<%BANNER%>

Three Essays on Education Policies and Child Health

Permanent Link: http://ufdc.ufl.edu/UFE0041993/00001

Material Information

Title: Three Essays on Education Policies and Child Health
Physical Description: 1 online resource (153 p.)
Language: english
Creator: Yin, Lu
Publisher: University of Florida
Place of Publication: Gainesville, Fla.
Publication Date: 2010

Subjects

Subjects / Keywords: abstinence, nclb, obesity, sexeducation
Economics -- Dissertations, Academic -- UF
Genre: Economics thesis, Ph.D.
bibliography   ( marcgt )
theses   ( marcgt )
government publication (state, provincial, terriorial, dependent)   ( marcgt )
born-digital   ( sobekcm )
Electronic Thesis or Dissertation

Notes

Abstract: Studies in education literature have been focusing on the impact of policies on children's academic performances. However, under single incentive, educators and parents pursuing higher test score gains may undertake ways that unintentionally harm children's health. Understanding how and whether certain education policies post negative impacts on children?s health is therefore very important, and has become the main research focus of this dissertation. In the first essay, we assess whether adolescent obesity has, in part, been driven by a factor previously overlooked: the school accountability movement. Exploiting the variation in the timing of the introduction of state accountability systems and in the grades to which those systems applied across states and over time, we find robust and consistent evidence across models that school accountability systems significantly increase students' Body Mass Index and explain a significant amount of the growth in adolescent obesity. These findings suggest that while accountability systems may promote students academic performance, this could occur at the expense of students' physical health. Policies that mitigate these effects could help to slow the rise in adolescent obesity. In the second essay, we undertake a comprehensive study of the effects of teachers' evaluation standards on children's probability of being obese and BMI using a rich student-level data from the Early Childhood Longitudinal Study-Kindergarten cohort of 1998 (ECLS-K) including interviews with parents, data from principals and teachers as well as direct child assessments. In models in which we control for student-level fixed effects, we find strong evidence that with the increase of teacher evaluation standards students tend to have higher BMI and are more likely to be overweight. The third essay examines the impact of State-Sex-Education policies as well as the new Personal Responsibility and Work Opportunity Reconciliation Act of 1996, the Title V, Section 510 Abstinence Education Program on adolescent risk behaviors. To account for the potential differential impacts of 1996 Title V Section 510 had upon state sex education mandates which will subsequently bias our analysis, we thus employ an Interrupted Time-Series design that first exploits the impact of 1996 reform on state sex education legislative and identify their effects on adolescent sexual behaviors subsequently. First, we find that neither abstinence-only nor comprehensive sex education decrease the probability of being sexually active or increase the likelihood of performing safe sex. Instead, we find that abstinence-only lower the probability of using condoms and birth control pills relative to not using any birth control method. Second, using the ITS model, we find that the trend in percentage of students who had sex (percentage of students who had sex before 13) decreases by 0.5% in high-implement states relative to low-implement states post 1996 reform.
General Note: In the series University of Florida Digital Collections.
General Note: Includes vita.
Bibliography: Includes bibliographical references.
Source of Description: Description based on online resource; title from PDF title page.
Source of Description: This bibliographic record is available under the Creative Commons CC0 public domain dedication. The University of Florida Libraries, as creator of this bibliographic record, has waived all rights to it worldwide under copyright law, including all related and neighboring rights, to the extent allowed by law.
Statement of Responsibility: by Lu Yin.
Thesis: Thesis (Ph.D.)--University of Florida, 2010.
Local: Adviser: Figlio, David N.
Electronic Access: RESTRICTED TO UF STUDENTS, STAFF, FACULTY, AND ON-CAMPUS USE UNTIL 2012-08-31

Record Information

Source Institution: UFRGP
Rights Management: Applicable rights reserved.
Classification: lcc - LD1780 2010
System ID: UFE0041993:00001

Permanent Link: http://ufdc.ufl.edu/UFE0041993/00001

Material Information

Title: Three Essays on Education Policies and Child Health
Physical Description: 1 online resource (153 p.)
Language: english
Creator: Yin, Lu
Publisher: University of Florida
Place of Publication: Gainesville, Fla.
Publication Date: 2010

Subjects

Subjects / Keywords: abstinence, nclb, obesity, sexeducation
Economics -- Dissertations, Academic -- UF
Genre: Economics thesis, Ph.D.
bibliography   ( marcgt )
theses   ( marcgt )
government publication (state, provincial, terriorial, dependent)   ( marcgt )
born-digital   ( sobekcm )
Electronic Thesis or Dissertation

Notes

Abstract: Studies in education literature have been focusing on the impact of policies on children's academic performances. However, under single incentive, educators and parents pursuing higher test score gains may undertake ways that unintentionally harm children's health. Understanding how and whether certain education policies post negative impacts on children?s health is therefore very important, and has become the main research focus of this dissertation. In the first essay, we assess whether adolescent obesity has, in part, been driven by a factor previously overlooked: the school accountability movement. Exploiting the variation in the timing of the introduction of state accountability systems and in the grades to which those systems applied across states and over time, we find robust and consistent evidence across models that school accountability systems significantly increase students' Body Mass Index and explain a significant amount of the growth in adolescent obesity. These findings suggest that while accountability systems may promote students academic performance, this could occur at the expense of students' physical health. Policies that mitigate these effects could help to slow the rise in adolescent obesity. In the second essay, we undertake a comprehensive study of the effects of teachers' evaluation standards on children's probability of being obese and BMI using a rich student-level data from the Early Childhood Longitudinal Study-Kindergarten cohort of 1998 (ECLS-K) including interviews with parents, data from principals and teachers as well as direct child assessments. In models in which we control for student-level fixed effects, we find strong evidence that with the increase of teacher evaluation standards students tend to have higher BMI and are more likely to be overweight. The third essay examines the impact of State-Sex-Education policies as well as the new Personal Responsibility and Work Opportunity Reconciliation Act of 1996, the Title V, Section 510 Abstinence Education Program on adolescent risk behaviors. To account for the potential differential impacts of 1996 Title V Section 510 had upon state sex education mandates which will subsequently bias our analysis, we thus employ an Interrupted Time-Series design that first exploits the impact of 1996 reform on state sex education legislative and identify their effects on adolescent sexual behaviors subsequently. First, we find that neither abstinence-only nor comprehensive sex education decrease the probability of being sexually active or increase the likelihood of performing safe sex. Instead, we find that abstinence-only lower the probability of using condoms and birth control pills relative to not using any birth control method. Second, using the ITS model, we find that the trend in percentage of students who had sex (percentage of students who had sex before 13) decreases by 0.5% in high-implement states relative to low-implement states post 1996 reform.
General Note: In the series University of Florida Digital Collections.
General Note: Includes vita.
Bibliography: Includes bibliographical references.
Source of Description: Description based on online resource; title from PDF title page.
Source of Description: This bibliographic record is available under the Creative Commons CC0 public domain dedication. The University of Florida Libraries, as creator of this bibliographic record, has waived all rights to it worldwide under copyright law, including all related and neighboring rights, to the extent allowed by law.
Statement of Responsibility: by Lu Yin.
Thesis: Thesis (Ph.D.)--University of Florida, 2010.
Local: Adviser: Figlio, David N.
Electronic Access: RESTRICTED TO UF STUDENTS, STAFF, FACULTY, AND ON-CAMPUS USE UNTIL 2012-08-31

Record Information

Source Institution: UFRGP
Rights Management: Applicable rights reserved.
Classification: lcc - LD1780 2010
System ID: UFE0041993:00001


This item has the following downloads:


Full Text

PAGE 1

THREE ESSAYS ON EDUCATION POLICIES AND CHILD HEALTH By LU YIN A DISSERTATION PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLORID A IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY UNIVERSITY OF FLORIDA 2010 1

PAGE 2

2010 Lu Yin 2

PAGE 3

I dedicate this dissertation to my wonderfu l family. Particularly to my loving and supportive fianc, Burhan Ogut, w ho has never left my side. 3

PAGE 4

ACKNOWLEDGMENTS I want to thank my dissert ation committee members, including David Figlio (chair), Larry Kenny, Mark Rush, and Paul Sindelar for their invaluable guidance, support and encouragement throughout the years. Without them, this dissertation will not be possible. I am especially grateful for t heir understanding and patience, while I was living away from campus. I feel deeply indebted to my advisor Davi d Figlio, who taught me to put my heart into research and motivated me to strive for bette r. I cannot hope for a better advisor. As I executed my dissertation research, I received the help of many people. Alan Guttmacher Institute provided dat a for the second chapter of this dissertation. Vivian Weis generously shared the report Between the Lines: States Im plementation of the Federal Governments Section 510(b) Abst inence Education Program in Fiscal Year 1998 issued by Sexuality Information and Education Council of the United States. My dissertation has greatly benefited from the comments of those who have read various versions of it, including David Denslow, Damon Clark, Peter Steiner, Steven Slutsky, Jon Hamilton, David Sappington, Ken Mease, and seminar participants at Northwestern University, University of Missouri, Abt. Associates, Urban Instit utes, and the American Institutes for Research. I alone am responsible for all errors and oversights. Finally, the acknowledgements would not be complete without a heartfelt thank you to my family. My fianc Burhan has a lways believed in me even when I could not quite believe in myself and has provided tr emendous help throughout my research. His patience and encouragement helped me to make the transit ion from where I came from to where I am now. My parents gave me th e courage to pursue my degree in the United States five years ago and wit hout their generous help, I w ould not have completed my 4

PAGE 5

study. Over the course of this five-year journey, my aunt, Lijun Yang, has been not only my closest family in Florida but also my mentor. She walked me through the highs and lows of my entire doctoral life. Beyond th is, I thank her for sharing her academic experiences with me, for listeni ng to my complaints and frustrations, and for believing in me. 5

PAGE 6

TABLE OF CONTENTS page ACKNOWLEDG MENTS .................................................................................................. 4LIST OF TABLES ............................................................................................................ 8LIST OF FI GURES ........................................................................................................ 10LIST OF ABBR EVIATION S ........................................................................................... 11ABSTRACT ................................................................................................................... 12 CHAPTER 1 INTRODUC TION .................................................................................................... 142 ARE SCHOOL ACCOUNTABILITY SYSTEMS CONTRIBUTING TO ADOLESCENT O BESITY? ..................................................................................... 17Introducti on ............................................................................................................. 17Accountability Systems ........................................................................................... 22Data ........................................................................................................................ 2 4Empirical St rategy ................................................................................................... 31Results .................................................................................................................... 38Do school accountability systems increase adolescent BMI? ........................... 38A Difference-in-Differences De sign ............................................................ 38POLS and FE models ................................................................................ 39Do school accountability systems reduc e adolescent activi ty levels? .............. 44Falsificatio n test s .................................................................................................... 45Testing school accountability system e ffects using adult BMI (ages 18 to 30) ................................................................................................................. 45The Effects of School Accountability System on Adolesc ent Hei ght ................ 46Specification test: ident ifying the effects of st udent accountabi lity .......................... 46Conclusi on .............................................................................................................. 483 HIGHER STANDARDS MAY NOT ALWAYS BE BETTER: WILL HIGH EVALUATION STANDARDS CONTRIBUTE TO CHILDHOOD OBESITY ............. 74Introducti on ............................................................................................................. 74Data ........................................................................................................................ 7 8Empirical me thods .................................................................................................. 82Results .................................................................................................................... 86Conclusi on .............................................................................................................. 894 IS IT WISE TO INVEST IN SEX EDUCATION? ABSTINENCE EDUCATION PROGRAM AND ADOLESCENT RISKY BEHAV IORS ........................................ 101 6

PAGE 7

Introducti on ........................................................................................................... 101A Brief Background on Absti nence-only prog rams ............................................... 106Theoretical Fr amework ......................................................................................... 108Empirical Fram ework ............................................................................................ 111Data ...................................................................................................................... 117Results .................................................................................................................. 120Nonlinear Panel Da ta Analysi s ....................................................................... 120Interrupted Time Series Model ....................................................................... 123Conclusion ............................................................................................................ 1265 CONCLUSION ...................................................................................................... 143 APPENDIX VARIABLES USED IN THE PRINCI PAL COMPONENT AN ALYSIS .......................... 147LIST OF REFE RENCES ............................................................................................. 148BIOGRAPHICAL SK ETCH .......................................................................................... 153 7

PAGE 8

LIST OF TABLES Table page 2-1 Summary statisti cs of all variables ...................................................................... 512-2 Participation map and date of introduction of school/student accountability system ................................................................................................................ 532-3 Summary statistics for differ ence-in-differenc es analysi s ................................... 552-4 Weighted average students' BMI in states with school accountability versus states without school accountability .................................................................... 562-5 Effects of school accountability on students' BMI: Results from DID controlling for individual characteri stics .............................................................. 572-6 Estimated effects of school accountability on student BM I ................................. 582-7 Estimated effects of school accountability on the probability of being overweig ht .......................................................................................................... 592-8 Estimated effects of school accountability on the probability of being underweight ........................................................................................................ 602-9 Estimated effects of SA using fix ed-trend model with du mmy variables ............. 612-10 Estimated effects of school accountability by gender ......................................... 622-11 Estimated effects of school acco untability by raci al group/ grade........................ 632-12 Estimated effect of school accountability on students' physi cal activities ........... 652-13 Estimated effects of school accountabi lity on adults' BMI: results from DID controlling for individual characteri stics .............................................................. 662-14 Falsification test: If student account ability systems affect student height ........... 662-15 Falsification test: If student accountability contributes to student overweight ..... 673-1 Summary Statistics of All Vari ables .................................................................... 913-2 Comparison of BMI and Obes ity Across Year and Gender ................................ 943-3 Teacher grading standards and obser ved teacher char acteristics ..................... 953-4 Means of dependent variables and selected student char acteristics .................. 96 8

