3.2) fits the data while Marshall tried to show that model (3.6) fits the data. The major difference between the two approaches concerns the operationalization of the concept that the 64 gender-option dependency is the same from year to year. In this study/ a three dimensional table was considered to exhibit the same year to year pattern of gender-option dependency if / conditioned on gender/ the same proportion of students chose each incorrrect response over the three year period. This seems to be a straightforward and natural way to operationalize the concept. To illustrate how Marshall operationalized the concept in question/ a hypothetical set of probabilities was constructed conforming to the pattern specified by Marshall. This is displayed in Table 3.3. Table 3 .3 Hypothetical Joint Probabilities of Year/ Option/ and Sex Year Sex 1 Option 2 3 First M .120 .045 Â¡ .015 F .060 .060 Â¡ .030 Second M .144 .022 | .019 F .072 .030 Â¡ .039 Third M .132 .040 Â¡ .016 F .066 .054 Â¡ J .039 Note: Probabilities reported are truncated to three places. 65 One important characteristic of this table involves two-by-two subtables of sex cross classified with two of the options for each year. For example* consider the three two-by-two tables obtained by cross classifying sex and option choices one and two for each year. These tables are indicated by dotted lines on Table 3.3. For this table* the ratio of the odds of a male choosing option one to the odds of a female choosing option one is the same (within truncating error) for each year. For example* this odds ratio for the first year is (.120/.045) / (.060/.060) = 2.67. Within the error caused by reporting truncated probabilities* the odds ratio for years two and three is the same as that for year one. The equality of these odds ratios is Marshall's operationalization stage of the year- to-year gender-option dependency. To show that the odds ratio can be constant over years* but that the probabilities of option choice conditional on sex and year can change from year to year* for both males and females* the probabilities in Table 3.3 were converted to the probabilities of option choice conditional upon sex and year. These conditional probabilities are reported in Table 3.4. Unlike the probabilities in Table 3.2* those in Table 3.4 change from year to year for both males and females. 66 Table 3.4 Hypothetical Probabilities of Option Choice Conditional on Sex and Year Year Sex 1 Option 2 3 First M .666 .250 .083 F .400 .400 .200 Second M .774 .120 .104 F .510 .212 .276 Third M .698 .214 .087 F .431 .352 .215 Note: Probabilities reported are truncated to three places. Which procedure is more appropriate to test the claim that the pattern of male and female option choice remains the same from year to year? It seems more reasonable to test for a pattern like that in Table 3.2 than to test for a pattern like that in Table 3.4 / and/ consequently/ this was the strategy adopted in this study. Comparison of Males and Females in Problem Solving Performance One object of the study was to compare the performance of males and females on problem solving. Two questions were addressed. Firstr do males and females differ in problem solving performance? Second/ do these differences persist when computational skill is controlled for/ and do these differences depend on the level of computational skill? Seven analyses of covariance (ANCOVA) tests were 67 conducted/ one with each of addition/ subtraction/ multiplication/ division/ addition of fractions/ decimals in subtraction/ and equivalence as covariates. A well known problem that arises in the use of ANCOVA is that an unreliable covariate can cause spurious differences between the sex groups. To solve this problem/ Porter (1967/1968) proposed the use of estimated true scores for observed scores. Porter (1967/1968) conducted a simulation that gave empirical support to the adequacy of this strategy. Hunter and Cohen (1974) have provided theoretical support for this strategy. CHAPTER IV RESULTS Introduction The data gathered from 6th grade students from the public schools in Puerto Rico who took the Basic Skills Test in Mathematics-6 during three consecutive years were analyzed in this study. The first objective was to investigate whether boys and girls differ in the selection of incorrect responses, and if the pattern of differences was consistent throughout the three years in which the test was administered. The second objective was to investigate whether males and females differ in problem solving performance, if these differences persist after accounting for computational skills, and if the male-female differences depend on the level of computational skills. The results are discussed in two sections. Study findings in the area of sex differences in the selection of incorrect responses are discussed in the first section. The second section is devoted to an exposition of the findings in the area of sex-related differences in problem solving. Sex-Related Differences in the Selection of Incorrect Responses As indicated in Chapter III, there are two models of interest. The first model indicates that year and option 68 69 are independent/ conditional upon sex* log m i j k = Ui + Uj + U^+Uij^+Uji^. (3.2) Substantively/ this model implies that the pattern of male and female option choices is consistent over the three years of test administration. The second model indicates that option is independent of sexr log = U i+ u j + uk+ ui k â€¢ (3.3) This model implies that the pattern of option choice is the same for males and females. In order to determine if males and females differed in the selection of incorrect responses and if these differences were stable across the three years of test administration/ a two-step test was performed. First/ a likelihood ratio test of the adequacy of model (3.2) was conducted for each of the 111 items of the test. Under this test a model fits the data when the value obtained is nonsignificant at a specified alpha level. For 50 (45%) of the items/ model (3.2) fitted the data adequately. For these items the pattern of male and female option choices was consistent for the three years. The values for the 61 items for which the likelihood ratio 70 tests were significant at the .01 level are reported in Table 4.1. Also reported in this table are the actual sample sizes and the minimum sample sizes necessary for the likelihood ratio tests to be significant. Of the 61 items* 59 had minimum sample sizes greater than 3*000. Thus* although for the three years both males and females samples had inconsistent option choices on these 61 items* on 59 of the items the inconsistency was relatively minor. Consequently* these 59 items were included in step 2 along with the initial 50 items. In step 2* a likelihood ratio test was performed comparing the adequacy of the models in (3.2) and (3.3). The X2 values associated with models (3.2) and (3.3) for each of the 109 items subjected to step 2 are reported in columns b and c of Table 4.2. Also reported in Table 4.2 is the difference between the two X2 values (see column d). This latter figure is the test statistic for comparing the adequacies of models (3.2) and (3.3). Significant X2 statistics are indicated by asterisks. The X2 statistics were significant for 100 of the items indicating a male-female difference in option choice for these items. In column e of Table 4.2 the actual sample sizes are reported. In column f* the minimum sample sizes necessary for significance are reported. For those 100 items which had significant X2 statistics reported in column d* 94 (94%) had minimum sample sizes greater than 3*000. lal 8 13 23 25 28 29 30 31 34 36 37 42 45 48 49 52 54 56 57 71 TABLE 4 .1 Summary of Significant Tests of Model 3.2 X1 for odel 3.2* Actual Sample Size Min. Sample Size for Significance (b) (c) (d) 46 .15 16,796 3,352 25.74 15,718 5,624 29.25 60,377 19,011 35.90 26 ,707 6,852 46 .00 84 ,847 16,988 61.48 88,396 13,242 25.19 87 ,473 31,982 23 .26 32,870 13 ,015 58.25 82 ,257 13 ,006 84.36 86 ,137 9,404 23 .70 37,068 14,405 23 .26 59,140 23 ,417 22.41 63,098 25,932 179.25 97 ,593 5,014 27.82 16,766 5,550 26.01 39,858 14,114 41.41 74,502 16,570 21.62 44,107 18,789 28.51 53 ,075 17 ,146 laL 59 60 61 62 63 64 65 66 67 68 69 70 71 73 74 76 77 78 79 72 TABLE 4.1 - Continued X2 for Actual Sample Min. Sample Model 3.2* Size Size for Significance (b) (c) LÃš) 50.96 41,829 7,560 66.43 60,494 8,387 21.39 35,632 15,342 55.03 50,699 8,485 158.36 52,068 3 ,028 88.01 73,196 7,660 56 .28 42,429 6 ,943 50.51 78,708 14,352 34.28 51,862 13,935 48.01 63,223 12,128 75.29 62,511 7 ,647 44.39 67 ,528 14,011 14,425.97 84 ,368 54** 24.61 53 ,358 19,969 22.75 84 ,175 34,077 99.68 70,183 6,485 79.38 66,304 7,693 64.56 76,806 10,957 23 .98 48,690 18,700 la) 80 81 82 83 84 85 86 87 88 89 90 91 93 94 95 96 98 101 102 104 73 TABLE 4.1 - Continued X* for Model 3.2* Actual Sample Size Min. Sample Size for Significance (b) (c) (d) 75.01 55/277 6,787 95 .51 57 ,710 5,565 51.99 71/315 12,633 20.12 81/021 37 ,088 38.62 78/959 18,830 1240 .78 41,180 306** 102.85 39/903 3 ,573 66.75 48,194 6,650 63.01 38,742 5/663 43 .26 39,648 8,441 85 .53 53 ,557 5,767 82.53 62,954 7/025 78.39 80 ,723 9,484 70.15 87,241 11,454 40.59 83 ,969 19,053 29.31 74,287 23 ,343 29.87 69,052 21,291 23 .77 83 ,782 32/462 61.06 75/708 11/419 40.47 72 ,517 16 ,503 7 4 TABLE 4.1 - Continued Item Number X2 for Model 3.2* Actual Sample Size Min. Sample Size for Significance la) (b) (c) (d) 105 28.66 60/712 19/510 108 26 .56 86/261 29/912 111 30.71 86/190 25/849 * p<.01 ** Minimum sample size less than 3/000 75 TABLE 4.2 Chi-square Values for the Comparison of Models (3.2) and (3.3) with Actual Sample Sizes and with the Corresponding Number of Subjects Needed for Significant Results Item NumÂ¬ ber Xa- for Model 3.2 X1 for Model 3.3 Xa- for Difference Actual Sample Size Minimum Sample Size for Signif. (a) (b) (c) (d) (e) (f) 1 16.11 30.94 14.83* 7*427 10*061 2 17.54 103.22 85 .68* 11*018 2*583** 3 20.06 47.51 27.45* 11*497 8*733 4 7.54 21.87 14.33* 6*016 8*414 5 14.75 70.40 55.65* 7*706 2*782** 6 11.86 23 .13 11.27* 13*046 23 *256 7 4.60 35.73 31.13* 8*832 5*700 8 46.15 329.98 283.83* 16*796 1*189** 9 10.13 109.99 99.86* 23 *075 4*642 10 9.54 65.29 55.75* 35*060 12*634 11 5.89 19.06 13.17* 24*908 37 *996 12 12.58 198.46 185.88* 40 *423 4*369 13 25.74 62.37 36.63* 15*718 8*621 14 11.19 80 .50 69.31* 30*238 8*765 15 16.63 17.33 0.70 42*889 1*230*914 16 14.51 84.29 69.78* 35*114 10*109 17 17.90 44.20 26 .30* 43 *962 33 *582 76 TABLE 4.2. - Continued Item NumÂ¬ ber X2 for Model 3.2 X2 for Model 3.3 X2 for Difference Actual Sample Size Minimum Sample Size for Signif. (a) (b) (c) (d) (e) (f) 18 11.71 15.68 3.97 52/3 40 264/864 19 9.37 18.77 9.40* 41/363 88/402 20 12.37 13.46 1.09 55 /707 1/026 /7 46 21 11.51 19.03 7.52 49/240 131/547 22 9.63 83 .74 74.11* 51/7 24 14/022 23 29.25 153.74 124.49* 60/377 9/744 24 10.35 42.93 32.58* 68/426 42/194 25 35.90 147.40 111.50* 26 /707 4/812 26 18.84 47 .30 28.46* 56/717 40 /037 27 8.78 222.87 214.09* 78/353 7/353 28 46.00 242.78 196.78* 84/847 8/662 29 61.48 170.22 108.74* 88/396 16/331 30 25.19 53 .84 28.65* 87/473 61/338 31 23 .26 47.42 24.16* 32/870 27/333 32 15.51 41.69 26.18* 48/777 37/430 33 7.57 48.84 41.27* 39/470 19/214 34 58.25 324.67 266.42* 82 /257 6/203 35 6.91 225 .66 218.75* 88/069 8/088 77 TABLE 4.2 - Continued Item NumÂ¬ ber Xa for Model 3.2 Xa for Model 3 .3 Xa for Difference Actual Sample Size Minimum Sample Size for Signif. (a) (b) (c) (d) (e) (f) 36 84.36 134.56 50 .20* 86 f137 34/472 37 23.70 76.85 53 .15* 37 f06 8 14/011 38 13.64 84 .24 70.60* 38/870 11/0618 39 13.35 757.60 744.25* 61/057 1/648** 40 6.67 701.39 694.72* 56 /438 1/630** 41 12.65 603.42 590.77* 57 /145 1/943** 42 23 .26 189.11 165.85* 59/140 7/154 43 16.29 21.65 5.36 33/712 126 ,357 44 3.55 115.84 112.29* 67/192 12/021 45 22.41 55.90 33.49* 13/098 37/851 46 6.56 16.30 9.74* 56 ,121 117 /007 47 7.82 76.15 68.33* 56 /555 16/628 48 179.25 452.64 273.39* 97 /593 7/172 49 27 .82 59.75 31.93* 16/766 10/549 50 14.23 240.27 226.04* 25/410 2/258** 51 16.37 224.54 208.17* 24 /715 2/385** 52 26.01 96 .14 70.13* 39/858 11/418 53 6.69 27.49 20.80* 44/694 43 /168 54 41.41 95.41 54.00* 74/502 27/718 78 TABLE 4.2 - Continued Item NumÂ¬ ber Xz for Model 3.2 X1 for Model 3.3 Xa for Difference Actual Sample Size Minimum Sample Size for Signif. (a) (b) (c) (d) (e) (f) 55 19.53 90 .87 71.34* 41,7 96 11,770 56 21.62 70.14 48.52* 44,107 18,263 57 28.51 35.78 7.27 53 ,07 5 146,668 58 17.49 51.69 34.20* 44,044 25,873 59 50.96 73.68 22.72* 41,829 36 ,987 60 66.43 148.39 81.96* 60,494 14,828 61 21.39 394.54 373.15* 35,632 1,918** 62 55.03 300.95 245.92* 50,699 4,142 63 158.36 278.34 119.98* 52 ,068 8,719 64 88.01 246.48 158.47* 73,196 9,279 65 56 .28 94.11 37.83* 42,429 22,532 66 50.51 113.88 63.37* 78,708 24,953 67 34.28 180 .60 146.32* 51,862 7,121 68 48.01 225.46 177.45* 63,223 7,158 69 75.29 251.66 176.37* 62,511 7,121 70 44.39 271.35 226 .96* 67,528 5,977 72 17.41 208.90 191.49* 84 ,175 8,831 73 24.61 37.80 13.19* 53 ,358 81,271 74 22.75 163.01 140 .26* 84 ,175 12,057 79 TABLE 4.2 - Continued Item NumÂ¬ ber Xa for Model 3.2 Xz for Model 3 .3 Xz for Difference Actual Sample Size Minimum Sample Size for Signif. (a) (b) (c) (d) (e) (f) 75 8.69 237.07 228.38* 67 /783 5/963 76 99.68 167 .72 68.04* 70/183 20,723 77 79.38 432.36 352.98* 66/304 3,774 78 64.56 127 .04 62.48* 76/806 24,696 79 23.98 72.59 48.61* 48/690 20/123 80 75.01 128.13 53 .12* 55/277 20,906 81 95.51 106.82 11.31* 57 /710 102/511 82 51.99 108.12 56.13* 71/315 25/525 83 20.12 95.18 75.06* 81 /021 21/685 84 38.62 69.05 30.43* 78,959 52,129 86 102.85 433.47 330.62* 39/903 2/425** 87 66.75 430.40 363.65* 48/194 2/662** 88 63.01 126 .27 63.26* 38/742 12/304 89 43 .26 289.14 245.88* 39/648 3 ,240 90 85.53 405.04 319.51* 53 ,557 3,368 91 82 .53 85.48 2.95 62/954 428,727 92 16.36 93 .12 76.76* 88/231 23 ,092 93 78.39 229.02 150.63* 80 ,723 10,766 94 70.15 137 .27 67.12* 87 ,241 26 ,113 80 TABLE 4.2 - Continued Item NumÂ¬ ber X2 for Model 3.2 X2 for Model 3.3 X2 for Difference Actual Sample Size Minimum Sample Size for Signif. (a) (b) (c) (d) (e) (f) 95 40 .59 68.41 27 .82* 83 /969 60/638 96 29.31 212.34 183.03* 74/287 8/154 97 8.60 126.96 118.36* 50/028 8/492 98 29.87 41.77 11.90* 69/052 116/576 99 9.44 9.94 0.50 72/647 2/918/956 100 11.35 41.48 30.13* 60/931 40 /627 101 23.77 315.62 291.85* 83 /782 5/767 102 61.06 437.09 376.03* 75/708 4/045 103 6.12 7.97 1.85 50/663 550/173 104 40 .47 283.94 243 .47* 72/517 5/984 105 28.66 48.59 19.93* 60/712 61/199 106 13 .47 1/682 .71 1/669.34* 71/946 866*' 107 13.13 99.20 86.07* 72/482 16 /918 108 26 .56 79.33 52.77* 86/261 32/840 109 5.73 18.33 12.60* 76/800 112/453 110 17.80 49.11 31.31* 70/980 45 /5 44 111 30.71 477.44 446 .73* 86/190 3/876 pC.oi Minimum sample size less than 3/000 Sex-related Differences in Problem Solving Performance 81 In this part/ results for the two questions related to the second objective of the study are presented. These questions serve as the framework for the presentation. Each question is stated/ followed by the results pertaining to that question. Question 1: Do males and females differ in problem solving performance? The responses of 492 males and 510 females who took the Puerto Rico Basic Skills Test-6 during the spring of the first year were analyzed in this study. Also/ data from 504 males and 509 females tested in the second year and from 509 males and 504 females tested in the third year were included in the analysis. The mean performance scores and the standard deviations for each of the eight variables are presented in Table 4.3. Results of t-tests are also presented in this table. Females outperformed males in problem solving/ a finding consistently present in all the three years of test administration. Over the three-year period the mean differences favored females in all variables except equivalence. The sex-related differences in problem solving were significant (p 82 TABLE 4.3 Means* Standard Deviations* and t-Tests for the Eight Mathematical Subtests Year/ Subtest Males X SD Females X SD t First Problem Solving 3.436 2.32 3.862 2.45 2.82* Addition 5.412 1.00 5.592 .86 3.05* Subtraction 4.550 1.66 4.852 1.56 1.22 Multi plication 4.014 1.80 4.433 1.68 3.81* Division 3.136 1.87 3.580 1.91 3.72* Fracadd 2.475 1.75 2.743 1.81 2.32** Decsub 2.648 1.83 3.321 1.86 5.77* Equivalence .77 .77 .73 .76 - .83 N = 492 N = 510 Second Problem Solving 3.632 2.28 4.015 2.36 2.65* Addition 5.470 .98 5.603 .86 2.29** Subtraction 4.843 1.52 4.923 1.44 .86 Multi plication 4.148 1.73 4.550 1.67 3.76* Division 3.242 1.77 3.632 1.86 3.42* Fracadd 2.565 1.74 2.903 1.91 2.94* 83 TABLE 4.3 - Continued Year/ Subtest Males X SD Females X SD t Decsub 2.863 1.87 3.440 1.80 5.00* Equivalence .76 .76 .65 .75 -2.32** N = 504 N = 509 Third Problem Solving 3.927 2.49 4.341 2.44 2.67* Addition 5 .536 .84 5.704 .64 3.59* Subtraction 4.836 1.52 4.958 1.46 1.30 Multiplication 4.168 1.74 4.541 1.68 3 .47* Division 3.343 1.88 3.795 1.83 3.87* Fracadd 2.819 1.89 3 .117 1.85 2.53** Decsub 3.021 1.93 3.448 1.85 3.60* Equivalanee .830 .82 .800 .78 - .60 N = 509 N = 504 Note: The number of items in the problem solving subtest was 9 . In each computation subtestr the number of items was 6. An item was included in the computation subtest only if it measured a computation skill required to solve a problem solving item. * ** p <.01 P <.05 84 Consistent significant differences were also found for addition/ multi plication/ division addition of fractions/ and subtraction of decimals. For subtraction the difference was not statistically significant. Question 2: Do sex-related differences in problem solving persist when computational skills are controlled forÂ» and is the male-female differences in problem solving dependent on level of computational skills? To address the question of dependence of male-female problem solving differences in computational skills for each year and computation subtest/ the possibility of an interaction was investigated. For the first year/ statistically significant interactions were found between sex and multiplication/ F (1/998) = 8.59/ p second year/ F (1 /1009) = 6 .39 / p<].05. No significant interactions were found in the third year. Analysis of covariance summary tables are shown as Tables 4.4/ 4.5/ and 4.6. Also/ the three interactions are depicted in Figures 4.1/ 4.2/ and 4.3. Each figure indicates that at lower levels of computational skills/ males outperformed females in problem solving/ with the reverse happening at higher levels of computational skills. 85 TABLE 4.4 ANCOVA Summary Table: Multiplication Covariate First Year Source df SS MS F Multiplication (M) 1 1195.90 1195.90 265.66* Sex (S) 1 31.95 31.95 7.07* M x S 1 38.80 38.80 8.59* Error 998 4509.00 4.51 * p <.01 TABLE 4.5 ANCOVA Summary Table: Divison Covariate First Year Source df SS MS F Division (D) 1 1195.90 1195.90 264.66* Sex (S) 1 13.94 13.94 3.22 D x S 1 18.39 18.39 4.25* Error 998 4317.00 4.32 * p<.01 86 TABLE 4.6 ANCOVA Summary Table: Subtraction Covariate Second Year Source df SS MS F Subtraction (S) 1 93 .14 593 .14 122.80* Sex (S) 1 18.58 18.58 3 .85* SxS 1 30.85 30.85 6.39* Error 1009 4873.40 4.83 *P <â€¢ 05 Problem Solving 87 9 â€ž 8 - 6 5 - 4 - 3 - 2 - 1 - T 1 â€œI 1 1 1 1 2 3 4 5 6 Multiplication Fig. 4.1 Sex by Multiplication Interaction. Problem Solving 88 9 8. 7- 5- 4- 3- 2. 1 ' r r t 1 2 3 Division Fig. 4.2 Sex by Division Interaction 4 5 6 Problem Solving 89 Subtraction Fig. 4.3 Sex by Subtraction Interaction 90 The significant interactions found between sex and multiplication^ and sex and division (first year)/ and sex and subtraction (second year)/ answered/ in part/ the question of whether male-female differences in problem solving performance depend on computational ability. However/ the evidence is quite weak. Of 21 possible interactions/ only three were significant. No variable exhibited a significant interaction for each of the three years. The analysis of covariance was also used to determine if sex-related differences exist after controlling for computational skills. Analyses were conducted for those variables that did not exhibit significant interactions with sex. As discussed in Chapter III/ estimated true scores were used for observed scores to adjust for unreliability of the covariates (the computational subtests). Reliability coefficients calculated for each covariate are shown in Table 4.7. Summaries of the analyses of covariance for the first year are reported in Table 4.8. The results show that females retained their superiority in problem solving performance when equivalence was the controlling variable in the analysis of covariance/ the only variable in which males outperformed females (nonsignificant). When the controlling variables were addition/ subtraction/ addition of fractions/ and subtraction with decimals/ female 91 TABLE 4.7 Reliability Three of the Years of Covariates for Each of Test Administration the Year Covariate First Second Third Addition .579 .610 .454 Subtraction .756 .724 .737 Multiplication .729 .729 .734 Division .716 .680 .714 Fracadd .709 .728 .748 Decsub .720 .716 .742 Equivalence .421 .394 .488 92 TABLE 4.8 ANCOVA Summary Table: Other Covariates First Year Source df SS MS F Addition 1 271.57 271.57 49.74* Sex 1 15.54 15.54 2.85 Error 999 5453.90 5.46 Subtraction 1 771.78 771.78 155.60* Sex 1 10.73 10.73 2.16 Error 999 4953.70 4.96 Fracadd 1 1412.11 1412.11 326.87* Sex 1 7.54 7.54 1.75 Error 999 4313.30 4.32 Decsub 1 1281.10 1281.10 287.89* Sex 1 4.94 4.94 1.11 Error 999 4444.30 4.45 Equivalence 1 811.44 811.44 164.93* Sex 1 69.69 69.69 14.16* Error 999 4914.00 4.92 * p .05 93 superiority in problem solving lost statistical significance. Adjusted means, associated with the analyses, are reported in Table 4.9. Summaries of the analyses of covariance for the second year are reported in Table 4.10. The analyses show that females retained their superiority in problem solving after controlling for equivalence. When the controlling variables were addition, multiplication, division, addition of fractions, and subtraction with decimals females superiority in problem solving lost statistical significance. Adjusted means are reported in Table 4.11. The analyses of covariance for the third year are summarized in Table 4.12. The results of the analyses show that females maintained their superiority in problem solving when performance on either subtraction or equivalence was controlled. When addition, multiplication, division, addition of fractions, and subtraction of decimals were the controlling variables, female superiority in problem solving lost statistical significance. Adjusted means are reported in Table 4.13. Summary Sex-related differences in mathematics achievement were the subject of this study. The first objective was to investigate whether males and females differ in the selection of incorrect responses, and if the pattern of 94 TABLE 4.9 Adjusted Means on Problem Solvingf by Covariate and Sex First Year Covariate Adjusted Means Males Females Addition 3 .53 3 .78 Subtraction 3 .55 3.76 Fracadd 3.56 3.74 Decsub 3.73 3.58 Equivalence 3 .38 3.91 95 TABLE 4.10 ANCOVA Summary Table: Other Covariates Second Year Source df ss MS F Addition 1 206.97 206.97 39.57* Sex 1 19.12 19.12 3.66 Error 1010 5277 .90 5.23 Multiplication 1 1026.68 1026.68 232.80* Sex 1 .80 .80 .18 Error 1010 4458.30 4.41 Division 1 1117 .46 1117.46 258.67* Sex 1 .64 .64 .15 Error 1010 4367.50 4.32 Fracadd 1 1341.13 1341.13 327.10* Sex 1 2.03 2.03 .50 Error 1010 4143.80 4.10 Decsub 1 1453 .91 1453.91 364.39* Sex 1 4.88 4.88 1.22 Error 1010 4031.10 3 .99 Equivalence 1 797.11 797.11 171.79* Sex 1 127.33 127.33 27.44* Error 1010 4687.90 4.64 * P<-05 96 TABLE 4 .11 Adjusted Means on Problem Solving/ by Covariate and Sex Second Year Covariate Adjusted Means Males Females Addition 3 .69 3 .96 Multiplication 3 .80 3.85 Division 3.80 3.85 Fracadd 3.78 3.87 Decsub 3.90 3 .75 Equivalence 3.46 44.18 97 TABLE 4.12 ANCOVA Summary Table Third Year Source df ss MS F Addition 1 205.96 205 .96 34.85* Sex 1 1.88 1.88 .32 Error 1010 5967.60 5.91 Subtraction 1 886.42 886.42 169.48* Sex 1 24.39 24.39 4.66* Error 1010 5287.2 5.23 Multi pi icati on 1 1402.42 1402.42 297.12* Sex 1 1.05 1.05 .22 Error 1010 4771.20 4.72 Division 1 2028.49 2028.49 494.75* Sex 1 1.11 1.11 .27 Error 1010 4145 .10 4.10 Fracadd 1 1733.72 1733.72 393.97* Sex 1 4.64 4.64 1.05 Error 1010 4439.89 4 .40 Decsub 1 2116.84 2116.84 526 .57* Sex 1 .15 .15 .04 Error 1010 4056 .76 4.02 Equivalence 1 1186.92 1186.92 240.26* Sex 1 59.85 59.85 12.12* Error 1010 4986.69 4.94 * p<. 05 98 TABLE 4.13 Adjusted Means on Problem Solving/ by Covariate and Sex Third Year Covariate Adjusted Means Males Females Addition 4.09 4.18 Subtraction 3.98 4.29 Multiplication 4.10 4.17 Division 4.17 4.16 Addf rac 4.07 4.20 Subdec 4 .15 4.12 Equivalence 3 .79 4.45 99 responses was consistent throughout the years in which the test was administered. Males and females selected different incorrect responses in 100 of the 111 items of the test. The pattern of male-female differences in the selection of incorrect responses was consistently found in each of the years of test administration. However/ for the vast majority of the 100 items/ male-female differences were relatively small in magnitude/ considering the fact that the number of subjects needed to obtain significance was very large. Therefore/ these findings lack educational significance. A second objective of the study was to investigate whether males and females differ in problem solving performance/ if these differences persist after controlling for computational skills/ and if the male-female differences depend on the level of computational skills. Results of the analyses reported in this section generally indicated that females outperformed males not only in problem solving/ but in six of the seven computational variables. Males surpassed females in equivalence/ but statistical significance was obtained for only one of the three years covered in the study. The results also tended to show that/ for examinees with similar levels of computational skills/ sex differences in problem solving did not exist. The only 100 exception was that/ when equivalence was controlled/ sex-related differences in favor of females/ persisted. The question of whether male-female differences in problem solving depend on computational skills was answered partially/ in the affirmative. Interactions/ indicating the dependence/ were significant only for multiplication and division in the first year and for subtraction in the second year. CHAPTER V DISCUSSION This cnapter comprises two sections: a) summary and interpretation of the results of the study/ and b) implications of the findings/ with suggestions for further research. Summary and Interpretation of the Results The main objective of this study was to investigate whether there were sex-related differences in incorrect responses/ and if the pattern of response was consistent over three consecutive administrations of the mathematics test. Males and females selected different incorrect responses in 100 of the 111 items of the test and this pattern of response was consistenly found in each of the three years of test administration. The results tend to support/ in part/ Marshall's findings (1981/ 1983) of observed sex-related differences in incorrect responses/ but in no way do they point to a conclusion that 6th grade boys and girls from the public schools in Puerto Rico exhibit different problem solving strategies. The significant differentiated pattern of 101 102 response was related to the use of a large number of subjects in the analysis. One must question whether analysis of incorrect responses has any educational value. Although the results of this study do not support the use of this analysis (error analysis) for the study of sex-related differences* it continues to be a promising technique in the discovery of how children approach mathematics problem solving* more so than analyses that look only at the items answered correctly or at total test scores. The second objective of the study was to investigate sex-related differences in problem solving performance* to find out if significant differences persisted after controlling for computational skills* and if the differences depended on the level of computational skills. The results showed that females outperformed males in problem solving and in six of the seven computational skills. Males showed superiority in equivalence in each of the three years* but statistical significance was obtained only for the second year. The results also showed that females retained their superiority in problem solving* only when equivalence (for the three years data) and subtraction (for the second year) were the controlling variables. Analysis of covariance was also used to answer the question of whether male-female differences in problem solving depend on the level of computational skills. The 103 question was answered/ partially/ in the affirmative. Interactions/ indicating the dependence/ were significant for multiplication and division in the first year and for subtration in the second year. In general/ the sex-related differences in mathematics problem-solving/ in favor of males performance/ reported by Marshall (1981/ 1984)/ and Fennema and Tartree (1983) were not supported in this investigation. Female performance was equal or higher than male performance in problem solving. The findings of the study have something in common with other investigations where mathematics problem solving has been the subject of interest/ the fact that males and females do poorly in word-story problem items (problem solving) . The mean score in problem solving for males and females in the 9-item problem solving subtest was 3.436 and 3.862 respectively for the first year/ 3.632 and 4.015 for the second year/ and 3.927 and 4.341 for the third year. Implications of the Findings and Suggestions for Further Research The main purpose of this study was to investigate sex-related differences in the selection of incorrect responses. Another objective was to compare male and female performance after accounting for computational ability. This type of research is conducted in an attempt 104 to contribute to the body of knowledge regarding such differences in mathematical learning and achievement. This section of Chapter V comprises implications of the findings and recommendations evolving from the present investigation. Sex-related Differences in Incorrect Responses The main purpose of education is to impart formal education equally for the sexes. Analysis of the data on incorrect responses showed that for 100 of the 111 items of the test/ males and females selected different incorrect responses/ and this pattern of responses was consistently found during the three years of test administration. However/ with a sample of less than 3/000 subjects/ statistical significant results were obtained for only 6 items. Therefore/ it seems that/ at least for 6th grade children/ educators need not concern themselves with varying teaching techniques for the sexes. This study/ however/ has value apart from the investigation of sex-related differences. It was found that/ year after year/ 6th graders from the public schools in Puerto Rico tend to make the same errors in the solution of mathematical problems. It is important/ therefore/ that teachers provide the learning strategies necessary to enable students to obtain a better understanding of the concepts of numbers. 105 Â¿exr_related.,Differences in Problem Solving Performance The results of this study do not give strong support to Marshall's (1981/ 1984) findings that the effect of computational skills on problem solving performance is different for each sex. However/ the findings of the present investigation are consistent with those found by Meyer (1978) and Whitaker (1976): female performance in problem solving is not significantly different from male performance. Females showed superiority in problem solving performance in all three years of test administration although their superiority was retained only when performance in equivalence was controlled for (for the three years data) and when performance in subtraction was controlled for (second year) (statistically significant only after accounting for equivalence) . The findings of this study are also in agreement with those reported by the National Assessment of Educational Progress (NAEP) (Carpenter et al.z 1980). Related to problem solving/ they conclude/ if it were necessary to single out one area that demands urgent attention/ it would clearly be problem solving. At all age levels/ and in virÂ¬ tually every content area/ performance was extreÂ¬ mely low on exercises requiring problem solving or application of mathematical skills. In geneÂ¬ ral/ respondents demonstrated a lack of the most basic problem-solving skills. Rather than attemptÂ¬ ing to think through a problem and figure out what needed to be done to solve the problem/ most resÂ¬ pondents simply tried to apply a single arithmetic 106 operation to the numbers in the problem. The reÂ¬ sults indicate that students are not familiar with such basic problem-solving strategies as drawing a picture of a figure described in a problem or checking the reasonableness of a result, (p. 338) The fact that, in generalÂ» after controlling for computational skillsÂ» there were no statistically significant differences between the problem solving performance of the males and females who participated in this study would seem to imply that teachers need not worry about designing different teaching strategies for the sexes. However^ because the performance in problem-solving is low for both sexesÂ» teachers may be encouraged to emphasize problem solving in their daily teaching. This study was designed to investigate sex-related differences in mathematical performance. The findings clearly demonstrated that 6th grade females and males from the public schools in Puerto Rico are equally good or equally bad in the solution of word-story items (problem solving). Also/ in general/ the effect of computational skills is similar for each sex. Based on these findings/ it appears that greater differences exist within each sex than between the sexes. In conclusion/ the findings of the present study highlight the need for investigations of different approaches to instructional designs aimed at improving both male and female performance in problem solving. REFERENCES AikenÂ» L. R. (1970). Attitudes toward mathematics. Review of Educational Resarch. IQ./ 551-596 . AlienÂ» R. H.Â» & ChambersÂ» D. L. (1977). A comparison of the mathematics achievement of males and females (Bulletin No. 9194). Madison: Wisconsin State Department of Public InstructionÂ» Division for Management and Planning Service. (ERIC Document Reproduction Services No. ED 159 073) ArmstrongÂ» J. M. (1979) . Achievement and participation of W.pme_n_in_.math.ematiÂ£sj Â¿Ln_Q_y.ejyi-e.W-. (Report of a two year study). DenverÂ» CO: Educational Commission of the StatesÂ» National Assessment of Educational Progress. (ERIC Document Reproduction Service No. ED 184 878) ArmstrongÂ» J. M.Â» & PriceÂ» R. A. (1982). Correlates and predictors of women's mathematics participation. Journal for Research in Mathematics EducationÂ» 11Â» 99-109. AstinÂ» H. S. ( 197 4) . Overview of the findings. In H. AstinÂ» H. SuniewickÂ» & S. Dweck (Eds.)Â» Women: A Bibliography on their Education and Careers. (pp. 1-10). New York: Behavorial Publications. BackmanÂ» M. E. (1972). Patterns of mental abilities: EthnicÂ» socioeconomic and sex differences. American Educational Research JournalÂ» 1Â» 1-11. BallewÂ» H.Â» & CunninghamÂ» W. (1982). Diagnosing strengths and weaknesses of sixth-grade students in solving word problems. Journal for Research in Mathematics EducationÂ» 13Â» 202-210. BalowÂ» I. H. (1964). Reading and computation ability as determinants of problem solving. The Arithmetic TeacherÂ» 11Â» 18-22. Bar-TalÂ» D. Â» & FriezeÂ» J. H. ( 1977 ) . Achievement motivation for males and females as a determinant of attribution for success and failure. Sex roles. 1Â» 301-313. 107 108 Bern* S. L. / & Bern D. V. (1970). We are all nonconscious sexists. Psychology Todays 22-24/ 26 / 115-116. Benbowf C. P./ & Stanley/ J. C. (1980). Sex differences in mathematical ability: Fact or artifact? Science/ 210/ 1262-1264. Benbow/ C. P./ & Stanley/ J. C. (1982). Consequences in high school and college of sex differences in mathemaÂ¬ tical reasoning ability: A longitudinal perspective. American Educational Research Journal/ 12/ 598 - 622. Bobbie/ C. N. (1971). Sex-role preference and academic achievement (Doctoral dissertation/ Yeshiva UniverÂ¬ sity) . Â£iss.ejtation Abstracts International/ 22/ 1818B-1819B. Carey/ G. L. (1958). Sex differences in problem solving performance as a function of attitude differences. Journal of Abnormal and Social Psychology/ 2Â£./ 256-260. Carpenter/ T. P./ Corbitt/ M. K./ Kepner/ H. S./ Lindquist/ M. M. / & Reys/ R. E. (1980). Results of the second NAEP mathematics assessment: Secondary school. Mathematics Teacher/ 73, 329-338. Carpenter/ T. P. / Lindquist/ M. M. / Mathews/ W. / & Silver/ E. A. (1984). Results of the third NAED mathematics assessment: Secondary school. Mathematics Teacher/ 2Â£/ 652-659. Chase/ C. I. (1960). The position of certain variables in the prediction of problem solving in arithmetic. Journal of Educational Research/ 21/ 9-15. Cohen/ D./ & Wilkie/ F. (1979). Sex-related differences in cognition among the elderly. In M. A. Witting and A. C. Petersen (Eds.)/ Sex-related differences in cognitive functioning: Developmental Issues/ New York: Academic Press. Connor/ J. M. / Serbin/ L. A./ & Schakman/ M. (1977). Sex differences in children's response to training on a visual-spatial test. Developmental Psychology/ 12./ 392-394 . 109 Creswell/ J. L. (1982/ February). Sex-related differences i..n_problem solving in rural black/ anglo and Chicago adolescents. Paper presented at the annual meeting of the Southwest Educational Research Association. Austin/ TX. (ERIC Document Reproduction Service No. ED 216 895) Dwyer/ C. A. ( 197 4) . Influence of children's sex role standards on reading and arithmentic achievement. Journal of Educational Psychology/ Â£Â£/ 811-816. Eccles/ J. ( 1 9 83 ) . Sex differences in mathematics participation. In M. Steinkamp & M. Maehr (Eds.)/ Women in Science (pp. 80-100). Greenwich/ CT: JAI Press. Elton/ C. F. / & Rose/ H. A. ( 1967). Traditional sex attitudes and discrepant ability measures in college women. Journal of Counseling Psycology/ M* 538-548. Ernest/ J. (1976). Mathematics and Sex. Santa Barbara: University of California. Exedisisz R. H. (1983). An investigation of the relationÂ¬ ship of reading comprehension/ vocabulary/ mathematical concepts/ and computation on problem solving among anglo/ black/ and Chicago male and female middle school adolescents (Doctoral Dissertation/ University of Houston/ 1982). Dissertation,Abstracts International/ 42., 226 4A-226 5A. Fennema/ E. (1974). Mathematics learning and the sexes: A review. Journal for Research in Mathematics Education/ Â¿/ 126-139. Fennema/ E. (1975). Mathematics/ spatial ability and the sexes. In E. Fennema/ (Ed.)/ Mathematics: What research savs about sex differences (pp. 33-44). Columbus: Ohio State University/ College of Education. (ERIC Document Reproduction Service No. ED 128-195) Fennema/ E. (1976) . Influences of selected cognitive affective/ and educational variables on sex-related djff_exgnc.e.s^,.n-mathematics learning and studying. Madison: University of Wisconsin/ Department of Curriculum and Instruction. (ERIC Document Reproduction Service No. ED 142 906) 110 Fennemar E. * & Shennanr J. (1977). Sex-related differences in mathematics achievements spatial visualizations and affective factors. American Educational.Research Journals 14' 51-71. Fennemas E.s & Shermans J. A. (1978). Sex-related differences in mathematics achievement and related factors. A further study. Journal for Research in Mathematics .Education/ 9s 189-204. Fennemas E.s & Tartrees L. A. (1983). Research on relationship of spatial visualization and confidence to male/female mathematics achievement in grades 6-8. Washingtons DC: National Science Foundation. (ERIC Document Reproduction Service No. ED 232 853) Fienbergs S. E. (1980). The Analysis of .Cross-Classified Categorical Data. Cambridges MA: The MIT Press. Foxs L. H. (1975a). Mathematically precocious: Male or female. In E. Fennema (Ed.) Mathematics learning: What research says.about sex differences (pp. 1-12). Columbus: Ohio State Universitys College of Education. (ERIC Document Reproduction Service No. ED 128 195) Foxs L. H. (1975b). Sex differences: Implications for program planning for the academically gifted. Paper presented at the Lewis M. Terman Memorial Symposium on Intellectual Talent held at the Johns Hopkins University* Baltimore* MD. Fox* L. H. (1977). The effects of sex role socialization on mathematics participation and achievement. In J. Shoemaker (Ed.)* Women_and_Mathe.mati,csâ€¢ Research Perspectives for Change* (N.I.E. Papers in EducaÂ¬ tion and Work: No. 8). Washington* D. C. : Eduation and Work Group* The National Institute of Education. Glennon* V. J.* & Callahan* L. G. (1968) Elementary school mathematics: A_guide._tp_..cur rent research* Washington* D. C.: Association for Supervision and Curriculum Development. Guay* R. B.* & McDaniel* D. (1977). The relationship between mathematics achievement and space ability among elementary school children. Journal for Research in Mathematics Education* 8* 211-215. Ill Hansen/ C. W. (1944). Factors associated with successful achievement in problem solving in sixth grade arithmentic. Journal of Educational Research# 38 # 111-119. Harnisch# D. L. (1983). Item response patterns: Application for educational practice. Journal of Educational Measurement# 21/ 191-205. Hilton# T. L. # & Berglund# G. W. (1974). Sex differences in mathematics achievement: A longitudinal study. Journal of. Educational Research# Â£2# 231-237. Hunter# J. E. # & Cohen# S. H. ( 1974). Correcting for unreliability in nonlinear models of attitude change. Psychometrika# 39 # 445-468. Jarvis# 0. T. (1964). Boy-girl ability differences in elementary school arithmentic. School Science and Mathematics# 64 # 657-659. Kaufman# A. (1984). Sex differences in mathematics reasoning in mathematically gifted and average fifth and sixth graders (Doctoral dissertation# Hofstra University# 1983) . Dissertation Abstracts InterÂ¬ national# 45 # 1094A-1095A. Kilpatrick I. (1969). Problem solving and creative behavior in mathematics. In J. W. Wilson# & L. R. Carry (Eds.)# Studies in mathematics (Vol. No. 19). Stanford# CA: School Mathematics Study Group. Kloosterman# P. W. (1985). Attributional theory# learned helplessness# and achievement in ninth grade mathematics (Doctoral dissertation# University of Wisconsin# 1984) . Dissertation Abstracts InternatioÂ¬ nal# 41/ 919A-920A. Knifong# J. D. # & Holtan# B. ( 1976) . An analysis of children's written solutions to word problems. Journal for Research in Mathematics_Education# 2# 106-112. Knifong# J. D.# & Holtan# B. (1977). A search for reading difficulties among erred word problems. Journal for Research in Mathematics Education# 1# 227-230. 112 LandaUf M. (1984). The effects of spatial ability and problem presentation format in mathematical problem solving performance on middle school students (Doctoral dissertation/ Northwestern University). Dissertation Abstracts International/ 4Â¿/ 442A- 443 A. Lindgren/ H. C. / Silva/ I./ FaracoÂ» I./ & DaRocha/ N. S. (1964). Attitudes toward problem solving as a function of success in arithmetic in Brazilian elementary schools. Journal of Educational Research/ Â¿2., 44-45 . Maccoby/ E. E. (1966). Sex differences in intellectual functioning. In E. E. Maccoby (Ed.)/ The Development of Sex Differences (pp. 25-55). Stanford/ CA: The Stanford University Press. Maccoby/ E. E. / & Jacklin/ C. N. (1972). Sex differences in intellectual functioning. In A. Anastasi/ (Ed.)/ Assessment in a pluralistic society, (pp. 37-55) . Proceedings of the 1972 Invitational Conference on Testing Problems/ Educational Testing Service/ Princeton/ NJ. Maccoby/ E. E./ & Jacklin/ C. N. (1974) . The Psychology of Sex Differences/ Stanford/ CA: Stanford University Press. Marshall/ S. P. (1981) . Sex Differences in Sixth Grade Children's Problem Solving. Paper presented at the annual meeting of the American Educational Research Association/ Los Angeles/ CA. Marshall/ S. P. (1983). Sex differences in mathematical errors: Analysis of distractor choice. Journal for Research in Mathematics Education/ 14./ 325-336 . Marshall/ S. 'p. (1984). Sex differences in children's mathematics achievement: Solving computations and story problems/ Journal of Educational Psychology/ lAf 194-204. Martin/ M. D. (1964). Reading comprehension/ abstract verbal reasoning and computation as factors in arithmetic problem solving (Doctoral dissertation/ University of Iowa/ 1963) . Dissertation Abstracts International/ 24/ 4547A-4548A. 113 Meyer? R. A. (1978). Sex-related differences in mathemaÂ¬ tical problem solving performance and intellectual abilities. Madison: University of Wisconsin? Research and Development Center for Cognitive Learning. (ERIC Document Reproduction Service No. ED 166 023). Milton? G. A. (1957) . The effects of sex-role identifica- cation upon problem solving skill. Journal of Abnormal & Social,Psychology? 5Â¿? 208-212. Moller? N. J. (1983). The impact of gender? masculinity? and feminity on math achievement and course decisions (Doctoral dissertion? Purdue University? 1982). Â£is_sertation_Abstracts International? 42/ 2584A. Muth? K. D. (1984). Solving arithmetic word problems: Role of reading and computational skills. Journal of Educational Psychology? 76 ? 205-210. Pallas? A. M.? & Alexander? K. L. (1983). Sex differences in quantitative SAT performance: New evidence on the differential coursework hypothesis. American Educational Research Journal? 21/ 165-182. Plank? E.? & Plank? R. (1954). Emotional components in arithmetical learning as seen through autobiographies. Psychological Studies of the Child? 2.? 274-296. Porter? A. C. (1968). The effects of using fallible variables in the analysis of covariance (Doctoral dissertation? University of Wisconsin? 1967). Dissertation Abstract International? 2Â£? 3517B. Probert? B. S. (1984). Math confidence workshops: A multimodal group intervention strategy in mathematics anxiety/avoidance. (Doctoral dissertation? University of Florida? 1983). Dissertation Abstracts International? 44 ? 2231B. Puerto Rico Department of Education. (1980). Basic skills test in mathematics-6. Hato Rey: Center of Evaluation. Radatz? H. (1979) . Error analysis in mathematics educa- tion. Journal for Research in Mathematics EducaÂ¬ tion ? 10 ? 163â€”172. Riedesel? A. C. (1969). Problem solving: Some suggestions from research. Arithmetic Teacher? 16 ? 54-58. 114 Sherman/ J. (1979) . Predicting mathematics performance in high school girls and boys. Journal of Educational Psychology/ 71/ 242-249. Sherman/ J. (1980). Mathematics/ spatial visualization/ and related factors: Changes in girls and boys/ grades 8-11. Journal of Educational Psychology/ 22/ 476-482. Stein/ A. H. (1971). The effects of sex role standards for achievement and sex role preference on three determinants of achievement motiviation. Developmental Psychology/ 4./ 219-231. Stein/ A. H./ & Smithless/ J. (1969). Age and sex differences in children's sex role standards about achievement/ Developmental Psychology/ 2/ 252-259. Suydam/ M. N. (1971) . Research on mathematics education (K-12) reported in 1970. Journal for Research in Mathematics Education/ 2/ 257-298. Suydam/ M. N. (1972). Research on mathematics education (K-12) reported in 1971. Journal for Research in Mathematics Education/ 2, 196-232. Suydam/ M. N. (1973) . Research on mathematics education (K-12) reported in 1972. Journal for Research in Mathematics Education/ 1/ 205-242. Suydam/ M. N. (1974). Research on mathematics education (K-12) recorded in 1973. Journal for Research in Mathematics Education/ Â¿/ 238-272. Suydam/ M. N. / & Weaver/ J. F. ( 1975) . Research on mathematics education (K-12) recorded in 1974. Journal for Research in Mathematics Education/ Â£/ 253-282. Verbeke/ K. A. (1983). Sex-related differences in mathematically gifted secondary students: An investigation of selected cognitive/ affective/ and educational factors (Doctoral dissertation/ University of Maryland/ 1982) . Dissertation Abstracts International/ 43/ 2267A-2268A. Weiner/ B. (1974). Achievement motivation and attribution theory. Morristown/ NJ: General Learning Press. 115 Whitaker, D. R. (1976). A study of the relationship between selected noncognitive factors and the problem solving performance of fourth-grade children. (Tech. Rep. No. 396). Madison: University of Wisconsin/ Research and Development Center for Cognitive Learning. Wiser L. K.r SteelÂ» L.r & MacDonald/ C. (1979). Origins and career consequences of sex differences in high school mathematics achievement (Grant No. NEI-G 7 8-001 ) . Palo Alto/ CA : American Institutes for Research. Zalewski/ D. L. (1974). An exploratory study to compare two performance measures: An interview-coding scheme of mathematical problem solving and a written test (Doctoral Dissertation/ University or Wisconsin/ Madison). Dissertation Abstracts International/ 1Â£# 3797A. BIOGRAPHICAL SKETCH Sonia Feliciano was born on March 17/ 1938/ in Mayaguez/ Puerto Rico. She graduated from Eugenio Ma. de Hostos High School/ in her hometown. She received a degree in business administration/ with a major in finance/ from the University of Puerto Rico in June/ 1961. After graduation she worked as junior accountant for Pittsburgh Plate Glass/ International. During the academic year of 1969-70 she started her career in education/ obtaining the degree of Master in Education from the University of Puerto Rico in June/ 1975. From January/ 1971 to June/ 1985/ she served in the public school system of Puerto Rico in different positions. She worked as educational researcher until July/ 1977/ when was appointed Director of the Research Center of the Commonwealth Department of Education. During the academic year of 1978-79/ she initiated doctoral studies in the Faculty of Philosophy and Educational Sciences of the Universidad Complutense de Madrid/ in Madrid/ Spain. She completed the coursework requirements and then transferred to University of Florida during the fall of 1979. 116 117 From May* 1982 to July* 1983 / she worked as graduate research assistant at the Foundations of Education Department/ University of Florida. She returned to Puerto Rico in August/ 1983 / to serve as Special Aide to the Assistant Secretary for the Vocational/ Technical/ and High Skills Educational Programs of the Commonwealth Department of Education. I certify that I have read this study and that in my opinion it conforms to acceptable standards of scholarly presentation and is fully adequatef in scope and qualityf as a dissertation for the degree of Doctor of Philosophy. Jam\es Algina* Chairman Processor of Foundations of Education I certify that I have read this study and that in my opinion it conforms to acceptable standards of scholarly presentation and is fully adequatef in scope and qualityf as a dissertation for the degree of Doctor of Philosophy. Linda Crocker Professor of Foundations of Education I certify that I have read this study and that in my opinion it conforms to acceptable standards of scholarly presentation and is fully adequatef in scope and qualityf as a dissertation for the degree of Doctor of Philosophy. , JÃ¼yfrnl njaery of Educational Leadership lichael Professor This dissertation was submitted to the Graduate Faculty of the College of Education and to the Graduate Schoolf and was accepted as partial fulfillment for the degree of Doctor of Philosophy. Augustf 1986 Chairmanf Foundations of Education Deanf College of Education Deanf Graduate School UNIVERSITY OF FLORIDA 3 1262 08554 6660 84 Consistent significant differences were also found for addition* multiplication* division* addition of fractions* and subtraction of decimals. For subtraction the difference was not statistically significant. Question 2: Do sex-related differences in problem solving persist when computational skills are controlled for* and is the male-female differences in problem solving dependent on level of computational skills? To address the question of dependence of male-female problem solving differences in computational skills for each year and computation subtest* the possibility of an interaction was investigated. For the first year* statistically significant interactions were found between sex and multiplication* F (1*998) = 8.59* p second year* F (1 *1009) = 6 .39 p<].05. No significant interactions were found in the third year. Analysis of covariance summary tables are shown as Tables 4.4* 4.5* and 4.6. Also* the three interactions are depicted in Figures 4.1* 4.2* and 4.3. Each figure indicates that at lower levels of computational skills* males outperformed females in problem solving* with the reverse happening at higher levels of computational skills. 98 TABLE 4.13 Adjusted Means on Problem Solving* by Covariate and Sex Third Year Covariate Adjusted Males Means Females Addition 4.09 4.18 Subtraction 3.98 4.29 Multiplication 4.10 4.17 Division 4.17 4.16 Addfrac 4.07 4.20 Subdec 4.15 4.12 Equivalence 3.79 4.45 18 students who had taken a similar number of mathematics courses and were in the top 85% of the class in mathematics achievement. They were tested in 1976. Four high school districts were included. In only one of the high school districts were sex-related differences in application and problem solving found# in favor of males. They concluded that when relevant factors are controlled/ sex-related differences in favor of males do not appear often/ and when they do/ they are not large. Sherman (1980) investigated the causes of the emerging sex-related differences in mathematics performance/ in favor of males/ during adolescence (grades 8-11). She wanted to know if these differences emerge as a function of sex-related differences in spatial visualization and sociocultural influences that consider math as a male domain. In grade 8/ she used the Romberg-Wearne Test and/ in grade 11/ a mathematical problem solving test derived from the French Kit of Tests. The analysis showed that for girls/ problem solving performance remained stable across the years. Mean problem solving performance for boys/ however/ was higher in grade 11 than in grade 8. No sex-related differences were found in grade 8/ but boys outperformed girls in grade 11/ where the Stafford test was used. Sherman found that for both sexes problem solving performance in grade 8 was the best predictor of problem 40 Achievement Test (subtests of reading/ arithmetic/ and reasoning) and the California Short-Form test of mental ability to a group of 1/400 children from the 6th grade. All levels of achievement were included in the analysis. Analysis of variance and covariance were used and compared. He confirmed the findings of other researchers to the effect that there is a direct relationship between I.Q. and reading ability/ and between I.Q. and computational skills. The results of the analysis of variance revealed that increases in computation ability were associated with higher achievement in problem solving. A relationship between reading ability and problem solving was also found/ but it was not as strong. Significant differences in problem solving performance associated with computational ability were found when intelligence was controlled. Balow concluded that computation is a much more important factor in problem solving than reading ability/ and that when I.Q. is taken into consideration/ the degree of the relationship between reading and problem solving ability becomes less pronounced. Intelligence tends to confound the relationship between these two variables. Knifong and Holtan (1976/ 1977) attempted to investigate the types of difficulties children have in solving word problems. They administered the word problem section of the Metropolitan Achievement Test to 35 children from the 6th grade. Errors were classified in two 74 TABLE 4.1 Continued Item Number Xa for Model 3.2* Actual Sample Size Min. Sample Size for Significance (a) ... (b) Cc) (d) 105 28.66 60/712 19/510 108 26.56 86/261 29/912 111 30.71 86/190 25/849 * p<.01 ** Minimum sample size less than 3/000 laL 59 60 61 62 63 64 65 66 67 68 69 70 71 73 74 76 77 78 79 72 TABLE 4.1 Continued X2 for Actual Sample Min. Sample Model 3.2* Size Size for Significance (b) (c) L) 50.96 41,829 7,560 66.43 60,494 8,387 21.39 35,632 15,342 55.03 50,699 8,485 158.36 52,068 3 ,028 88.01 73,196 7,660 56 .28 42,429 6 ,943 50.51 78,708 14,352 34.28 51,862 13,935 48.01 63,223 12,128 75.29 62,511 7 ,647 44.39 67 ,528 14,011 14,425.97 84 ,368 54** 24.61 53 ,358 19,969 22.75 84 ,175 34,077 99.68 70,183 6,485 79.38 66,304 7,693 64.56 76,806 10,957 23 .98 48,690 18,700 61 A problem arises in interpreting this analysis with the data used in this research. Over the three yearsr there were responses available from 135*340 students. Even if only 5% of the students answered an item incorrectly* the responses of 7*767 students would be used in analyzing this item. On the other hand* if 90% answered an item incorrectly* the responses of 121*806 students would be used in analyzing the data. As a result of the large sample size* the tests described above are likely to be very powerful. In step 1 of the analysis* then* even a very small change from year to year in the proportion of males or females who choose an option is apt to be detected* and the results will indicate that hypothesis 1 is not supported. For step 2* even a very small dependence of option choice on sex is likely to be detected and hypothesis 2 is likely to be supported. In brief* the problem caused by the large sample size is that practically insignificant differences may yield statistical significance. Fortunately* the form of the test statistic used in the likelihood ratio test suggests a reasonable solution to this problem. The test statistic is (3.4) 15 The test is composed of 23 items designed to yield 3 scores: a comprehension scorer an application scorer and a problem solving score. The comprehension question ascertains whether a child understands the information given in the item stem. The application question assesses the child's mastery of a prerequisite concept or skill of the problem solving question. The problem solving question poses a question whose solution is not immediately availabler that isr a situation which does not lend itself to the immediate application of a rule or algorithm. The application and problem solving parts of the test may refer to a common unit of information (the item stem) but the questions are independent in that the response to the application question is not used to respond to the problem solving question. Meyer (1978) / Whitaker (1976) / Fennema and Sherman (1978)/ and Sherman (1979) have used the Romberg-Wearne test in their studies. Meyer (1978) investigated whether males and females differ in problem solving performance and examined their prerequisite computational skills and mathematical concepts for the problem solving questions. A sample of 179 students from the 4th grade were administered 19 "reference tests" for intellectual abilities and the Romberg-Wearne test. The analysis showed that males and females were not significantly different in the comprehension/ application/ and problem solving questions 55 1/000 students were selected randomly for each year. For the first yearr 1/002 were selected (492 boys and 510 girls); for the second year/ 1/013 students were selected (504 boys and 509 girls); and/ for the third year/ 1/013 students were selected (509 boys and 504 girls). The student population in Puerto Rico includes children from the urban and rural zones and comprises children from low and middle socioeconomic levels. Findings can be generalized only to this population. The Instrument The Basic Skills Test in Mathematics-6 is a criter ion-referenced test used in the Department of Education of Puerto Rico as part of the annual assessment program. The test measures academic achievement in operations/ mathematical concepts/ and story problems. It has a reported split half reliability or .95. The test was designed specifically for Puerto Rico. Its contents and the procedures followed for its development were formulated and reviewed by educators from the mathematics department of the Department of Education of Puerto Rico/ in coordination with the Evaluation Center of the Department of Education and mathematics teachers from the school districts. The emphasis placed on each skill area is depicted in Figure 3.1. Problem Solving 88 9 8. 7- 5- 4- 3- 2. 1 r r t 1 2 3 Division Fig. 4.2 Sex by Division Interaction 4 5 6 5 Females were better on items of computation and males were more successful on word-story problem items (problem solving). She also found that females successful performance in the problem solving items was more dependent on their successful performance in the computation items. Males did not need as much as females to succeed in the computation items in order to answer correctly the problem solving items. Although the general findings seem to support sex differences in mathematics learning and achievement the research done does not consistently support superiority for either sex. Most of the research has been concerned with how the sexes differ on subtests or total test scores in mathematics. Moreover the great majority of the studies deal with correct responses. Sex differences in incorrect responses at the item level have not been fully researched. Only two studies dealing with sex differences in incorrect responses at the item level were found in the research literature (Marshall 1981 1983). Marshall investigated whether boys and girls made similar errors in computation and story problems. She analyzed boys' and girls' answers to six mathematics items and found that the sexes made different errors possibly reflecting different problem solving strategies. Her original findings were supported when she studied the same problem using a large number of items three years later. 70 tests were significant at the .01 level are reported in Table 4.1. Also reported in this table are the actual sample sizes and the minimum sample sizes necessary for the likelihood ratio tests to be significant. Of the 61 items# 59 had minimum sample sizes greater than 3#000. Thus# although for the three years both males and females samples had inconsistent option choices on these 61 items# on 59 of the items the inconsistency was relatively minor. Consequently# these 59 items were included in step 2 along with the initial 50 items. In step 2# a likelihood ratio test was performed comparing the adequacy of the models in (3.2) and (3.3). The X2 values associated with models (3.2) and (3.3) for each of the 109 items subjected to step 2 are reported in columns b and c of Table 4.2. Also reported in Table 4.2 is the difference between the two X2 values (see column d) This latter figure is the test statistic for comparing the adequacies of models (3.2) and (3.3). Significant X2 statistics are indicated by asterisks. The X2 statistics were significant for 100 of the items indicating a male-female difference in option choice for these items. In column e of Table 4.2 the actual sample sizes are reported. In column f# the minimum sample sizes necessary for significance are reported. For those 100 items which had significant X2 statistics reported in column d# 94 (94%) had minimum sample sizes greater than 3 #000 100 exception was that/ when equivalence was controlled/ sex-related differences in favor of females/ persisted. The question of whether male-female differences in problem solving depend on computational skills was answered partially/ in the affirmative. Interactions/ indicating the dependence/ were significant only for multiplication and division in the first year and for subtraction in the second year. 2 Many variables* cognitive* affective* and educational* have been investigated since 1974 in relation to sex differences in mathematics learning and achievement. Fennema and Sherman (1977) investigated the effect of differential formal mathematics education. After controlling for the number of years of exposure to the study of mathematics* they found sex differences in only two of the four schools under study. However* in those schools where boys scored higher than girls* differences were also found in their attitudes toward mathematics. Hilton and Berglund (1974) found significant sex differences after controlling for the number of mathematics courses taken* and attributed them to sex differences in interests. "As the boys' interests in science increase related to the girls'* their achievement in mathematics increases relative to that of the girls" (p. 234) . Wise* Steel* and McDonald (1979) reanalyzed test data collected in a longitudinal study of 400*000 high school students (Project Talent). They found that when the effect of the number of high school mathematics courses was not controlled* no sex differences emerged for 9th graders* but that gains made by boys during the next three years were more than twice that of the girls. These differences between the sexes disappeared when the number of mathematics courses taken was controlled. Results of the 1978 Women and Mathematics National Survey* Survey I* 13 III) to investigate sex related differences in the selection of incorrect responses/ and the consistency of such differences over three years of administration of the test. Based on her findings that sex differences were found in 80% of the items/ Marshall classified the students' errors according to Radatz' (1979) five-category error classification. The categories are language (errors in semantics)/ spatial visualization/ mastery/ association/ and use of irrelevant rules. It was found that girls' errors are more likely to be due to the misuse of spatial information/ the use of irrelevant rules/ or the choice of an incorrect operation. Girls also make relatively more errors of negative transfer and keyword association. Boys seem more likely than girls to make errors of perseverance and formula interference. Both sexes make language-related errors/ but the errors are not the same. Available research is not extensive enough to make definite judgments about the sex-related differences observed in incorrect responses. Clearly more research is needed. CHAPTER IV RESULTS Introduction The data gathered from 6th grade students from the public schools in Puerto Rico who took the Basic Skills Test in Mathematics-6 during three consecutive years were analyzed in this study. The first objective was to investigate whether boys and girls differ in the selection of incorrect responses, and if the pattern of differences was consistent throughout the three years in which the test was administered. The second objective was to investigate whether males and females differ in problem solving performance, if these differences persist after accounting for computational skills, and if the male-female differences depend on the level of computational skills. The results are discussed in two sections. Study findings in the area of sex differences in the selection of incorrect responses are discussed in the first section. The second section is devoted to an exposition of the findings in the area of sex-related differences in problem solving. Sex-Related Differences in the Selection of Incorrect Responses As indicated in Chapter III, there are two models of interest. The first model indicates that year and option 68 86 TABLE 4.6 ANCOVA Summary Table: Subtraction Covariate Second Year Source df SS MS F Subtraction (S) 1 93.14 593.14 122.80* Sex (S) 1 18.58 18.58 3.85* SxS 1 30.85 30.85 6.39* Error 1009 4873.40 4.83 *p<.05 25 emerged in spatial visualization scores became non-significant. In the two schools where sex differences in mathematics achievement were found/ differences between the sexes were also found in their attitudes toward mathematics. Researchers like Backman (1972) / who analyzed data from Project Talent/ and Allen and Chambers (1977) have also hypothesized that sex-related differences in mathematics achievement may be related to different curricula followed by males and females. Allen and Chambers attributed male superiority in mathematics problem solving to differences in the number of mathematics courses taken in high school. This issue has been seriously questioned by Astin (1974)/ Fox (1975a/ 1975b)/ and Benbow and Stanley (1980)/ among others. Astin and Fox have reported large differences in favor of males among gifted students taking the Scholastic Achievement Test. These differences occur as early as grade 7/ when there are no sex differences in the number of courses taken. Benbow and Stanley (1980) compared mathematically precocious boys and girls in the 7th grade/ with similar mathematics background/ and found sizeable sex-related differences favoring boys in mathematical reasoning ability. Five years later/ they conducted a follow-up study which showed that boys 95 TABLE 4.10 ANCOVA Summary Table: Other Covariates Second Year Source df ss MS F Addition 1 206.97 206.97 39.57* Sex 1 19.12 19.12 3.66 Error 1010 5277 .90 5.23 Multiplication 1 1026.68 1026.68 232.80* Sex 1 .80 .80 .18 Error 1010 4458.30 4.41 Division 1 1117 .46 1117.46 258.67* Sex 1 .64 .64 .15 Error 1010 4367.50 4.32 Fracadd 1 1341.13 1341.13 327.10* Sex 1 2.03 2.03 .50 Error 1010 4143.80 4.10 Decsub 1 1453 .91 1453.91 364.39* Sex 1 4.88 4.88 1.22 Error 1010 4031.10 3 .99 Equivalence 1 797.11 797.11 171.79* Sex 1 127.33 127.33 27.44* Error 1010 4687.90 4.64 * P<-05 25 emerged in spatial visualization scores became non-significant. In the two schools where sex differences in mathematics achievement were found differences between the sexes were also found in their attitudes toward mathematics. Researchers like Backman (1972) r who analyzed data from Project Talent and Allen and Chambers (1977) have also hypothesized that sex-related differences in mathematics achievement may be related to different curricula followed by males and females. Allen and Chambers attributed male superiority in mathematics problem solving to differences in the number of mathematics courses taken in high school. This issue has been seriously questioned by Astin (1974) Fox (1975a 1975b) and Benbow and Stanley (1980) among others. Astin and Fox have reported large differences in favor of males among gifted students taking the Scholastic Achievement Test. These differences occur as early as grade 7 when there are no sex differences in the number of courses taken. Benbow and Stanley (1980) compared mathematically precocious boys and girls in the 7th grade with similar mathematics background and found sizeable sex-related differences favoring boys in mathematical reasoning ability. Five years later they conducted a follow-up study which showed that boys skills* sex-related differences in problem solving performance do not exist. Females retained their superiority in problem solving when equivalence (in all three years) and subtraction (in one year) were the controlling variables. The question of whether male-female differences in problem solving depend on computational skills was answered* partially* in the affirmative. x 22 Approximately 270/000 students from the 6th grade were administered the Survey of Basic Skills of the California Assessment Program/ during the years 1977/ 1978/ and 1979. Responses were analyzed using log-linear models. Successful solving of computation items was positively associated with successful solving of story problems. Girls were more successful in computation than boys/ and boys were more successful than girls in solving story problems. This finding supports reports from the National Assessment of Educational Progress (NAEP) (Armstrong/ 1979). To investigate the effects of reading/ SES/ language/ and chronological age/ only those test forms containing 2 computation items and 2 story problems were considered for analysis; 32 items from 8 test forms were included in the analysis. The results of these analyses showed that at every level of reading score/ 6th grade children were more successful in computation than in story problems. Although the differences were not large/ at every reading score boys consistently had higher probabilities of success in story problems than did girls/ and girls consistently showed higher probabilities of success in computation than boys. Also/ as the reading score increased/ the difference between the probability of success in story problems and the probability of success in ACKNOWLEDGMENTS I would like to acknowledge ray gratitude to several people who have influenced my formal education and/or made this study possible. My special thanks to Dr. James Algina* Chairman of the doctoral committee* who contributed to the development of my love for research and statistics. He has been insuperable as professor and valued friend; his help and guidance in the preparation and completion of this study were invaluable. I extend my thanks to Dr. Linda Crocker for her advice and help during my doctoral studies at University of Florida. Thanks also go to Dr. Michael Nunnery* member of the doctoral committee. To Dr. Wilson Guertin* who was a friend for me and my family* I extend my special thanks. Thanks are also extended to Dr. Amalia Charneco* past Undersecretary of Education of the Puerto Rico Department of Education* for her continuous support. To my sister Nilda Santaella* who typed the thesis* I give my sincere thanks. Special thanks go to my family and to those friends who provided encouragement throughout this critical period of my life. ii 42 children whose errors fell under the category or "other errors." Students were asked to read each problem aloud and answer these questions: What kind of situation does the problem describe? What does the problem ask you to find? How would you work the problem? Ninety five percent of the students read the problem correctly; 98% explained the kind of situation the problem described in a correct manner; 92% correctly answered what the problem was asking them to find* and 36% correctly answered the question of how to work the problem. The fact that a large percent of the students whose errors were classified as "other errors" (in which reading skills might have been a factor) correctly stated how to work the problem is strong evidence of their ability to read and interpret the problems correctly. The errors made by this group of students had a distinct origin unrelated to reading ability. Zalewski (1974) investigated the relative contribution of verbal intelligence reading comprehension vocabulary interpretation of graphs and tables mathematical concepts number sentence selection and computation to successful mathematical word problem solution and the relationship of the dependent variable to the eight independent variables. She worked with a group of 4th grade children who were administered the subtests of the Iowa Test of Basic Skills (ITBS) and the Weschler Intelligence Scale for 64 gender-option dependency is the same from year to year. In this study/ a three dimensional table was considered to exhibit the same year to year pattern of gender-option dependency if / conditioned on gender/ the same proportion of students chose each incorrrect response over the three year period. This seems to be a straightforward and natural way to operationalize the concept. To illustrate how Marshall operationalized the concept in question/ a hypothetical set of probabilities was constructed conforming to the pattern specified by Marshall. This is displayed in Table 3.3. Table 3 .3 Hypothetical Joint Probabilities of Year/ Option/ and Sex Year Sex 1 Option 2 3 First M .120 .045 Â¡ .015 F .060 .060 Â¡ .030 Second M .144 .022 | .019 F .072 .030 Â¡ .039 Third M .132 .040 ! .016 F .066 .054 Â¡ J .039 Note: Probabilities reported are truncated to three places. 67 conducted/ one with each of addition/ subtraction/ multiplication/ division/ addition of fractions/ decimals in subtraction/ and equivalence as covariates. A well known problem that arises in the use of ANCOVA is that an unreliable covariate can cause spurious differences between the sex groups. To solve this problem/ Porter (1967/1968) proposed the use of estimated true scores for observed scores. Porter (1967/1968) conducted a simulation that gave empirical support to the adequacy of this strategy. Hunter and Cohen (1974) have provided theoretical support for this strategy. 104 to contribute to the body of knowledge regarding such differences in mathematical learning and achievement. This section of Chapter V comprises implications of the findings and recommendations evolving from the present investigation. Sex-related Differences in Incorrect Responses The main purpose of education is to impart formal education equally for the sexes. Analysis of the data on incorrect responses showed that for 100 of the 111 items of the test/ males and females selected different incorrect responses/ and this pattern of responses was consistently found during the three years of test administration. However/ with a sample of less than 3/000 subjects/ statistical significant results were obtained for only 6 items. Therefore/ it seems that/ at least for 6th grade children/ educators need not concern themselves with varying teaching techniques for the sexes. This study/ however/ has value apart from the investigation of sex-related differences. It was found that/ year after year/ 6th graders from the public schools in Puerto Rico tend to make the same errors in the solution of mathematical problems. It is important/ therefore/ that teachers provide the learning strategies necessary to enable students to obtain a better understanding of the concepts of numbers. Table Page 4.11 Adjusted Means on Problem Solving* by Covariate and Sex 96 4.12 ANCOVA Third Year 97 4.13 Adjusted Means on Problem Solving* by Covariate and Sex 98 vi 90 The significant interactions found between sex and multiplication* and sex and division (first year)* and sex and subtraction (second year)* answered* in part* the question of whether male-female differences in problem solving performance depend on computational ability. However* the evidence is quite weak. Of 21 possible interactions* only three were significant. No variable exhibited a significant interaction for each of the three years. The analysis of covariance was also used to determine if sex-related differences exist after controlling for computational skills. Analyses were conducted for those variables that did not exhibit significant interactions with sex. As discussed in Chapter III* estimated true scores were used for observed scores to adjust for unreliability of the covariates (the computational subtests). Reliability coefficients calculated for each covariate are shown in Table 4.7. Summaries of the analyses of covariance for the first year are reported in Table 4.8. The results show that females retained their superiority in problem solving performance when equivalence was the controlling variable in the analysis of covariance* the only variable in which males outperformed females (nonsignificant). When the controlling variables were addition* subtraction* addition of fractions* and subtraction with decimals* female 59 Inspection of the year by option contingency tables shows that year and option are independent for each gender. Thus/ hypothesis 1 is equivalent to the hypothesis that year and option are independent conditional upon gender. Hypothesis 2 is also true for Table 3.1 and Table 3.2. Therefore/ when hypothesis 1 is correct/ hypothesis 2 is equivalent to the hypothesis that sex and option choice are dependent. In his discussion of the analysis of three dimensional contingency tables/ Fienberg (1980) presents the following saturated model for the data: log m i j k = i + Uj + Uk+Uij + Uik+Ujk+U i j k. (3.1) In this model/ mijk is the expected value of the frequency in cell ijk of the three dimensional table. The model states that all three classification factors for a three dimensional contingency table are mutually dependent. In the present research i is the year index/ j is the option index/ and k is the sex index. Fienberg shows that deleting the terms Uij and Uijk yields a model in which year and option are independent conditional upon sex. This model is log mij k Ui + Uj + Uk+Uik+Uj k. (3.2) 75 TABLE 4.2 Chi-square Values for the Comparison of Models (3.2) and (3.3) with Actual Sample Sizes and with the Corresponding Number of Subjects Needed for Significant Results Item Num ber Xa- for Model 3.2 X1 for Model 3.3 Xa- for Difference Actual Sample Size Minimum Sample Size for Signif. (a) (b) (c) (d) (e) (f) 1 16.11 30.94 14.83* 7*427 10*061 2 17.54 103.22 85 .68* 11*018 2*583** 3 20.06 47.51 27.45* 11*497 8*733 4 7.54 21.87 14.33* 6*016 8*414 5 14.75 70.40 55.65* 7*706 2*782** 6 11.86 23 .13 11.27* 13*046 23 *256 7 4.60 35.73 31.13* 8*832 5*700 8 46.15 329.98 283.83* 16*796 1*189** 9 10.13 109.99 99.86* 23 *075 4*642 10 9.54 65.29 55.75* 35*060 12*634 11 5.89 19.06 13.17* 24*908 37 *996 12 12.58 198.46 185.88* 40 *423 4*369 13 25.74 62.37 36.63* 15*718 8*621 14 11.19 80 .50 69.31* 30*238 8*765 15 16.63 17.33 0.70 42*889 1*230*914 16 14.51 84.29 69.78* 35*114 10*109 17 17.90 44.20 26 .30* 43 *962 33 *582 41 categories. Category I included clerical and computational errors. Category II included other types of errors such as average and area errors/ use of wrong operation/ no response/ and erred responses offering no clues. It the student's work indicated the correct procedure and yet the problem was missed because of a computational or clerical error/ it was assumed that the problem was read and understood. An analysis of frequencies showed that clerical errors were responsible for 3% of the problems incorrectly solved/ computational errors accounted for 49%/ and other errors for 48% of the erred problems. Knifong and Holtan concluded that "improved computational skills could have eliminated nearly half ot the word problem errors" (p. Ill). These computational errors were made in a context where other skills such as reading/ interpretation of the problem/ and integration of these skills necessary for the solution of word problems/ might interact. However/ Knifong and Holtan state that their findings neither confirm nor deny that improvement ot reading skills will lead to improvement in problem solving. They conclude that "it is difficult to attribute major importance to reading as a source of failure" (p. 111). In a later analysis/ looking for evidence of poor reading abilities affecting children's success in word problems/ Knifong and Holtan (1977) interviewed the 69 are independent/ conditional upon sex* log m j j k = U + U j + U k+U k+U j k (3.2) Substantively/ this model implies that the pattern of male and female option choices is consistent over the three years of test administration. The second model indicates that option is independent of sex/ log mijk = Ui+Uj+Uk+Uik. (3.3) This model implies that the pattern of option choice is the same for males and females. In order to determine if males and females differed in the selection of incorrect responses and if these differences were stable across the three years of test administration/ a two-step test was performed. First/ a likelihood ratio test of the adequacy of model (3.2) was conducted for each of the 111 items of the test. Under this test a model fits the data when the value obtained is nonsignificant at a specified alpha level. For 50 (45%) of the items/ model (3.2) fitted the data adequately. For these items the pattern of male and female option choices was consistent for the three years. The X^ values for the 61 items for which the likelihood ratio 97 TABLE 4.12 ANCOVA Summary Table Third Year Source df ss MS F Addition 1 205.96 205.96 34.85* Sex 1 1.88 1.88 .32 Error 1010 5967.60 5.91 Subtraction 1 886.42 886.42 169.48* Sex 1 24.39 24.39 4.66* Error 1010 5287.2 5.23 Multiplication 1 1402.42 1402.42 297.12* Sex 1 1.05 1.05 .22 Error 1010 4771.20 4.72 Division 1 2028.49 2028.49 494.75* Sex 1 1.11 1.11 .27 Error 1010 4145.10 4.10 Fracadd 1 1733.72 1733.72 393.97* Sex 1 4.64 4.64 1.05 Error 1010 4439.89 4.40 Decsub 1 2116.84 2116.84 526.57* Sex 1 .15 .15 .04 Error 1010 4056.76 4.02 Equivalence 1 1186.92 1186.92 240.26* Sex 1 59.85 59.85 12.12* Error 1010 4986.69 4.94 * p<. 05 66 Table 3.4 Hypothetical Probabilities of Option Choice Conditional on Sex and Year Year Sex 1 Option 2 3 First M .666 .250 .083 F .400 .400 .200 Second M .774 .120 .104 F .510 .212 .276 Third M .698 .214 .087 F .431 .352 .215 Note: Probabilities reported are truncated to three places. Which procedure is more appropriate to test the claim that the pattern of male and female option choice remains the same from year to year? It seems more reasonable to test for a pattern like that in Table 3.2 than to test for a pattern like that in Table 3.4 / and/ consequently/ this was the strategy adopted in this study. Comparison of Males and Females in Problem Solving Performance One object of the study was to compare the performance of males and females on problem solving. Two questions were addressed. Firstr do males and females differ in problem solving performance? Second/ do these differences persist when computational skill is controlled for/ and do these differences depend on the level of computational skill? Seven analyses of covariance (ANCOVA) tests were 85 TABLE 4.4 ANCOVA Summary Table: Multiplication Covariate First Year Source df SS MS F Multiplication (M) 1 1195.90 1195.90 265.66* Sex (S) 1 31.95 31.95 7.07* M x S 1 38.80 38.80 8.59* Error 998 4509.00 4.51 * p <;.oi TABLE 4.5 ANCOVA Summary Table: Divison Covariate First Year Source df SS MS F Division (D) 1 1195.90 1195.90 264.66* Sex (S) 1 13.94 13.94 3.22 D x S 1 18.39 18.39 4.25* Error 998 4317.00 4.32 * p<.01 64 gender-option dependency is the same from year to year. In this study/ a three dimensional table was considered to exhibit the same year to year pattern of gender-option dependency if/ conditioned on gender/ the same proportion of students chose each incorrrect response over the three year period. This seems to be a straightforward and natural way to operationalize the concept. To illustrate how Marshall operationalized the concept in question/ a hypothetical set of probabilities was constructed conforming to the pattern specified by Marshall. This is displayed in Table 3.3. Table 3.3 Hypothetical Joint Probabilities of Year/ Option/ and Sex Year Sex 1 Option 2 3 First M .120 .045 | .015 F .060 .060 Â¡ .030 Second M .144 .022** .019 F .072 .030 ; .039 Third M .132 .040*1 .016 F .066 .054 Â¡ .039 Note: Probabilities reported are truncated to three places. 102 response was related to the use of a large number of subjects in the analysis. One must question whether analysis of incorrect responses has any educational value. Although the results of this study do not support the use of this analysis (error analysis) for the study of sex-related differences* it continues to be a promising technique in the discovery of how children approach mathematics problem solving* more so than analyses that look only at the items answered correctly or at total test scores. The second objective of the study was to investigate sex-related differences in problem solving performance* to find out if significant differences persisted after controlling for computational skills* and if the differences depended on the level of computational skills. The results showed that females outperformed males in problem solving and in six of the seven computational skills. Males showed superiority in equivalence in each of the three years* but statistical significance was obtained only for the second year. The results also showed that females retained their superiority in problem solving* only when equivalence (for the three years data) and subtraction (for the second year) were the controlling variables. Analysis of covariance was also used to answer the question of whether male-female differences in problem solving depend on the level of computational skills. The 12 test form/ no sex differences were found in response patterns. Approximately 45% of each sex selected option c. The next popular choice for both sexes was option d/ selected by approximately 35% of both boys and girls. On the story problem of Form If males and females responded alike. Their most popular incorrect response choice was option a for both males and females. The second most popular incorrect choice was option c for both sexes. Response to the story problem in Form 2 showed sex differences in response choice. Including the correct option/ 33% of the girls selected option a./ 20% chose option C/ and 20% option d. For males/ approximately 25% selected option a and the same percent selected option d. Marshall concluded that although the analysis of incorrect responses does not explain why boys and girls differ in their responses/ the analysis shows that boys and girls approach problems in different ways and these varying strategies can be useful in indentifying how the sexes differ in reasoning abilities. Two years later/ Marshall (1983) analyzed the responses of approximately 300/000 boys and girls to mathematics items contained in the 16 test forms of the Survey of Basic Skills during the years 1977/ 1978/ and 1979. She used log-linear models (explained in Chapter 42 children whose errors fell under the category or "other errors." Students were asked to read each problem aloud and answer these questions: What kind of situation does the problem describe? What does the problem ask you to find? How would you work the problem? Ninety five percent of the students read the problem correctly; 98% explained the kind of situation the problem described in a correct manner; 92% correctly answered what the problem was asking them to find* and 36% correctly answered the question of how to work the problem. The fact that a large percent of the students whose errors were classified as "other errors" (in which reading skills might have been a factor) correctly stated how to work the problem* is strong evidence of their ability to read and interpret the problems correctly. The errors made by this group of students had a distinct origin* unrelated to reading ability. Zalewski (1974) investigated the relative contribution of verbal intelligence* reading comprehension* vocabulary* interpretation of graphs and tables* mathematical concepts* number sentence selection* and computation to successful mathematical word problem solution* and the relationship of the dependent variable to the eight independent variables. She worked with a group of 4th grade children who were administered the subtests of the Iowa Test of Basic Skills (ITBS) and the Weschler Intelligence Scale for 34 achievement and interest in mathematics result from identification with the masculine role. A study related to the differential cultural reinforcement hypothesis is that of Dwyer (1974). Dwyer examined the relationship between sex role standards (the extent to which an individual considers certain activities appropriate to males or females) and achievement in reading and arithmetic. Students from grades 2/ 4 > 6 > 81 10 r and 12 participated in this study. She found that sex role standards contributed significant variance to reading and arithmetic achievement test scores and that the effect was stronger for males than for females. This led to her conclusion that sex-related differences in reading and arithmetic are more a function of the child's perception of these areas as sex-appropriate or sex-inappropriate than of the child's biological sex individual preference for masculine and feminine sex roles/ or liking or disliking reading or mathematics. In a study which agrees with the masculine identification hypothesis* Milton (1957) found that individuals who had received strong masculine orientation performed better in problem solving than individuals who received less masculine orientation. Elton and Rose (1967) found that women with high mathematical aptitude and average verbal aptitude scored higher on the masculinity scale of the Omnibus Personality Inventory (OPI) than those with average scores on both tasks. 67 conducted* one with each of addition* subtraction* multiplication* division* addition of fractions* decimals in subtraction* and equivalence as covariates. A well known problem that arises in the use of ANCOVA is that an unreliable covariate can cause spurious differences between the sex groups. To solve this problem* Porter (1967/1968) proposed the use of estimated true scores for observed scores. Porter (1967/1968) conducted a simulation that gave empirical support to the adequacy of this strategy. Hunter and Cohen (1974) have provided theoretical support for this strategy. 103 question was answered/ partially/ in the affirmative. Interactions/ indicating the dependence/ were significant for multiplication and division in the first year and for subtration in the second year. In general/ the sex-related differences in mathematics problem-solving/ in favor of males performance/ reported by Marshall (1981/ 1984)/ and Fennema and Tartree (1983) were not supported in this investigation. Female performance was equal or higher than male performance in problem solving. The findings of the study have something in common with other investigations where mathematics problem solving has been the subject of interest/ the fact that males and females do poorly in word-story problem items (problem solving) The mean score in problem solving for males and females in the 9-item problem solving subtest was 3.436 and 3.862 respectively for the first year/ 3.632 and 4.015 for the second year/ and 3.927 and 4.341 for the third year. Implications of the Findings and Suggestions for Further Research The main purpose of this study was to investigate sex-related differences in the selection of incorrect responses. Another objective was to compare male and female performance after accounting for computational ability. This type of research is conducted in an attempt 46 Their study is important because it represents an attempt to demonstrate that multiple factors can interact in the correct solution of a mathematics word problem. They constructed three graded tests from a basal mathematics series for grades 3 through 8. For test lr the problems were set in pure computational form (the effects or reading/ interpretation/ as well as the necessity for integration were removed in an effort to measure the computational skills required by the word problems) . For test 2/ the effects of reading and computation were removed by reading the problems to the students and by giving scores based on whether or not the students set them up properly/ in an attempt to measure problem interpretation alone. For test 3/ the effect of computation was removed. The test yielded two scoresone by grading the students on whether or not they set up the problems properly and another by grading on the basis of the correct answer. The tests were administered to all 244 students from the 6th grade in two different schools. A diagnostic profile was obtained for each of the 217 students for which complete data were available: a computational score/ a probiem-interpretation score/ a reading score/ and a reading-problem solving score. They assumed that if the reading-prob1em interpretation score was lower (one or more levels lower) 106 operation to the numbers in the problem. The re sults indicate that students are not familiar with such basic problem-solving strategies as drawing a picture of a figure described in a problem or checking the reasonableness of a result, (p. 338) The fact that, in general after controlling for computational skills there were no statistically significant differences between the problem solving performance of the males and females who participated in this study would seem to imply that teachers need not worry about designing different teaching strategies for the sexes. However^ because the performance in problem-solving is low for both sexes teachers may be encouraged to emphasize problem solving in their daily teaching. This study was designed to investigate sex-related differences in mathematical performance. The findings clearly demonstrated that 6th grade females and males from the public schools in Puerto Rico are equally good or equally bad in the solution of word-story items (problem solving). Also/ in general/ the effect of computational skills is similar for each sex. Based on these findings/ it appears that greater differences exist within each sex than between the sexes. In conclusion/ the findings of the present study highlight the need for investigations of different approaches to instructional designs aimed at improving both male and female performance in problem solving. 32 attributes success to an internal, stable attribute, such as ability, then one is confident of being successful in the future and will continue to strive in that area. If one attributes success to an external factor such as a teacher, or to an unstable one, such as effort, then one will not be as confident of success in the future and will cease to strive. Failure attribution patterns work this way: if failure is attributed to unstable causes, such as effort, failure can be avoided in the future and the tendency will be to persist in the task. However, if failure is attributed to a stable cause, such as ability, the belief that one cannot avoid failure will remain. Studies reported by Bar-Tal and Frieze (1977) suggest that males and females tend to exhibit different attributional patterns of success and failure. Males tend to attribute their success to internal causes and their failures to external or unstable ones. Females show a different pattern; they tend to attribute success to external or unstable causes and failures to internal ones. The pattern of attributions, success attributed externally and failure attributed internally, has become hypothesized to show a strong effect on mathematics achievement in females. K1oosterman' s (1985) study supported these findings. According to Kloosterman, attributional variables appear to be more important achievement mediators for females than for males, as measured by mathematics word problems. More research is needed in this area. 108 Bern/ S. L. / & Bern D. V. (1970). We are all nonconscious sexists. Psychology Today/ 22-24, 261 115-116. Benbow/ C. P./ & Stanley/ J. C. (1980). Sex differences in mathematical ability: Fact or artifact? Science/ 210/ 1262-1264. Benbow/ C. P./ & Stanley/ J. C. (1982). Consequences in high school and college of sex differences in mathema tical reasoning ability: A longitudinal perspective. American .Educational Research Journal/ 12./ 598 622. Bobbie/ C. N. (1971) Sex-role preference and academic achievement (Doctoral dissertation/ Yeshiva Univer sity) Dissertation Abstg^ 1818B-1819B. Carey/ G. L. (1958) Sex differences in problem solving performance as a function of attitude differences. Journal Of Abnormal. $tn.d Social psychology/ 22/ 256-260. Carpenter/ T. P./ Corbitt/ M. K./ Kepner/ H. S./ Lindquist/ M. M./ & Reys/ R. E. (1980). Results of the second NAEP mathematics assessment: Secondary school. Mathematics Teacher/ 22/ 329-338. Carpenter/ T. P. / Lindquist/ M. M./ Mathews/ W./ & Silver/ E. A. (1984). Results of the third NAED mathematics assessment: Secondary school. Mathematics Teacher/ 22/ 652-659. Chase/ C. I. (1960). The position of certain variables in the prediction of problem solving in arithmetic. Journal of Educational Research/ 14./ 9-15. Cohen/ D./ & Wilkie/ F. (1979). Sex-related differences in cognition among the elderly. In M. A. Witting and A. C. Petersen (Eds.)/ Sex-related differences in C.ognjtjve_J.unct,iojiin.qj Developmental Issues/ New York: Academic Press. Connor/ J. M. / Serbin/ L. A./ & Schakmanz M. (1977). Sex differences in children's response to training on a visual-spatial test. Developmental Psychology/ 12/ 392-394. CHAPTER IV RESULTS Introduction The data gathered from 6th grade students from the public schools in Puerto Rico who took the Basic Skills Test in Mathematics-6 during three consecutive years were analyzed in this study. The first objective was to investigate whether boys and girls differ in the selection of incorrect responses/ and if the pattern of differences was consistent throughout the three years in which the test was administered. The second objective was to investigate whether males and females differ in problem solving performance/ if these differences persist after accounting for computational skills/ and if the male-female differences depend on the level of computational skills. The results are discussed in two sections. Study findings in the area of sex differences in the selection of incorrect responses are discussed in the first section. The second section is devoted to an exposition of the findings in the area of sex-related differences in problem solving. Sex-Related Differences in the Selection of ,Incorr_egt.. Response As indicated in Chapter III/ there are two models of interest. The first model indicates that year and option 68 94 TABLE 4.9 Adjusted Means on Problem Solving* by Covariate and Sex First Year Covariate Adjusted Means Males Females Addition 3.53 3.78 Subtraction 3.55 3.76 Fracadd 3.56 3.74 Decsub 3.73 3.58 Equivalence 3.38 3.91 REFERENCES Aiken/ L. R. (1970). Attitudes toward mathematics. Review of Educational Resarch/ IQ./ 551-596 . Allen/ R. H./ & Chambers/ D. L. (1977). A comparison of the mathematics achievement of males and females (Bulletin No. 9194). Madison: Wisconsin State Department of Public Instruction/ Division for Management and Planning Service. (ERIC Document Reproduction Services No. ED 159 073) Armstrong/ J. M. (1979) Achievement and participation of W.pme_n_in_.math.ematiÂ£sj Ln_Q_y.ejyi-e.W-. (Report of a two year study). Denver/ CO: Educational Commission of the States/ National Assessment of Educational Progress. (ERIC Document Reproduction Service No. ED 184 878) Armstrong/ J. M./ & Price/ R. A. (1982). Correlates and predictors of women's mathematics participation. Journal for Research in Mathematics Education/ 12/ 99-109. Astin/ H. S. ( 197 4) Overview of the findings. In H. Astin/ H. Suniewick/ & S. Dweck (Eds.) / Women: A Bibliography on their Education and Careers. (pp. 1-10). New York: Behavorial Publications. Backman/ M. E. (1972). Patterns of mental abilities: Ethnic/ socioeconomic and sex differences. American Educational Research Journal/ 1/ 1-11. Ballew/ H./ & Cunningham/ W. (1982). Diagnosing strengths and weaknesses of sixth-grade students in solving word problems. Journal for Research in Mathematics Education/ 13/ 202-210. Balow/ I. H. (1964). Reading and computation ability as determinants of problem solving. The Arithmetic Teacher/ 11/ 18-22. Bar-Tal / D. / & Frieze/ J. H. ( 1977 ) Achievement motivation for males and females as a determinant of attribution for success and failure. Sex roles. 1/ 301-313. 