PAGE 9

3-5 Estimated effects of teacher grading standards on st udents' BMI and the probability of being overwei ght ........................................................................... 973-6 Estimated differential effects of t eacher grading standards on students' BMI and probability of being overweig ht .................................................................... 994-1 Predicted Effects of Different types of Sex Education on the Probability of becoming Sexually active and Condom Us age ................................................ 1284-2 Summary Statistics of All Variab les .................................................................. 1294-3 Estimated Effects of State Sex Education on adolescent Pr(sexual intercourse)....................................................................................................... 1314-4 Estimated Effects of State Sex Education on Performing Safe sex (using condoms) .......................................................................................................... 1324-5 Estimated Effects of State Policie s on Pr(Sexual intercourse) and Pr(Condom Usage) .............................................................................................................. 1334-6 Estimated Effects of State Sex Education on Birth Control Method .................. 1344-7 Estimated Effects of State Sex Education on Birth Control Method .................. 1354-8 Estimated Effects of State Sex Education on number of sex partners .............. 1364-9 Estimated Effects of State Sex Educat ion on the age of firs t time had sex ...... 1374-10 Estimated Effects of State Se x Education using ITS design ............................. 138 9

PAGE 10

LIST OF FIGURES Figure page 2-1 States with school acc ountability systems and rate of overweight, 1993-2005 .. 682-2 States with different te sting grades when initiated the accountability systems ... 682-3 Trends in Body Mass Index and Percentage Overwe ight, 1999 5 ............... 692-4 BMI density for states Miss ouri and Tennessee, 1999 vs. 2005 ......................... 712-5 YRBSS Participat ion Map 19992005 ................................................................. 722-6 States with the Date of Introducti on of School Accountability before 1999, Between 1999 and 2001 and After 2001 S hown in Differ ent Legends ............... 722-7 States with the Date of Introduction of High School Exit Exam Before 1999, Between 1999 and 2001 and After 2001 Show n in Different Legends. .............. 733-1 Student BMI by teacher grading st andards ...................................................... 1003-2 Student overweight rate by teacher grading standards .................................... 1004-1 Percentage of students who had sexu al intercourse Pre-Post 1996 Reform (State leve l) ...................................................................................................... 1394-2 Percentage of students who had sexu al intercourse Pre-Post 1996 Reform (Aggregate leve l) .............................................................................................. 1404-3 Percentage of students who had sexu al intercourse Pre-Post 1996 Reform (Aggregate leve l) .............................................................................................. 1404-4 Percentage of students used condoms last time had sex Pre-Post 1996 Reform (Aggregate level). ................................................................................ 1414-5 Percentage of students used condoms last time had sex Pre-Post 1996 Reform (Aggregate level). ................................................................................ 1414-6 Percentage of students who had sex before 13 years of age Pre-Post 1996 Reform (Aggregate level) ................................................................................. 1424-7 Percentage of students who had sex before 13 years of age Pre-Post 1996 Reform (Aggregate level). ................................................................................ 142 10

PAGE 11

LIST OF ABBREVIATIONS BMI Body Mass Index ECLS-K Early Childhood Longitudinal Study Kindergarten cohort SIECUS Sexuality Information and Educ ation Council of the United States ITS Interrupted Time Series YRBSS Youth Risk Behavior Surveillance System NCLB No Child Left Behind Act CDC Centers for Disease Control and Prevention DID Difference-in-Differences AYP Average Yearly Progress POLS Pooled Ordinary Least Square BRFSS Behavioral Risk Factor Surveillance System NCES National Center for Education Statistics IRT Item Response Theory ARS Academic Rating Scale SES Socioeconomic Status PSS Private School Survey CCD Common Core of Data 11

PAGE 12

Abstract of Dissertation Pr esented to the Graduate School of the University of Florida in Partial Fulf illment of the Requirements for t he Degree of Doctor of Philosophy THREE ESSAYS ON EDUCATION POLICIES AND CHILD HEALTH By Lu Yin August 2010 Chair: David Figlio Major: Economics Studies in education literat ure have been focusing on the impact of policies on childrens academic performances. Howeve r, under single incent ive, educators and parents pursuing higher test score gains ma y undertake ways that unintentionally harm childrens health. Understanding how and whether certain education policies post negative impacts on childrens health is therefore very important, and has become the main research focus of this dissertation. In the first essay, we assess whether adolescent obesity has, in part, been driven by a factor previously overlooked: the school accountability movement. Exploiting the variation in the timing of the introduction of state accountability systems and in the grades to which those systems applied across states and over time, we find robust and cons istent evidence across models that school accountability systems significantly increas e students' Body Mass Index and explain a significant amount of the growth in adole scent obesity. These findings suggest that while accountability systems may promote students academic performance, this could occur at the expense of students' physical h ealth. Policies that mitigate these effects could help to slow the rise in adolescent obesity. In the second essay, we undertake a comprehensive study of the effects of t eachers' evaluation standards on children's 12

PAGE 13

13 probability of being obese and BM I using a rich student-level data from the Early Childhood Longitudinal Study-Kindergarten cohort of 1998 (ECLS-K) including interviews with parents, data from principals and teachers as well as direct child assessments. In models in which we contro l for student-level fixed effects, we find strong evidence that with the increase of teacher evaluation standards students tend to have higher BMI and are more likely to be overweight. The third essay examines the impact of State-Sex-Education policies as we ll as the new Personal Responsibility and Work Opportunity Reconciliation Act of 1996, the Title V, Section 510 Abstinence Education Program on adole scent risk behaviors. To account for the potential differential impacts of 1996 Title V Secti on 510 had upon state se x education mandates which will subsequently bias our analysis, we thus employ an Interrupted Time-Series design that first exploits the impact of 1996 reform on state sex education legislatives and identify their effects on adolescent sexual behaviors subsequently. First, we find that neither abstinence-only nor comprehens ive sex education decrease the probability of being sexually active or increase the lik elihood of performing safe sex. Instead, we find that abstinence-only lower the probability of using condoms and birth control pills relative to not using any birth control method Second, using the ITS model, we find that the trend in percentage of students who had sex (percentage of students who had sex before 13) decreases by 0.5% in high-implement states relati ve to low-implement states post 1996 reform.

PAGE 14

CHAPTER 1 INTRODUCTION Education policies are designed to pr omote student academ ic performance, narrow racial and ethnicity achievement gap and improve student readiness for college. Therefore, studies in education literature have been focusing on the impact of policies on childrens academic performances. However, under such single incentive educators and parents may undertake ways that unintentionally harm childrens health. Understanding how and whether certain education policies post negative impacts on childrens health is therefor e very important, and has become the main research focus of this dissertation. The adolescent obesity rate is tripled over the last three decades to 17%. In the second chapter, we assess whether this has, in part, been driven by a factor previously overlooked: the school accountability movement. Three potential reasons may underlie this phenomenon, increased stress, reduced physical activity or worsened diets. Exploiting the variation in the timing of the introduction of stat e accountability systems and in the grades to which those systems applied across st ates and over time, this paper estimates the effects of school account ability on students' weight status using a rich panel of student-level data. Three identification strategies are adopted in the empirical analysis, a Difference-in-Differences design to identify the short run treatment effect, a state fixed effects strategy and a state fixed-trend effects model. We find robust and consistent evidence across models that school accountability systems significantly increase students' Body Mass Index and explain a significant amount of the growth in adolescent obesity. We further estimate the causal effects on the treated of different lengths and frequencies of exposure to the sc hool accountability systems and find that 14

PAGE 15

the effects are increasing in the length/frequency of expo sure to the systems with a decreasing marginal effect. Placebo tests are performed to gauge the credibility of these conclusions. Consistent with these fi ndings, we also find that at least among females, school accountability decreases t he number of times that students participate in PE classes. These findings suggest that while accountability systems may promote students academic performance, this could o ccur at the expense of students' physical health. Policies that mitigate these effects c ould help to slow the rise in adolescent obesity. High performance standards have been advocated by school administrators, teachers, parents, and even by students themselves in education for centuries, however, little is known about their effect s on student health outcomes. In the third chapter, we undertake a comprehensive study of the effects of teachers' evaluation standards on children's probability of being obese and BMI using a rich student-level data from the Early Childhood Longitudinal Study-Kindergarten cohort of 1998 (ECLSK) including interviews with parents, data from principals and teachers as well as direct child assessments. To account for the potential endogenei ty problem due to simultaneity between BMI and teachers' gradi ng standards, we first matched students' direct assessments scores horizontally and then divide students into high standard cohort and low standard cohort based on thei r teachers' ratings of English, mathematics, and science skills. We then estimate the effects of teachers' heterogeneous standards on student s' BMI and the probability of being overweight given that students perform equally in direct assessment which includes reading, math, and science. Furthermore, we evaluate the im pact of students' self-assessment of their 15

PAGE 16

16 academic and social skills on their own BMI and the probability of being overweight using the same method. The fourth chapter examines the impact of StateSex-Education policies as well as the new Personal Responsibilit y and Work Opportunity Reconciliation Act of 1996, the Title V, Section 510 Abstinence Education Program on adolescent risk behaviors. While intended to prevent student risky behaviors and hence promote adolescent health and quality of education, there is little empirical evidence about the effectiveness and efficacy of such state level policies, especially after t he implementation of Title V Section 510. Using a rich panel of student-level data between year 1993 and 2005, available from Youth Risk Behavior Surveillanc e System (YRBSS), we first explore the effects of different state sex education mand ates on adolescent sexual behaviors using various panel data techniques. To account fo r the potential different ial impacts of 1996 Title V Section 510 had upon state sex educat ion mandates which will subsequently bias our analysis, we thus employ an Interr upted Time Series design that first exploits the impact of 1996 reform on state sex educati on legislatives and identify their effects on adolescent sexual behaviors subsequently. Fi rst, we find that nei ther abstinence-only nor comprehensive sex education decrease the probability of being sexually active or increase the likelihood of performing safe sex. Instead, we find t hat abstinence-only lower the probability of using condoms and birt h control pills relative to not using any birth control method. Second, using the IT S model, we find that the trend in percentage of students who had sex (perc entage of students who had sex before 13) decreases by 0.5% in high-implement st ates relative to low-impl ement states post 1996 reform.

PAGE 17

CHAPTER 2 ARE SCHOOL ACCOUNTABILITY SYSTEM S CONTRIBUTING TO ADOLESCENT OBESITY? Introduction The striking rise in childhood obesity has a ttracted national attention over the last several decades. Between the 1970s and 1990s, the fraction of overweight children and adolescents tripled from 5 percent to 17 percent. There are several public health and economic reasons to be concerned with this sharp increase. Being overweight in childhood increases the risk of being an overweight adult (Guo et al. 2002), coronary calcification in adulthood (Freedman et al. 2001), and cardiovascular disease mortality (Must et al.1999; Lakka et al. 2002; Mokdad et al. 2003). Additionally, greater health, social, and economic costs are known to be a ssociated with obesity. Each year, obesity causes at least 400,000 deaths in the United States and the costs of providing health care to obese American adults and children have been estimated to be approximately $147 billion in 2008 (Finkelstein et al. 2009). The existing literature in medicine and soci al sciences on the causes of obesity can be divided into studies directly examining genetic1 (Perusse et al. 2001; Katzmarzyk et al.1999) or behavioral fact ors (Stunkard et al. 1990; Dietz 1994), a studies investigating environmental fact ors, for example, family structure and socioeconomic status (Locard et al. 1992; Woolston 1987; Bar-Or et al. 1998). Recentl researchers have applied the tools of economics to explore the causes and consequences of rising obesity in the United States. Among the most prominent causes nd y, 1 Research on human genetic mapping has identified se veral specific genes and gene mutations that are believed to cause human obesity (Perusse et al. 200 1). Research found that a child born to a parent in the upper five percent of the BMI distribution has a 60 to100 percent greater risk of being obese compared to a child born to parents with nor mal, healthy weights (K atzmarzyk et al.1999). 17

PAGE 18

of obesity examined in the literature, the leading thr ee are lack of physical activity2 (Lakdawalla and Philipson 2002), increased television viewing3 (Eisenmann, Bartee and Wang, 2002; Reilly et al. 2005) and fast food consumption et te r inds se in e ite mother. 4 (Cutler et al. 2003; Chou al. 2004). Only a few studies focus spec ifically on childhood and adolescent obesity. Cawley, Meyerhoefer, and Newhouse (2006) ex plore the impact of state Physical Education (PE) requirements on youth physical activity and obesity. Although the sta requirements do increase student PE time, they find no evidence that PE lowers BMI o reduces the probability that a student is overweight. Schanzenbach (forthcoming) f that children who consume school lunches ar e more likely to be obese than those who brown bag their lunches. Anderson, Butcher and Levine (2003) examine the increa maternal employment as a potential source of childhood overweig ht. As the increased value of time may lead more people to shift to fast-food alternatives that are often calorie and fat laden. They conclude that the intensity of a mothers employment over a childs lifetime has a positive effect on a child s likelihood of being overweight, given th child is in a high income family, with a well-educated or wh While the previous research provides impor tant insight into the actions and forces that induce individuals to gain excessive weight over time, they are unlikely to be the sole causes of this epidemic. This paper considers a potentially important additional 2 Lakdawalla and Philipson (2002) estimate that one-third of the increase in BMI in recent decades can be attributed to falling food prices and two-thirds due to decreased exertion in normal work activity. 3 Eisenmann, Bartee and Wang (2001) examined Physical Activity, TV Viewing, and Weight in U.S. Youth using 1999 Youth Risk Behavior Survey and found that physical inactivity and frequent television viewing may contribute to obsity. Reilly et al. 2005 showed that sedentary activities such as watching television and playing video games significantly increase rates of obesity in children as young as three years old. 4 Cutler et al. (2003) and Chou et al. (2003) found that body weight and obesity prevalence increase significantly as the per capita number of restaurants and the real price of cigarettes go up, suggesting that more eating out and less smoking may have contributed to the rise in obesity. 18

PAGE 19

explanation: the increased high-stakes testing systems for students and schools. Increased accountability for students began in earnest in the early 1970s, and since then many states have used high-stakes ac hievement or profic iency tests to hold students accountable for meeti ng state-mandated educational achievement standards. Beginning in the early 1990s, this account ability based on high-stakes testing was applied to schools in addition to students, and many states started their own test-based accountability systems. The No Child Left Behind Act of 2001 furthered this national trend toward increased school accountability, one in which dozens of states formally grade or evaluate their schools based on st udent test performance. Rather than evaluating the academic achievement intended to be accomplished through the systems5, this paper contributes a new angle by focusing on an overlooked issue associated with school accountabilit y: the possibility that incr eases in implementation of school accountability across states raise the prevalence of overweight among adolescents. Three potential reasons may underlie this phenomenon. First, school accountability may increase weight gain by imposing additional stress on students, which has been shown to be associated with obes ity in small scale studies. Over the last several decades, social scientists have dev oted a considerable amount of effort to identifying the effects of va rious psychological states on obesity (Friedman and Brownell 1995). Their findings revealed that obesity is gener ally the result of a lifestyle that is plagued by stress, pressure, boredom, and poor self image. With the increased course work at school and the unavoidable pressures from various tests, young people facing 5 See Koretz and Barron (1998), Clark (2003), Haney (2000, 2002), Figlio and Rouse (2006), West and Peterson (2006) and Chakrabarti (2006). 19