107 83 TABLE 4.3 Continued Year/ Subtest Males X SD Females X SD t Decsub 2.863 1.87 3.440 1.80 5.00* Equivalence .76 .76 .65 .75 -2.32** N = 504 N = 509 Third Problem Solving 3.927 2.49 4.341 2.44 2.67* Addition 5 .536 .84 5.704 .64 3.59* Subtraction 4.836 1.52 4.958 1.46 1.30 Multiplication 4.168 1.74 4.541 1.68 3 .47* Division 3.343 1.88 3.795 1.83 3.87* Fracadd 2.819 1.89 3 .117 1.85 2.53** Decsub 3.021 1.93 3.448 1.85 3.60* Equivalanee .830 .82 .800 .78 - .60 N = 509 N = 504 Note: The number of items in the problem solving subtest was 9 . In each computation subtestr the number of items was 6. An item was included in the computation subtest only if it measured a computation skill required to solve a problem solving item. * ** p <.01 P <.05 38 At present# no set of variables has been clearly established as a determinant of problem difficulty. Several researchers have investigated the effect of reading and computation on problem solving performance. Others have studied the effect of student attitudes toward problem solving in problem solving learning and achievement. Typically# correlational methods have been used to investigate these questions. Computational_S_k_U.ls, and_.Pr,QbJ.etn Solving Performance One of the first researchers to study the effect of computation and reading on problem solving performance was Hansen (1944). He investigated the relationship of arithmetical factors# mental factors# and reading factors to achievement in problem solving. Sixth grade students were administered tests in problem solving and categorized as superior achievers (best problem solvers) and inferior archievers (poorest problem solvers) The two groups were compared in selected factors believed to be related to success in arithmetic problem solving: arithmetical# mental and reading factors. After controlling for mental and chronological age# the superior achievers in problem solving surpassed the inferior achievers in mental and arithmetical factors. The superior group did better in only two of the six items under the reading factors: general language ability and the reading of graphs# charts# and tables. 58 Table 3.1 Hypothetical Probabilities of Option Choice Conditional on Year and Sex Year Sex 1 Option 2 3 First M .7 .2 .1 F .5 .3 .2 Second M .7 .2 .1 F .5 .3 .2 Third M .7 .2 .1 F .5 .3 .2 Hypothetical Conditional on Table 3.2 Probabilities of Year and Sex and Option Choice Arranged by Sex Option Sex Year 1 2 3 M First .7 .2 .1 Second .7 .2 .1 Third .7 .2 .1 F First .5 .3 .2 Second .5 .3 .2 Third .5 .3 .2 5 Females were better on items of computation and males were more successful on word-story problem items (problem solving). She also found that females successful performance in the problem solving items was more dependent on their successful performance in the computation items. Males did not need/ as much as females/ to succeed in the computation items in order to answer correctly the problem solving items. Although the general findings seem to support sex differences in mathematics learning and achievement/ the research done does not consistently support superiority for either sex. Most of the research has been concerned with how the sexes differ on subtests or total test scores in mathematics. Moreover/ the great majority of the studies deal with correct responses. Sex differences in incorrect responses at the item level have not been fully researched. Only two studies dealing with sex differences in incorrect responses at the item level were found in the research literature (Marshall/ 1981/ 1983). Marshall investigated whether boys and girls made similar errors in computation and story problems. She analyzed boys' and girls' answers to six mathematics items and found that the sexes made different errors/ possibly reflecting different problem solving strategies. Her original findings were supported when she studied the same problem using a large number of items three years later. Problem Solving 87 9 . 8 7 6 . 5 - 4 - 3 - 2 - 1 - T 1 1 1 I 2 3 4 5 6 Multiplication Fig. 4.1 Sex by Multiplication Interaction 78 TABLE 4.2 Continued Item Num ber X1 for Model 3.2 Xa- for Model 3.3 X* for Difference Actual Sample Size Minimum Sample Size for Signif. (a) . (b) (c) (d) (e) (f) 55 19.53 90.87 71.34* 41/796 11/770 56 21.62 70.14 48.52* 44/107 18/263 57 28.51 35.78 7.27 53/075 146/668 58 17.49 51.69 34.20* 44/044 25/873 59 50.96 73.68 22.72* 41/829 36/987 60 66.43 148.39 81.96* 60/494 14/828 61 21.39 394.54 373.15* 35/632 1/918** 62 55.03 300.95 245.92* 50/699 4/142 63 158.36 278.34 119.98* 52/068 8/719 64 88.01 246.48 158.47* 73/196 9/279 65 56.28 94.11 37.83* 42/429 22/532 66 50.51 113.88 63.37* 78/708 24/953 67 34.28 180.60 146.32* 51/862 7/121 68 48.01 225.46 177.45* 63/223 7/158 69 75.29 251.66 176.37* 62/511 7/121 70 44.39 271.35 226.96* 67/528 5/977 72 17.41 208.90 191.49* 84/175 8/831 73 24.61 37.80 13.19* 53/358 81/271 74 22.75 163.01 140.26* 84/175 12,057 66 Table 3.4 Hypothetical Probabilities of Option Choice Conditional on Sex and Year Year Sex 1 Option 2 3 First M .666 .250 .083 F .400 .400 .200 Second M .774 .120 .104 F .510 .212 .276 Third M .698 .214 .087 F .431 .352 .215 Note: Probabilities reported are truncated to three places. Which procedure is more appropriate to test the claim that the pattern of male and female option choice remains the same from year to year? it seems more reasonable to test for a pattern like that in Table 3.2 than to test for a pattern like that in Table 3.4* and* consequently* this was the strategy adopted in this study. C-Omparisoa of_Males_and Females in Problem Solving Pejfbrmance One object of the study was to compare the performance of males and females on problem solving. Two questions were addressed. First* do males and females differ in problem solving performance? Second* do these differences persist when computational skill is controlled for* and do these differences depend on the level of computational skill? Seven analyses of covariance (ANCOVA) tests were BIOGRAPHICAL SKETCH Sonia Feliciano was born on March 17/ 1938/ in Mayaguez/ Puerto Rico. She graduated from Eugenio Ma. de Hostos High School/ in her hometown. She received a degree in business administration/ with a major in finance/ from the University of Puerto Rico in June/ 1961. After graduation she worked as junior accountant for Pittsburgh Plate Glass/ International. During the academic year of 1969-70 she started her career in education/ obtaining the degree of Master in Education from the University of Puerto Rico in June/ 1975. From January/ 1971 to June/ 1985/ she served in the public school system of Puerto Rico in different positions. She worked as educational researcher until July/ 1977/ when was appointed Director of the Research Center of the Commonwealth Department of Education. During the academic year of 1978-79/ she initiated doctoral studies in the Faculty of Philosophy and Educational Sciences of the Universidad Complutense de Madrid/ in Madrid/ Spain. She completed the coursework requirements and then transferred to University of Florida during the fall of 1979. 116 27 Research findings in this area have been inconsistent. In 1966 / Maccoby stated that "by early school years/ boys consistently do better (than girls) on spatial tasks and this difference continues through the high school and college years" (p.26). In 1972/ Maccoby and Jacklin said that the differences in spatial ability between the sexes "remain minimal and inconsistent until approximately the ages of 10 or 11/ when the superiority of boys becomes consistent in a wide range of populations and tests" (p- 41) In 1974 / after a comprehensive literature search/ Maccoby and Jacklin concluded that sex differences in spatial visualization become more pronounced between upper elementary years and the last year of high school/ the years when sex-related differences in mathematics achievement favoring boys emerge. Guay and McDaniel (1977) supported in part Maccoby and Jacklin's 1974 findings. They found that among elementary school children/ males had greater high level spatial ability than females/ but that males and females were equal in low level spatial ability. This finding is inconsistent with that portion of Maccoby and Jacklin's review that suggests that sex differences become evident only during early adolescence. Cohen and Wilkie (1979) however/ stated that in tests measuring distinct spatial tasks/ males perform better than females in early adolescence and throughout their life span. Most studies carried out after 36 they show less interest in the subject than counterpart males. These differences in interest are what Hilton and Berglund (1974) suggest to account for sex-related differences in mathematics achievement. Although the perception of the usefulness of mathematics is still an important predictor of course taking for girls there is a growing similarity between males and females regarding the usefulness of mathematics (Armstrong & Price/ 1982; Fennema & Sherman/ 1977; Moller/ 1982/1983) Armstrong and Price investigated the relative influence of selected factors in sex-related differences in mathematics participation. Both males and females selected usefulness of mathematics as the most important factor in deciaing whether or not to take more mathematics in high school. Moller's study revealed that both males and females based mathematics course-taking decisions on career usefulness. A Fennema and Sherman (1977) study showed only slight differences between males and females in their feelings about the usefulness of mathematics. In her study of this variable among college students/ Probert (1983/1984) did not find any sex-related differences either. These have been the main affective variables researched in attempting to explain the underlying causes of sex-related differences observed in mathematics learning and achievement. In spite of the great diversity of studies TABLE OF CONTENTS Page ACKNOWLEDGMENTS LIST OF TABLES V LIST OF FIGURES V ABSTRACT viii CHAPTER I INTRODUCTION 1 Purpose of the Study 6 Significance of the Study 6 Organization of the Study 8 II REVIEW OF THE LITERATURE 9 Sex-related Differences in Incorrect Response Patterns 10 Sex-related Differences in Problem Solving 14 Cognitive and Affective Variables that Influence Sex Differences in Mathematics Learning and Achievement 23 Differences in Formal Mathematics Education 24 Differences in Spatial Ability 26 Differentiated Effect of Affective Variables 30 Problem Solving Performance and Related Variables 37 Computational Skills and Problem Solving Performance 38 Reading and Problem Solving Perfor mance 44 Attitudes Toward Problem Solving and Problem Solving Performance 50 iii CHAPTER I INTRODUCTION Sex differences in mathematics learning and achievement have been the subject of intensive research. Research done before 1974 has shown that male performance on mathematical achievement tests is superior to female performance by the time they reach upper elementary or junior high school (Fennema/ 1976/ p. 2). The literature strongly suggests that at the elementary level females outperform males in computation and males excel in mathematical reasoning (Glennon & Callahan/ 1968; Jarvis/ 1964; Maccoby/ 1966). Since 1974/ research findings have been less consistent. Fennema (1974)/ after reviewing 36 studies/ found that during secondary school or earlier/ sex-related differences in mathematics achievement are not so evident/ but that when differences are found/ they favor males in high level cognitive tasks (problem solving) and females in low level cognitive tasks (computation) As a result of a further review of the literature/ Fennema (1977) concluded that at the elementary level/ sex-related differences do not exist at all cognitive levels/ from computation to problem solving. 1 S^x^-elajted-Diiferences in Problem Solving Performance 81 In this part* results for the two questions related to the second objective of the study are presented. These questions serve as the framework for the presentation. Each question is stated* followed by the results pertaining to that question. Question 1: Do males and females differ in problem solving performance? The responses of 492 males and 510 females who took the Puerto Rico Basic Skills Test-6 during the spring of the first year were analyzed in this study. Also* data from 504 males and 509 females tested in the second year and from 509 males and 504 females tested in the third year were included in the analysis. The mean performance scores and the standard deviations for each of the eight variables are presented in Table 4.3. Results of t-tests are also presented in this table. Females outperformed males in problem solving* a finding consistently present in all the three years of test administration. Over the three-year period the mean differences favored females in all variables except equivalence. The sex-related differences in problem solving were significant (p<.01) for all three years. 86 TABLE 4.6 ANCOVA Summary Table: Subtraction Covariate Second Year Source df SS MS F Subtraction (S) 1 93 .14 593 .14 122.80* Sex (S) 1 18.58 18.58 3 .85* SxS 1 30.85 30.85 6.39* Error 1009 4873.40 4.83 *P < 05 14 jLexrxe-latÂ£_d_Differ enees in Problem Solving It has already been acknowledged that the subject of problem solving has been extensively researched. However/ as early as 1969/ Kilpatrick criticized the fact that the study of problem solving has not been systematic? some researchers have studied the characteristics of the problem while others have given their attention to the characteristics of the problem solvers. Moreover/ differences in the tests used to measure problem solving performance also constitute an obstacle when trying to compare the results of the studies carried out. In order to avoid this pitfall and provide a basis for comparison/ the studies reviewed in this section/ dealing with sex-related differences in problem solving/ have been divided in two groups. The first comprises those studies that used the Romberg-Wearne Problem Solving Test. The second contains other relevant studies in which problem solving performance has been measured by means of other instruments. The Romberg-Wearne Problem Solving Test merits special mention because it was the first attempt "to develop a test to overcome the inadequacies of total test scores in explaining the reasons why some students are successful problem solvers and others are not" (Whitaker/ 1976/ pp. 9/ 10) 63 model tests were conducted using a .01 level of significance. Since this research is based on Marshall (1981/ 1983)/ it is important to compare the method of analysis used in this study to the one used by Marshall. Marshall also used a two-step analysis. In the first step of her analysis she deleted the Uijk term from (3.1) and tested the adequacy of the model/ 1 g m j. j k = i + U j + o k+ U i j + Ui k+ j k (3.6) Following this/ she deleted the Ujk term to obtain log m j k = (Ji + Uj + Uk+Ui j + Uik. (3.7) and compared these two models using a likelihood ratio test. If the first test was nonsignificant and the second significant/ Marshall claimed that option choice was dependent on sex and the pattern of dependency was the same from year to year. This is the same claim that this study sought to establish. However/ the approach used here was to present evidence that model (3.2) fits the data while Marshall tried to show that model (3.6) fits the data. The major difference between the two approaches concerns the operationalization of the concept that the 16 of the test. The sexes differed in only 2 of the 19 reference tests Spatial Relations and Picture Group Name-Selection. A factor analysis however showed differences in the number and composition of the factors. For females/ a general mathematics factor was determined by mathematics computation/ comprehension/ application/ and problem solving. For males/ the comprehension and application parts determined one factor; problem solving with two other reference tests (Gestalt and Omelet) determined another factor. Meyer concluded that comprehension of the data and mastery of the prerequisite mathematical concepts did not guarantee successful problem solving either for males or for females. Problem solving scores for both sexes were about one third their scores in comprehension and one fourth their scores in application. She also concluded that the sexes may have approached the problem solving questions differently. The methods used by females for solving problem situations may have paralleled their approach to the application parts. Males may have used established rules and algorithms for the application parts/ but may have used more of a Gestalt approach to the problem solving situation. Whitaker (1976) investigated the relationship between the mathematical problem performance of 4th grade children and their attitudes toward problem solving/ their teachers' 45 He found that in the 4th grade the correlations between problem solving and abstract verbal reasoning/ reading comprehension/ arithmetic concepts/ and computation were .61/ .64/ .66/ and .60 respectively/ and .56/ .68/ .69/ and .63 in the 8th grade. When computation was held constant/ the correlation between problem solving and reading was .52 in grade 4 and .54 in grade 8. When reading was held constant the correlation between problem solving and computation was .43 in grade 4 and .42 in grade 8. Creswell (1982) worked with a sample of anglo and black adolescents from Chicago. Each subject was administered the California Achievement test. Multiple regression was used to analyze the data. The analysis showed that reading is more important than computation in predicting student performance in problem solving. Reading accounted for 49.5% of the variance; computation accounted for 14.6% of the variance. Ballew and Cunningham (1982) worked with 6th grade students in an attempt to find what proportion of students have as their main source of difficulty with word problems each of the following factors: a) computation skills/ b) interpretation of the problem/ c) reading and/ d) integrating these skills in the solution of problems. They also wanted to know if a student can be efficiently diagnosed as having one of the four categories as his/her main difficulty with mathematics word problems. 112 Landau* M. (1984). The effects of spatial ability and problem presentation format in mathematical problem solving performance on middle school students (Doctoral dissertation* Northwestern University). Dissertation. Abstracts., mteEP3ti.onal* * 442a- 443 A. Lindgren* H. C. Silva* I.* Faraco* I.* & DaRocha* N. S. (1964). Attitudes toward problem solving as a function of success in arithmetic in Brazilian elementary schools. Journal of Educational Research* 5&* 44-45. Maccoby* E. E. (1966) Sex differences in intellectual functioning. In E. E. Maccoby (Ed.)* The Development of Sex Differences (pp. 25-55) Stanford* CA: The Stanford University Press. Maccoby* E. E. & Jacklin* C. N. (1972). Sex differences in intellectual functioning. In A. Anastasi* (Ed.)* Asseestnent.in a pluralistic, sopi-elx* (pp. 37-55). Proceedings of the 1972 Invitational Conference on Testing Problems* Educational Testing Service* Princeton* NJ. Maccoby* E. E.* & Jacklin* C. N. (1974). The Psychology of Sex Differences* Stanford* CA: Stanford University Press. Marshall* S. P. (1981) Sex Differences in Sixth Grade Children's Problem Solving. Paper presented at the annual meeting of the American Educational Research Association* Los Angeles* CA. Marshall* S. P. (1983). Sex differences in mathematical errors: Analysis of distractor choice. Journal for Research in Mathematics Education* 14.* 325-336. Marshall* S. *P. (1984). Sex differences in children's mathematics achievement: Solving computations and story problems* Journal. Qf Ed.ucati.Q.P.^l_Pgy.ch.Q.l.P.gy* 2Â£* 194-204. Martin* M. D. (1964). Reading comprehension* abstract verbal reasoning and computation as factors in arithmetic problem solving (Doctoral dissertation* University of Iowa* 1963) pissertatiPn_Ab.strac.t.5 International* 24 4547A-4548A. Problem Solving 87 9 8 - 6 5 - 4 - 3 - 2 - 1 - T 1 I 1 1 1 1 2 3 4 5 6 Multiplication Fig. 4.1 Sex by Multiplication Interaction. 24 learning and achievement arise from sex differences in spatial visualization; and that sex-related differences result from a differentiated effect of affective variables on the mathematics performance of males and females. D.,iffÂ£JC-epcgg.. in. F,Qfmal_ M3th.em.3tics_JÂ¡ducation (Differential Coursework Hypothesis) The basis for the differential coursework hypothesis is the fact that sex-related differences in mathematical learning and achievement show up when comparing groups which are not equal in previous mathematics learning. After the 8th grade/ boys tend to select mathematics courses more often than girls. Therefore/ girls show lower achievement scores in mathematics tests because their mathematics experience is not as strong as the boys' (Fennema/ 1975; Fennema & Sherman/ 1977; Sherman/ 1979). Fennema and Sherman's study (1977) lends additional support to the feasibility of viewing sex differences in mathematics learning and achievement as reflecting something other than a difference in mathematics aptitude. After controlling for previous study of mathematics/ they found significant sex differences in mathematics achievement in only two of the four schools under study/ making the attribution to sex per se less likely. Controlling for the number of space visualization-related courses/ the sex-related differences which originally 79 TABLE 4.2 Continued Item Num ber X1 for Model 3.2 Xz for Model 3 .3 Xz for Difference Actual Sample Size Minimum Sample Size for Signif. (a) (b) (c) (d) (e) (f) 75 8.69 237.07 228.38* 67 /783 5/963 76 99.68 167 .72 68.04* 70/183 20/723 77 79.38 432.36 352.98* 66/304 3/774 78 64.56 127 .04 62.48* 76/806 24/696 79 23.98 72.59 48.61* 48/690 20/123 80 75.01 128.13 53 .12* 55/277 20 /906 81 95.51 106.82 11.31* 57 /710 102/511 82 51.99 108.12 56.13* 71/315 25/525 83 20.12 95.18 75.06* 81 /021 21/685 84 38.62 69.05 30.43* 78/959 52/129 86 102.85 433.47 330.62* 39/903 2/425** 87 66.75 430.40 363.65* 48/194 2/662** 88 63.01 126 .27 63.26* 38/742 12/304 89 43 .26 289.14 245.88* 39/648 3 /240 90 85.53 405.04 319.51* 53 /557 3/368 91 82 .53 85.48 2.95 62/954 428/727 92 16.36 93 .12 76.76* 88/231 23 /092 93 78.39 229.02 150.63* 80 ,122 10/766 94 70.15 137 .27 67.12* 87 /241 26 /113 80 TABLE 4.2 Continued Item Num ber Xa for Model 3.2 Xa for Model 3.3 X2 for Difference Actual Sample Size Minimum Sample Size for Signif. la) lb) (c) (d) lell . Cf) 95 40.59 68.41 27.82* 83/969 60/638 96 29.31 212.34 183.03* 74/287 8/154 97 8.60 126.96 118.36* 50/028 8/492 98 29.87 41.77 11.90* 69/052 116/576 99 9.44 9.94 0.50 72/647 2/918/956 100 11.35 41.48 30.13* 60/931 40/627 101 23.77 315.62 291.85* 83/782 5/767 102 61.06 437.09 376.03* 75/708 4/045 103 6.12 7.97 1.85 50/663 550/173 104 40.47 283.94 243.47* 72/517 5/984 105 28.66 48.59 19.93* 60/712 61/199 106 13.47 1/682.71 1/669.34* 71/946 866*1 107 13.13 99.20 86.07* 72/482 16/918 108 26.56 79.33 52.77* 86/261 32/840 109 5.73 18.33 12.60* 76/800 112/453 110 17.80 49.11 31.31* 70/980 45/544 111 30.71 477.44 446.73* 86/190 3/876 PC.01 Minimum sample size less than 3/000 112 LandaUf M. (1984). The effects of spatial ability and problem presentation format in mathematical problem solving performance on middle school students (Doctoral dissertation* Northwestern University). Dissertation Abstracts International? 4/ 442A- 443 A. Lindgren/ H. C./ Silva/ I./ Faraco I./ & DaRocha/ N. S. (1964). Attitudes toward problem solving as a function of success in arithmetic in Brazilian elementary schools. Journal of Educational Research/ 2., 44-45 . Maccoby/ E. E. (1966). Sex differences in intellectual functioning. In E. E. Maccoby (Ed.)/ The Development of Sex Differences (pp. 25-55). Stanford/ CA: The Stanford University Press. Maccoby/ E. E. / & Jacklin/ C. N. (1972). Sex differences in intellectual functioning. In A. Anastasi/ (Ed.)/ Assessment in a pluralistic society, (pp. 37-55) . Proceedings of the 1972 Invitational Conference on Testing Problems/ Educational Testing Service/ Princeton/ NJ. Maccoby/ E. E./ & Jacklin/ C. N. (1974) The Psychology of Sex Differences/ Stanford/ CA: Stanford University Press. Marshall/ S. P. (1981) Sex Differences in Sixth Grade Children's Problem Solving. Paper presented at the annual meeting of the American Educational Research Association/ Los Angeles/ CA. Marshall/ S. P. (1983). Sex differences in mathematical errors: Analysis of distractor choice. Journal for Research in Mathematics Education/ 14./ 325-336 . Marshall/ S. 'p. (1984). Sex differences in children's mathematics achievement: Solving computations and story problems/ Journal of Educational Psychology/ lAf 194-204. Martin/ M. D. (1964). Reading comprehension/ abstract verbal reasoning and computation as factors in arithmetic problem solving (Doctoral dissertation/ University of Iowa/ 1963) Dissertation Abstracts International/ 24/ 4547A-4548A. 47 than the problem-interpretation score? the difficulty was due to reading ability. If the score of the lowest of the three areas (computation? problem interpretation? and reading-problem interpretation score) was the same as the reading-problem solving score? the student's area of greatest immediate need was either computation? problem interpretation? or reading. If the reading-problem solving score was lower than the lowest of the other three scores? the student's area of greatest immediate need was integration. Analysis of the data revealed that for 19% of the students? problem interpretation was their major difficulty; for 26% of the students? integration (total problem solving) was their greatest immediate need; for another 26%? computation was the major weakness; and for 29%? reading was their greatest immediate need. Seventy five percent of the students demonstrated clear strength in computation? 21% in problem interpretation? and 4% in reading-problem interpretation. An analysis across all students (including those without complete data) showed that 26% of the subjects could not work word problems at a level as high as that at which they could compute? interpret problems? and read and interpret problems? when those areas were measured separately. This led them to conclude that knowing the skills or the components of solving word problems is not sufficient for 6 Purpose of the Study The purpose of this study was twofold. The first was to investigate sex-related differences in the selection of incorrect responses in a multiple-choice achievement test in mathematics. For each test item* the research questions were as follows: 1. Is there a difference in the proportion of males and females choosing each incorrect option? 2. Is the same pattern of differences found in data obtained in three different administrations of the test? The second objective was to investigate sex-related differences in test scores in mathematics problem solving. The following questions were studied: 1. Do males and females differ in problem solving perf ormance? 2. Do sex differences in problem solving persist after controlling for computational skillsf and does the differential success of males and females on problem solving items depend on their success on the computation items? Significance of the Study Item response patterns are very useful techniques in the assessment of mathematics learning and achievement. Total test scores can be very misleading in the assessment of student performance and provide no diagnostic information about the nature and seriousness of student 65 One important characteristic of this table involves two-by-two subtables of sex cross classified with two of the options for each year. For example* consider the three two-by-two tables obtained by cross classifying sex and option choices one and two for each year. These tables are indicated by dotted lines on Table 3.3. For this table* the ratio of the odds of a male choosing option one to the odds of a female choosing option one is the same (within truncating error) for each year. For example* this odds ratio for the first year is (.120/.045) / (.060/.060) * 2.67. Within the error caused by reporting truncated probabilities* the odds ratio for years two and three is the same as that for year one. The equality of these odds ratios is Marshall's operationalization stage of the year- to-year gender-option dependency. To show that the odds ratio can be constant over years* but that the probabilities of option choice conditional on sex and year can change from year to year* for both males and females* the probabilities in Table 3.3 were converted to the probabilities of option choice conditional upon sex and year. These conditional probabilities are reported in Table 3.4. Unlike the probabilities in Table 3.2* those in Table 3.4 change from year to year for both males and females. 77 TABLE 4.2 Continued Item Num ber Xa for Model 3.2 Xa for Model 3 .3 Xa for Difference Actual Sample Size Minimum Sample Size for Signif. (a) (b) (c) (d) (e) (f) 36 84.36 134.56 50 .20* 86 f137 34*472 37 23.70 76.85 53 .15* 37 *06 8 14*011 38 13.64 84 .24 70.60* 38*870 11*0618 39 13.35 757.60 744.25* 61*057 1*648** 40 6.67 701.39 694.72* 56 *438 1*630** 41 12.65 603.42 590.77* 57 *145 1*943** 42 23 .26 189.11 165.85* 59*140 7*154 43 16.29 21.65 5.36 33*712 126 *357 44 3.55 115.84 112.29* 67*192 12*021 45 22.41 55.90 33.49* 13*098 37*851 46 6.56 16.30 9.74* 56 *727 117 *007 47 7.82 76.15 68.33* 56 *555 16*628 48 179.25 452.64 273.39* 97 *593 7*172 49 27 .82 59.75 31.93* 16*766 10*549 50 14.23 240.27 226.04* 25*410 2*258** 51 16.37 224.54 208.17* 24 *715 2*385** 52 26.01 96 .14 70.13* 39*858 11*418 53 6.69 27.49 20.80* 44*694 43 *168 54 41.41 95.41 54.00* 74*502 27*718 92 TABLE 4.8 ANCOVA Summary Table: Other Covariates First Year Source df SS MS F Addition 1 271.57 271.57 49.74* Sex 1 15.54 15.54 2.85 Error 999 5453.90 5.46 Subtraction 1 771.78 771.78 155.60* Sex 1 10.73 10.73 2.16 Error 999 4953.70 4.96 Fracadd 1 1412.11 1412.11 326.87* Sex 1 7.54 7.54 1.75 Error 999 4313.30 4.32 Decsub 1 1281.10 1281.10 287.89* Sex 1 4.94 4.94 1.11 Error 999 4444.30 4.45 Equivalence 1 811.44 811.44 164.93* Sex 1 69.69 69.69 14.16* Error 999 4914.00 4.92 * p .05 UNIVERSITY OF FLORIDA 3 1262 08554 6660 77 TABLE 4.2 Continued Item Num ber X* for Model 3.2 Xa for Model 3.3 X2- for Difference Actual Sample Size Minimum Sample Size for Signif. (a) (b) (c) . (d) . (e) (f) 36 84.36 134.56 50.20* 86,137 34,472 37 23.70 76.85 53.15* 37,068 14,011 38 13.64 84.24 70.60* 38,870 11,0618 39 13.35 757.60 744.25* 61,057 1,648** 40 6.67 701.39 694.72* 56,438 1,630** 41 12.65 603.42 590.77* 57,145 1,943** 42 23 .26 189.11 165.85* 59,140 7,154 43 16.29 21.65 5.36 33,712 126,357 44 3.55 115.84 112.29* 67,192 12,021 45 22.41 55.90 33.49* 13,098 37,851 46 6.56 16.30 9.74* 56,727 117,007 47 7.82 76.15 68.33* 56,555 16,628 48 179.25 452.64 273.39* 97,593 7,172 49 27.82 59.75 31.93* 16,766 10,549 50 14.23 240.27 226.04* 25,410 2,258** 51 16.37 224.54 208.17* 24,715 2,385** 52 26.01 96.14 70.13* 39,858 11,418 53 6.69 27.49 20.80* 44,694 43,168 54 41.41 95.41 54.00* 74,502 27,718 CHAPTER II REVIEW OF THE LITERATURE Sex differences in mathematics learning and achievement have been a subject of concern for educators and psychologists. Many studies found in the literature support the existence of these differences. Boys show superiority in higher level cognitive tasks (problem solving or mathematical reasoning) in the upper elementary years and in the early high school years (Fennema/ 1974; Maccoby & Jacklin/ 1974) . Almost all the research carried out has dealt with analysis of total correct scores in mathematics aptitude and achievement tests or scores in subtests. The literature related to sex differences in incorrect responses/ the main subject of the present study/ is surprisingly sparse. For the most part/ the studies have investigated the differences between the sexes in mathematics learning and achievement and the underlying variables causing the differences. Cognitive and affective variables have been the matter at issue in the establishment of sex differences. Although research in mathematics problem solving/ the secondary subject of this investigation/ is extensive/ most 9 48 success* since the components must be integrated into a whole process (mastery learning of the components cannot assure mastery of the process). Their analysis also led them to conclude that* in the case of 6th graders* inability to read problems is a major obstacle in solving word problems. Only 12% of the subjects could read and set up problems correctly at a higher level than they could compute* while 60% could compute correctly at a higher level than they could read and set up problems; 44% could set up problems better when they heard them read than when they read the problem themselves. Only 13% could set up problems better when they read them than when they heard them read. Muth ( 1984 ) investigated the role of reading and computational skills in the solution of word problems. A group of 200 students from the 6th grade were administered a test of basic skills and a mathematics word problem test. The word problem test consisted of 15 sample items supplied by the National Assessment of Educational Progress. The items were adapted to include some extraneous information and complex syntactic structure. Four versions of the test were constructed by combining two versions of problem information (absence vs. presence of extraneous information) with two versions of syntactic structure (simple vs. complex syntax) Task performance was measured by means of the number of problems answered correctly* 104 to contribute to the body of knowledge regarding such differences in mathematical learning and achievement. This section of Chapter V comprises implications of the findings and recommendations evolving from the present investigation. ex-related,,Differences jp incorrect Responses The main purpose of education is to impart formal education equally for the sexes. Analysis of the data on incorrect responses showed that for 100 of the 111 items of the test/ males and females selected different incorrect responses/ and this pattern of responses was consistently found during the three years of test administration. However/ with a sample of less than 3/000 subjects/ statistical significant results were obtained for only 6 items. Therefore/ it seems that/ at least for 6th grade children/ educators need not concern themselves with varying teaching techniques for the sexes. This study/ however/ has value apart from the investigation of sex-related differences. It was found that/ year after year/ 6th graders from the public schools in Puerto Rico tend to make the same errors in the solution of mathematical problems. It is important/ therefore/ that teachers provide the learning strategies necessary to enable students to obtain a better understanding of the concepts of numbers. 23 computation grew larger. This difference grew larger for girls than for boys. Although SES was a major factor in solving computation and story problem items successfully/ the effect was similar for each sex. Sex-related differences by primary language or chronological age were not large. This research carried out by Marshall with elementary grade children supports previous research findings that males are better than females in mathematics problem solving (a higher order skill) and females are better than their counterpart males in computation (a lower level skill). Marshall's research also brought out a different aspect of this question: the notion that girls find it more necessary than boys to succeed in the computation items in order to successfully solve the story problem items. Cognitive and Affective Factors That Influence_Sex Differences in Mathematics Learning and Achievement The research reviewed in the literature does not provide evidence of any unique variable that could serve as an explanation for the observed sex-related differences in mathematics learning and achievement. However/ some issues have been discussed/ among which the most prevalent are that sex-related differences in mathematics learning and achievement are a result of differences in formal education; that sex-related differences in mathematics 30 The issue of sex-related differences in spatial visualization ability as an explanation for sex differences observed in mathematics achievement is less convincing and the findings more contradictory than in the issue of sex differences in formal education. Besides these cognitive issues/ other issues/ mostly affective in nature/ have also been studied in trying to explain the origin of these sex differences in mathematics achievement and learning. The studies dealing with these affective variables are reviewed in the next section. Differentiated Effect of Affective Variables Researchers have attempted to explain the effect of sex differences in internal beliefs/ interests/ and attitudes (affective variables) on mathematics learning and achievement. A brief statement of each explanation precedes the summary of studies conducted that support the explanation. ;. Females/ more than males/ lack confidence in their ability to learn mathematics and this affects their achievement in mathematics and their election of more advanced mathematics courses. Maccoby and Jacklin (1974) reported that self- confidence in terms of grade expectancy and success in particular tasks was found to be consistently lower in women than in men. In 1978/ Fennema and Sherman reported 43 Children (WISC). Multiple regression analysis was performed. A correlation of .769 was found between word problem solving and the eight independent variables. Correlations between word problem solving and the independent variables ranged from .363 (verbal intelligence) to .674 (mathematical concepts). Correlations between the independent variables ranged from .369 (verbal intelligence and computation) to .749 (reading comprehension and vocabulary). Mathematical concepts* computation* and number sentence selection were almost as effective as all eight independent variables in predicting achievement in mathematical word problem solving. Mathematical concepts* computation* number sentence selection* and reading comprehension accounted for 58% of the variance* whereas all eight predictors accounted for 59% of the variance. The two best predictors were mathematical concepts and computation* which accounted for 54% variance. Other variables accounted for about 40% of the variance. The author recommends that the findings of this study be interpreted cautiously because the correlation between the eight independent variables was high* and* according to Zalewski* in a study of this nature where the interest is primarily in the influence of several variables on one dependent variable* a low correlation between the independent variables is required. (p. 2804) 99 responses was consistent throughout the years in which the test was administered. Males and females selected different incorrect responses in 100 of the 111 items of the test. The pattern of male-female differences in the selection of incorrect responses was consistently found in each of the years of test administration. However/ for the vast majority of the 100 items/ male-female differences were relatively small in magnitude/ considering the fact that the number of subjects needed to obtain significance was very large. Therefore/ these findings lack educational significance. A second objective of the study was to investigate whether males and females differ in problem solving performance/ if these differences persist after controlling for computational skills/ and if the male-female differences depend on the level of computational skills. Results of the analyses reported in this section generally indicated that females outperformed males not only in problem solving/ but in six of the seven computational variables. Males surpassed females in equivalence/ but statistical significance was obtained for only one of the three years covered in the study. The results also tended to show that/ for examinees with similar levels of computational skills/ sex differences in problem solving did not exist. The only 103 question was answered/ partially/ in the affirmative. Interactions/ indicating the dependence/ were significant for multiplication and division in the first year and for subtration in the second year. In general/ the sex-related differences in mathematics problem-solving/ in favor of males performance/ reported by Marshall (1981/ 1984)/ and Fennema and Tartree (1983) were not supported in this investigation. Female performance was equal or higher than male performance in problem solving. The findings of the study have something in common with other investigations where mathematics problem solving has been the subject of interest/ the fact that males and females do poorly in word-story problem items (problem solving) The mean score in problem solving for males and females in the 9-item problem solving subtest was 3.436 and 3.862 respectively for the first year/ 3.632 and 4.015 for the second year/ and 3.927 and 4.341 for the third year. Implications of the Findings and Suggestions for Further Research The main purpose of this study was to investigate sex-related differences in the selection of incorrect responses. Another objective was to compare male and female performance after accounting for computational ability. This type of research is conducted in an attempt 43 Children (WISC). Multiple regression analysis was performed. A correlation of .769 was found between word problem solving and the eight independent variables. Correlations between word problem solving and the independent variables ranged from .363 (verbal intelligence) to .674 (mathematical concepts). Correlations between the independent variables ranged from .369 (verbal intelligence and computation) to .749 (reading comprehension and vocabulary). Mathematical concepts computation and number sentence selection were almost as effective as all eight independent variables in predicting achievement in mathematical word problem solving. Mathematical concepts computation number sentence selection and reading comprehension accounted for 58% of the variance whereas all eight predictors accounted for 59% of the variance. The two best predictors were mathematical concepts and computation which accounted for 54% variance. Other variables accounted for about 40% of the variance. The author recommends that the findings of this study be interpreted cautiously because the correlation between the eight independent variables was high and according to Zal ewski in a study of this nature where the interest is primarily in the influence of several variables on one dependent variable a low correlation between the independent variables is required. (p. 2804) 65 One important characteristic of this table involves two-by-two subtables of sex cross classified with two of the options for each year. For example* consider the three two-by-two tables obtained by cross classifying sex and option choices one and two for each year. These tables are indicated by dotted lines on Table 3.3. For this table* the ratio of the odds of a male choosing option one to the odds of a female choosing option one is the same (within truncating error) for each year. For example* this odds ratio for the first year is (.120/.045) / (.060/.060) = 2.67. Within the error caused by reporting truncated probabilities* the odds ratio for years two and three is the same as that for year one. The equality of these odds ratios is Marshall's operationalization stage of the year- to-year gender-option dependency. To show that the odds ratio can be constant over years* but that the probabilities of option choice conditional on sex and year can change from year to year * for both males and females* the probabilities in Table 3.3 were converted to the probabilities of option choice conditional upon sex and year. These conditional probabilities are reported in Table 3.4. Unlike the probabilities in Table 3.2* those in Table 3.4 change from year to year for both males and females. MATH CONCEPTS Fractions j 2 items) j / Numbers j \ \ (18 items ) \ / Fig. 3.1 Number of Questions by Skill Area in the Basic Skills Test in Mathematics 6 CHAPTER III METHOD The first objective of this study was to investigate sex differences in the selection of incorrect responses in a mathematics multiple-choice achievement test* and to determine whether these differences were consistent over three consecutive administrations of the test. The second objective was to investigate whether males and females differ in problem solving performance* if these differences persist after accounting for computational skills* and if the male-female differences depend on the level of computational skills. This chapter contains descriptions of the sample* the test instrument* and the statistical analysis used in achieving the above mentioned objectives. The Sample To achieve the first objective of the study* all the students who took the Basic Skills Test in Mathematics-6 ("Prueba de Destrezas Bsicas en Matemticas-6") in each or the three years were included in the study. To achieve the second objective of the study* approximately 54 REFERENCES Aiken/ L. R. (1970). Attitudes toward mathematics. Re.Y,iÂ£M_Qf. Education2_Reg.ar.Â£h/ 10, 551-596. Allen, R. H., & Chambers, D. L. (1977). A comparison of the mathematics achievement of males and females (Bulletin No. 9194). Madison: Wisconsin State Department of Public Instruction, Division for Management and Planning Service. (ERIC Document Reproduction Services No. ED 159 073) Armstrong, j. m. (1979). ftgh.ie-yejnent_and..participatiQn pf women in mathematics: An overview. (Report of a two year study). Denver, CO: Educational Commission of the States, National Assessment of Educational Progress. (ERIC Document Reproduction Service No. ED 184 878) Armstrong, J. M., & Price, R. A. (1982). Correlates and predictors of women's mathematics participation, journal f. 99-109. Astin, H. S. (1974). Overview of the findings. In H. Astin, H. Suniewick, & S. Dweck (Eds.), Women: A Bibliography on their Education and Careers. (pp. 1-10). New York: Behavorial Publications. Backman, M. E. (1972). Patterns of mental abilities: Ethnic, socioeconomic and sex differences. American Educational Research Journal, Â£, 1-11. Ballew, H., & Cunningham, W. (1982). Diagnosing strengths and weaknesses of sixth-grade students in solving word problems. Journal for Research in Mathematics Education, 11, 202-210. Balow, I. H. (1964). Reading and computation ability as determinants of problem solving. The Arithmetic Teacher, 11, 18-22. Bar-Tal, D., & Frieze, J. H. (1977). Achievement motivation for males and females as a determinant of attribution for success and failure. Sex roles, 1, 301-313. 107 Table Page 4.11 Adjusted Means on Problem Solving* by Covariate and Sex 96 4.12 ANCOVA Third Year 97 4.13 Adjusted Means on Problem Solving* by Covariate and Sex 98 vi 35 It is not until adolescence that sex differences in the perception of mathematics as a male domain are found (Fennema* 1976; Stein* 1971? Stein & Smithless* 1969; Verbeke* 1983). In a study with 2nd* 6th and 12th graders* Stein and Smithless (1969) found that students' perceptions of spatial* mechanical* and arithmetic skills as masculine became more defined as these students got older. Fennema (1976) considers that the influence each sex exerts upon the other on all aspects of behavior is stronger during adolescence. Since during these years males stereotype mathematics as a male domain* they send this message to females who* in turn* tend to be influenced in their willingness to study or not to study mathematics. Before that stage* girls consider arithmetic feminine* while boys consider it appropriate for both sexes (Bobbe* 1971) . Usefulness of mathematics. Females perceive mathema tics as less useful to them than males do* and this perception occurs at a very young age. As a result* females exert less effort than males to learn or elect to take advanced mathematics courses. Many studies reported before 1976 found that the perception of the usefulness of mathematics for one's future differs for males and females* and is related to course taking plans and behavior (Fox* 1977). If females do not perceive mathematics as useful for their future* 92 TABLE 4.8 ANCOVA Summary Table: Other Covariates First Year Source df SS MS F Addition 1 271.57 271.57 49.74* Sex 1 15.54 15.54 2.85 Error 999 5453.90 5.46 Subtraction 1 771.78 771.78 155.60* Sex 1 10.73 10.73 2.16 Error 999 4953.70 4.96 Fracadd 1 1412.11 1412.11 326.87* Sex 1 7.54 7.54 1.75 Error 999 4313.30 4.32 Decsub 1 1281.10 1281.10 287.89* Sex 1 4.94 4.94 1.11 Error 999 4444.30 4.45 Equivalence 1 811.44 811.44 164.93* Sex 1 69.69 69.69 14.16* Error 999 4914.00 4.92 * p .05 114 Sherman/ J. (1979) Predicting mathematics performance in high school girls and boys. Journal of Educational Psychology/ 71/ 242-249. Sherman/ J. (1980). Mathematics/ spatial visualization/ and related factors: Changes in girls and boys/ grades 8-11. Journal of Educational Psychology/ 22./ 476-482. Stein/ A. H. (1971). The effects of sex role standards for achievement and sex role preference on three determinants of achievement motiviation. Developmental Psychology/ 4./ 219-231. Stein/ A. H./ & Smithless/ J. (1969). Age and sex differences in children's sex role standards about achievement/ Developmental Psychology/ 1/ 252-259. Suydam/ M. N. (1971) Research on mathematics education (K-12) reported in 1970. Journal for Research in Mathematics Education/ Z> 257-298. Suydam/ M. N. (1972). Research on mathematics education (K-12) reported in 1971. Journal for Research in Mathematics Education/ 196-232. Suydam/ M. N. (1973) Research on mathematics education (K-12) reported in 1972. Journal for Research in Mathematics Education/ 1/ 205-242. Suydam/ M. N. (1974). Research on mathematics education (K-12) recorded in 1973. Journal for Research in Mathematics Education/ / 238-272. Suydam/ M. N. / & Weaver/ J. F. ( 1975) Research on mathematics education (K-12) recorded in 1974. Journal for Research in Mathematics Education/ Â£/ 253-282. Verbeke/ K. A. (1983). Sex-related differences in mathematically gifted secondary students: An investigation of selected cognitive/ affective/ and educational factors (Doctoral dissertation/ University of Maryland/ 1982) Dissertation Abstracts International/ 43/ 2267A-2268A. Weiner/ B. (1974). Achievement motivation and attribution theory. Morristown/ NJ: General Learning Press. Ill Hansen/ C. W. (1944). Factors associated with successful achievement in problem solving in sixth grade arithmentic. Journal of Educational Research# 38 # 111-119. Harnisch# D. L. (1983). Item response patterns: Application for educational practice. Journal of Educational Measurement# 2SL' 191-205. Hilton# T. L. # & Berglund# G. W. (1974). Sex differences in mathematics achievement: A longitudinal study. Journal of. Educational Research# Â£2# 231-237. Hunter# J. E. # & Cohen# S. H. ( 1974). Correcting for unreliability in nonlinear models of attitude change. Psychometrika# 39 # 445-468. Jarvis# 0. T. (1964). Boy-girl ability differences in elementary school arithmentic. School Science and Mathematics# 64 # 657-659. Kaufman# A. (1984). Sex differences in mathematics reasoning in mathematically gifted and average fifth and sixth graders (Doctoral dissertation# Hofstra University# 1983) Dissertation Abstracts Inter national# 45 # 1094A-1095A. Kilpatrick I. (1969). Problem solving and creative behavior in mathematics. In J. W. Wilson# & L. R. Carry (Eds.)# Studies in mathematics (Vol. No. 19). Stanford# CA: School Mathematics Study Group. Kloosterman# P. W. (1985). Attributional theory# learned helplessness# and achievement in ninth grade mathematics (Doctoral dissertation# University of Wisconsin# 1984) Dissertation Abstracts Internatio nal# 4Â£/ 919A-920A. Knifong# J. D. # & Holtan# B. ( 1976) An analysis of children's written solutions to word problems. Journal for Research in Mathematics_Education# 2# 106-112. Knifong# J. D.# & Holtan# B. (1977). A search for reading difficulties among erred word problems. Journal for Research in Mathematics Education# Â£# 227-230. 24 learning and achievement arise from sex differences in spatial visualization; and that sex-related differences result from a differentiated effect of affective variables on the mathematics performance of males and females. Differences in Formal Mathematics Education (Differential Coursework Hypothesis) The basis for the differential coursework hypothesis is the fact that sex-related differences in mathematical learning and achievement show up when comparing groups which are not equal in previous mathematics learning. After the 8th grade/ boys tend to select mathematics courses more often than girls. Therefore/ girls show lower achievement scores in mathematics tests because their mathematics experience is not as strong as the boys' (Fennema/ 1975; Fennema & Sherman/ 1977; Sherman/ 1979). Fennema and Sherman's study (1977) lends additional support to the feasibility of viewing sex differences in mathematics learning and achievement as reflecting something other than a difference in mathematics aptitude. After controlling for previous study of mathematics/ they found significant sex differences in mathematics achievement in only two of the four schools under study/ making the attribution to sex per se less likely. Controlling for the number of space visualization-related courses/ the sex-related differences which originally Problem Solving 88 T T T 1 1 2 3 Division Fig. 4.2 Sex by Division Interaction 4 5 6 95 TABLE 4.10 ANCOVA Summary Table: Other Covariates Second Year Source df ss MS F Addition 1 206.97 206.97 39.57* Sex 1 19.12 19.12 3.66 Error 1010 5277.90 5.23 Multiplication 1 1026.68 1026.68 232.80* Sex 1 .80 .80 .18 Error 1010 4458.30 4.41 Division 1 1117 .46 1117.46 258.67* Sex 1 .64 .64 .15 Error 1010 4367.50 4.32 Fracadd 1 1341.13 1341.13 327.10* Sex 1 2.03 2.03 .50 Error 1010 4143.80 4.10 Decsub 1 1453 .91 1453.91 364.39* Sex 1 4.88 4.88 1.22 Error 1010 4031.10 3.99 Equivalence 1 797.11 797.11 171.79* Sex 1 127.33 127.33 27.44* Error 1010 4687.90 4.64 * P<.05 33 Mathematics.as. a male domain. Mathematics is an activity more closely related to the male sex domain than to the female sex domain (Eccles et al.# 1983). Thus* the mathematical achievement ot boys is higher than that of girls. According to John Ernest (1976) # in his study Mathematics and Sex/ mathematics is a sexist discipline. He attributed sex-related differences in mathematical achievement to the creation by society of sexual stereotypes and attitudes# restrictions# and constraints that promote the idea of the superiority of boys in mathematics. Ernest reported that boys# girls# and teachers# all believe that boys are superior in mathematics# at least by the time students reach adolescence. Bern and Bern (1970) agree and argue that an American woman is trained to "know her place" in society because ot the pervasive sex-role concept which results in differential expectations and socialization practices. Plank and Plank (1954) were more specific. They discussed two hypotheses related to this view: the differential cultural reinforcement hypothesis and the masculine identification hypothesis. The differential cultural reinforcement hypothesis states that society in general perceives mathematics as a male domain# giving females less encouragement for excelling in it. The masculine identification hypothesis establishes that CHAPTER V DISCUSSION This cnapter comprises two sections: a) summary and interpretation of the results of the study/ and b) implications of the findings/ with suggestions for further research. Summary and Interpretation of the Results The main objective of this study was to investigate whether there were sex-related differences in incorrect responses/ and if the pattern of response was consistent over three consecutive administrations of the mathematics test. Males and females selected different incorrect responses in 100 of the 111 items of the test and this pattern of response was consistenly found in each of the three years of test administration. The results tend to support/ in part/ Marshall's findings (1981/ 1983) of observed sex-related differences in incorrect responses/ but in no way do they point to a conclusion that 6th grade boys and girls from the public schools in Puerto Rico exhibit different problem solving strategies. The significant differentiated pattern of 101 44 In a more recent investigation# Exedisis (1983) studied the contribution of reading ability# vocabulary# mathematical concepts# computation# sex# and race on problem-solving performance. The Iowa Test of Basic Skills was administered to a group of 6th# 7th# and 8th grade anglo and black Chicago male and female adolescents. Problem solving was highly correlated to an understanding of basic mathematical concepts# somewhat correlated to race# and weakly correlated to computational and vocabulary skills# sex# and reading ability. Although the findings of these studies show a relationship between computational skills and problem solving achievement# this relationship is not strong enough to be considered the most determinant factor in problem solving achievement# as some of the researchers have been careful to point out. In spite of the dismissal of reading as a determinant factor in problem solving achievement by some of these same researchers# more recent studies in this area have led others to hold different views. Reading end problem. SQlvinfl-JLgj.f.Qirniangg Martin (1964) studied the contribution of reading comprehension# computation# abstract verbal reasoning# and arithmetic concepts to arithmetic problem solving performance. Fourth and 8th grade students were administered the Iowa tests of Basic Skills and the Lorge-Thorndlike intelligence test (verbal). I certify that I have read this study and that in my opinion it conforms to acceptable standards of scholarly presentation and is fully adequate/ in scope and quality/ as a dissertation for the degree of Doctor of Philosophy. Jaimes Algina/Chai rman Processor of Foundations of Education I certify that I have read this study and that in my opinion it conforms to acceptable standards of scholarly presentation and is fully adequate/ in scope and quality/ as a dissertation for the degree of Doctor of Philosophy. 2U J- Linda Crocker Professor of Foundations of Education I certify that I have read this study and that in my opinion it conforms to acceptable standards of scholarly presentation and is fully adequate/ in scope and quality/ as a dissertation for the degree of Doctor of Philosophy. Leadership This dissertation was submitted to the Graduate Faculty of the College of Education and to the Graduate School/ and was accepted as partial fulfillment for the degree of Doctor of Philosophy. August/ 1986 Cnairman/ Founp lations. of Education A/uJ Cl zzUfh Dean/ College of EducSrtion Dean/ Graduate School (a) 80 81 82 83 84 85 86 87 88 89 90 91 93 94 95 96 98 101 102 104 73 TABLE 4.1 Continued X* for Model 3.2* Actual Sample Size Min. Sample Size for Significance (b) (c) (d) 75.01 55/277 6,787 95 .51 57 ,710 5,565 51.99 71/315 12,633 20.12 81/021 37 ,088 38.62 78/959 18,830 1240 .78 41,180 306** 102.85 39/903 3 ,573 66.75 48,194 6,650 63.01 38,742 5/663 43 .26 39,648 8,441 85 .53 53 ,557 5,767 82.53 62,954 7/025 78.39 80 ,723 9,484 70.15 87,241 11,454 40.59 83 ,969 19,053 29.31 74,287 23 ,343 29.87 69,052 21,291 23 .77 83 ,782 32/462 61.06 75/708 11/419 40.47 72 ,517 16 ,503 CHAPTER III METHOD The first objective of this study was to investigate sex differences in the selection of incorrect responses in a mathematics multiple-choice achievement test, and to determine whether these differences were consistent over three consecutive administrations of the test. The second objective was to investigate whether males and females differ in problem solving performance* if these differences persist after accounting for computational skills* and if the male-female differences depend on the level of computational skills. This chapter contains descriptions of the sample* the test instrument* and the statistical analysis used in achieving the above mentioned objectives. The Sample To achieve the first objective of the study* all the students who took the Basic Skills Test in Mathematics-6 ("Prueba de Destrezas Bsicas en Matemticas-6") in each or the three years were included in the study. To achieve the second objective of the study* approximately 54 7 errors (Harnisch? 1983). Item response patterns are valuable for the identification of large group differences/ including d i st rict-to-district/ school-to-school/ and classroom-to-classroom variations on different subsets of items. The response patterns can provide diagnostic information about the type of understanding the student has on various mathematics topics (e.g. / problem solving). Marshall (1981/ 1984) has used the item response pattern technique and her findings indicate sex differences in mathematics performance at the item level. Females outperformed males in computation and males outscored females in problem solving. Also/ the success of girls in the problem solving items was dependent upon their success in the computation items; for boys/ success in the problem solving items did not depend as much on their computational performance. Marshall (1981/ 1983) has also reported that males and females differ in the selection of incorrect responses/ reflecting differences in reasoning abilities. In Puerto Rico/ a high percentage of children promoted to the 7th grade in the public schools does not master the basic skills in mathematics. If 6th grade male and female children can be diagnosed as having different problem solving abilities/ as Marshall found with California children/ teachers may need to provide tailor-made mathematics instruction for each sex/ in order to ensure equal access to formal education and enhance mathematics 49 number of problems set up correctly/ and amount of time spent taking the test. Reading ability and computational ability were both positively correlated with number of correct answers and with number of problems correctly set up/ and negatively correlated with test-taking time. Presence of extraneous information was negatively correlated with correct answers and correct set ups and positively correlated with test-taking time. Syntactic complexity was not significantly correlated with any of the performance measures. Results of a multiple regression analysis showed that reading accounted for 46% of the variance in total correct answers and computation accounted for 8%. Reading ability and computational ability uniquely accounted for 14% and 8% of the variance in the number of correct answers/ respectively. Extraneous information added significantly to the variance explained in the number of correct answers/ but syntactic structure did not. Reading ability accounted for 5% of the variance in test-taking time/ but computation did not add significantly to the variance explained by reading. Muth concluded that reading and computation both contribute significantly to success in solving arithmetic word problems/ but that reading plays a more significant role than does computation. 39 The findings suggest that reading factors are not as important as arithmetic and mental factors in problem solving performance. However/ these findings should be taken cautiously as the content of the Gates tests (used to measure reading) is literary and does not include mathematical material. Chase (1960) studied 15 variables in an effort to find out which ones have significant influence on the anility to solve verbal mathematics problems. Only computation reading to note details and fundamental knowledge were primarily related to problem solving. Computation accounted for 20.4% of the 32% variance directly associated with problem solving. Chase concluded that a pupil's ability in the mechanics of computation comprehension of the principles that underline the number systems and the extent to which important items of information are noticed when reading are good predictors of the student's ability in solving verbal problems. Balow (1964) investigated the importance of reading ability and computation ability in problem solving performance. He objected to the approaches used by other researchers who in their analyses dichotomized research subjects as "poor" or "good" students and who ignored the recognizea effect of intelligence on reading and on mathematics achievement. Balow administered the Stanford 31 that in their study involving students from grades 6 through 12, boys showed a higher level of confidence in mathematics at each grade level. These differences between the sexes occurred in most instances even when no sex-related differences in mathematics achievement were found. The correlation between confidence in mathematics performance and mathematics achievement in this study was higher than for any other affective variable investigated. Sherman reported a similar finding in 1980; in males, the most important factor related to continuation in theoretical mathematics courses was confidence in learning mathematics. This variable weighed more than any of the cognitive variables: mathematics achievement, spatial visualization, general ability, and verbal skill. In the case of females, among the affective variables, confidence in learning mathematics was found to be second in importance to perceived usefulness of mathematics. Probert (1983) supported these findings with college students. A variable that needs discussion within the context of sex differences in confidence as learners of mathematics is causal attribution. Causal attribution models attempt to classify those factors to which one attributes success or failure. The model proposed by Weiner (1974) categorizes four dimensions of attribution of success and failure: stable and internal, unstable and internal, stable and external, and unstable and external. For example, if one 93 superiority in problem solving lost statistical significance. Adjusted meansr associated with the analyses/ are reported in Table 4.9. Summaries of the analyses of covariance for the second year are reported in Table 4.10. The analyses show that females retained their superiority in problem solving after controlling for equivalence. When the controlling variables were additionr multiplication/ division/ addition of fractions/ and subtraction with decimals females superiority in problem solving lost statistical significance. Adjusted means are reported in Table 4.11. The analyses of covariance for the third year are summarized in Table 4.12. The results of the analyses show that females maintained their superiority in problem solving when performance on either subtraction or equivalence was controlled. When addition/ multiplication/ division/ addition of fractions/ and subtraction of decimals were the controlling variables/ female superiority in problem solving lost statistical significance. Adjusted means are reported in Table 4.13. Summary Sex-related differences in mathematics achievement were the subject of this study. The first objective was to investigate whether males and females differ in the selection of incorrect responses/ and if the pattern of 98 TABLE 4.13 Adjusted Means on Problem Solving/ by Covariate and Sex Third Year Covariate Adjusted Means Males Females Addition 4.09 4.18 Subtraction 3.98 4.29 Multiplication 4.10 4.17 Division 4.17 4.16 Addf rac 4.07 4.20 Subdec 4 .15 4.12 Equivalence 3 .79 4.45 Problem Solving 89 Subtraction Fig. 4.3 Sex by Subtraction Interaction 51 student attitudes and performance in the area of problem solving. Carey (1958) constructed a scale to measure attitudes toward problem solving. Her interest was in general problem solving rather than in mathematical problem solving. Her work constitutes the first attempt to construct a measure of attitudes toward problem solving. The scale was used with a group of college students/ and she foundr among other things/ that problem solving performance is positively related to problem solving attitudes and that/ in the case of females/ positive modification of attitudes toward problem solving brings a significant gain in problem solving performance. Lindgren/ Silva/ Faracoz and DaRocha (1964) adapted Carey's scale of attitudes toward problem solving and applied it to a group of 4th grade Brazilian children. Students also answered an arithmetic achievement test/ a general intelligence test/ and a socioeconomic (SE) scale. A low but significant positive correlation was found between arithmetic achievement and attitudes toward problem solving. A near zero correlation was found between attitudes toward problem solving and intelligence. Since problem solving is one aspect of the discipline of mathematics/ this correlation between attitudes and arithmetic achievement can lead to a conclusion or a strong 2 Many variables# cognitive/ affective# and educational# have been