PAGE 20

school accountability are now under higher levels of stress, which may directly relate to their mental and physical health. A recent study shows that children residing in states with more stringent accountability laws are more likely to be diagnosed with Attention Deficit/Hyperactivity Disorder (ADHD) and cons equently are more likely to be prescribed psychostimulant drugs for controlling the symptoms6 (Bokhari and Schnedier 2009). Other studies assessing the relation between homework and school exams and stress indicate that pressure from teachers grades, and homework is one of the primary sources of stress and depressi on for teens (Lindblad 2 006; Kang 1997; Kouzma 2002). Depressed feelings in adolescence are, in turn, associated with an increased risk for the development and persistence of obesity over one year (Goodman and Whitaker 2002). Second, students, facing likely punishment for poor test outcomes, such as repeating the same grade or graduating wit hout honors, may sacrifice much-needed physical activity time for more study ti me. In addition, facing a need to respond to improve academic test scores and fend off potential funding losses, school administrators may change policies in order to stimulate student performance. Rouse, Hannaway, Goldhaber, and Figlio (2007 NBER working paper) show that schools under higher pressure to perform well tend to change their policies in order to improve students performances, such as requiring summer school, before/after school tutoring and Saturday classes. The direct effect of these policies will be decreased physical activity time in or after school and longer study/homework time, all of which may be correlated with young peopl e being overweight. 6 One should notice that another potential reas on for the increase in ADHD diagnosis and psychostimulant consumption rates may due to schools labeling marginal students such that to reshape the testing pool or students are now receiving more appropriate diagnosis. 20

PAGE 21

Third, to respond to the academic impr ovement focused testing system schools may reduce the nutritional content of students food consumption in schools. Figlio and Winicki (2005) find evidence that schools respond to accountability pressures, by substantially increasing the calorific content of school menus on testing days. Anderson and Butcher (2005) find that sch ools under financial pressure due to the accountability system7 are more likely to make junk food available to their students. To investigate the relationship between school accountability and adolescent obesity I exploit unevenness in the timing of the introduction of school accountability systems and the grade levels to which these applied using a rich panel of student-level data, available from the Youth Risk Behavior Surveillance System (YRBSS) developed by Centers for Disease Control and Prevention (CDC). Three identification strategies are adopted in the empirical anal ysis, a Difference-in-Differ ences design to identify the short run treatment effect, a state fixed effects strategy and a state fixed-trend effects model where I construct a state BMI trend index for each state to control for potential heterogeneous BMI growth trends. Regr ession results suggest that school accountability systems have positive and statisti cally significant effects on student BMI and growth in adolescent obesity controlling for individual-level measures, state policy covariates and state BMI growth trend. This effect is bigger in the short run (within 4 years) than in the long run (more than 4 years), am ong female students and among Asian and Hispanic students compared to White and AfricanAmerican students. Additionally, school accountability decreases the number of times female students 7 Schools that got failing grades will suffer consequence s; hence schools may try to raise money in order to strengthen core academics to meet the achievemen t goals without cutting elective courses. They could achieve it through soft drink and vending contra cts, or through other snack food sales. 21

PAGE 22

participate in physical education classes, but has no statistically significant effect on male students. These results are consistent with a simult aneous work by Anderson, Butcher and Schanzenbach (2009) where they us e school level data from Arkansas and find that schools that were below the average yearly progress (AYP) threshold under NCLB in year t-1 have a small, but statistically significantly higher rate of overweight and obesity in year t. Findings of this study suggest that a part of the recent rise in adolescent obesity can be attributed to the school accountabilit y movement. Although this represents only a small portion of the overall growth, it is a portion that may be responsive to policies. Since many of the determinants of this ri se may be invariant to education or health policies, this is nevertheless an important conclusion. The rest of the paper is organized as follows. The next section presents a br ief background of the accountability systems and section three describes the data set followed by the discussion of empirical framework and methods used in the analysis. Section IV presents the results followed by falsification tests and Section VI concludes. Accountability Systems Increased accountability for students began in 1970s and these testing programs are generally designed to measure the degr ee of skills possessed by students. For example, high school exit examinations were first integrated into the student accountability reform movement in 1970s afte r unfavorable international comparisons of mathematics and science achievement among U.S. students with t hat of students in other industrialized nations. Since then educators and policy makers are working to bring value to the high school diploma by raising the rigor of high school standards, assessments and curriculum and aligning expectations with the demands of 22

PAGE 23

postsecondary education and work. This accountability system was applied to schools in the early 1990s. In 1993, Wisconsin, North Carolina, Taxe s and Connecticut were the first four states to implement account ability at different grades. 12 states had accountability systems at the school level by 1996, and 39 st ates did so by 2000. The No Child Left Behind Act of 2001 (NCLB) provided more ex tensive federal requi rements for states, student assessment, and school accountabi lity systems than had ever previously existed. Each state develops its own a ccountability system based upon the states content and achievement standards, valid and reliable measures of academic achievement, and other key indicators of school and district performance such as attendance and graduation rates. By 2003, all t he states were required to implement school accountability that formally grade or evaluate their schools based on student test performance and make report cards publicly avai lable. To better illustrate the different timing of school accountability introduction dates, these data are presented in the form of maps that appears in Figure 2-6. Additionally, accountability system was not applied to all grades when it was initially introduced. Specific ally, around 14 states introduced the system into 1 to 3 different grades, 13 states applied it to 4 to 6 grades and other states selected more than 7 grades to apply the system. Under NCLB however, all states were required to administer assessments in reading and mathem atics every year to all students in grades 3 through 8 and once during high school by the end of school year 2005-06 as illustrated in Figure 2-2. 23

PAGE 24

In the current accountability environment, policymakers want tests to have stakes for test-takers attached to them so that students will exert greater effort to pass them and use test results to reward schools with additional funding from the government or to punish the schools that failed to meet t he academic yearly progress (AYP) goals. Different from previous education policy main ly focused on providing resources, school accountability systems provide additional in centive for schools to maintain high academic performances by requiring school s publish their annual test scores to the public. To respond to school accountabilit y pressures, schools may reallocate resources, both time and financial assets, toward testing subjects and unintentionally harm students physical and mental health. Data To test the proposed hypothesis, I make use of the Youth Risk Behavior Surveillance System (YRBSS) for t he years 1999, 2001, 2003, and 2005 which has been administered biennially since 1991 by the Centers for Disease Control and Prevention (CDC). Starting in year 1999, CDC collected data on students weight and height; therefore data before 1999 are not included in this analysis. The YRBSS system, established in 1991, monitors high school students risky behaviors, including those relating to obesity, eating hab its, physical activities, and other risky behaviors. It includes national, state, and local school-based surveys of nationally representative samples of 9th through 12th gr ade students. In order to im plement this across-state analysis, the restricted versi on of the YRBSS data with state identifiers was acquired. In the analysis, observations were dropped if a students weight or height was missing, reducing the pooled sample size from 57,826 to 54,065, a reduction of around 7%. Schools with relatively high numbers of Af rican-American and Hispanic students were 24

PAGE 25

oversampled to ensure adequate representation of these groups. A weight is applied to each record to adjust for non-response and the distribution of student s by grade, sex, and race/ethnicity in each state. Self-reported data on height and weight allow me to construct three outcome variables: the body mass index (BMI) of each respondent and indicators of whether he or she is overweight or under weight. Brener et al. (2003) s how that self reported height and weight by high school students are valid proxies for measured values, but these might underestimate the prevalence of overweight. Nevertheless, weight status variables will be the main dependent variables in all the regression analyses, and measurement error present at left hand side will not affect the accuracy of the estimates. According to the CDC growth char ts in the United Stat es, the 97th percentile of student BMI, ages between 15 and 18 years old, is below 34. The 5th percentile of student BMI, ages between 15 and 18 years old, is above 14 (Centers for Disease Control and Prevention, 2000) Therefore, observations with BMI>40 as well as BMI<10 are excluded from the data se t which may be subject to measurement errors and this again reduced the sample size to 53,499, around a 1% fall. The trends in student BMI and the percentages of the student obese and overweight8 in the YRBSS are plotted in Figure 2-39. Between 1999 and 2005, BMI increased by 0.4 kg/m2 or by 2.8% for all students whil e the average BMI is higher for male students and the percentages of over weight and at the risk of becoming 8 Children who are overweight or at risk of bec oming overweight are defined as a BMI above the 85th percentile for children of the same age and gender in 2000 (defined by Centers for Disease Control and Prevention). 9 The values of BMI and overweight are comput ed based on YRBSS sampling weights which produce nationally representative figures as of each survey year. 25

PAGE 26

overweight increased over 3% for female students. One might notice that we observe a small dip in the BMI and obesity trends in year 2001 which might due to the fact that not every state participated in YRBSS in every r ound of the survey as showed in Figure 2510. Therefore, I plot the same graphs only using data from states that conducted all 4 waves of YRBSS. Reassuringly, the BMI and obesity trends increase consistently over the 4 survey years in this case. Relati ve to Asian students, White, Hispanic, and African-American students have higher BMI while White students have lower BMI compared to African-American and Hispanic students. On average, 7.8% of Asian students are overweight, yet this number is significantly hi gher, 20.5%, for Hispanic and African-American students. From the graph, we can see that BMI increased monotonically as the student progressed from grade 9 to grade 12, while the percentage of student ov erweight did not show any system atic pattern across grades. In 1999, 15.5% of ninth graders we re overweight, which was hi gher than any other grades. In 2005, however, 18% of twelfth graders were overweight, highest among all grades. In addition to main individual control measur es, such as gender, race, and grade, four variables related to food consumption11 are also included in the later regressions under the hypothesis that unhealthy eating habits c ontribute to extra weight gain. I also examine three variables pertaining to student s' physical activities: numbers of times a student participated in vigorous exercise, moder ate exercise, and PE classes in the last seven days. These variables are used as outcome variables under the hypothesis that 10 States apply for funding to conduct the YRBS every survey year. However, for various reasons not every state will apply for the money. 11 Questions on the YRBSS survey are: during the pa st 7 days, how many times did you eat fruit/green salad/potatoes/carrots etc.? Regression results excluding these variables are not significantly different. (Results are available upon request.) 26

PAGE 27

with the introduction of acc ountability, students may reduce the amount of physical activity. Following prior studies, I also added TV viewing, measured as number of hours of TV watched per day for each individual, into the accountability and BMI equation to control for its potential positive effects on adolescent obesity. Definitions, means, and standard deviation s of all variables employed in the regressions appear in Table 2-1. Except where noted, they are based on the sample of 53,499 that emerges when obser vations with missing values are deleted. The summary statistics in the table and a ll regressions in this paper are computed based on YRBSS sampling weights and are representative of the population at large. According to YRBSS, the target population consisted of all public, Catho lic, and other private school students in grades 9 through 12 (Methodology of the Youth Risk Behavior Surveillance System 2004). The main concern with this is that only public schools are legally bound by accountability related policies; includi ng both public and private schools in the dataset will potentially underesti mate the impact of accountability on student BMI and obese prevalence. As CDC does not release school-level identifiers and the data does not contain any internal information t hat may help me to identify school types12, to determine in which states private schools were part of the sample becomes impossible. However, sampling schools with any of gr ades 9 to 12 were selected with probability proportional to school enrollment size. Thus most of the surveyed schools are public schools due to the bigger enrollment size co mpared with private schools. In addition, YRBSS is the only nationwide survey that includes questions on BMI, eating habits, and 12 For instance, we may use school size or grades s pan as proxies for public schools. Unfortunately, the YRBSS does not include such information. 27

PAGE 28

physical activities starting from year 1999, which makes YRBSS the most appropriate dataset for this study. The key components of the analysis are the variables used to measure the accountability systems. I obtain the date of introduction of school accountability system from the Consortium for Policy Research in Education (CPRE) and then collect data on testing grades from the websit e of Department of Education for each state, shown in Table 2-2 column 2 and 3. As treatment may differ not only in the length of accountability treatment, but also vary by the intensity of treatment, exploiting the variations in the introduction dates of school accountability and in testing grades across states, I construct two micro level independent variables whic h equal to the number of years an individual has been exposed to the system (LENGTH_gr ade) and the number of times an individual has been tested (FREQUENCY_grade) at each survey year since the first time she/he has been tested for t he accountability purpose, respectively. For example, in Texas where school accountabilit y was implemented in 1993 for grades 3 to 8 and grade 10, if a student fr om 9th grade was surveyed in year 2001, the number of years she has been exposed to the system equals to 7 and the number of times she has been tested for the accountability purpose, in this case, equals to 613. One potential problem with the LENGTH variable is that the official implement ation date of school accountability policies for each state was us ed in calculating the exposure variable; however some states adopted an unofficial but similar system at an earlier time. For instance, the Florida school accountability system was announced to be implemented in 13 A 9th grader in 2001 would be in 1st grade in 1993, so the first time she/he be treated under SA would be in 1995 when she/he was in 3rd grade. LE NGTH_grade=2001-1995+1=7. FREQUENCY_grade=6 (grade 3 in 1995 to grade 8 in 2000. This student is not tested in grade 9 in 2001.) 28