investigated since 1974 in relation to sex differences in mathematics learning and achievement. Fennema and Sherman (1977) investigated the effect of differential formal mathematics education. After controlling for the number of years of exposure to the study of mathematics# they found sex differences in only two of the four schools under study. However# in those schools where boys scored higher than girls# differences were also found in their attitudes toward mathematics. Hilton and Berglund (1974) found significant sex differences after controlling for the number of mathematics courses taken# and attributed them to sex differences in interests. "As the boys' interests in science increase related to the girls'# their achievement in mathematics increases relative to that of the girls" (p. 234). Wise# Steel# and McDonald (1979) reanalyzed test data collected in a longitudinal study of 400#000 high school students (Project Talent). They found that when the effect of the number of high school mathematics courses was not controlled# no sex differences emerged for 9th graders# but that gains made by boys during the next three years were more than twice that of the girls. These differences between the sexes disappeared when the number of mathematics courses taken was controlled. Results of the 1978 Women and Mathematics National Survey# Survey I# 28 1974 have failed to support these sex differences in spatial abilities (Armstrong* 1979; Connor* Serbin* & Schackman* 1977; Fennema & Sherman* 1978; Sherman* 1979). Fennema and Sherman (1978) and Sherman (1979) have explored sex-related differences in mathematical achievement and cognitive and affective variables related to these differences. In a study involving students from grades 6* 7* and 8* from four school districts* Fennema and Sherman found that spatial visualization and problem solving were highly correlated for both sexes (.59 and .60). Even in the school district where sex differences were found in problem solving* no significant sex-related differences were found in spatial visualization. When Sherman (1980) compared groups of males and females in two different grades* 8 and 11* she found no sex-related differences in problem solving or in spatial visualization in grade 8. In grade 11* however* although the sexes differed in their problem solving performance* no sex-related differences were found in spatial visualization. Even though spatial visualization in grade 8 was the second best predictor of problem solving performance in grade 11* sex differences in grade 11 were not a result of spatial visualization since no differences were found in that skill. In spite of the fact that no sex differences were found in spatial abilities* it is evident that males and 83 TABLE 4.3 - Continued Year/ Subtest Males X SD Females X SD t Decsub 2.863 1.87 3.440 1.80 5.00* Equivalence .76 .76 .65 .75 -2.32** N = 504 N = 509 Third Problem Solving 3.927 2.49 4.341 2.44 2.67* Addition 5.536 .84 5.704 .64 3.59* Subtraction 4.836 1.52 4.958 1.46 1.30 Multiplication 4.168 1.74 4.541 1.68 3.47* Division 3.343 1.88 3.795 1.83 3.87* Fracadd 2.819 1.89 3.117 1.85 2.53** Decsub 3.021 1.93 3.448 1.85 3.60* Equivalance .830 .82 .800 .78 - .60 N = 509 N = 504 Note: The number of items in the problem solving subtest was 9. In each computation subtest the number of items was 6. An item was included in the computation subtest only if it measured a computation skill required to solve a problem solving item. * p <.01 ** p <.05 12 test form/ no sex differences were found in response patterns. Approximately 45% of each sex selected option Â£. The next popular choice for both sexes was option d/ selected by approximately 35% of both boys and girls. On the story problem of Form 1/ males and females responded alike. Their most popular incorrect response choice was option a. for both males and females. The second most popular incorrect choice was option js for both sexes. Response to the story problem in Form 2 showed sex differences in response choice. Including the correct option/ 33% of the girls selected option ji/ 20% chose option c / and 20% option d. For males/ approximately 25% selected option s and the same percent selected option d. Marshall concluded that although the analysis of incorrect responses does not explain why boys and girls differ in their responses/ the analysis shows that boys and girls approach problems in different ways and these varying strategies can be useful in indentifying how the sexes differ in reasoning abilities. Two years later/ Marshall (1983) analyzed the responses of approximately 300/000 boys and girls to mathematics items contained in the 16 test forms of the Survey of Basic Skills during the years 1977/ 1978/ and 1979. She used log-linear models (explained in Chapter 106 operation to the numbers in the problem. The re sults indicate that students are not familiar with such basic problem-solving strategies as drawing a picture of a figure described in a problem or checking the reasonableness of a result, (p. 338) The fact that/ in general/ after controlling for computational skills/ there were no statistically significant differences between the problem solving performance of the males and females who participated in this study would seem to imply that teachers need not worry about designing different teaching strategies for the sexes. However/ because the performance in problem-solving is low for both sexes/ teachers may be encouraged to emphasize problem solving in their daily teaching. This study was designed to investigate sex-related differences in mathematical performance. The findings clearly demonstrated that 6th grade females and males from the public schools in Puerto Rico are equally good or equally bad in the solution of word-story items (problem solving). Also/ in general/ the effect of computational skills is similar for each sex. Based on these findings/ it appears that greater differences exist within each sex than between the sexes. In conclusion/ the findings of the present study highlight the need for investigations of different approaches to instructional designs aimed at improving both male and female performance in problem solving. 82 TABLE 4.3 Means* Standard Deviations* and t-Tests for the Eight Mathematical Subtests Year/ Subtest Males X SD Females X SD t First Problem Solving 3.436 2.32 3.862 2.45 2.82* Addition 5.412 1.00 5.592 .86 3.05* Subtraction 4.550 1.66 4.852 1.56 1.22 Multi plicati on 4.014 1.80 4.433 1.68 3.81* Division 3.136 1.87 3.580 1.91 3.72* Fracadd 2.475 1.75 2.743 1.81 2.32** Decsub 2.648 1.83 3.321 1.86 5.77* Equivalence .77 .77 .73 .76 - .83 N = 492 N = 510 Second Problem Solving 3.632 2.28 4.015 2.36 2.65* Addition 5.470 .98 5.603 .86 2.29** Subtraction 4.843 1.52 4.923 1.44 .86 Multiplication 4.148 1.73 4.550 1.67 3.76* Division 3.242 1.77 3.632 1.86 3.42* Fracadd 2.565 1.74 2.903 1.91 2.94* 20 As a result of her investigations/ Marshall (1981) concluded that sex-related differences in mathematics performance may be the result of comparing the sexes on total test scores. If the test contains more computation items than problem solving items/ girls will perform better than boys/ but if the test contains more problem solving items than computation ones/ boys will outperform girls. With this in mind/ Marshall investigated sex-related differences in computation and problem solving by analyzing the responses of approximately 18/000 students from grade 6 who had been administered the Survey of Basic Skills Test: Grade 6/ during the academic year 1978-79. Two of the 16 test forms of the Survey were used to assess skills such as concepts of whole numbers/ fractions/ and decimals. These skills were tested both as simple computations and as story problems (problem solving). Two computation items and one story problem item were selected because they were particularly related; both computation items required skills needed in solving the corresponding story problem. It was assumed that correct solution of the computation item correlates with solving the story problem because the story problem requires a similar computation. Marshall found that girls were better in computation and boys were better in problem solving. She also found that boys were much more likely than girls to answer the 51 student attitudes and performance in the area of problem solving. Carey (1958) constructed a scale to measure attitudes toward problem solving. Her interest was in general problem solving rather than in mathematical problem solving. Her work constitutes the first attempt to construct a measure of attitudes toward problem solving. The scale was used with a group of college students* and she found* among other things* that problem solving performance is positively related to problem solving attitudes and that* in the case of females* positive modification of attitudes toward problem solving brings a significant gain in problem solving performance. Lindgren* Silva* Faraco* and DaRocha (1964) adapted Carey's scale of attitudes toward problem solving and applied it to a group of 4th grade Brazilian children. Students also answered an arithmetic achievement test* a general intelligence test* and a socioeconomic (SE) scale. A low but significant positive correlation was found between arithmetic achievement and attitudes toward problem solving. A near zero correlation was found between attitudes toward problem solving and intelligence. Since problem solving is one aspect of the discipline of mathematics* this correlation between attitudes and arithmetic achievement can lead to a conclusion or a strong 47 than the problem-interpretation scorer the difficulty was due to reading ability. If the score of the lowest of the three areas (computationr problem interpretation/ and reading-problem interpretation score) was the same as the reading-problem solving score/ the student's area of greatest immediate need was either computation/ problem interpretation/ or reading. If the reading-problem solving score was lower than the lowest of the other three scores/ the student's area of greatest immediate need was integration. Analysis of the data revealed that for 19% of the students/ problem interpretation was their major difficulty; for 26% of the students/ integration (total problem solving) was their greatest immediate need; for another 26%/ computation was the major weakness; and for 29%/ reading was their greatest immediate need. Seventy five percent of the students demonstrated clear strength in computation/ 21% in problem interpretation/ and 4% in reading-problem interpretation. An analysis across all students (including those without complete data) showed that 26% of the subjects could not work word problems at a level as high as that at which they could compute/ interpret problems/ and read and interpret problems/ when those areas were measured separately. This led them to conclude that knowing the skills or the components of solving word problems is not sufficient for 62 Here the summation is over all the cells in the contingency table/ F0 refers to the observed frequency in a cell/ and Fe refers to the estimated expected frequency in a cell. Denoting the observed proportion in a cell by P0 and the estimated expected proportion in a cell by Pe/ the test statistic can be written G2 = 2 N J2Po 1 9e (Po/pe}- (3.5) where N is the total number of subjects. This form of the test statistic suggests the following strategy. For any significant G2/ using P0 and Pe calculated from the total data set available for an item/ calculate the minimum N required for G2 to be significant. If the minimum N is very large/ this suggests that the statistically significant result is not practically significant since it can only be detected in very large samples. Of course/ the question remains as to what can be considered a minimum large N. Although there is room for argument/ it seems reasonable to claim that if an average of 1000 subjects per year is required to show significance/ then the result is not likely to be practically significant. On the basis of this reasoning/ it was proposed to ignore all significant results that would be nonsignificant if there were less than 3000 subjects available. In addition/ all log-linear BIOGRAPHICAL SKETCH Sonia Feliciano was born on March 17/ 1938/ in Mayaguez/ Puerto Rico. She graduated from Eugenio Ma. de Hostos High School/ in her hometown. She received a degree in business administration/ with a major in finance/ from the University of Puerto Rico in June/ 1961. After graduation she worked as junior accountant for Pittsburgh Plate Glass/ International. During the academic year of 1969-70 she started her career in education/ obtaining the degree of Master in Education from the University of Puerto Rico in June/ 1975. From January/ 1971 to June/ 1985/ she served in the public school system of Puerto Rico in different positions. She worked as educational researcher until July/ 1977/ when was appointed Director of the Research Center of the Commonwealth Department of Education. During the academic year of 1978-79/ she initiated doctoral studies in the Faculty of Philosophy and Educational Sciences of the Universidad Complutense de Madrid/ in Madrid/ Spain. She completed the coursework requirements and then transferred to University of Florida during the fall of 1979. 116 52 correlation between attitudes toward problem solving and problem solving performance. Whitaker (1976) constructed a student attitude scale to measure some aspects of 4th grade student attitudes toward mathematic problem solving. He included statements reflecting children's beliefs about the nature of various types of mathematical problems the nature of the problem solving process the desirability or persevering when solving a problem and the value of generating several ideas for solving a problem. He correlated student attitudes toward problem solving with their scores in a mathematical test which yielded a comprehension score an application score and a problem solving score. He found a significant positive relationship between problem solving performance and student attitude scores on the subscale which measured reactions to such things as problem solving techniques or problem situations or to the frustation or anxiety experienced when confronted with problem solving situations. In another part of this study Whitaker investigated the relationship between the attitudes of 4th grade teachers toward problem solving and their students' performance in problem solving. A very weak and nonsignificant negative correlation was found between the teacher's attitudes toward problem solving and student performance. 20 As a result of her investigations Marshall (1981) concluded that sex-related differences in mathematics performance may be the result of comparing the sexes on total test scores. If the test contains more computation items than problem solving items girls will perform better than boys but if the test contains more problem solving items than computation ones boys will outperform girls. With this in mind Marshall investigated sex-related differences in computation and problem solving by analyzing the responses of approximately 18000 students from grade 6 who had been administered the Survey of Basic Skills Test: Grade 6 during the academic year 1978-79. Two of the 16 test forms of the Survey were used to assess skills such as concepts of whole numbers fractions and decimals. These skills were tested both as simple computations and as story problems (problem solving). Two computation items and one story problem item were selected because they were particularly related; both computation items required skills needed in solving the corresponding story problem. It was assumed that correct solution of the computation item correlates with solving the story problem because the story problem requires a similar computation. Marshall found that girls were better in computation and boys were better in problem solving. She also found that boys were much more likely than girls to answer the 91 TABLE 4.7 Reliability Three of the Years of Covariates for Each of Test Administration the Year Covariate First Second Third Addition .579 .610 .454 Subtraction .756 .724 .737 Multiplication .729 .729 .734 Division .716 .680 .714 Fracadd .709 .728 .748 Decsub .720 .716 .742 Equivalence .421 .394 .488 LIST OF TABLES Table Page 3.1 Hypothetical Probabilities of Option Choice Conditional on Year and Sex 58 3.2 Hypothetical Probabilities of Option Choice Conditional on Year and Sex and Arranged by Sex 58 3.3 Hypothetical Joint Probabilities of Year/ Option/ and Sex 64 3.4 Hypothetical Probabilities of Option Choice Conditional on Sex and Year 66 4.1 Summary of Significant Tests of Model 3.2 71 4.2 Chi-square Values for the Comparison of 75 Models (3.2) and (3.3) with Actual Sample Sizes and with the Corresponding Number of Subjects Needed for Significant Results 75 4.3 Means/ Standard Deviations/ and t-Test for the Eight Mathematical Variables 82 4.4 ANCOVA First Year: Multiplication Covariate 85 4.5 ANCOVA First Year: Division Covariate 85 4.6 ANCOVA Second Year: Subtraction Covariate... 86 4.7 Reliability of the Covariates for Each of the Three Years of Test Administra tion 91 4.8 ANCOVA First Year: Other Covariates 92 4.9 Adjusted Means on Problem Solving/ by Covariate and Sex 94 4.10 ANCOVA Second Year: Other Covariates 95 v LIST OF TABLES Table Page 3.1 Hypothetical Probabilities of Option Choice Conditional on Year and Sex 58 3.2 Hypothetical Probabilities of Option Choice Conditional on Year and Sex and Arranged by Sex 58 3.3 Hypothetical Joint Probabilities of Yearr Option/ and Sex 64 3.4 Hypothetical Probabilities of Option Choice Conditional on Sex and Year 66 4.1 Summary of Significant Tests of Model 3.2 71 4.2 Chi-square Values for the Comparison of 75 Models (3.2) and (3.3) with Actual Sample Sizes and with the Corresponding Number of Subjects Needed for Significant Results 75 4.3 Means/ Standard Deviations/ and t-Test for the Eight Mathematical Variables 82 4.4 ANCOVA First Year: Multiplication Covariate 85 4.5 ANCOVA First Year: Division Covariate 85 4.6 ANCOVA Second Year: Subtraction Covariate... 86 4.7 Reliability of the Covariates for Each of the Three Years of Test Administra tion 91 4.8 ANCOVA First Year: Other Covariates 92 4.9 Adjusted Means on Problem Solving/ by Covariate and Sex 94 4.10 ANCOVA Second Year: Other Covariates 95 v 22 Approximately 270/000 students from the 6th grade were administered the Survey of Basic Skills of the California Assessment Program/ during the years 1977/ 1978/ and 1979. Responses were analyzed using log-linear models. Successful solving of computation items was positively associated with successful solving of story problems. Girls were more successful in computation than boys/ and boys were more successful than girls in solving story problems. This finding supports reports from the National Assessment of Educational Progress (NAEP) (Armstrong/ 1979) . To investigate the effects of reading/ SES/ language/ and chronological age/ only those test forms containing 2 computation items and 2 story problems were considered for analysis; 32 items from 8 test forms were included in the analysis. The results of these analyses showed that at every level of reading score/ 6th grade children were more successful in computation than in story problems. Although the differences were not large/ at every reading score boys consistently had higher probabilities of success in story problems than did girls/ and girls consistently showed higher probabilities of success in computation than boys. Also/ as the reading score increased/ the difference between the probability of success in story problems and the probability of success in 94 TABLE 4.9 Adjusted Means on Problem Solvingf by Covariate and Sex First Year Covariate Adjusted Means Males Females Addition 3 .53 3 .78 Subtraction 3 .55 3.76 Fracadd 3.56 3.74 Decsub 3.73 3.58 Equivalence 3 .38 3.91 17 attitude toward mathematical problem solving/ and related sex and program-type differences. Although his main objective was to construct an attitude scale to measure attitudes toward problem solving/ his study is important because his findings support Meyer's regarding the lack of significant sex-related differences in problem solving performance. Performance in the problem solving questions/ for both males and females/ was much lower than performance in the application questions/ and much lower than performance in the comprehension questions. In fact/ the mean score for each part of the item/ for both males and females/ was almost identical to the mean scores obtained by males and females in Meyer's study. Whitaker noted that each application item is more difficult than its preceding comprehension item/ and that each problem solving item is more difficult than its preceding application item. No significant sex-related differences were found for any of the three parts of the item (comprehension/ application/ or problem solving). Fennema and Sherman (1978) investigated sex-related differences in mathematics achievement and cognitive and affective variables related to such differences. They administered the Romberg-Wearne Problem Solving Test to a representative sample of 1320 students (grades 6-8) from Madison/ Wisconsin/ predominantly middle-class/ but including great diversity in SES. The sample consisted of 19 solving performance in grade 11. Spatial visualization was a stronger predictor for girls than for boys. Mathematics as a male domain was a good predictor for girls only; the less a girl stereotyped mathematics as a male domain in grade 8/ the higher her problem solving score in grade 11. Attitudes toward success in mathematics in grade 8 was a more positive predictor of problem solving performance for boys than for girls; the more positive the attitudes toward success in mathematics in grade 8/ the higher their performance in problem solving in grade 11. None of these four studies all of which used the Romberg-Wearne Mathematics Problem Solving Test show statistically significant sex-related differences in problem solving performance. In later studies other tests were used to measure this variable (Kaufman 1984; Marshall 1981 1984) . Kaufman (1983/1984) investigated if sex differences in problem solving favoring males exist in the 5th and 6th grades and if these differences were more pronounced in mathematically gifted students than in students of average mathematical ability. The Iowa Test of Basic Skills and a mathematics problem-solving test were administered to 504 subjects. Males in the average group as well as males in the gifted group outperformed females but only the gifted group showed statistically significant differences. 7 4 TABLE 4.1 Continued Item Number X2 for Model 3.2* Actual Sample Size Min. Sample Size for Significance la) (b) (c) (d) 105 28.66 60/712 19/510 108 26 .56 86/261 29/912 111 30.71 86/190 25/849 * p<.01 ** Minimum sample size less than 3/000 Abstract of Dissertation Presented to the Graduate School of the University of Florida in Partial Fulfillment of the Requirements for the Degree of Doctor of Philosophy COMPARISON OF MALES AND FEMALES ON MATH ITEM PERFORMANCE: ANALYSIS OF RESPONSE PATTERNS By Sonia Feliciano August/ 1986 Chairman: James Algina Major Department: Foundations of Education The first objective of this study was to investigate sex differences in the selection of incorrect responses on a mathematics multiple-choice test/ and to determine whether these differences were consistent over three consecutive administrations of the test. A second objective was to compare male and female performance in problem solving after controlling for computational skills. The responses of all 6th grade students from the public schools in Puerto Rico who took the Basic Skills Test in Mathematics-6 ("Prueba de Destrezas Bsicas en Matemticas-6") during three academic years were used in the analyses relevant to the first objective. VXll 14 Sex-related Differences in Problem Solving It has already been acknowledged that the subject of problem solving has been extensively researched. However* as early as 1969* Kilpatrick criticized the fact that the study of problem solving has not been systematic; some researchers have studied the characteristics of the problem while others have given their attention to the characteristics of the problem solvers. Moreover* differences in the tests used to measure problem solving performance also constitute an obstacle when trying to compare the results of the studies carried out. In order to avoid this pitfall and provide a basis for comparison* the studies reviewed in this section* dealing with sex-related differences in problem solving* have been divided in two groups. The first comprises those studies that used the Romberg-Wearne Problem Solving Test. The second contains other relevant studies in which problem solving performance has been measured by means of other instruments. The Romberg-Wearne Problem Solving Test merits special mention because it was the first attempt "to develop a test to overcome the inadequacies of total test scores in explaining the reasons why some students are successful problem solvers and others are not" (Whitaker* 1976* pp. 9* 10) . 57 Analysis of the Data Analysis of Sex bv Option by Year Cross Classifications Log-linear models were used to analyze the sex by option by year cross classifications for each item. Two topics are addressed in this section/ the hypotheses tested using log-linear models and a comparison of the hypotheses tested in this study with those tested by Marshall (1981/ 1983) . The object of the analysis was to test two hypotheses: 1. The proportion of males and the proportion of females choosing each incorrect option does not vary from year to year. Note that this hypothesis is stated in the null form. 2. Assuming that the first hypothesis is correct/ the proportion of males who choose each incorrect option is different from the proportion of females who choose each incorrect option. This hypothesis is stated in the alternate form. In Table 3.1 a hypothetical cross classification of sex/ option/ and year is presented. Hypothesis 1 is true for this three dimensional contingency table. In Table 3.2 the three dimensional contingency table is rearranged to show the year by option contingency table for each gender. 82 TABLE 4.3 Means* Standard Deviations* and t-Tests for the Eight Mathematical Subtests Year/ Subtest Males X SD Females X SD t First Problem Solving 3.436 2.32 3.862 2.45 2.82* Addition 5.412 1.00 5.592 .86 3.05* Subtraction 4.550 1.66 4.852 1.56 1.22 Multi plication 4.014 1.80 4.433 1.68 3.81* Division 3.136 1.87 3.580 1.91 3.72* Fracadd 2.475 1.75 2.743 1.81 2.32** Decsub 2.648 1.83 3.321 1.86 5.77* Equivalence .77 .77 .73 .76 - .83 N = 492 N = 510 Second Problem Solving 3.632 2.28 4.015 2.36 2.65* Addition 5.470 .98 5.603 .86 2.29** Subtraction 4.843 1.52 4.923 1.44 .86 Multi plication 4 .148 1.73 4.550 1.67 3.76* Division 3.242 1.77 3.632 1.86 3.42* Fracadd 2.565 1.74 2.903 1.91 2.94* CHAPTER V DISCUSSION This cnapter comprises two sections: a) summary and interpretation of the results of the study; and b) implications of the findings/ with suggestions for further research. Summary and Interpretation of the Results The main objective of this study was to investigate whether there were sex-related differences in incorrect responses/ and if the pattern of response was consistent over three consecutive administrations of the mathematics test. Males and females selected different incorrect responses in 100 of the 111 items of the test and this pattern of response was consistenly found in each of the three years of test administration. The results tend to support/ in part/ Marshall's findings (1981/ 1983) of observed sex-related differences in incorrect responses/ but in no way do they point to a conclusion that 6th grade boys and girls from the public schools in Puerto Rico exhibit different problem solving strategies. The significant differentiated pattern of 101 31 that in their study involving students from grades 6 through 12 boys showed a higher level of confidence in mathematics at each grade level. These differences between the sexes occurred in most instances even when no sex-related differences in mathematics achievement were found. The correlation between confidence in mathematics performance and mathematics achievement in this study was higher than for any other affective variable investigated. Sherman reported a similar finding in 1980; in males the most important factor related to continuation in theoretical mathematics courses was confidence in learning mathematics. This variable weighed more than any of the cognitive variables: mathematics achievement spatial visualization general ability and verbal skill. In the case of females among the affective variables confidence in learning mathematics was found to be second in importance to perceived usefulness of mathematics. Probert (1983) supported these findings with college students. A variable that needs discussion within the context of sex differences in confidence as learners of mathematics is causal attribution. Causal attribution models attempt to classify those factors to which one attributes success or failure. The model proposed by Weiner (1974) categorizes four dimensions of attribution of success and failure: stable and internal unstable and internal stable and external and unstable and external. For example if one 97 TABLE 4.12 ANCOVA Summary Table Third Year Source df ss MS F Addition 1 205.96 205 .96 34.85* Sex 1 1.88 1.88 .32 Error 1010 5967.60 5.91 Subtraction 1 886.42 886.42 169.48* Sex 1 24.39 24.39 4.66* Error 1010 5287.2 5.23 Multi pi icati on 1 1402.42 1402.42 297.12* Sex 1 1.05 1.05 .22 Error 1010 4771.20 4.72 Division 1 2028.49 2028.49 494.75* Sex 1 1.11 1.11 .27 Error 1010 4145 .10 4.10 Fracadd 1 1733.72 1733.72 393.97* Sex 1 4.64 4.64 1.05 Error 1010 4439.89 4 .40 Decsub 1 2116.84 2116.84 526 .57* Sex 1 .15 .15 .04 Error 1010 4056 .76 4.02 Equivalence 1 1186.92 1186.92 240.26* Sex 1 59.85 59.85 12.12* Error 1010 4986.69 4.94 * p<. 05 35 It is not until adolescence that sex differences in the perception of mathematics as a male domain are found (Fennemar 1976; Stein/ 1971; Stein & Smithless/ 1969; Verbeke/ 1983). In a study with 2nd/ 6th and 12th graders/ Stein and Smithless (1969) found that students' perceptions of spatial/ mechanical/ and arithmetic skills as masculine became more defined as these students got older. Fennema (1976) considers that the influence each sex exerts upon the other on all aspects of behavior is stronger during adolescence. Since during these years males stereotype mathematics as a male domain/ they send this message to females who/ in turn/ tend to be influenced in their willingness to study or not to study mathematics. Before that stage/ girls consider arithmetic feminine/ while boys consider it appropriate for both sexes (Bobbe/ 1971). Usefulness of mathematics. Females perceive mathema tics as less useful to them than males do/ and this perception occurs at a very young age. As a result/ females exert less effort than males to learn or elect to take advanced mathematics courses. Many studies reported before 1976 found that the perception of the usefulness of mathematics for one's future differs for males and females/ and is related to course taking plans and behavior (Fox/ 1977). If females do not perceive mathematics as useful for their future/ 52 correlation between attitudes toward problem solving and problem solving performance. Whitaker (1976) constructed a student attitude scale to measure some aspects of 4th grade student attitudes toward mathematic problem solving. He included statements reflecting children's beliefs about the nature of various types of mathematical problems/ the nature of the problem solving process/ the desirability or persevering when solving a problem/ and the value of generating several ideas for solving a problem. He correlated student attitudes toward problem solving with their scores in a mathematical test which yielded a comprehension score/ an application score/ and a problem solving score. He found a significant positive relationship between problem solving performance and student attitude scores on the subscale which measured reactions to such things as problem solving techniques or problem situations/ or to the frustation or anxiety experienced when confronted with problem solving situations. In another part of this study/ Whitaker investigated the relationship between the attitudes of 4th grade teachers toward problem solving and their students' performance in problem solving. A very weak and nonsignificant negative correlation was found between the teacher's attitudes toward problem solving and student performance. I certify that opinion it conforms presentation and is as a dissertation for I have read this study and that in my to acceptable standards of scholarly fully adequatef in scope and qualityf the degree of Doctor of Philosophy. is Algina/ Cl iessor of Foi irman dations of Education I certify that opinion it conforms presentation and is as a dissertation for I have read this study and that in my to acceptable standards of scholarly fully adequatef in scope and qualityf the degree of Doctor of Philosophy. Linda Crocker Professor of Foundations of Education I certify that opinion it conforms presentation and is as a dissertation for I have read this study and that in my to acceptable standards of scholarly fully adequatef in scope and qualityf the degree of Doctor of Philosophy. .chael Professor r njary of Educational Leadership This dissertation was submitted to the Graduate Faculty of the College of Education and to the Graduate Schoolf and was accepted as partial fulfillment for the degree of Doctor Augustf Deanf Graduate School 10 of the studies consider sex differences incidental to the major study findings. The available literature offers very little research directly related to the problem of sex differences in this area. The review of the literature has been divided in four sections. The first section consists of a detailed summary of the available research on sex diferences in incorrect responses. The second section deals with sex-related differences in problem solving performance. These sections are directly related to the objectives of the study. The third section is more peripheral/ and contains a discussion of the more prevalent issues about the influence of cognitive and affective variables on sex differences in mathematics learning and achievement. The fourth is a summary of the research dealing with variables considered as influential to mathematics problem solving performance. Sex-related Differences in incorrect^Response Patterns Research findings tend to suggest that boys and girls may be approaching problem solving differently (e.g./ Fennema and Sherman# 1978; Marshall# 1981# 1983; Meyer# 1978# among others). Marshall (1981) investigated whether 6th grade boys and girls approach mathematical problem solving with different strategies. Her specific interest was whether the sexes made the same errors. 60 Fienberg also shows that deleting the Ujk term from (3.2) to obtain log mij k = Ui + Uj+Uk + Uik. (3.3) yields a model that specifies that option is independent of sex. Based on Fienberg1 s presentation an appropriate analysis for testing the hypotheses is 1. Conduct a likelihood ratio test of the adequacy of model (3.2). If this test is nonsignificant then the data support the adequacy of the model and the hypothesis that conditional on gender option and year are independent. Because model (3.1) is a saturated model testing the adequacy of model (3.2) is the same as comparing the adequacies of models (3.1) and (3.2). 2. Conduct a likelihood ratio test comparing the adequacies of models (3.2) and (3.3). If this test is significant then model (3.2) fits the data better than model (3.3) and the data support the hypothesis that the choice of incorrect option is dependent on sex. To summarize if the first test is nonsignificant and the second test is significant then the choice of option is dependent on sex and the pattern of dependency is the same for all three years. 117 From May 1982 to July* 1983 she worked as graduate research assistant at the Foundations of Education Department University of Florida. She returned to Puerto Rico in August 1983 to serve as Special Aide to the Assistant Secretary for the Vocational Technical and High Skills Educational Programs of the Commonwealth Department of Education. 110 Fennema E.r & Sherman/ J. (1977). Sex-related differences in mathematics achievement/ spatial visualization/ and affective factors. American Educational Research Journal/ 14/ 51-71. Fennema/ E. / & Sherman/ J. A. (1978). Sex-related differences in mathematics achievement and related factors. A further study. Journal for Research in Mathematics...Education* h 189-204. Fennema/ E./ & Tartree/ L. A. (1983). Research on relationship of._spatisl...visualizatiQD,_and.confidence to male/female mathematics achievement in grades 6-8. Washington/ DC: National Science Foundation. (ERIC Document Reproduction Service No. ED 232 853) Fienberg/ s. e. (1980). Th.e...Analys.is..pf.CcQS.s=.ciaasified Categorical Data. Cambridge/ MA: The MIT Press. Fox/ L. H. (1975a). Mathematically precocious: Male or female. In E. Fennema (Ed.) Mathematics learning: what_rÂ£SÂ£arch--say-S-3bQut sea;_differences (pp. 1-12). Columbus: Ohio State University/ College of Education. (ERIC Document Reproduction Service No. ED 128 195) Fox/ L. H. (1975b). Sex differences: Implications for program_planning for..the academically giftedt Paper presented at the Lewis M. Terman Memorial Symposium on Intellectual Talent held at the Johns Hopkins University/ Baltimore/ MD. Fox/ L. H. (1977) The effects of sex role socialization on mathematics participation and achievement. In J. shoemaker (Ed.)/ women_and...Mathema.tics: Research Perspectives for Change/ (N.I.E. Papers in Educa tion and Work: No. 8). Washington/ D. C.: Eduation and Work Group/ The National Institute of Education. Glennon/ V. J./ & Callahan/ L. G. (1968) Elementary school mathematics; A^guide tc current research/ Washington/ D. C.: Association for Supervision and Curriculum Development. Guay/ R. B. / & McDaniel/ D. (1977). The relationship between mathematics achievement and space ability among elementary school children. Journal for Research.-in.-Mathamatics Education* Â£* 211-215. 18 students who had taken a similar number of mathematics courses and were in the top 85% of the class in mathematics achievement. They were tested in 1976. Four high school districts were included. In only one of the high school districts were sex-related differences in application and problem solving found* in favor of males. They concluded that when relevant factors are controlled* sex-related differences in favor of males do not appear often* and when they do* they are not large. Sherman (1980) investigated the causes of the emerging sex-related differences in mathematics performance* in favor of males* during adolescence (grades 8-11). She wanted to know if these differences emerge as a function of sex-related differences in spatial visualization and sociocultural influences that consider math as a male domain. In grade 8* she used the Romberg-Wearne Test and* in grade 11* a mathematical problem solving test derived from the French Kit of Tests. The analysis showed that for girls* problem solving performance remained stable across the years. Mean problem solving performance for boys* however* was higher in grade 11 than in grade 8. No sex-related differences were found in grade 8* but boys outperformed girls in grade 11* where the Stafford test was used. Sherman found that for both sexes problem solving performance in grade 8 was the best predictor of problem TABLE OF CONTENTS Page ACKNOWLEDGMENTS ii LIST OF TABLES V LIST OF FIGURES vii ABSTRACT viii CHAPTER I INTRODUCTION 1 Purpose of the Study 6 Significance of the Study 6 Organization of the Study 8 II REVIEW OF THE LITERATURE 9 Sex-related Differences in Incorrect Response Patterns 10 Sex-related Differences in Problem Solving 14 Cognitive and Affective Variables that Influence Sex Differences in Mathematics Learning and Achievement 23 Differences in Formal Mathematics Education 24 Differences in Spatial Ability 26 Differentiated Effect of Affective Variables 30 Problem Solving Performance and Related Variables 37 Computational Skills and Problem Solving Performance 38 Reading and Problem Solving Perfor mance 44 Attitudes Toward Problem Solving and Problem Solving Performance 50 iii MATH CONCEPTS Fig. 3.1 Number of Questions by Skill Area in the Basic Skills Test in Mathematics 6 xml version 1.0 encoding UTF-8 REPORT xmlns http:www.fcla.edudlsmddaitss xmlns:xsi http:www.w3.org2001XMLSchema-instance xsi:schemaLocation http:www.fcla.edudlsmddaitssdaitssReport.xsd INGEST IEID E7GIIQLHZ_4RJU7A INGEST_TIME 2012-09-24T12:49:43Z PACKAGE AA00011802_00001 AGREEMENT_INFO ACCOUNT UF PROJECT UFDC FILES 28 1974 have failed to support these sex differences in spatial abilities (Armstrong/ 1979; Connor/ Serbin, & Schackmanr 1977; Fennema & Sherman/ 1978; Sherman/ 1979). Fennema and Sherman (1978) and Sherman (1979) have explored sex-related differences in mathematical achievement and cognitive and affective variables related to these differences. In a study involving students from grades 6/ 7/ and 8/ from four school districts/ Fennema and Sherman found that spatial visualization and problem solving were highly correlated for both sexes (.59 and .60). Even in the school district where sex differences were found in problem solving/ no significant sex-related differences were found in spatial visualization. When Sherman (1980) compared groups of males and females in two different grades/ 8 and 11/ she found no sex-related differences in problem solving or in spatial visualization in grade 8. In grade 11/ however/ although the sexes differed in their problem solving performance/ no sex-related differences were found in spatial vizualization. Even though spatial visualization in grade 8 was the second best predictor of problem solving performance in grade 11/ sex differences in grade 11 were not a result of spatial visualization since no differences were found in that skill. In spite of the fact that no sex differences were found in spatial abilities/ it is evident that males and 3 indicated no significant sex differences for 8th grade students on measures of problem solving or algebra. However* females outperformed males in computation and spatial visualization. For the 12th grade students* statistically significant sex differences favoring males were found in problem solving* but not in algebra* computation* or spatial visualization. For males and females who had enrolled in courses beyond general mathematics and who had taken or were enrolled in courses such as pre-calculus* calculus* or geometry* differences in problem solving or spatial visualization did not exist. Sex differences favoring males were found on a total score obtained summing across the computation* problem solving* and algebra subtests (Armstrong* 1979). The mathematics data collected in the second survey by the 1978 National Assessment of Educational Progress showed significant sex differences for both 13- and 17-year-old students. The 13-year-old females outperformed males in the computational subtest and males outscored females by 1 1/2 percentage points in problem solving (statistically significant). No statistically significant differences were found in algebra. No sex differences were found for the 17-year-old group either in the computation subtest or in the algebra subtest. Males surpassed females in problem solving. A reanalysis of the data from the 17-year-old group confirmed male superiority in problem solving after controlling for 113 Meyer# R. A. (1978). Â£g.xrJg.la.ted_diffgJ.gi3c.es.j,p_math.ema:- ti,cal_problem_ solving performance and intellectual abilities. Madison: University of Wisconsin# Research and Development Center for Cognitive Learning. (ERIC Document Reproduction Service No. ED 166 023) . Milton# G. A. (1957) The effects of sex-role identifica- cation upon problem solving skill. Journal of Abnormal & Social.Psychology# 55.# 208-212. Moller# N. J. (1983). The impact of gender# masculinity# and feminity on math achievement and course decisions (Doctoral dissertion# Purdue University# 1982). Hiss ej.t a.tion,Abstrac.te_ International# l# 2584A. Muth# K. D. (1984). Solving arithmetic word problems: Role of reading and computational skills. Journal of Educational. Psychology# 15# 205-210. Pallas# A. M.# & Alexander# K. L. (1983). Sex differences in quantitative SAT performance: New evidence on the differential coursework hypothesis. American Idji^atj^p.aJ.,Res.ea_rc.li.Journal* 21# 165-182. Plank# E.# & Plank# R. (1954). Emotional components in arithmetical learning as seen through autobiographies. Psychological Studies of the Child# Â£# 274-296. Porter# A. C. (1968). The effects of using fallible variables in the analysis of covariance (Doctoral dissertation# University of Wisconsin# 1967). Di.s.s.er.tatifih-^abgtjact.,.Inte.r.P,a-t.ip.nai# 25# 3517B. Probert# B. S. (1984). Math confidence workshops: A multimodal group intervention strategy in mathematics anxiety/avoidance. (Doctoral dissertation# University of Florida# 1983). nisÂ£.er.tatiQp_Abgtracts International# 2231B. Puerto Rico Department of Education. (1980). Basic skills,test in mathematics-6. Hato Rey: Center of Evaluation. Radatz# H. (1979) Error analysis in mathematics educa tion. jQumai fo.r.jtene.arch .in Majieinatl_c,s_fcjuc.a- tion# lfi.# 163-172. Riedesel# A. C. (1969) Problem solving: Some suggestions from research. Arithmetic Teacher# 15# 54-58. 13 III) to investigate sex related differences in the selection of incorrect responses/ and the consistency of such differences over three years of administration of the test. Based on her findings that sex differences were found in 80% of the items/ Marshall classified the students' errors according to Radatz' (1979) five-category error classification. The categories are language (errors in semantics)/ spatial visualization/ mastery/ association/ and use of irrelevant rules. It was found that girls' errors are more likely to be due to the misuse of spatial information/ the use of irrelevant rules/ or the choice of an incorrect operation. Girls also make relatively more errors of negative transfer and keyword association. Boys seem more likely than girls to make errors of perseverance and formula interference. Both sexes make language-related errors/ but the errors are not the same. Available research is not extensive enough to make definite judgments about the sex-related differences observed in incorrect responses. Clearly more research is needed. 105 exr_related.,Differences in Problem Solving Performance The results of this study do not give strong support to Marshall's (1981/ 1984) findings that the effect of computational skills on problem solving performance is different for each sex. However/ the findings of the present investigation are consistent with those found by Meyer (1978) and Whitaker (1976): female performance in problem solving is not significantly different from male performance. Females showed superiority in problem solving performance in all three years of test administration although their superiority was retained only when performance in equivalence was controlled for (for the three years data) and when performance in subtraction was controlled for (second year) (statistically significant only after accounting for equivalence) . The findings of this study are also in agreement with those reported by the National Assessment of Educational Progress (NAEP) (Carpenter et al.z 1980). Related to problem solving/ they conclude/ if it were necessary to single out one area that demands urgent attention/ it would clearly be problem solving. At all age levels/ and in vir tually every content area/ performance was extre mely low on exercises requiring problem solving or application of mathematical skills. In gene ral/ respondents demonstrated a lack of the most basic problem-solving skills. Rather than attempt ing to think through a problem and figure out what needed to be done to solve the problem/ most res pondents simply tried to apply a single arithmetic 17 attitude toward mathematical problem solving* and related sex and program-type differences. Although his main objective was to construct an attitude scale to measure attitudes toward problem solving* his study is important because his findings support Meyer's regarding the lack of significant sex-related differences in problem solving performance. Performance in the problem solving questions* for both males and females* was much lower than performance in the application questions* and much lower than performance in the comprehension questions. In fact* the mean score for each part of the item* for both males and females* was almost identical to the mean scores obtained by males and females in Meyer's study. Whitaker noted that each application item is more difficult than its preceding comprehension item* and that each problem solving item is more difficult than its preceding application item. No significant sex-related differences were found for any of the three parts of the item (comprehension* application* or problem solving). Fennema and Sherman (1978) investigated sex-related differences in mathematics achievement and cognitive and affective variables related to such differences. They administered the Romberg-Wearne Problem Solving Test to a representative sample of 1320 students (grades 6-8) from Madison* Wisconsin* predominantly middle-class* but including great diversity in SES. The sample consisted of 26 maintained their superiority in mathematics ability during high school. While Fox attributed sex-related differences in mathematical achievement to differential exposure to mathematical games and activities outside school Benbow and Stanley suggested that sex-related differences in mathematics performance stem from superior mathematical ability in males not from differences in mathematics formal education. The differential coursework hypothesis is not totally convincing and as reporteu berore it has been challenged by researchers such as Benbow and Stanley (1980). However Pallas and Alexander (1983) have questioneu the generalizanility of Benbow and Stanley's findings based on the fact that they used highly precocious learners. The differential coursework hypothesis can be accepted only as a partial explanation of differences in mathematics performance found between the sexes. JUJ-Â£Â£j^gÂ£.s-Jn_Sp.fltial Ability The basic premise in this issue is that males and females differ in spatial visualization and this explains differential mathematics learning and achievement. Until recently sex differences in spatial ability in favor of males were believed to be a fact and were thought by some to be related to sex differences in mathematical achievement CHAPTER II REVIEW OF THE LITERATURE Sex differences in mathematics learning and achievement have been a subject of concern for educators and psychologists. Many studies found in the literature support the existence of these differences. Boys show superiority in higher level cognitive tasks (problem solving or mathematical reasoning) in the upper elementary years and in the early high school years (Fennema/ 1974; Maccoby & Jacklin/ 1974) . Almost all the research carried out has dealt with analysis of total correct scores in mathematics aptitude and achievement tests or scores in subtests. The literature related to sex differences in incorrect responses the main subject of the present study/ is surprisingly sparse. For the most part/ the studies have investigated the differences between the sexes in mathematics learning and achievement and the underlying variables causing the differences. Cognitive and affective variables have been the matter at issue in the establishment of sex differences. Although research in mathematics problem solving/ the secondary subject of this investigation/ is extensive/ most 9 3 indicated no significant sex differences for 8th grade students on measures of problem solving or algebra. However/ females outperformed males in computation and spatial visualization. For the 12th grade students/ statistically significant sex differences favoring males were found in problem solving/ but not in algebra/ computation/ or spatial visualization. For males and females who had enrolled in courses beyond general mathematics and who had taken or were enrolled in courses such as pre-calculus/ calculus/ or geometry/ differences in problem solving or spatial visualization did not exist. Sex differences favoring males were found on a total score obtained summing across the computation/ problem solving/ and algebra subtests (Armstrong/ 1979). The mathematics data collected in the second survey by the 1978 National Assessment of Educational Progress showed significant sex differences for both 13- and 17-year-old students. The 13-year-old females outperformed males in the computational subtest and males outscored females by 1 1/2 percentage points in problem solving (statistically significant) No statistically significant differences were found in algebra. No sex differences were found for the 17-year-old group either in the computation subtest or in the algebra subtest. Males surpassed females in problem solving. A reanalysis of the data from the 17-year-old group confirmed male superiority in problem solving after controlling for 79 TABLE 4.2 Continued Item Num ber Xa for Model 3.2 Xz for Model 3.3 Xz for Difference Actual Sample Size Minimum Sample Size for Signif. (a) lb) . (c) . (d)... (e) (f) 75 8.69 237.07 228.38* 67/783 5/963 76 99.68 167.72 68.04* 70/183 20/723 77 79.38 432.36 352.98* 66/304 3/774 78 64.56 127.04 62.48* 76/806 24/696 79 23.98 72.59 48.61* 48/690 20/123 80 75.01 128.13 53.12* 55/277 20/906 81 95.51 106.82 11.31* 57/710 102/511 82 51.99 108.12 56.13* 71/315 25/525 83 20.12 95.18 75.06* 81/021 21/685 84 38.62 69.05 30.43* 78/959 52/129 86 102.85 433.47 330.62* 39/903 2/425** 87 66.75 430.40 363.65* 48/194 2/662** 88 63.01 126.27 63.26* 38/742 12/304 89 43.26 289.14 245.88* 39/648 3/240 90 85.53 405.04 319.51* 53/557 3/368 91 82.53 85.48 2.95 62/954 428/727 92 16.36 93.12 76.76* 88/231 23/092 93 78.39 229.02 150.63* 80/723 10/766 94 70-15 137.27 67.12* 87/241 26/113 41 categories. Category I included clerical and computational errors. Category II included other types of errors* such as average and area errors/ use of wrong operation* no response* and erred responses offering no clues. It the student's work indicated the correct procedure and yet the problem was missed because of a computational or clerical error* it was assumed that the problem was read and understood. An analysis of frequencies showed that clerical errors were responsible for 3% of the problems incorrectly solved* computational errors accounted for 49%* and other errors for 48% of the erred problems. Knifong and Holtan concluded that "improved computational skills could have eliminated nearly half of the word problem errors" (p. Ill). These computational errors were made in a context where other skills such as reading* interpretation of the problem* and integration of these skills necessary for the solution of word problems* might interact. However* Knifong and Holtan state that their findings neither confirm nor deny that improvement of reading skills will lead to improvement in problem solving. They conclude that "it is difficult to attribute major importance to reading as a source of failure" (p. 111). In a later analysis* looking for evidence of poor reading abilities affecting children's success in word problems* Knifong and Holtan (1977) interviewed the 59 Inspection of the year by option contingency tables shows that year and option are independent for each gender. Thus/ hypothesis 1 is equivalent to the hypothesis that year and option are independent conditional upon gender. Hypothesis 2 is also true for Table 3.1 and Table 3.2. Therefore/ when hypothesis 1 is correct/ hypothesis 2 is equivalent to the hypothesis that sex and option choice are dependent. In his discussion of the analysis of three dimensional contingency tables/ Fienberg (1980) presents the following saturated model for the data: log mijk = Ui + U j + jc+i j + Uik+Uj k+Ui j k- (3.1) In this model/ mjk is the expected value of the frequency in cell ijk of the three dimensional table. The model states that all three classification factors for a three dimensional contingency table are mutually dependent. In the present research i is the year index/ j is the option index/ and k is the sex index. Fienberg shows that deleting the terms Uj and U^jk yields a model in which year and option are independent conditional upon sex. This model is log mij k Ui+Uj+Uk+Uik+Ujk (3.2) 29 females may use them in a different way. Meyer (1978) with an elementary grade sample and Fennema and Tartree (1983) with an intermediate level sample found that the influence of spatial visualization on solving mathematics problems is subtle and that males and females use their spatial skills differently in solving word story problems (problems that measure problem solving ability or reasoning). Fennema and Tartree (1983) carried out a three-year longitudinal study which showed that girls and boys with equivalent spatial visualization skills did not solve the same number of items nor did they use the same processes in solving problems. The results also suggested that a low level of spatial visualization skills was a more debilitating factor for girls than for boys in problem solving performance. Landau (1984) also investigated the relationship between spatial visualization and mathematics achievement. She studied the performance of middle school children in mathematical problems of varying difficulty and the extent to which a diagramatic representation is likely to facilitate solution. She found that spatial ability was strongly correlated to mathematical problem solving and that the effect of spatial ability was more influential for females. Females made more use of diagrams in the solution of problems reducing the advantage of males over females in problem solving performance. Problem Solving 89 Subtraction Fig. 4.3 Sex by Subtraction Interaction 44 In a more recent investigation? Exedisis (1983) studied the contribution of reading ability* vocabulary* mathematical concepts* computation* sex* and race on problem-solving performance. The Iowa Test of Basic Skills was administered to a group of 6th* 7th* and 8th grade anglo and black Chicago male and female adolescents. Problem solving was highly correlated to an understanding of basic mathematical concepts* somewhat correlated to race* and weakly correlated to computational and vocabulary skills* sex* and reading ability. Although the findings of these studies show a relationship between computational skills and problem solving achievement* this relationship is not strong enough to be considered the most determinant factor in problem solving achievement* as some of the researchers have been careful to point out. In spite of the dismissal of reading as a determinant factor in problem solving achievement by some of these same researchers* more recent studies in this area have led others to hold different views. Headin_g_and_J)r obiem_ Solving Performance Martin (1964) studied the contribution of reading comprehension* computation* abstract verbal reasoning* and arithmetic concepts to arithmetic problem solving performance. Fourth and 8th grade students were administered the Iowa tests of Basic Skills and the Lorge-Thorndlike intelligence test (verbal). 38 At present no set of variables has been clearly established as a determinant of problem difficulty. Several researchers have investigated the effect of reading and computation on problem solving performance. Others have studied the effect of student attitudes toward problem solving in problem solving learning and achievement. Typically/ correlational methods have been used to investigate these questions. Computational Skills and Problem.Solving Performance One of the first researchers to study the effect of computation and reading on problem solving performance was Hansen (1944). He investigated the relationship of arithmetical factors/ mental factors/ and reading factors to achievement in problem solving. Sixth grade students were administered tests in problem solving and categorized as superior achievers (best problem solvers) and inferior archievers (poorest problem solvers). The two groups were compared in selected factors believed to be related to success in arithmetic problem solving: arithmetical/ mental and reading factors. After controlling for mental and chronological age/ the superior achievers in problem solving surpassed the inferior achievers in mental and arithmetical factors. The superior group did better in only two of the six items under the reading factors: general language ability and the reading of graphs/ charts/ and tables. 26 maintained their superiority in mathematics ability during high school. While Fox attributed sex-related differences in mathematical achievement to differential exposure to mathematical games and activities outside school/ Benbow and Stanley suggested that sex-related differences in mathematics performance stem from superior mathematical ability in males not from differences in mathematics formal education. The differential coursework hypothesis is not totally convincing and/ as reporteu berore/ it has been challenged by researchers such as Benbow and Stanley (1980). However/ Pallas and Alexander (1983) have questioneu the generalizaoility of Benbow and Stanley's findings based on the fact that they used highly precocious learners. The differential coursework hypothesis can be accepted only as a partial explanation of differences in mathematics performance found between the sexes. Differenees in Spatial Ability The basic premise in this issue is that males and females differ in spatial visualization and this explains differential mathematics learning and achievement. Until recently/ sex differences in spatial ability in favor of males were believed to be a fact and were thought by some to be related to sex differences in mathematical achievement. 108 Bern* S. L. & Bern D. V. (1970). We are all nonconscious sexists. Psychology Todays 22-24* 26/ 115-116. Benbow* C. P.* & Stanley* J. C. (1980). Sex differences in mathematical ability: Fact or artifact? Science* 210 1262-1264. Benbow* C. P.* & Stanley* J. C. (1982). Consequences in high school and college of sex differences in mathema tical reasoning ability: A longitudinal perspective. American Educational Research Journal* 12* 598 622. Bobbie* C. N. (1971). Sex-role preference and academic achievement (Doctoral dissertation* Yeshiva Univer sity) Â£iss.ejtation Abstracts International* 22* 1818B-1819B. Carey* G. L. (1958). Sex differences in problem solving performance as a function of attitude differences. Journal of Abnormal and Social Psychology* 2Â£* 256-260. Carpenter* T. P.* Corbitt* M. K.* Kepner* H. S.* Lindquist* M. M. & Reys* R. E. (1980). Results of the second NAEP mathematics assessment: Secondary school. Mathematics Teacher* 73 329-338. Carpenter* T. P. Lindquist* M. M. Mathews* W. & Silver* E. A. (1984). Results of the third NAED mathematics assessment: Secondary school. Mathematics Teacher* 16.* 652-659. Chase* C. I. (1960). The position of certain variables in the prediction of problem solving in arithmetic. Journal of Educational Research* 24/ 9-15. Cohen* D.* & Wilkie* F. (1979). Sex-related differences in cognition among the elderly. In M. A. Witting and A. C. Petersen (Eds.)* Sex-related differences in cognitive functioning: Developmental Issues* New York: Academic Press. Connor* J. M. Serbin* L. A.* & Schakman* M. (1977). Sex differences in children's response to training on a visual-spatial test. Developmental Psychology* 12* 392-394. 37 dealing with both cognitive and affective variables* there are no clear-cut findings to render unequivocal support to a particular variable as accounting for these sex-related differences. However* everything seems to point to the fact that affective* rather than cognitive variables play a more significant role in the sex-related differences observed in mathematics performance and learning. In most of the studies dealing with affective variables* findings consistently show that these factors influence mathematics performance in females more than in males. In at least one area* confidence as learners of mathematics* Sherman (1980) found that this variable influenced course election more than all the cognitive variables previously discussed. The case for the societal influences on sex roles and expectations to account for the differences in mathematics learning is also supported in one way or another in the studies reported in the literature. Problem Solving Performance _and_Rela.ted Variables Problem solving has been perhaps the most extensively researched area in mathematics education. Published reviews by Kilpatrick (1969)* Riedesel (1969)* and Suydam and Weaver (1970-1975) attest to this. Much of the research done has focused on identifying the determinants of problem difficulty and the problem features that influence the solution process. LIST OF FIGURES Figure Page 3.1 Number of Questions by Skill Area in a Basic Skills Test in Mathematics-6 56 4.1 Sex by Multiplication Interaction 87 4.2 Sex by Division Interaction 88 4.3 Sex by Subtraction Interaction 89 vii 8 achievement. Since there are no investigations reported in sex differences in item response patterns in Puerto Rico research is needed. Organization of the Study A review of the literature on sex differences in mathematics performance is reported in Chapter II. The research methodology is presented in Chapter III. Research questions sample instrument and data analysis are discussed in that chapter. Chapter IV is an exposition of the results of the study. Chapter V contains a summary and interpretation of the results of the study and the implications of the findings together with suggestions for further research. 23 computation grew larger. This difference grew larger for girls than for boys. Although SES was a major factor in solving computation and story problem items successfully/ the effect was similar for each sex. Sex-related differences by primary language or chronological age were not large. This research carried out by Marshall with elementary grade children supports previous research findings that males are better than females in mathematics problem solving (a higher order skill) and females are better than their counterpart males in computation (a lower level skill). Marshall's research also brought out a different aspect of this question: the notion that girls find it more necessary than boys to succeed in the computation items in order to successfully solve the story problem items. Â£ognitive_and_Affective Factors That Influence Sex Differences in Mathematics Learning and Achievement The research reviewed in the literature does not provide evidence of any unique variable that could serve as an explanation for the observed sex-related differences in mathematics learning and achievement. However/ some issues have been discussed/ among which the most prevalent are that sex-related differences in mathematics learning and achievement are a result of differences in formal education; that sex-related differences in mathematics 115 Whitaker, D. R. (1976). A study of the relationship between selected noncognitive factors and the problem solving performance of fourth-grade children. (Tech. Rep. No. 396). Madison: University of Wisconsin/ Research and Development Center for Cognitive Learning. Wiser L. K.r Steel L.r & MacDonald/ C. (1979). Origins and career consequences of sex differences in high school mathematics achievement (Grant No. NEI-G 7 8-001 ) Palo Alto/ CA : American Institutes for Research. Zalewski/ D. L. (1974). An exploratory study to compare two performance measures: An interview-coding scheme of mathematical problem solving and a written test (Doctoral Dissertation/ University or Wisconsin/ Madison). Dissertation Abstracts International/ 15./ 3797A. 91 TABLE 4.7 Reliability of the Covariates for Each of the Three Years of Test Administration Covariate First Year Second Third Addition .579 .610 .454 Subtraction .756 .724 .737 Multiplication .729 .729 .734 Division .716 .680 .714 Fracadd .709 .728 .748 Decsub .720 .716 .742 Equivalence .421 .394 .488 skills/ sex-related differences in problem solving performance do not exist. Females retained their superiority in problem solving when equivalence (in all three years) and subtraction (in one year) were the controlling variables. The question of whether male-female differences in problem solving depend on computational skills was answered/ partially/ in the affirmative. x 110 Fennemar E. & Shermanr J. (1977). Sex-related differences in mathematics achievements spatial visualizations and affective factors. American Educational.Research Journals 14' 51-71. Fennemas E.s & Shermans J. A. (1978). Sex-related differences in mathematics achievement and related factors. A further study. Journal for Research in Mathematics .Education/ 9s 189-204. Fennemas E.s & Tartrees L. A. (1983). Research on relationship of spatial visualization and confidence to male/female mathematics achievement in grades 6-8. Washingtons DC: National Science Foundation. (ERIC Document Reproduction Service No. ED 232 853) Fienbergs S. E. (1980). The Analysis of .Cross-Classified Categorical Data. Cambridges MA: The MIT Press. Foxs L. H. (1975a). Mathematically precocious: Male or female. In E. Fennema (Ed.) Mathematics learning: What research says.about sex differences (pp. 1-12). Columbus: Ohio State Universitys College of Education. (ERIC Document Reproduction Service No. ED 128 195) Foxs L. H. (1975b). Sex differences: Implications for program planning for the academically gifted. Paper presented at the Lewis M. Terman Memorial Symposium on Intellectual Talent held at the Johns Hopkins University* Baltimore* MD. Fox* L. H. (1977). The effects of sex role socialization on mathematics participation and achievement. In J. Shoemaker (Ed.)