PAGE 29

1999, yet a similar system actua lly started running several year s earlier, in 1994. If this is the case, the estimated effects of sc hool accountability might be downward biased or I might not observe an adjus tment period of student BMI fo r these states. One should also notice that to comply with the NCLB 2005-06 deadline, many st ates introduced new tests or tests in additional grades during the past few school years. Without taking into account of such changes, both LENG TH and FREQUENCY variables will be misspecified. I therefore co llected data about major changes made in the states testing grades since 2002 from State Achievement Pr ofiles conducted by C enter on Education Policy and calculate the two main independent variables to account for these changes. Further, it is likely that not only students who are directly tested under school accountability systems experience higher wei ght gain, but also students who anticipate future tests. Therefore, usi ng the variation across states and over time in introduction of accountability laws, I construct a separate va riable which equals to the number of years a state had implemented the school accountability system to capture the overall effects on the entire student body (LENGTH_general). My research design exploits the substant ial variation in students weight status across states in the timing of the enactment of pre and post school accountability laws. Controlling for state trend in the growth of BMI becomes cruc ial when trying to identify a cross state effect of school accountability on students BMI. On one hand, the fact that not every state participated in YRBSS in every round of the survey hinders the estimation of state trend in years 1999 to 2005 directly using YRBSS. On the other hand, BMI trend calculated using adolescent data will be plagued by the impact of school accountability and hence result in biased estimates. Instead, I employ data from 29

PAGE 30

the Behavioral Risk Factor Surveillance S ystem (BRFSS), a large-scale state-based system of health surveys of approximately 350,000 adults annually that is conducted by the CDC and by 1994, all states were par ticipating the BRFSS. The BRFSS is a consistent source of information on health risk behaviors at state level, including respondents height and weight information, wh ich allow the author to calculate her/his BMI and information on demographic factors, educ ation and eating habits which are the same as the questions asked in YRBSS. I use BRFSS from 1999 through 2005 which spans the same period as YRBSS, to estimate a BMI growth rate for each state which is later used as a control variable in the accountability-weight status analysis. Since the school accountability systems are ul timately measured at the state level, I incorporate additional statespecific measures in the analyses to capture time-varying trends within areas. Specifically, three state level control variables are augmented with YRBSS and state level measur es pertaining to school acc ountability policy and testing grades in later analysis. They are percentage of individuals below poverty line used to control for states socioeconom ic status and the percentage of the state population with a bachelors degree used to control for paren ts education effects on youth BMI (both obtained from the US Census Bureau, Population Division) both of which in some studies show negative relation with adult BMI. The hypothesis is more schooling of a parent leads to lower levels of obesity for he r children as she may influence kids choice of food and lifestyle. Last, I obtain data on percentage of public schools providing AP courses for each state in each survey year (from College Board's Advanced Placement (AP) Program website)14. States established various high school honor diploma 14 College Board's Advanced Placement (AP) Program was started nearly four decades ago to enable students to complete college-level studies while still in high school and to obtain college credit or 30

PAGE 31

programs based on students regular classes as well as AP course performance to encourage high school students to take challengi ng and rigorous courses. Therefore, the variable, which is coded as percentage of public schools providing AP courses, measures the effect of increased high school academic options on student BMI. Empirical Strategy The primary aim of this study is to eval uate the impact of school accountability on students weight status. As the implement ation map showed, twenty-seven states implemented school accountability after 1999 and among them ten states did not adopt this policy until 2003. This uneven timing al lows me to employ a Difference-indifferences (DID) design to identify the shor t run effect that school accountability had upon the student BMI of those w ho are in the treated states. Specifically, using data from years 1999 and 2003, I cla ssify states with school a ccountability implementation date between 1999 and 2002 as a treatment group. On average, students in the treatment group are exposed to the school accountability systems for 2.4 years. The choice of comparison group is crucial in DID analysis. I use students who have not been affected by the school accountability systems and reside in states which started the policy in 2003 as a control group based on the assumption that students weight status will not change immediately a fter the implementation of school accountability systems. A DID estimator exploits the existence of a comparison group in an attempt to estimate the impact of the treatment (i n this case, school accountability) on the eligible group (in this case, states which employed accountability systems before 2003). This relies on the assumption that there are no other cont emporaneous shocks affecting the relative placement. Today more than 500,000 students in about half of the nation's high schools take at least one AP course. 31

PAGE 32

BMI of treatment and comparison groups. This assumption is fairly satisfied because during the time period there were no major heal th or education reforms that might affect student BMI to a great extent. Even if the assumption of common shocks is not satisfied fully, it does become more reasonable once I condition upon observable characteristics. The DID model estimates: (1) istst ist t is t is istState XYear Treat Year Treat BMI '') (3 2 1 Equation 1 posits that the BM I, for the ith student residing in state s during year t, is a function of pre-post school accountabi lity indicators, measured here by Treat a dummy for treatment group (1 if treated, 0 if control), Year a dummy for before or after policy (1 if after, 0 if before) and the interaction term capturing the variations to the treated students (relative to the untreated student s) in the year after the policy (relative to the year before the policy). In addition, BMI depends on a vector of individual characteristics (X) and a vector of time-var ying state characteristics (State). This DID estimator illustrates the unique identificati on strategy made possible by the YRBSS data. This approach differences out the time-i nvariant cross-sectional heterogeneity and relies instead on comparing the changes withi n the groups that did or did not face school accountability treatment over this period. For this estimator to identify the impact of school accountability it must be true that the average change in student BMI for the treatment group without the introduction of school accountability systems would have been on the same trajectory as the student BMI for the control group. In this case, both control and treatment groups must be randomly selected. Students should not be self selected to be in either group, yet the implementation of school accountability must be exogenous to their eligibility. However, 32

PAGE 33

the exogeneity assumption, 0)( ist istTreatE, may not hold in this case. For instance, most states in the southeast regions adopted school accountability relatively earlier than other states, and this is the case partially because students in the South are traditionally have ranked at the bottom on National Asse ssment of Educatio nal Progress (NAEP) scores, college entrance tests, and other indicato rs of educational ac hievement. If this is the case, the DID design will not be the most reliable strategy. To overcome the difficulties stated above to the most extent, the data obtai ned from Youth Risk Behavior Survey enables the use of panel data techniques. Consider a simple model of adolescent BMI: (2) istst ist ist istState XSA BMI ''0(3) istsist Where BMI is student is body mass index in state s at time t, which is later replaced by the probability of an individual is overweight or underweight when using a Probit model, SA represents either the number of year s or number of times student i has been exposed to school accountability or tested in state s at time t and ist is the error term. In this study, if the introduction of accountability is exogenous and uncorrelated with any local policies or socioeconomic st atus, then Pooled Ordinary Least Square (POLS) regression of student BMI on school accountability in each state would produce an unbiased estimate. However, if under (3) and s is correlated with any st ate policies other than accountability which may be correlated with student BMI, either POLS or GLS (RE) estimates will be biased, which to a great extent, would be the case. For example, various physical education programs implemented across states over the years which 33

PAGE 34

may coincide with school accountability im plementation. Between years 1991 and 2003, the percentage of high school students enrolled in daily PE classes declined by 13% from 41.6% to 28.4% (Gunbaum et al. 2004). Because physical activity level is believed to be associated with body weight (Heitmann et al.1997; Stone et al. 1998), state PE policy may bias SA in a POLS model. Se cond, as mentioned ear lier, states with relatively worse performance on national tests have a higher possibility of adopting a school accountability system before 1999 which results in the endogeneity problem in the variable of SA, 0)( ististSAE. Nevertheless, if the correl ation is restricted to the state-specific component, i.e. )(),,,(ist sst istististE StateXSAE fixed effects estimation is the preferred procedure to elim inate any other bias from the policies that lead to up or down trends in student BM I in each state. Fixed effects model estimates: (4) istsst ist ist istState XSA BMI ''0The various dates of introducti on of the accountabi lity systems and testing grades over the periods across states during which student BMI are observed enable the fixed effects analysis. In equation 4, s refers to the state fixed effects. One potential problem with the fixed effects model is that it fails to control for the potential state BMI trend over time which causes endogeneity problem due to omitted variable. For instance, Mississippi has been t he fattest state in years 2005 to 2007, with the average percentage of obese adult popul ation 31.6%, while the adult obesity rate is about 13% lower in Colorado. Between year 2007 and 2008, the obesity rate increases 0.8% from 32% to 32.8% in Missi ssippi while the obesity rate decreases by 0.2% from 18.7% to 18.5% (Overweigh t and obesity Trends by State 1985-2008, CDC 34

PAGE 35

2008). Figure 2-3 further validates the existenc e of differential state trends by plotting densities of BMI for states Tennessee and Missouri in year 1999 and 2005 using data from YRBSS. As one can see that the kern el density is more peaked than the normal for Missouri while the kernel density is ve ry right-skewed for Tennessee. From 1999 to 2005, the density of BMI for Miss ouri noticeably shifted to th e right where the shift is much milder for Tennessee. The fixed effe cts model would fail to address this statespecific growth pattern and leads to an overesti mation of the causal effect of SA. I thus use a fixed-trend model as specified in equation (5) to solve this potential problem to the most extent. (5) isttstsst ist ist istState XSA BMI ''0where t represents time dummies incremented in years and ts captures different state trend over time effects. The fi xed-trend model can resolve the endogeneity concern since the adolescent BMI depends only on historical level of state BMI rather than its growth. One drawback of including the full set of year dummies, however, is that the SA measured by LENGTH will not be accurately estimated. Particularly, LENGTH variable is constructed as the number of years a student has been exposed to the system, in which case the incremental change will be ex actly two years in the following survey period for students from a sa me state progressing from one grade to the next. The multicollinearity between LENGTH measur es and the year dummies made it very difficult to disentangle their separate e ffects. Therefore the only measure left is FREQUENCY. Furthermore, time trend picks up the effect of permanent factors in national and state shocks and leaves only trans itory variation from its trend to be 35

PAGE 36

explained by changes in SA, the coeffici ent of FREQUENCY may be underestimated with larger standard errors in comparison to a fixed effects model. Alternatively, to control for the unobserved nationwide events not measured by the other independent variables, I could firstly estimate the state by year trend for each state and then include them into the main regression. To estima te the unbiased state BM I trend, I use BRFSS data for adults, aged younger than 35 years-ol d, and calculate BMI trend for each state which is then incorporated in the main acco untability-weight status equations. I choose adults age 18-35, because adults BMI in this age group are less likely to be affected by medical or other social stat us factors that could not be observed and hence controlled by the regression. Also, this age group is most likely to resemble the BMI trend of students for each state. I further divide this sample into two subgroups aged younger or older than 25 years to test if the state trends differ across age groups. The original fixed effects model becomes: (6) ists s t st ist ist istageTrend Year State XSA BMI )25(* ''1 (7) ists s t st ist ist istage Trend Year State XSA BMI )30 25(* ''1 Where Trends is obtained from: (8) isttss st ist a ist aTrend State X BMI ''0where a B M I is the outcome variable for adult i residing in state s during year t Xa is a set of adult characteristics and State is a set of state characteristics which include the same controls as X and State in equations (1)-(7). So far, I assume that school accountabi lity has linear effects on student weight status once controlling for individual characteristics and various state policy characteristics, however, it might be the case that student s weights respond to school 36

PAGE 37

accountability systems in a non-linear fashion, i.e. istist istististSA SABMIE ), ( (assuming )(ististXSAEis linear in function of X ). If this is the case, the linearity assumption is violated and hence jeopardize s the analysis. Additionally, students weight status may not change dramatically in a relatively short period, for instance one or two semesters, rather an adjustment period for student s to respond to the policy reform may exist. One might also be worri ed that state level programs implemented earlier or historical state socioeconomic status whose effects become apparent at the same time as the accountability policy was instituted. I thus em ployed an additional method by creating a set of dummy variables to signify separate length ranges for school accountability, namely: Exposed to School Accountability 1-2 years, Exposed to School Accountability 3-4 years, etc. Specifically, I then divided the school accountability LENGTH variable into six twoyear intervals, which created five dummy variables (using students had not been or had been exposed to the system for less than one year as baseline), to test for nonlinearity in the effects. Similarly, to test the differential impacts of the various int ensity of SA on students weight status, I use the same method s eparating FREQUENCY_grade into 5 mutually exclusive groups where the higher number of times a student had been tested represents a more stringent accountability system. Last, I run separate regressions using t he probability of being overweight and the probability of being under weight as dependent variables for each proposed equation, though I use BMI as the outcome variable in every equation showed above. Marginal effects are presented in the next section. Various probit regressions are estimated: (9) ] '' [)1Pr(1ists st ist ist istMeasure StateTrend State XSA Y 37

PAGE 38

where the dependent variable is a dummy variable which equals one if student i is overweight or underweight in state s during year t StateTrendMeasure is substituted with several measures described above in different models. Results Do school accountability systems increase adolescent BMI? A Difference-in-Differences Design In this section, I present the DID result s based on the model discussed in Section III. The means of the data are presented in Table 2-3 for the accountability states and control states (both for the before policy year and after policy year). There are not many striking differences across groups except that non-accountab ility states have a lower percentage of African-American st udents and a higher percentage of White students than the accountability states. One reason is that African-American students usually have worse academic performances rela tive to other racial groups and school accountability systems, in turn, are more likely to be implemented in states where academic performance is poor. Hence, stat es with higher perc entages of AfricanAmerican students may adopt school accountability systems in an earlier year to promote students academic perfo rmances. Another differenc e between accountability and non-accountability states wort h mentioning is that most states implemented school accountability system in grades 3 to 11, therefore the post policy percentages of eleventh graders who have not been exposed to the systems in the non-accountability states decreased significantly lower to ar ound 7% compared to their accountability counterparts, around 25%. Table 4 illustrates DID estimation of the s hort run effect of sc hool accountability on student BMI. The panel compares the change in BMI for students in the states that 38

PAGE 39

implemented the policies to the change for students in the control states. Each cell contains the mean average student BMI, along with the standard error and the number of observations. There is a 0.17 kg/m2 decrease in the BMI of students in the nonaccountability states over this period, compared to a 0.2 kg/m2 increase in the BMI of students in other states. Thus there was a total 0.37 kg/m2 relative increase in the BMI of states implemented t he policies. However, the difference is not statistically significant. As the summary statistics revealed, there is an imbalanced distri bution of racial groups across accountability and non-accoun tability states; hence a regression framework will allow me to control for t hese observables that affect the outcome variables of interest. The third row of Table 2-5 shows the es timates of the interaction term from equation (1) and for brevity, some of the regression output has been omitted from the presentation. The coefficients indicate that student BMI incr eased significantly by 0.79 kg/m2 for the treatment group with the average 2.4 years treatment Female students BMI increased by 1.17 kg/m2 for the treatment group while there is no statistically significant increase of BMI for male students due to the treatment. POLS and FE models Starting with the specificati on using continuous variables, I tested the effects of school accountability on student BMI, along with variables that control for individual characteristics and state characteristics. T he first column of Table 2-6 to Table 2-8 report the estimated effects of school accountability on student BMI and the probability of being overweight and underweight using the POLS of equation (2) where top panel used LENGTH measures and bottom panel used FREQUENCY measure. They are provided here primarily as a benchmark to g auge the possible bias in comparison to the 39