* Women_and_Mathe.mati,csResearch Perspectives for Change* (N.I.E. Papers in Educa tion and Work: No. 8). Washington* D. C. : Eduation and Work Group* The National Institute of Education. Glennon* V. J.* & Callahan* L. G. (1968) Elementary school mathematics: A_guide._tp_..cur rent research* Washington* D. C.: Association for Supervision and Curriculum Development. Guay* R. B.* & McDaniel* D. (1977). The relationship between mathematics achievement and space ability among elementary school children. Journal for Research in Mathematics Education* 8* 211-215. 61 A problem arises in interpreting this analysis with the data used in this research. Over the three years? there were responses available from 135/340 students. Even if only 5% of the students answered an item incorrectly/ the responses of 7/767 students would be used in analyzing this item. On the other hand/ if 90% answered an item incorrectly/ the responses of 121/806 students would be used in analyzing the data. As a result of the large sample size/ the tests described above are likely to be very powerful. In step 1 of the analysis/ then/ even a very small change from year to year in the proportion of males or females who choose an option is apt to be detected/ and the results will indicate that hypothesis 1 is not supported. For step 2/ even a very small dependence of option choice on sex is likely to be detected and hypothesis 2 is likely to be supported. In brief/ the problem caused by the large sample size is that practically insignificant differences may yield statistical significance. Fortunately/ the form of the test statistic used in the likelihood ratio test suggests a reasonable solution to this problem. The test statistic is - 2 ^ F0 l9e (F0/Fe). (3.4) Ill Hansen* C. W. (1944) Factors associated with successful achievement in problem solving in sixth grade arithmentic. journal qÂ£ Educational,-Researoh' 22.1 111-119. Harnisch* D. L. (1983). Item response patterns: Application for educational practice. Journal of Educational Measurement* 22* 191-205. Hilton* T. L. & Berglund* G. W. (1974). Sex differences in mathematics achievement: A longitudinal study. Journal of Educational Research* Â£2* 231-237. Hunter* J. E. & Cohen* S. H. (1974). Correcting for unreliability in nonlinear models of attitude change. Psvchometrika* 22* 445-468. Jarvis* O. T. (1964). Boy-girl ability differences in elementary school arithmentic. School Science and Mtkemati_c_s/ 4, 657-659. Kaufman* A. (1984). Sex differences in mathematics reasoning in mathematically gifted and average fifth and sixth graders (Doctoral dissertation* Hofstra University* 1983) Dissertation Abstracts Inter- JjafcionflJU 15-' 1094A-1095A. Kilpatrick I. (1969). Problem solving and creative behavior in mathematics. In J. W. Wilson* & L. R. Carry (Eds.)* Studies in mathematics (Vol. No. 19). Stanford* CA: School Mathematics Study Group. Kloosterman* P. W. (1985). Attributional theory* learned helplessness* and achievement in ninth grade mathematics (Doctoral dissertation* University of Wisconsin* 1984) Dissertation Abstracts Internatio- nal, 4Â£/ 919A-920A. Knifong* J. D.* & Holtan* B. (1976). An analysis of children's written solutions to word problems. Journal .for Research in Mathematics Education* 1* 106-112. Knifong* J. D.* & Holtan* B. (1977). A search for reading difficulties among erred word problems. Journal for 2esÂ£.arch_, in Mathematics .Education* 2* 227-230. 50 The studies reviewed in this section show a positive relationship between reading and problem solving performance/ but in the case of Ballew and Cunningham (1982)/ this relationship is not viewed singly but rather as one among the interacting factors that produce successful problem solving. The third variable reviewed is the effect of student attitudes toward problem solving on problem solving performance. Many researchers have tried to demonstrate that this variable is a determinant factor in problem solving achievement. Attitudes Toward Problem Solving and Problem Solving Performance Research studies support the existence of positive and rather stable relationships between student attitudes and achievement in mathematics. Aiken (1970) has suggested that an individual's attitude toward one aspect of the discipline (mathematics) / such as problem solving/ may be entirely different from his/her attitude toward another phase of the discipline/ such as computation. Research/ however/ has been directed to the use of single/ global measures of attitudes toward mathematics rather than to the investigation of attitudes toward a particular phase of the discipline. The studies described below are only part of the few investigations which have examined the relationship between 96 TABLE 4.11 Adjusted Means on Problem Solving* by Covariate and Sex Second Year Adjusted Means Covariate Males Females Addition 3.69 3.96 Multiplication 3.80 3.85 Division 3.80 3.85 Fracadd 3.78 3.87 Decsub 3.90 3.75 Equivalence 3.46 44.18 Log-linear models were used in the analysis of incorrect responses. The results of the analyses showed that for 100 of the 111 items of the testr males and females selected different incorrect options/ and this pattern of responses was consistently found during the three years of test administration. However/ for the vast majority of the 100 items the male-female differences were relatively small/ considering the fact that the number of subjects needed to obtain statistical significance was very large. The responses of approximately 1/000 randomly selected students per academic year were analyzed in the comparison of male and female performance in problem solving. Females outperformed males in problem solving and in six of the seven computational variables. Males showed superiority in equivalence in all the three years/ but statistical significance was obtained in only one of the years. Analysis of covariance (ANCOVA) was used in the comparison of male and female performance in problem solving after controlling for computational skills. Seven analyses of covariance tests were conducted/ one for each of the covariates. Estimated true scores for observed scores were used in the analyses. The results tend to show that for examinees with similar levels of computational IX 11 She analyzed the responses of 9>000 boys and 9>000 girls to 6 selected items> 2 computation items> and 1 story problem item from each of 2 of the 16 forms of the Survey of Basic Skills test administered during the academic year 1978-79. The Survey is a 30-item achievement test administered every year to all 6th grade children in California through the California Assessment Program. There are 16 forms of the test> to which approximately 9 >000 boys and 9 >000 girls respond each year. Of the 160 mathematics items contained in the 16 forms of the test> 32 are on measurement and graphing> 28 on number concepts> 28 on whole number arithmetics> 20 on fraction arithmetics> 20 on decimal arithmeti cs > 20 on geometry > and 12 on probability and statistics. The item analysis performed on the 1978-79 data showed that boys and girls tended to select different incorrect responses. In the first computation item (Form 1 of the test) both sexes reflected similar mistakes in carrying> but in different columns. In the second computation item> both sexes ignored the decimal points and selected the same incorrect response. However> more girls than boys chose this response. In the first computation item (Form 2 of the test) the incorrect choice of both sexes was option Â£> but the second most frequently selected option was for boys and b for girls. In the second computation item of this 63 model tests were conducted using a .01 level of significance. Since this research is based on Marshall (1981/ 1983)/ it is important to compare the method of analysis used in this study to the one used by Marshall. Marshall also used a two-step analysis. In the first step of her analysis she deleted the Uijk term from (3.1) and tested the adequacy of the model/ 1 g m j. j k = Ui + Uj + Ufc+Uij + Uik+Ujk. (3.6) Following this/ she deleted the Ujk term to obtain log mij = Ui+Uj + Uk+ij + Uik. (3.7) and compared these two models using a likelihood ratio test. If the first test was nonsignificant and the second significant/ Marshall claimed that option choice was dependent on sex and the pattern of dependency was the same from year to year. This is the same claim that this study sought to establish. However/ the approach used here was to present evidence that model (3.2) fits the data while Marshall tried to show that model (3.6) fits the data. The major difference between the two approaches concerns the operationalization of the concept that the 115 Whitaker/ D. R. (1976). A study of the relationship between selected noncognitive factors and the problem solving performance of fourth-grade children. (Tech. Rep. No. 396). Madison: University of Wisconsin/ Research and Development Center for Cognitive Learning. Wise/ L. K./ Steel/ L./ & MacDonald/ C. (1979). Origins and career conseguences of sex differences in bigh__s_chool mathematics achievement (Grant No. NEI-G 78-001). Palo Alto/ CA : American Institutes for Research. Zalewski/ D. L. (1974). An exploratory study to compare two performance measures: An interview-coding scheme of mathematical problem solving and a written test (Doctoral Dissertation/ University or Wisconsin/ Madison). Dissertation Abstracts International/ 15./ 3797A. 27 Research findings in this area have been inconsistent. In 1966 / Maccoby stated that "by early school years/ boys consistently do better (than girls) on spatial tasks and this difference continues through the high school and college years" (p.26). In 1972/ Maccoby and Jacklin said that the differences in spatial ability between the sexes "remain minimal and inconsistent until approximately the ages of 10 or 11/ when the superiority of boys becomes consistent in a wide range of populations and tests" (p.41). In 1974/ after a comprehensive literature search/ Maccoby and Jacklin concluded that sex differences in spatial visualization become more pronounced between upper elementary years and the last year of high school/ the years when sex-related differences in mathematics achievement favoring boys emerge. Guay and McDaniel (1977) supported in part Maccoby and Jacklin's 1974 findings. They found that among elementary school children/ males had greater high level spatial ability than females/ but that males and females were equal in low level spatial ability. This finding is inconsistent with that portion of Maccoby and Jacklin's review that suggests that sex differences become evident only during early adolescence. Cohen and Wilkie (1979) however/ stated that in tests measuring distinct spatial tasks/ males perform better than females in early adolescence and throughout their life span. Most studies carried out after 78 TABLE 4.2 Continued Item Num ber Xz for Model 3.2 X1 for Model 3.3 Xa for Difference Actual Sample Size Minimum Sample Size for Signif. (a) (b) (c) (d) (e) (f) 55 19.53 90 .87 71.34* 41,7 96 11,770 56 21.62 70.14 48.52* 44,107 18,263 57 28.51 35.78 7.27 53 ,075 146,668 58 17.49 51.69 34.20* 44,044 25,873 59 50.96 73.68 22.72* 41,829 36 ,987 60 66.43 148.39 81.96* 60,494 14,828 61 21.39 394.54 373.15* 35,632 1,918** 62 55.03 300.95 245.92* 50,699 4,142 63 158.36 278.34 119.98* 52 ,068 8,719 64 88.01 246.48 158.47* 73,196 9,279 65 56 .28 94.11 37.83* 42,429 22,532 66 50.51 113.88 63.37* 78,708 24,953 67 34.28 180 .60 146.32* 51,862 7,121 68 48.01 225.46 177.45* 63,223 7,158 69 75.29 251.66 176.37* 62,511 7,121 70 44.39 271.35 226 .96* 67,528 5,977 72 17.41 208.90 191.49* 84 ,175 8,831 73 24.61 37.80 13.19* 53 ,358 81,271 74 22.75 163.01 140 .26* 84,175 12,057 75 TABLE 4.2 Chi-square Values for the Comparison of Models (3.2) and (3.3) with Actual Sample Sizes and with the Corresponding Number of Subjects Needed for Significant Results Item Num ber Xa- for Model 3.2 Xa- for Model 3.3 Xa for Difference Actual Sample Size Minimum Sample Size for Signif. (a) lb) (c) (d) (e) (f) 1 16.11 30.94 14.83* 7 r 427 10,061 2 17.54 103.22 85.68* 11,018 2,583** 3 20.06 47.51 27.45* 11,497 8,733 4 7.54 21.87 14.33* 6,016 8,414 5 14.75 70.40 55.65* 7,706 2,782** 6 11.86 23.13 11.27* 13,046 23,256 7 4.60 35.73 31.13* 8,832 5,700 8 46.15 329.98 283.83* 16,796 1,189** 9 10.13 109.99 99.86* 23,075 4,642 10 9.54 65.29 55.75* 35,060 12,634 11 5.89 19.06 13.17* 24,908 37,996 12 12.58 198.46 185.88* 40 ,423 4,369 13 25.74 62.37 36.63* 15,718 8,621 14 11.19 80.50 69.31* 30,238 8,765 15 16.63 17.33 0.70 42,889 1,230,914 16 14.51 84.29 69.78* 35,114 10,109 17 17.90 44.20 26.30* 43,962 33,582 113 Meyer? R. A. (1978). Sex-related differences in mathema tical problem solving performance and intellectual abilities. Madison: University of Wisconsin? Research and Development Center for Cognitive Learning. (ERIC Document Reproduction Service No. ED 166 023). Milton? G. A. (1957) The effects of sex-role identifica- cation upon problem solving skill. Journal of Abnormal & Social,Psychology? 5? 208-212. Moller? N. J. (1983). The impact of gender? masculinity? and feminity on math achievement and course decisions (Doctoral dissertion? Purdue University? 1982). nis_sertation_Abstracts International? 41/ 2584A. Muth? K. D. (1984). Solving arithmetic word problems: Role of reading and computational skills. Journal of Educational Psychology? 76 ? 205-210. Pallas? A. M.? & Alexander? K. L. (1983). Sex differences in quantitative SAT performance: New evidence on the differential coursework hypothesis. American Educational Research Journal? 21/ 165-182. Plank? E.? & Plank? R. (1954). Emotional components in arithmetical learning as seen through autobiographies. Psychological Studies of the Child? 2.? 274-296. Porter? A. C. (1968). The effects of using fallible variables in the analysis of covariance (Doctoral dissertation? University of Wisconsin? 1967). Dissertation Abstract International? 2Â£? 3517B. Probert? B. S. (1984). Math confidence workshops: A multimodal group intervention strategy in mathematics anxiety/avoidance. (Doctoral dissertation? University of Florida? 1983). Dissertation Abstracts International? 44 ? 2231B. Puerto Rico Department of Education. (1980). Basic skills test in mathematics-6. Hato Rey: Center of Evaluation. Radatz? H. (1979) Error analysis in mathematics educa- tion. Journal for Research in Mathematics Educa tion ? 10 ? 163172. Riedesel? A. C. (1969). Problem solving: Some suggestions from research. Arithmetic Teacher? 16 ? 54-58. 48 success since the components must be integrated into a whole process (mastery learning of the components cannot assure mastery of the process). Their analysis also led them to conclude that* in the case of 6th graders inability to read problems is a major obstacle in solving word problems. Only 12% of the subjects could read and set up problems correctly at a higher level than they could compute while 60% could compute correctly at a higher level than they could read and set up problems; 44% could set up problems better when they heard them read than when they read the problem themselves. Only 13% could set up problems better when they read them than when they heard them read. Muth (1984) investigated the role of reading and computational skills in the solution of word problems. A group of 200 students from the 6th grade were administered a test of basic skills and a mathematics word problem test. The word problem test consisted of 15 sample items supplied by the National Assessment of Educational Progress. The items were adapted to include some extraneous information and complex syntactic structure. Four versions of the test were constructed by combining two versions of problem information (absence vs. presence of extraneous information) with two versions of syntactic structure (simple vs. complex syntax). Task performance was measured by means of the number of problems answered correctly 102 response was related to the use of a large number of subjects in the analysis. One must question whether analysis of incorrect responses has any educational value. Although the results of this study do not support the use of this analysis (error analysis) for the study of sex-related differences# it continues to be a promising technique in the discovery of how children approach mathematics problem solving# more so than analyses that look only at the items answered correctly or at total test scores. The second objective of the study was to investigate sex-related differences in problem solving performance# to find out if significant differences persisted after controlling for computational skills# and if the differences depended on the level of computational skills. The results showed that females outperformed males in problem solving and in six of the seven computational skills. Males showed superiority in equivalence in each of the three years# but statistical significance was obtained only for the second year. The results also showed that females retained their superiority in problem solving# only when equivalence (for the three years data) and subtraction (for the second year) were the controlling variables. Analysis of covariance was also used to answer the question of whether male-female differences in problem solving depend on the level of computational skills. The 117 From May* 1982 to July* 1983 / she worked as graduate research assistant at the Foundations of Education Department/ University of Florida. She returned to Puerto Rico in August/ 1983 / to serve as Special Aide to the Assistant Secretary for the Vocational/ Technical/ and High Skills Educational Programs of the Commonwealth Department of Education. 46 Their study is important because it represents an attempt to demonstrate that multiple factors can interact in the correct solution of a mathematics word problem. They constructed three graded tests from a basal mathematics series for grades 3 through 8. For test 1* the problems were set in pure computational form (the effects or reading* interpretation* as well as the necessity for integration were removed in an effort to measure the computational skills required by the word problems). For test 2* the effects of reading and computation were removed by reading the problems to the students and by giving scores based on whether or not the students set them up properly* in an attempt to measure problem interpretation alone. For test 3* the effect of computation was removed. The test yielded two scoresone by grading the students on whether or not they set up the problems properly and another by grading on the basis of the correct answer. The tests were administered to all 244 students from the 6th grade in two different schools. A diagnostic profile was obtained for each of the 217 students for which complete data were available: a computational score* a problem-interpretation score* a reading score* and a reading-problem solving score. They assumed that if the reading-problem interpretation score was lower (one or more levels lower) 71 TABLE 4.1 Summary of Significant Tests of Model 3.2 Item Number X* for Model 3.2* Actual Sample Size Min. Sample Size for Significance (a) (b) (c) (d) 8 46.15 16/796 3/352 13 25.74 15/718 5/624 23 29.25 60/377 19/011 25 35.90 26/707 6/852 28 46.00 84/847 16/988 29 61.48 88/396 13/242 30 25.19 87/473 31/982 31 23.26 32/870 13/015 34 58.25 82/257 13/006 36 84.36 86/137 9/404 37 23.70 37/068 14/405 42 23.26 59/140 23/417 45 22.41 63/098 25/932 48 179.25 97/593 5,014 49 27.82 16/766 5/550 52 26.01 39/858 14/114 54 41.41 74/502 16/570 56 21.62 44/107 18/789 57 28.51 53/075 17/146 58 Table 3.1 Hypothetical Probabilities of Option Choice Conditional on Year and Sex Year Sex 1 Option 2 3 First H .7 .2 .1 F .5 .3 .2 Second H .7 .2 .1 F .5 .3 .2 Third M .7 .2 .1 F .5 .3 .2 Table 3.2 Hypothetical Conditional on Probabilities of Year and Sex and Option Choice Arranged by Sex Sex Year 1 Option 2 3 M First .7 .2 .1 Second .7 .2 .1 Third .7 .2 .1 F First .5 .3 .2 Second .5 .3 .2 Third .5 .3 .2 90 The significant interactions found between sex and multiplication^ and sex and division (first year)/ and sex and subtraction (second year)/ answered/ in part/ the question of whether male-female differences in problem solving performance depend on computational ability. However/ the evidence is quite weak. Of 21 possible interactions/ only three were significant. No variable exhibited a significant interaction for each of the three years. The analysis of covariance was also used to determine if sex-related differences exist after controlling for computational skills. Analyses were conducted for those variables that did not exhibit significant interactions with sex. As discussed in Chapter III/ estimated true scores were used for observed scores to adjust for unreliability of the covariates (the computational subtests). Reliability coefficients calculated for each covariate are shown in Table 4.7. Summaries of the analyses of covariance for the first year are reported in Table 4.8. The results show that females retained their superiority in problem solving performance when equivalence was the controlling variable in the analysis of covariance/ the only variable in which males outperformed females (nonsignificant). When the controlling variables were addition/ subtraction/ addition of fractions/ and subtraction with decimals/ female ACKNOWLEDGMENTS I would like to acknowledge my gratitude to several people who have influenced my formal education and/or made this study possible. My special thanks to Dr. James Alginar Chairman of the doctoral committee/ who contributed to the development of my love for research and statistics. He has been insuperable as professor and valued friend; his help and guidance in the preparation and completion of this study were invaluable. I extend my thanks to Dr. Linda Crocker for her advice and help during my doctoral studies at University of Florida. Thanks also go to Dr. Michael Nunnery/ member of the doctoral committee. To Dr. Wilson Guertin* who was a friend for me and my family/ I extend my special thanks. Thanks are also extended to Dr. Amalia Charneco/ past Undersecretary of Education of the Puerto Rico Department of Education/ for her continuous support. To my sister Nilda Santaella/ who typed the thesis/ I give my sincere thanks. Special thanks go to my family and to those friends who provided encouragement throughout this critical period of my life. ii 10 of the studies consider sex differences incidental to the major study findings. The available literature offers very little research directly related to the problem of sex differences in this area. The review of the literature has been divided in four sections. The first section consists of a detailed summary of the available research on sex diferences in incorrect responses. The second section deals with sex-related differences in problem solving performance. These sections are directly related to the objectives of the study. The third section is more peripheral and contains a discussion of the more prevalent issues about the influence of cognitive and affective variables on sex differences in mathematics learning and achievement. The fourth is a summary of the research dealing with variables considered as influential to mathematics problem solving performance. Sex-related Differences in Incorrect Response Patterns Research findings tend to suggest that boys and girls may be approaching problem solving differently (e.g. Fennema and Sherman 1978; Marshall 1981 1983; Meyer 1978 among others). Marshall (1981) investigated whether 6th grade boys and girls approach mathematical problem solving with different strategies. Her specific interest was whether the sexes made the same errors. UNIVERSITY OF FLORIDA 3 1262 08554 6660 Abstract of Dissertation Presented to the Graduate School of the University of Florida in Partial Fulfillment of the Requirements for the Degree of Doctor of Philosophy COMPARISON OF MALES AND FEMALES ON MATH ITEM PERFORMANCE: ANALYSIS OF RESPONSE PATTERNS By Sonia Feliciano Augustf 1986 Chairman: James Algina Major Department: Foundations of Education The first objective of this study was to investigate sex differences in the selection of incorrect responses on a mathematics multiple-choice test* and to determine whether these differences were consistent over three consecutive administrations of the test. A second objective was to compare male and female performance in problem solving after controlling for computational skills. The responses of all 6th grade students from the public schools in Puerto Rico who took the Basic Skills Test in Mathematics-6 ("Prueba de Destrezas Bsicas en Matemticas-6") during three academic years were used in the analyses relevant to the first objective. viii 53 The studies reviewed have confirmed the relationship between problem solving performance and attitudes toward problem solving (Carey 1958? Lindgren et al. 1964? Whitaker 1976). However the results reportea in the studies that investigated the relationship between problem solving performance and computation and between reading and problem solving fail to be consistent in their conclusions. Hansen (1944) Chase (I960) Balow (1964) Knifong and Holtan (1976 1977) and Zalewski (1974? concluded that computation is more strongly related to problem solving than is reading. Martin (1964) Creswell (1982) Ballew and Cunningham (1982) and Muth (1984) concluded that reading aoility and mathematical problem solving show a stronger relationship than computation and problem solving. Exedisis's (1983) findings led to the conclusion that the effect or reading and computation in problem solving performance is unimportant. IllMETHOD 54 The Sample 54 The Instrument 55 Analysis of the Data 57 Analysis of Sex by Option by Year Cross Classifications 57 Comparison of Males and Females in Problem Solving Performance 66 IV RESULTS 6 8 Introduction 68 Sex-related Differences in the Selection of Incorrect Responses 68 Sex-related Differences in Problem Solving Performance 81 Summary 93 V DISCUSSION 101 Summary and Interpretation of the Results 101 Implications of the Findings and Suggestions for Further Research 103 Sex-related Differences in Incorrect Responses 104 Sex-related Differences in Problem Solving Performance 105 REFERENCES 107 BIOGRAPHICAL SKETCH 116 IV Sex-related Differences in Problem Solving Performance 81 In this part/ results for the two questions related to the second objective of the study are presented. These questions serve as the framework for the presentation. Each question is stated/ followed by the results pertaining to that question. Question 1: Do males and females differ in problem solving performance? The responses of 492 males and 510 females who took the Puerto Rico Basic Skills Test-6 during the spring of the first year were analyzed in this study. Also/ data from 504 males and 509 females tested in the second year and from 509 males and 504 females tested in the third year were included in the analysis. The mean performance scores and the standard deviations for each of the eight variables are presented in Table 4.3. Results of t-tests are also presented in this table. Females outperformed males in problem solving/ a finding consistently present in all the three years of test administration. Over the three-year period the mean differences favored females in all variables except equivalence. The sex-related differences in problem solving were significant (p 4 mathematics preparation. Males who were enrolled or had completed algebra II outperformed the females in computation and problem solving but not in algebra. Males who studied beyond algebra II outscored females on all three subtests: computation? algebra? and problem solving (Armstrong? 1979). Carpenter? Lindquist? Mathews? and Silver (1984) analyzed the results of the Third National Assessment of Educational Progress (NAEP)? and compared them with the First and Second Surveys. Between 1978 and 1982? the differences between the average performance of males and females remained stable at each age level. At ages 9 and 13? the overall performance of males and females was not significantly different. At age 17? males scored higher than females by about 3 percentage points. When course background was held constant? achievement differences still existed at age 17. For each category of course background? male achievement exceeded female achievement. Consistent with previous assessments? sex differences in problem solving in favor of males were found for the 17-year-old sample. At ages 9 and 13? no large differences were found between the sexes within any level of course background. Marshall (1981? 1984) investigated sex differences in mathematics performance. She found that males and females excel each other in solving different types of problems. 105 Sexrrelated Dif fgrgnce_s_in. Problem Solving Performance The results of this study do not give strong support to Marshall's (1981/ 1984) findings that the effect of computational skills on problem solving performance is different for each sex. However/ the findings of the present investigation are consistent with those found by Meyer (1978) and Whitaker (1976): female performance in problem solving is not significantly different from male performance. Females showed superiority in problem solving performance in all three years of test administration although their superiority was retained only when performance in equivalence was controlled for (for the three years data) and when performance in subtraction was controlled for (second year) (statistically significant only after accounting for equivalence). The findings of this study are also in agreement with those reported by the National Assessment of Educational Progress (NAEP) (Carpenter et al.z 1980). Related to problem solving/ they conclude/ if it were necessary to single out one area that demands urgent attention/ it would clearly be problem solving. At all age levels/ and in vir tually every content area/ performance was extre mely low on exercises requiring problem solving or application of mathematical skills. In gene ral/ respondents demonstrated a lack of the most basic problem-solving skills. Rather than attempt ing to think through a problem and figure out what needed to be done to solve the problem/ most res pondents simply tried to apply a single arithmetic 40 Achievement Test (subtests of reading, arithmetic, and reasoning) and the California Short-Form test of mental ability to a group of 1,400 children from the 6th grade. All levels of achievement were included in the analysis. Analysis of variance and covariance were used and compared. He confirmed the findings of other researchers to the effect that there is a direct relationship between I.Q. and reading ability, and between I.Q. and computational skills. The results of the analysis of variance revealed that increases in computation ability were associated with higher achievement in problem solving. A relationship between reading ability and problem solving was also found, but it was not as strong. Significant differences in problem solving performance associated with computational ability were found when intelligence was controlled. Balow concluded that computation is a much more important factor in problem solving than reading ability, and that when I.Q. is taken into consideration, the degree of the relationship between reading and problem solving ability becomes less pronounced. Intelligence tends to confound the relationship between these two variables. Knifong and Holtan (1976, 1977) attempted to investigate the types of difficulties children have in solving word problems. They administered the word problem section of the Metropolitan Achievement Test to 35 children from the 6th grade. Errors were classified in two 99 responses was consistent throughout the years in which the test was administered. Males and females selected different incorrect responses in 100 of the 111 items of the test. The pattern of male-female differences in the selection of incorrect responses was consistently found in each of the years of test administration. However* for the vast majority of the 100 items* male-female differences were relatively small in magnitude* considering the fact that the number of subjects needed to obtain significance was very large. Therefore* these findings lack educational significance. A second objective of the study was to investigate whether males and females differ in problem solving performance* if these differences persist after controlling for computational skills* and if the male-female differences depend on the level of computational skills. Results of the analyses reported in this section generally indicated that females outperformed males not only in problem solving* but in six of the seven computational variables. Males surpassed females in equivalence* but statistical significance was obtained for only one of the three years covered in the study. The results also tended to show that* for examinees with similar levels of computational skills* sex differences in problem solving did not exist. The only |