PAGE 40

more reliable state fixed effects models di scussed below. First, note that both the LENGTH and FREQUENCY variables have positive effects on BMI and the probability of being overweight while the LENGTH_general which m easures the overall school accountability effects presents slightly higher effects implying that not only students in the tested graded were affect ed by the school accountabil ity systems. On the other hand, I do not observe any statistically signific ant effects of school accountability on the probability of being underweight. As one would suspect, POLS is not the ideal model under the circumstances where there may be omitted state characteristics that also affect BMI and the prevalence of obesity, or when the strict exogeneity assumption is violated. Therefore, columns (2) and (3) in Table 2-6 to 2-8 present the fixed effects model of the relationship between student wei ght status and school accountability from equation (4) without or with state covariates, respectively. All the coefficients of interest across model specification are pos itive and statistically signific ant at the 1% level except for the effects on the probabilit y of being underweight. The si zes of the coefficients, though estimated very precisely, also incr ease compared to POLS. The inclusion of state covariates as showed in Model (3) address the potential problem that unobservable conditions of state economic status might give rise to the increase of BMI. The coefficients decrease slightly but are not statistically different from Model (2) which assures me that the effect s of SA are not driven by unobserved state time-varying effects. Evaluated at sample means, an addi tional year of exposure to the school accountability system or an additional test is associated with a 0.06 kg/m2 increase in the BMI from 22.93 to 22.99 kg/m2 and a 0.5 percentage poi nt increase in the percentage overweight from 14.2 to 14.7%. The impact of accountability on the 40

PAGE 41

probability of being under weight is only statistically significant when using LENGTH_general and the probability of a student is underweight decreases by 0.2% when she/he exposed to the system for an additiona l year implying that if SA is related with the current eating disorder phenomenon, I do not observe it here. Although these coefficient estimates pr esent considerable and statistically significant effects of SA on adolescent obesity, one might suspect they are still somewhat overestimated due to the potential endogeneity problem elaborated earlier. So in column (6) in Tables 2-6 to 2-8, I show results from the fixed-trend specification adopted here to deal with the endogene ity problem due to omitted state BMI growth. As expected, the standard errors in all regressi ons increase to a great extent while the magnitudes for FREQUENCY meas ure did not change significant ly from model (3). To further test the possibility of endogeneity bias associated wit h LENGTH measure, I next present results estimated using equation (7)-(9) in column (4) and (5) in Tables 2-6 to 28. I first use data from BRFSS for adults aged between 18 and 24 to estimate BMI trend for each state which is then incorporated in equation (7). Similarly, I again estimate BMI trend using BRFSS for adults aged between 25 and 35 and then include them into equation (8). The magnitude for the estimated effects of LENGTH and FREQUENCY only drop slightly relative to the estimates in fixed effects model implying that the endogeneity bias due to omitted state BMI tr end does not exert practically important impact on my main results. The standard erro rs change little, and so do the estimates for the LENGTH and FREQUENCY measures when estimating the probability of being overweight. It seems reasonable at this point, to concl ude that students under NCLB tend to have higher BMI level and are more lik ely to be overweight, increasing BMI by 41

PAGE 42

0.6 kg/m2 and overweight rate by 0.5%. However, I do not observe any statistically significant effects of school accountability on students probability of being underweight. As discussed in the previous sections, early implementers may experience bigger effects on BMI and probability of being overweight compared to late implementers over time due to the potential non-line ar growing effects over ti me. Apart from that, school accountability implemented before 1999 is si gnificantly different from those adopted after 1999 and 2001 (NCLB) in ei ther measurement of school performances based on student test scores, or rewards and sanction s for high-performing and low-performing schools. Furthermore, there might be a lag between school accountability implementation and the full response on st udent BMI and the probabi lity of becoming overweight due to physical adjus tment time. Using a set of dummy variables, I test the hypothesis stated above. Logically, for the linearity assumption to hold a monotonic relationship with invariant growth rate rather than an increasi ng growth rate is expected, i.e. the incremental across LENGTH/FREQU ENCY categories should be relatively stable. The first specification using fixed tr end effects, which includes five levels of dummy variables representing the LENGTH of school accountability exposure, is reported in Table 2-9 columns (1) and (2). Re lative to the omitted category, states without accountability laws, t he impact of the system on st udent BMI is statistically significant and positive, and the longer the l ength, the larger the coefficients. On average, there is around a 0.1 kg/m2 increase in BMI and a 1% increase in the probability of being overweight for every two-year increase in the length of school accountability, suggesting a linear growth ra te of BMI and overweight rate under SA I then test FREQUENCY measures using the same model as displayed in columns 3 and 42

PAGE 43

4. Coefficients across categories increase slight ly but the growth rate of BMI is smaller than measured by LENGTH and present hig her effects when a st udent tested less than 4 times under accountability system. With the increase in the length/frequency of exposure to the systems, I observe a dec reasing marginal effect. However, the difference among growth rates across categories is not statistically significant. Similarly as measured by LENGTH, there is a 1% increas e in the probability of being overweight for every two times increase in the frequency of school accountability Results presented in Table 2-9 indicate that SA has slightly stronger short run effects on adolescent obesity and the marginal effect decreases gradually. Overall, however, linearity assumption holds. To further investigate the differential impact of accountability across gender, ethnicity, and grade, I ran st ate fixed effects, fixedtrend effects and state trends regressions using various subsamples. Results from different specifications do not vary significantly, therefore I only present estima tes obtained from fixed effects models and fixed effects with state*year trends (cal culated using adult aged younger than 24) models. Table 2-10 presents estimates using a subsample of female and male students respectively. Consistent with results us ing the whole sample, FREQUENCY measure has slightly larger effects on both BMI and the probability of being overweight for female. It is also important to point out that, the magnitude of the estimates on female BMI are slightly higher, while the probability of being overwei ght due to SA is higher for male students when measured by LENGTH. I also observe differential impacts of SA on student BMI across racial groups as show n in Table 2-11. Note that school accountability has the strongest effect on BMI for Asian students, followed by White 43

PAGE 44

students and African American st udents. Hispanic students BMI do not respond to the SA policy, though the probability of being overweight is highest for these students compared to all other racial groups under accountability systems. Last, using a subsample of students from gr ades 9 to 12, I adopted the same regression analyses to test if accountability was only bound to the af fected grade. Since school accountability is mostly implemented in grades 10 and 11 in public high schools15 and many states integrate high school exit exam into thei r school accountability systems, therefore students in these grades are more likely to be affected. In addition, since SA presents substantial lag effects on student BMI, I expect to see higher effects on 11th and 12th graders rather on 9th and 10th graders. Consistent with my expectation, as Table 2-11 shows, BMI is significantly higher for student s from grade 11 to 12, yet the probabilities of being overweight are only significantly higher for 12th graders. Do school accountability systems re duce adolescent activity levels? With the increased strength of accountabilit y, it might be the case that students substitute their physical activity time with studying time to avoid poor test outcomes. Hence, I tested the effects of the SA16 on the number of days students participated in physical activities using fixed effects model s. In Table 2-12 each panel represents a different version of the dependent variable: t he number of PE classes in the last seven days, or number of days partici pated in exercise that is long er than twenty minutes that made the respondent sweat in the last seven days (vigorous exercise), or number of days participated in exercise longer than thirty minutes that did not make the 15 43 states do not implement school accountability systems in 9th grade. 16 Both LENGTH and FREQUENCY present similar effects on students physical activities, we only show results using LENGTH measure here. 44

PAGE 45

respondent sweat in the last seven days (moderate exercise). With fixed-trend effects in place, the variable of interest is only negativ e and statistically significant for the female students participation in PE classes, which suggests that for each additional year of school accountability the nu mber of days participated in PE classes decreased 0.032 day for female students. As can be seen in Table 2-12, students who are under the system one year longer are slightly less physically active in both vigorous and moderate exercise, though none of these differences are statistically significant. The findings are important for policy makers since school a ccountability policies are designed to improve academic performance without harming student s physical health. Nevertheless, the results presented above suggest that with th e implementation of accountability across states, students spend less time on physical exercise. Especially, female students cut back on PE class time. Falsification tests Testing school accountability system effect s using adult BMI (ages 18 to 30) I have argued that accountability systems resu lt in higher student BMI, to further support this hypothesis I use adult BMI as dependent variable in a similar DID regression framework. One im portant distinction between adults and students is that adults are not directly affected by the school systems. Thus, the specifications with adult BMI as dependent variable should not produce st atistically significant coefficients on accountability variables. With these hypothes es in mind, I again use adult (ages 18-30) BMI data obtained from BRFSS to test the effects of SA on adults. Although more parameterized models are pref erred in the SA and adolescent obesity estimation, using the same LENGTH measure on adult obesit y seems unreasonable. Especially, I will not be able to calculate the exact number of years an adult has been treated and hence 45

PAGE 46

make the comparison inaccurate. Instead, em ploying the same DID strategy discussed above I use data in years 1999 and 2003 and classify states into treatment and control groups based on the implementation date of school account ability. For the control variables, I include same individual and state covariates as in the YRBSS regressions. Table 2-13 contains the results of the esti mated effect of school accountability on adult BMI. Different columns represent different regressions rest ricted by age. In terms of BMI, school accountability systems fail to show st atistically significant effect on adults BMI across all age groups. In other wo rds, adult BMI, ages between 18 and 3017, are not significantly affected by the implementat ion of school accountability which further supports the original hypothesis. The Effects of School Accountability System on Adolescent Height I examine next whether increases in the implementation of school accountability systems have similar effects on adolescents heights. To gauge the credibility of these conclusions, this placebo test uses an outco me that cannot possibly be affected by the accountability systems. Using the preferred fi xed-trend model, I perfor m this falsification test with the whole sample and with a series of subsamples. As showed in Table 2-14, I do not observe any statistically significant effects of school accountability systems on adolescents height across models, implying that my conclusion of the effects of school accountability on student weight status is legitimate. Specification test: identifying the effects of student accountability High-stake testing system s were firstly implemented at the student level after unfavorable international comparisons of mathematics and science achievement among 17 We also run several regressions with various age cutoffs, for instance 18-24 and 25-34. Results are not sensitive to how we group adults by age (not shown here). 46

PAGE 47

U.S. students with that of st udents in other industrialized nations. Since then, educators and policy makers have been working to bring value to the high school diploma by raising the rigor of high school standards, a ssessments, and curriculum. For example, if students cannot pass the test at a certain gr ade by meeting a pre-calculated national average, they run the risk of being held back a grade level or not being allowed to graduate. Not until the 1990s was this system applied to the school level. Therefore, one might suspect that student accountabilit y systems may also contribute to the increase of student BMI, since ultimately students are directly accountable for their grades and academic performances, and henc e school accountability will be upward biased. In this section, I test the effects of student accountability on student BMI. The student accountability reform mo vement began in 1970s and one of the main parts of the policy change in this movement is state requirement of a high school exit exam. As shown in figure 2-6, before 1990, 10 states required the HS exit exam, and the number increased to 24 by 2005 (National Center on Educational Outcomes surveys; Guy et al 2002). Now more states are planning to impl ement this policy in the near future to promote high school quality. To identify the potential effect of student accountability, th is paper uses the date of introduction of high school exit examinati ons as the main indepe ndent variable. As student accountability systems are bound to students in grade eleventh or twelfth, I might not observe a cumulative effect on stu dent BMI. Therefore, rather than using the same measure as school accountability, I generate a dummy variable coded as one if the state implemented high school exit exam before or in the survey year and zero otherwise. 47

PAGE 48

Using similar analysis strategies as in the previous sections, I start with the POLS and later focus on fixed effects and fixedtrend effects regressions. The results presented in Table 2-16 indicate that even after including student accountability into the models, school accountability still shows significant and positive effects on student BMI and the probability of being overweight consis tently across specifications while the effects are slightly smaller once controlling for state fixed effect s and state fixed-trend effects. The high school exit exam requirem ent also present statistically significant effect on both student BMI and the probability of being overweight indicating that both school and student accountability systems may unintentionally increase the prevalence of adolescent obesity. Result s do not change significantly compared to the estimates obtained from models excluding student account ability variable which provides further support for my hypothesis. Conclusion Existing literature on adolescent obesity blames lack of physical activity and the rise of calorie intake, while the evidence presented in this paper highlights three conclusions that the previous literature has not explored. Firs t, this paper contributes to the sparse literature on adolescent obesity by identifyi ng school accountability system as a strong factor that may contribute to higher adolescent BMI and the probability of being overweight. I construct three differ ent measures to ev aluate state school accountability over the years between 1999 and 2005. According to my estimates, all measures present statistically significant and positive effects on student BMI and the probability of being overweight. Specifically, an additional y ear of exposure to school accountability systems will lead to an increase in BMI by around 0.06 kg/m2 and this effect is bigger among females, and among As ian and White students, as compared to 48

PAGE 49

Hispanic and African-Amer ican students. One more year exposure to school accountability also increases the probabilit y of being overweight by around 0.5% and this effect is slightly bigger among male s and Hispanic students. Se cond, my empirical results show that school accountability systems present significant lag effects on adolescent BMI and the probability of being ov erweight. It appears that students need an adjustment period before the full respons e to changes of school environmental factor. Last, school accountability significant ly decreases the number of times female students participate in PE classe s. Although school accountability fails to show any statistically significant negative effects on students participation in vigorous and moderate exercises, it shows a negative sign on both vigor ous and moderate exercises. The reasons why young people in the nation experience a significant growth in obesity are complicated and studies in the ar ea usually can only answer the question from one of the angles. This paper attempts to shed new light on the possible cause of the adolescent overweight pr oblem from the perspective of the change in education accountability policies. One limitation of studies on causes of adolescent obesity is the low explanatory power. The main reason might be the undesirable dataset which fails to include parents BMI and ot her environmental variables. Therefore, obtaining an updated and richer dataset will enable researchers to get a better look of this currently very urgent health problem in future studies. Of course, administrators have the goals to improve math, reading, and science test scores, but in the meant ime we should not ignore the possibility that high stake testing systems in turn affect the test takers physical health Findings of this study imply that while school and student accountability systems may promote students academic 49

PAGE 50

performance, this could occur at the expense of students ph ysical health. It is time for accountability advocates to consider inco rporating policies addressing solutions to combat adolescents obesity whil e improving academic performance contemporaneously. 50

PAGE 51

Table 2-1. Summary stat istics of all variables Variables Mean (Standard deviation) Definition Body Mass Index 22.927 (4.216) Weight in kilograms divided by height in meters squared Overweight students 0.15 (0.357) Dichotomous variable that equals 1 if body mass index 85th percentile for children of the same gender and age Female 0.493 (0.5) Dichotomous variable that equals 1 if respondent is Female Asian 0.031 (0.175) Dichotomous variable that equals 1 if respondent is Asian African American 0.135 (0.342) Dichotomous variable that equals 1 if respondent is African American Hispanic 0.104 (0.305) Dichotomous variable that equals 1 if respondent is Hispanic White 0.636 (0.481) Dichotomous variable that equals 1 if respondent is White Grade 9 0.287 (0.453) Dichotomous variable that equals 1 if respondent is in Grade 9 Grade 10 0.26 (0.438) Dichotomous variable that equals 1 if respondent is in Grade 10 Grade 11 0.234 (0.424) Dichotomous variable that equals 1 if respondent is in Grade 11 Grade 12 0.218 (0.413) Dichotomous variable that equals 1 if respondent is in Grade 12 # Eat Fruit Last 7 days 2.856 (1.545) Number of times a respondent eats fruit during the past 7 days # Eat Green salad Last 7 days 2.109 (1.192) Number of times a respondent eats green salad during the past 7 days # Eat Potatoes Last 7 days 2.071 (1.076) Number of times a re spondent eats potatoes during the past 7 days # Eat Carrots Last 7 days 1.709 (1.024) Number of times a re spondent eats carrots during the past 7 days # Exercise-no Sweat Last 7 Days 4.719 (2.495) Number of times a respondent exercises or participates in physical activity for at least 30 minutes that did not made her/him sweat and breathe hard # Exercise Sweat Last 7 Days 3.653 (2.512) Number of times a respondent exercises or participates in physical activity for at least 20 minutes that made her/him sweat and breathe hard # Days have PE Class 3.21 (2.204) Number of times a respondent exercises or participates in physical education class 51

PAGE 52

52 Table 2-1. Continued # Hours Watching TV per day 3.02 (1.74) Number of hours a respondent watches TV on an average school day State Level Variables LENGTH_grade 4.519 (3.064) Number of years that the respondent has been exposed to school accountability system(based on testing grades) LENGTH_general 5.285 (3.359) Number of years that the respondent has been exposed to school accountability system FREQUENCY_grade 2.984 (2.366) Number of times that the respondent has been tested to school accountability system Length of High Sch. Exit Exam 8.05 (9.404) Number of years that High school exit exam has been implemented in respondents state of residence %Adult with Bachelor Degree 25.679 (4.185) Percentage of adults with bachelor degree in respondents state of residence % Individual below poverty 10.909 (3.027) Percentage of individu als below poverty line in respondents state of residence %Pub.Sch. with AP course 72.9 (0.188) Percentage of public schools provide AP courses in respondents state of residence No. of Obs. 53,499 Notes: *Summary statistics and standard dev iation (in parentheses) from authors calculations from the 1999,2001,2003, 2005 Youth Risk Behavior Surveillance System. Africa American and Hispanics are oversampl ed in the YRBSS, results above are wei ghted means non-weighted means are similar for all variables except for the proportion of blacks and Hispanics

PAGE 53

Table 2-2. Participation m ap and date of introduction of school/student accountability system States YRBSS participat ion map from year 1999 to 2005 Year of accountability system adoption Testing Grades Year of HighSchool Exit Exam adoption 1999 2001 2003 2005 Alabama 1997 3-8/3-8 & 11 1983 Arizona 2000 3-8 & 10 2001 Arkansas 1999 4,6,8/3-8 California 1999 2-8/2-11 2005 Colorado 2002 3-10 Connecticut 1993 4,6,8 Delaware 1998 3-8, 10/3-11 Florida 1999 4,5,8,10/3-10 1977 Georgia 2000 3-8, 11/ 1-8 & 11 1995 Hawaii 2001 3, 10/3-8 & 11 1983 Idaho 2003 3, 10 2000 Illinois 2003 4,8,10/3-8 & 11 Indiana 1995 3,6,8/4,5, 7,9/3-10 2000 Iowa 2003 3-8 & 11 Kansas 1995 3-8 Kentucky 1995 3-8,10-11/3-8 Louisiana 1999 4,8,/4,8,10,11 1991 Maine 1999 4,8,11 Maryland 1999 3-8 1979 Massachusetts 1998 4,8,10/3,4,7, 6/3,5,7,8/38 2003 Michigan 1998 4,5,7,8,11/3-8 & 10-11 Minnesota 1996 3-8 & 10-11 2000 Mississippi 1994 3-8/3-12 1989 Missouri 1997 3-8 & 10-11 Montana 1998 4,8,10 Nevada 1996 10-12/3,5,10 1980 53

PAGE 54

54 Table 2-2. Continued New Jersey 2003 3-8 & 11 1994 New Mexico 2003 3-9 & 11/3-8 & 11 1991 New York 1998 3-8 & 10/3-8 & 10-12 1979 North Carolina 1993 3-8 & 10 1981 Ohio 2002 3-8 & 10 1994 Oklahoma 1996 3, 10 Oregon 2000 3-8 & 10 Pennsylvania 2002 5,8,11/3-8 & 11 Rhode Island 1997 4,8,10/3-8, 11 South Carolina 1999 3-8/3-8,10 1990 South Dakota 2003 3-8 & 11 Tennessee 1996 3-8 & 11 1983 Texas 1993 3-8 & 10 1987 Utah 2003 3-8 & 10/10-12 Vermont 1999 3-8/3-8,10 Virginia 1998 3,5,8/3-8 2004 Washington 1998 4,7,10 West Virginia 1997 3-8 & 10 Wisconsin 1993 4,8,10 Notes:* The participation map is obtained from CDC YRBSS. The introduction dat es of school accountability systems are obtained from the Consortium for Policy Research in Education (CPRE) and testing grades are collected from the state depar tment of education website. To co mply with the NCLB 2005-06 deadline, many states introduced new test s or tests in additional grades during the past few school years. I therefore collected data about major changes made in the stat es testing grades since 2002 from State Achievement Profiles conducted by C enter on Education Policy. The in troduction date of high school ex it exam updated by author based on National Center On Educat ional Outcomes (NCEO) surveys.

PAGE 55

Table 2-3. Summary statistics fo r difference-in-differences analysis State School Accountability Policy Non-accountability states Accountability states Variables Before Policy After Policy Before Policy After Policy BMI 22.565 (5.675) 22.392 (4.763) 22.839 (7.844) 23.043 (4.774) Female 0.34 (0.474) 0.495 (0.500) 0.499 (0.500) 0.492 (0.5) Asian 0.054 (0.226) 0.015 (0.12) 0.046 (0.210) 0.027 (0.162) African American 0.035 (0.184) 0.086 (0.280) 0.366 (0.482) 0.278 (0.448) Hispanic 0.298 (0.458) 0.255 (0.437) 0.102 (0.303) 0.179 (0.383) White 0.576 (0.495) 0.591 (0.492) 0.405 (0.491) 0.440 (0.496) Grade 9 0.251 (0.434) 0.263 (0.441) 0.206 (0.404) 0.247 (0.431) Grade 10 0.221 (0.415) 0.334 (0.472) 0.189 (0.392) 0.250 (0.433) Grade 11 0.280 (0.45) 0.066 (0.248) 0.244 (0.430) 0.251 (0.434) Grade 12 0.248 (0.433) 0.335 (0.472) 0.360 (0.480) 0.253 (0.435) No. of Obs. 435 821 4,273 6,330 Notes: Summary statistics and standard deviations (in pa rentheses) from authors calc ulations from the 1999,2003 Youth Risk Behavior Surveillance System. See text fo r definitions of accountability and nonaccountability states an d before and after policy. 55

PAGE 56

Table 2-4. Weighted average students' BMI in states with school accountability versus states without school accountability State School Accountability Policy Time differences Students' BMI Before Policy After Policy for locations Non-accountability states 22.565 (0.272) 22.392 (0.166) -0.173 (0. 319) [435] [821] Accountability states 22.839 (0.121) 23.043 (0.070) 0.204 (0.140) [4273] [6330] Location difference at a point in time: 0.274 0.651 (0.298) (0.180) Difference-in-Differences 0.377 (0.348) Notes: Summary statistics, robust st andard errors are clus tered on state le vel (in parentheses) and number of observations [in squre br ackets] from authors calcul ations from the 1999 and 2003 Youth Risk Behavior Surveillance System. See te xt for definitions of account ability and non-accountability st ates and before and after policy. 56

PAGE 57

Table 2-5. Effects of school accountabi lity on students' BMI: Results from DID controlling for individual characteristics Variables Whole sample Female Male Year 2003 -0.690 (0.403) -1.117 (0.554)* -0.361 (0.429) Treated States 0.50 (0.255) -0.161 (0.272) 0.224 (0.376) Accountability Treatment Effects 0.794^ (0.341) 1.171^ (0.553) 0.482 (0.396) R 2 0.05 0.05 0.038 Sample size 11,381 5,587 5,794 Notes: The table lists regression coefficients and standard errors in parentheses. Standard errors are cluster-corrected by state. Statisti cal significance level: ***: p< .001; **:p<.01; *:p<.05; ^: p<.1 Addi tional controls include student's gender, grade, age square, ra ce and variables related to students' eating habit: number of time s eating fruit/ green salad/potat oes/carrots in the last 7d ays; the percentage of adults with bachelor degree in each stat e; percentage of individuals below poverty in each state and the percentage of public schools with AP course in each state. 57

PAGE 58

Table 2-6. Estimated effects of school accountability on student BMI Panel A Number of years exposed to school accountability (LENGTH) Specification (1) (2) (3) (4) (5) (6) LENGTH_grade 0.045* (0.020) 0.075*** (0.015) 0.062** (0.021) 0.058** (0.022) 0.056* (0.022) LENGTH_general 0.042* (0.018) 0.085*** (0.014) 0.093*** (0.026) 0.094** (0.023) 0.094** (0.024) State fixed effects No Year fixed effects No No No No No State-Specific Trends (State by year fixed effects) No No No No No State Covariates No No State Year Trends (using adults BMI age<=24) No No No No No State-Year Trends (using adults BMI 24
PAGE 59

Table 2-7. Estimated effects of school accountability on the probability of being overweight Panel A Number of years exposed to school accountability (LENGTH) Specification (1) (2) (3) (4) (5) (6) LENGTH_grade 0.003^ (0.001) 0.005*** (0.001) 0.005** (0.002) 0.005** (0.002) 0.005** (0.002) LENGTH_general 0.003* (0.001) 0.005*** (0.001) 0.004^ (0.002) 0.006* (0.003) 0.006** (0.002) State fixed effects No Year fixed effects No No No No No State-Specific Trends (State by year fixed effects) No No No No No State Covariates No No State Year Trends (using adults BMI age<=24) No No No No No State-Year Trends (using adults BMI 24
PAGE 60

Table 2-8. Estimated effects of school accountability on the probability of being underweight Panel A Number of years exposed to school accountability (LENGTH) Specification (1) (2) (3) (4) (5) (6) LENGTH_grade -0.0002 (0.0005) -0.0009 (0.0006) -0.0002 (0.001) 0.00004 (0.0009) 0.00006 (0.0009) LENGTH_general -0.0004 (0.0004) -0.002** (0.0006) -0.002 (0.001) -0.002 (0.001) -0.002 (0.001) State fixed effects No Year fixed effects No No No No No State-Specific Trends (State by year fixed effects) No No No No No State Covariates No No State Year Trends (using adults BMI age<=24) No No No No No State-Year Trends (using adults BMI 24
PAGE 61

61 Table 2-9. Estimated effects of SA using fixed-trend model with dummy variables Model LENGTH FREQUENCY Dependent Variables BMI Pr(Overweight) BMI Pr(Overweight) 1-2 years/times 0.238* (0.115) 0.025** (0.0093) 0.296** (0.108) 0.027** (0.009) 3-4 years/times 0.459** (0.136) 0.038** (0.014) 0.436** (0.132) 0.039** (0.012) 5-6 years/times 0.303^ (0.161) 0.025^ (0.013) 0.495** (0.177) 0.044*** (0.013) 7-8 years/times 0.541** (0.170) 0.043** (0.015) 0.477* (0.213) 0.051** (0.016) 9-10 years/times 0.624** (0.237) 0.065** (0.023) R 2 0.051 0.026 0.05 0.026 Sample size 50,148 50,146 50,176 50,146 Notes: Each column of the table pr esents coefficients and standard errors (cluster-corrected by state) in parentheses from a different regression. Statistica l significance level: ***: p<.001; **:p<.01; *:p<.05; ^: p<.1 All model s include individual characterist ics, state covariates, state fixed effects and state by year trends (calculated using adults BMI data age<=24).

PAGE 62

Table 2-10. Estimated effects of school accountability by gender Main Indep. Var. LENGTH FREQUENCY Female Male Female Male Dep. Var. BMI Pr(Overweight) BMI Pr(Overweight) BMI Pr(Overweight) BMI Pr(Overweight) (1) 0.074* (0.028) 0.004* (0.002) 0.048* (0.025) 0.005* (0.002) 0.090* (0.040) 0.004 (0.003) 0.040 (0.032) 0.005^ (0.003) (2) 0.073* (0.028) 0.004* (0.002) 0.042 (0.025) 0.005* (0.002) 0.090* (0.039) 0.005 (0.003) 0.027 (0.033) 0.005 (0.004) Sample size 25672 25672 24476 24476 25671 25671 24476 24475 Notes: Each column of the table presents coefficients and standard errors (cluster-corrected by state) in parentheses from a different regression. Statisti cal significance level: ***: p<.001; **:p<.01; *:p<.05; ^: p<.1 Model (1) includes individual characteristics, st ate covariates and state fixed effects. Model (2) includes individual characteristics, state covariates, state fixed e ffects and state by year trends (calculated using adu lts BMI data age<=24). Additi onal controls include student's gender, grade, age square, race and variables related to students' eating habit: number of times eating fruit/green salad/potat oes/carrots in the last 7days; the percentage of adults with bachelor degree in each state; percentage of indi viduals below poverty in each state and the pe rcentage of public schools with AP course in each state. Separ ate regressions are run for eac h sex group to allow for het erogeneous treatment effects. 62

PAGE 63

Table 2-11. Estimated effects of school accountability by racial group/grade Panel A LENGTH African American Hispanic White Asian Model BMI Pr(Overweight) BMI Pr(Overweight) BMI Pr(Overweight) BMI Pr(Overweight) (1) 0.058 (0.046) 0.001 (0.004) 0.039 (0.037) 0.007* (0.004) 0.070* (0.026) 0.005* (0.002) 0.168*** (0.043) 0.008* (0.005) (2) 0.049 (0.048) 0.003 (0.004) 0.039 (0.028) 0.007* (0.003) 0.058* (0.024) 0.005* (0.002) 0.168** (0.045) 0.009^ (0.005) Sample size 11803 11786 10812 10774 21629 21629 1497 1379 9th Grade 10th Grade 11th Grade 12th Grade Dependent Variables BMI Pr(Overweight) BMI Pr(Overweight) BMI Pr(Overweight) BMI Pr(Overweight) (1) 0.045 (0.049) 0.003 (0.005) 0.055* (0.027) 0.003 (0.002) 0.072* (0.034) 0.003 (0.003) 0.105*** (0.027) 0.009*** (0.002) (2) 0.026 (0.061) 0.003 (0.005) 0.051^ (0.027) 0.002 (0.002) 0.067* (0.035) 0.003 (0.003) 0.112*** (0.024) 0.011*** (0.002) Sample size 11,910 11,907 12,480 12,472 12,853 12,841 12,903 12,903 Panel B FREQUENCY African American Hispanic White Asian Model BMI Pr(Overweight) BMI Pr(Overweight) BMI Pr(Overweight) BMI Pr(Overweight) (1) 0.119^ (0.070) 0.005 (0.005) 0.019 (0.023) 0.008** (0.002) 0.066* (0.032) 0.003 (0.003) 0.211 (0.089)* 0.009 (0.007) (2) 0.118^ (0.065) 0.009 (0.006) 0.005 (0.032) 0.007* (0.003) 0.046 (0.033) 0.002 (0.003) 0.212* (0.086) 0.010 (0.007) Sample size 11803 11786 10812 10774 21629 21629 1497 1379 9th Grade 10th Grade 11th Grade 12th Grade Dependent Variables BMI Pr(Overweight) BMI Pr(Overweight) BMI Pr(Overweight) BMI Pr(Overweight) (1) 0.039 (0.053) 0.003 (0.005) 0.052^ (0.029) 0.001 (0.003) 0.101 (0.067) 0.006 (0.006) 0.212*** (0.047) 0.014** (0.004) 63

PAGE 64

64 Table 2-11. Continued (2) 0.017 (0.069) 0.002 (0.005) 0.045 (0.028) 0.001 (0.003) 0.095 (0.074) 0.007 (0.006) 0.261*** (0.051) 0.019*** (0.004) Sample size 11,910 11,907 12,480 12,472 12,853 12,841 12,903 12,903 Notes: Each column of the table present s coefficients and standard erro rs (cluster-corrected by st ate) in parentheses from a different regression. Statisti cal significance level: ***: p<. 001; **:p<.01; *:p<.05; ^:p<.1 Model (1) in cludes individual characteristics, state covariates and state fixed effects. Model (2) includes individual characteristics, state covariates, state fixed e ffects and state by year trends (calculated using adu lts BMI data age<=24). Addition al controls include student's gender, grade, age square, race and variables related to students' eating habit: number of times eating fruit/green salad/potatoes/carrots in t he last 7days; the percentage of adults wit h bachelor degree in each state; percent age of individuals below poverty in eac h state and the percentage of public schools with AP course in each state. Separat e regressions are run for each et hnicity group/grade to allow for het erogeneous treatment effects.

PAGE 65

Table 2-12. Estimated effect of school a ccountability on students' physical activities Variables PE Classes PE Classes (Female only) PE Classes (Male only) School Accountability -0.012 (0.01) -0.032* (0.013) 0.008 (0.015) R 2 0.18 0.223 0.14 Sample size 46,409 23,972 22,437 Variables Vigorous Exercise Vigorous Exercise (Female Only) Vigorous Exercise (Male Only) School Accountability -0.009 (0.012) -0.013 (0.016) -0.002 (0.017) R 2 0.134 0.125 0.08 Sample size 50,056 25,635 24,421 Variables Moderate Exercise Moderate Exe. (Female Only) Moderate Exe. (Male Only) School Accountability -0.003 (0.013) -0.014 (0.017) 0.010 (0.019) R 2 0.06 0.066 0.058 Sample size 50,054 25,633 24,745 Notes: The table lists FE regressi on coefficients and robust standard errors in parentheses. Statisti cal significance level: ***: p< .001; **:p<.01; *:p<.05; ^: p<.1 Vigorous Exercise: exercise longer than 20 minutes that make you sweat Moder ate Exercise: exercise longe r than 30 minutes that did not make you sweat PE class: school PE classes Additional controls include student's gender, grade, age sq uare, race and variables relat ed to students' eat ing habit: number of times eating fruit/g reen salad/potatoes/carrots in the la st 7days; the percentage of adults with bachelor degree in each state; percentage of individuals below poverty in each stat e and the percentage of public schools with AP course in each state. 65

PAGE 66

Table 2-13. Estimated effects of school a ccountability on adults' BM I: results from DID controlling for individual characteristics Variables Age 18-19 Age 20-24 Age 25-30 Year 2003 0.437 (0.161) 0.487 (0.132) 0.361 (0.081) Treated States 0.116 (0.237) -0.176 (0.157) -0.186 (0.204) Accountability Treatment Effects -0.119 (0.221) 0.17 (0.163) 0.159 (0.16) R 2 0.046 0.054 0.053 Sample size 6,377 12,304 15,244 Notes: The table lists regressi on coefficients and standard errors in parentheses. Standard errors are cluster-corrected by state. St atistical significance level: ***: p<.001; **:p<.01; p<.05 Table 2-14. Falsification test: If student accountability systems affect student height Model Whole sample Female Male Dependent Variables Height Height Height LENGTH_grade -0.00002 (0.0006) 0.0001 (0.0008) -0.0003 (0.0009) FREQUENCY_grade 0.0002 (0.0007) 0.0001 (0.0009) -0.0003 (0.001) Notes: Each column of the tabl e presents coefficients and standard errors (cluster-corrected by state) in parentheses from a different regression. Addi tional controls are listed in the notes to Table 5. Statis tical significance level: *** : p<.001; **:p<.01; *:p<.05 Models in cludes individual characterist ics, state covariates, state fixed effects and state by year trends 66

PAGE 67

Table 2-15. Falsification test: If student a ccountability contributes to student overweight Model (1) (2) (3) Dependent Variables BMI Pr(Overw eight) BMI Pr(Overwei ght) BMI Pr(Overwei ght) School Accountability 0.049*** (0.007) 0.004* (0.002) 0.041** (0.015) 0.003* (0.001) 0.041** (0.015) 0.003* (0.001) Student Accountability 0.019 (0.025) 0.009 (0.011) 0.068*** (0.011) 0.003* (0.001) 0.071* (0.03) 0.003* (0.001) R 2 0.051 0.02 0.051 0.03 0.05 0.026 Sample size 50,717 50,715 50,148 50,146 50,148 50,146 Notes: Each column of the table present s coefficients and standard errors (clustercorrec ted by state) in parentheses from a different regression. Addi tional controls are listed in the notes to Table 5. Stat istical significance level: ***: p<.001; **:p<.01; *:p<.05 Model (1) includes individual charac teristics and state covariates. Model (2) includes individual characteristics, state covariates and state fixed effects. Model (3) includes individual characteristics, state covariates, state fixed effects and state by year trends (calculated using adults BMI data age<=24). 67

PAGE 68

Figure 2-1 States with school accountability systems and rate of overweight, 1993-2005 Figure 2-2 States with differ ent testing grades when initia ted the accountability systems 68

PAGE 69

22.9 23 23.1 23.2 23.3 23.4 23.5 23.6 1999200120032005Weighted Aerage BMIWeighted Average Student BMI Balanced panel Unbalanced panel 0.15 0.16 0.17 0.18 0.19 0.2 1999200120032005Weighted Percentage Overweight StudentsWeighted Percentage Overweight Students Balanced panel Unbalanced panel 22.813 22.657 23.022 23.211 -0.156 0.365 0.19 -.2 0 .2 Change of BMI 22.5 23 23.5 Weighted Average Students BMI 1999 2001 2003 2005 YearWeighted Average Student BMI 0.143 0.133 0.154 0.171 1.1% 2.1% 1.7% -.02 0 .02 Change of %overweight students .1 .12 .14 .16 .18 .2 Percentage of Overweight Students 1999 2001 2003 2005 YearWeighted Percentage of Overweight Students 22.3 22.6 22.9 23.2 23.5 23.8 1999 2001 2003 2005Weighted Average Student BMI by Gender Balanced Panel:Female Unbalanced Panel:Female Balanced Panel:Male Unbalanced Panel:Male 0.14 0.15 0.16 0.17 0.18 0.19 0.2 0.21 1999 2001 2003 2005Weighted Average Percentage Overweight Students by Gender Balanced Panel:Female Unbalanced Panel:Female Balanced Panel:Male Unbalanced Panel:Male Figure 2-3. Trends in Body Mass I ndex and Percentage Overweight, 1999 69

PAGE 70

22.363 22.048 22.572 22.871 23.239 23.297 23.444 23.541 Male Female 0.875 1.249 0.872 0.67 .5 0 .5 1 1.5 Difference=Male BMI-Female BMI 22 22.5 23 23.5 24 Weighted Average Students BMI 1999 2001 2003 2005 YearWeighted Average Student BMI by Gender .146 .124 .152 .178 .14 .142 .156 .164 Female Male -0.6% 1.8% 0.4% -1.4% -.02 0 .02 Difference=Male overweight% Female overweight% .12 .14 .16 .18 .2 % of Students who are overweight by gender 1999 2001 2003 2005 YearWeighted Percentage of Overweight Students by Gender 21.3 21.8 22.3 22.8 23.3 23.8 24.3 1999 2001 2003 2005Weighted Average Student BMI by Racial Groups White Hispanic Asian African American 0.05 0.09 0.13 0.17 0.21 0.25 1999 2001 2003 2005Weighted Average Percentage Overweight Students by Racial Groups White Hispanic Asian African American 22.2 22.6 23 23.4 23.8 24.2 1999 2001 2003 2005Weighted Average Student BMI by Grade Grade 9 Grade 10 Grade 11 Grade 12 0.14 0.15 0.16 0.17 0.18 0.19 0.2 1999 2001 2003 2005Weighted Average Percentage Overweight Students by Grade Grade 9 Grade 10 Grade 11 Grade 12 Figure 2-3. Continued 70

PAGE 71

MO 99 MO 05 TN 99 TN 05 0 .05 .1 .15 Density for BMI 10 15 20 25 30 35 40 BMI Missouri 99 Missouri 05 Tennessee 99 Tennessee 05Years 1999 and 2005BMI Density for States Missouri and Tennessee Note: Average BMIMO 1999= 21.96 Average BMITN 1999= 23.74 Average BMIMO 2005= 22.51 Average BMITN 2005= 24.15 Figure 2-4 BMI density for states Missouri and Tennessee, 1999 vs. 2005 71

PAGE 72

Figure 2-5 YRBSS Part icipation Map 1999-2005 Figure 2-6. States with the Date of Introduction of Scho ol Accountability before 1999, Between 1999 and 2001 and After 2001 Shown in Different Legends LEGEND 1 Time 2 Times 3 Times 4 Times YRBSS Participation LEGEND 1 Time 2 Times 3 Times 4 Times WY ND NE NH RIMT CO MN IA A R KY A K CT DEWA ME O R ID NV UT NM SD KS O K LA MS VT WV MD IN TN AL SC NC VA PA MA NJ CA AZ TX MO WI NY MI OH IL GA FL 72

PAGE 73

73 Figure 2-7 States with the Da te of Introduction of High School Exit Exam Before 1999, Between 1999 and 2001 and After 2001 S hown in Different Legends.

PAGE 74

CHAPTER 3 HIGHER STANDARDS MAY NOT ALWAYS BE BETTER: WILL HIGH EVALUATION STANDARDS CONTRIBUTE TO CHILDHOOD OBESITY Introduction High performance standards have been advoc ated by school administrators, teachers, parents, and even by students themselves in education for centuries. While economists have studied the theoretical (Becker and Rosen 1990; Betts 1998; Costrell 1994) and empirical (Figlio and Lucas 2004; Lillard and DeCicca 2001; Betts and Grogger 2000; Betts 1995) determination a nd impacts of educational standards on students academic achievement, little is known about their unintended effects on student health outcomes. This lack of evidence is surprising considering the more than 20-year-long history of st andard-based education reform, in cluding the most recent reform plan of President O bama, uniformly raising academ ic standards in all states. The existing literature on education standards presents mixed effects on student academic performances both theoretically an d empirically. One stream of studies focuses on evaluating the impacts of differential institutional education standards on students labor market outcome s. For instance, Bagues, Labi ni, and Zinovyeva (2008) find a significant negative correlation bet ween departments average grades and the labor market outcomes of their graduates usi ng data from Italy, i. e. graduating from a high grading department leads to a higher unemployment probability and lower wages. Lillard and DeCicca examined the effects of graduation standards and find a positive impact on drop-out rates. Another stream of studies instead is inte rested in grading standards at the level of schools or teachers. Figlio and Lucas eval uate the effects of teacher-level grading standards on student achievement and find that higher teacher grading standards tend 74

PAGE 75

to have large, positive impacts on student te st score gains in mathematics and reading. Like Betts and Grogger (2000), they also find differential effects of grading standards, depending on student type. Highachieving students tend to benefit most from high standards in reading, while low-achieving students benefit from high standards only when their classmates are high-achievers. Although the issue of educational st andards has been widely discussed in the economics literature both from th eoretical and empirical perspec tives, there is so far no studies investigating the relationship bet ween teachers grading standards and students health outcomes. Understanding and measuri ng the unintended effects of educational standards on childrens health are important because teachers, schools and states under standard based policies are only held a ccountable for test scores, but not for other student outcomes, such as childrens health, schools facing pressure to improve students academic performance may make decisions that may have unintended consequences. They are especially relevant in the current education reform debate because many of the policy decisions, for instance raising academic standards, adopting a national common stan dard and linking teacher pay to student achievement are all predicated upon the assumption of large positive effects of standard-based education policies. However, two rec ent studies by Anderson, Butcher and Schanzenbach (2009) and Yin (2009) find st rong evidence that school accountability system, especially the No Child Left Behind Act, present significant negative effects on childrens health, especially obesity. Anderson, Butcher and Sc hanzenbach (2009) use school level data from Arkansas and find t hat schools that were below the average yearly progress (AYP) threshold under NCLB in year t-1 have a sma ll, but statistically 75

PAGE 76

significantly higher rate of overweight and obesity in year t. Yin (2009) examine the effects of school accountability movement on adolescent obesity, specifically age between 15 and 18, and find that school accountability policies lead to a higher obesity probability and higher BMI. This paper, therefore, is the first to fill the gap in the ex isting literature by investigating how teacher-level grading st andards may affect childrens health, specifically childhood obesity. Between the 1970s and 2000s, the fraction of overweight children increased dramatically from 5 percent to 25 percent. Previous studies in childhood obesity literature focuses on genetic (Perusse et al. 2001; Katzmarzyk et al.1999) and behavioral factors (Stunkard et al. 1990, Dietz 1994) which did not provide much policy implications. Children spend on aver age 6 to 8 hours at school, it is critical to understand how the school environment may contribute to obesity. Cawley, Meyerhoefer, and Newhouse (2006) explore t he impact of state Physical Education (PE) requirements on yout h physical activity and obesity. Although the state requirements do increase students PE time, t hey find no evidence that PE lowers BMI or reduces the probability t hat a student is overweight. Schanzenbach (forthcoming) finds that children who consume school lunches are more likely to be obese than those who brown bag their lunches. While the aforementioned papers provide important evidence of how school environment may contribute to childhood obes ity, there are remaining unanswered questions in the childhood obesity literature. First, the existing literature does not examine how teacher-level factors may affe ct childrens health, considering students spend the majority of time with their teachers at schools. An extensive empirical 76

PAGE 77

literature has documented a positive link between teacher expectations and student achievement (Gauthier 1982; Proctor 1984). As grading standards reflect teachers expectations to the students which in turn may change students future performance as extensively described elsewhere (Jussim, 1989; 1991). Teachers under the impression that higher grading standards may impr ove students academic performance may undertake more stringent standards that hav e unintended consequences on childrens health. High grading standards may cause lo w self-esteem which has been shown in the obesity literature as one of the main causes of childhood obesity. Second, aforementioned studies investigating the impact of school accountability policies on childhood obesity either uses school level data or focuses on adolescent obesity, the current paper instead uses individual level data that tracks children from kindergarten to grade eight. Specifically, we expand on the existing empirical work in educational standards and childhood obesity literatures in two im portant ways. First, we analyze the unintended effects of grading standards at the teacher level on childhood obesity. Second, using a large scale longitudinal dataset allows us to control for the potential endogeneity problem due to simultaneity by dividing direct students assessment test scores and teacher evaluation scores into ten deciles respectively and then categorizing teachers into high, median and low standards based on teachers evaluation scores for students performed equally in direct student assessment tests. Conditional on this extensive set of controls, we find that re lative to low grading standards, students with median or high grading standards teachers tend to have higher BMI by 0.07 kg/m2 and are 12 percentage points more likely to become overweight. We then st ratify the sample 77

PAGE 78

by gender, race/ethnicity and high/low achiev ers to analyze whether higher standards have adverse consequences for minorities. The results show that female, White and low achieving students present stronger respons e to teachers grading standards compared to their counterparts. A straightforward policy implication follo ws from the above results. Teachers, schools and policy makers should be very cautious about raising evaluation standards. If, as shown in this paper, grading standards vary across teachers, rewarding teachers with high grading standards may lead to undesirable consequences. The rest of the paper is organized as follo ws. Section 2 describes the data and the main variables use in the empirical part. Section 3 presents the empirical analysis method. Section 4 summarizes the main re sults and section 5 concludes and discusses policy implications. Data We investigate the potential casual re lation between teacher grading standards and childhood obesity using Early Childhood Longitudinal Study, Kindergarten Class of 1998 (ECLS-K) data, a very detailed nationally representative and individual-based study of childrens early school experienc es beginning with kindergarten through eighth grade, developed under the spons orship of the U.S. Department of Education, National Center for Education Statistics (NCES). A total of 21,260 children throughout the country participated in the base year data co llection in the fall of 1998 and spring of 1999. Five more waves of data were colle cted beyond kindergarten: fall and spring first grade, and spring third, fifth, and eighth grades. All data colle ction was completed in the 78

PAGE 79

spring of 2007 when most of the children were in eighth grade1. The current study used data from the weighted sample at baseline of spring 1999 and 4 other waves collected in springs of 2000, 2002, 2004, and 2007. Data were collected from parents, teachers, and schools to provide important contextual information about the environment for the sampled children which allow ed the author to control for children, family and school characteristics that might contribute to childhood obesity. We exclude observations with missing teacher grading standard information fo r the purpose of the hypothesis testing. A weight is applied to each record to adj ust for non-response and the distribution of students by grade, sex, and race /ethnicity in each state. The design of the ECLS-K emphasizes t he interrelationships between the child and family and the child and school and recognizes the importance of factors that represent the childs health status. Children sampled in the study are directly assessed in each round of data collection and also indirectly assessed by their teacher. The direct assessment that took approxim ately 50-70 minutes in the base year and 90 minutes in the following waves was designed to prov ide data on the developmental status of children in the United States at the start of their formal schooling through eighth grade. In all rounds, one-on-one direct child asse ssments were administered that included cognitive and physical components. The cogn itive assessment score s include measures that can be compared across waves to st udy childrens gains in reading and mathematics. We select Item Response Theor y (IRT) Scale Scores as the indicator for childrens academic achievements mainly because scores derived from the IRT model are based on all of the childs responses to a subject area assessment and can identify 1 Sample also includes students who were either hel d back (e.g., seventh-graders) or promoted ahead an extra year or more (e.g., ninth graders) in 2007, an d this sample frame applied to other survey years. 79

PAGE 80

cross-sectional differences among subgroups2. Teachers were asked to complete a questionnaire for each of the sampled childr en in their classrooms. Specifically, teachers were asked to respond to 39 questions about the childs academic performance. Using this information, an academ ic rating scale (ARS) was developed to measure teachers evaluations of student s academic achievement in reading and mathematics as the indirect cognitive asse ssment. The ARS Scores were rescaled to have a low of one and a high of five to corres pond to the five-point rating scale that teachers used in rating ch ildren on these items. Childrens height and weight were recor ded to measure their physical growth and development. A Shorr Board (f or measuring height) and a di gital bathroom scale were used to obtain the height and weight measurements, which were recorded on a height and weight recording form3. Childrens BMI4 was calculated based on height and weight. From these data, BMI percentiles we re calculated using the 2000 Centers for Disease Control and Prevention growth charts. Obesity wa s defined as a BMI (kg/m2) greater than or equal to the 95th percentile for age and gender, overweight as a BMI greater than or equal to t he 85th percentile but less than the 95th percentile. According to the CDC growth charts in the United States, the 97th perc entile of student BMI, between 4 and 16 years old, is below 28. The 5th percentile of student BMI, ages ages 2 IRT uses the pattern of right, wrong, and omitted res ponses to the items actually administered in a test and the difficulty, discriminating ability, and guess-abi lity of each item to place each child on a continuous ability scale. 3 Childrens height and weight measurements were each taken twice to prevent error and provide an accurate reading. 4 Composite Body Mass Index (BMI) was calculated by multiplying the composite weight in pounds by 703.0696261393 and dividing by the square of the childs composite height in inches. 80

PAGE 81

between 4 and 16 years old, is above 17 (Cent ers for Disease Control and Prevention, 2000). Therefore, observations with BMI>30 as well as BM I<10 are excluded from the data set which may be subject to measurement errors. This reduced the sample size to 66,718. Between 1999 and 2007, BMI increased by 5.09 kg/m2 or by 30% for all students while the average BMI is higher for male students and the percentages of overweight and increased over 10% for all students. According to the sample, approximately 25 percent children were overweight in 1999 and 35 percent in 2007 as shown in Table 3-1. Additionally, to control fo r other individual charac teristics that might have contributed to childhood obesity, we incl ude age, age square, gender, and race into the analysis. In addition, we also include fam ily, teacher and school characteristics to account for other observable demographics. For instance, ECLS-K calculated socioeconomic status (SES) at the household level using data for the set of parents who completed the parent interview in each survey year. The SES variable reflects the socioeconomic status of the household, which has shown to be linked with obesity, at the time of data collection and the components used to create the SES include father/male guardians educati on, mother/female guardians education, father/male guardians occupation5, mother/female guardians o ccupation, and household income6. Thus, each component was converted to a z-score with mean of 0 and a standard 5 Occupation was recoded to reflect the average of the 1989 General Social Survey (GSS) prestige score. This was computed as the average of the corresponding prestige scores for the 1980 Census occupational categories covered by the ECLS-K occupation. 6 For income, the component i is the logarithm of the income for i-th household. The logarithm of income was used because the distri bution of the logarithm of income is less skewed than the direct income values. Income was compared to Ce nsus poverty thresholds, which vary by household size. Households whose income fell below the appropriate threshold were classified as poor. 81

PAGE 82

deviation of one. As described, the SES co mposite is the average of up to five measures, each of which was standardiz ed to have a mean of 0 and a standard deviation of 1, hence the negative values. For analyses, we either include family characteristics or only include the continuous SES measure to avoid multicollinearity. To control for potential unobserved school factors that might affe ct childrens weight status, we also include school type, percentage of st udents eligible for free/reduced lunch, and total school enrollment which was created usi ng the school enrollment variable from the school administrator questionnaire. If this va riable was missing, data for private schools were taken from the Private School Survey (PSS) and data for public schools were taken from the Common Core of Da ta (CCD) public school universe. Table 3-1 depicts descriptive statisti cs for these key variables, including definitions, means, and standard errors. Except where noted, they are based on the sample of 66,718 that emer ges when observations with missing values are deleted. Empirical methods An important concern in determining if teacher grading standards has an effect on childhood obesity is the fact that overweight children are more likely to be low performing students, and in turn more likely to receive lower ratings from their teachers (Datar and Sturm, 2006; Datar, Sturm, and J ennifer 2003). Therefore, if we construct grading standards directly based on teachers ratings collected by ECLS-K, our estimates of grading standards might be biased. Instead, we first compare students academic performance horizontally using di rect student assessment scores and then define students with similar direct assessment scores but receiving a low rating from their teachers as high standard group (treatment group). This measure is different from the definitions used in previous literature (Figlio and Lucas 2004; Betts and Grogger 82

PAGE 83

2000) where the standards are defined by aggr egating teachers grades across years to capture the time-invariant tendencies of hi gh or low grading standards for the same teacher. The first reason why we define our standard differently is that the items and the metric for the eighth-grade teacher ratings ar e different from the Academic Rating Scale (ARS) ratings in earlier rounds of data colle ction, so the scores are not directly comparable to those for kinderga rten, first, third, or fift h grades. Secondly, the average number of students rated by the same teacher across years is 2.5 and the total number of teachers sampled is around 18,176. Theref ore, the average treatment effect of teacher grading standard is less likely to be biased by students non-academic characteristics. Last, in the preliminar y studies testing our standard measures, the standard deviation of common residuals is around 0.8, the standard deviation of individual teacher residual is around 0.5 and the fraction of variance due to common residual ( ) is 0.683. Accordingly, we were able to reject the null hypothesis that all variance is due to within teacher variation. Students in ECLS-K sample are given test s in reading and math in each survey year. Thus, we divide these direct assessment scores into deciles for each year and for reading and math, respectively. Teachers ar e asked to complete a survey which includes information about how they would rate the sampled students for both reading and math. We consider students fall into the same deciles perform equally in reading and math, and generate an indicator that equals to 2 if student received a lower teacher rating (high grading standard), 1 if student received a same teacher rating as their own score (median grading standard) and zero otherwise. 83

PAGE 84

In Table 3-3, we present teachers characte ristics by stratifying the sample into high, median, and low grading standards. One c an see that teacher characteristics do not differ significantly except that teachers with regular certifications are more likely to perform high grading standards versus teacher s with no teaching certifications. Student BMI is highest for high grading standard gr oup and lowest for low standard group with approximately 4 kg/m2 difference and overweight rate is 10% higher for high standard group relative to low standard gr oup as shown in Table 3-4. We begin our empirical work by estimating a simple pooled ordinary least squares (POLS) regression model, including an indicator for teacher grading standard and controls for other individual characte ristics, family backgrounds and school characteristics. Grading standards (GS) are further measured by reading standards and math standards, respectively. (1) istst it ist ist istSchool family XGS BMI ''1 In this study, if teacher grading st andard is exogeneous and uncorrelated with students weight status, then POLS would produce an unbiased estimate. However, as mentioned earlier, simultaneity is the main concern for the current study. Although scholars in education have em phasized that teachers c ontemporaneous perceptions of students performance as well as their expecta tions for students future performance are generally accurate (e.g., see Egan and Archer 1985; Good 1987; Hoge and Butcher 1984; Mitman 1985; Monk 1983; Pedulla, Airasian, and Madaus, 1980), one cannot guarantee that teachers gr ading standards will not be influenced by students appearances or teachers stereotypes will prevent them from forming inaccurate ratings for individual students. 84

PAGE 85

To further circumvent this potential empi rical problem, we utilize panel data from 1998 until 2007, five waves, so that we c an control for unobserved, time-invariant individual fixed effects. Specifica lly, fixed effects model estimates: (2) istt ist it ist ist istyear School family XGS BMI ''1 Equation 2 posits that the BMI, for the ith student in school s during year t is a function of teacher grading standard indicator, measured here by GS (2 if high standard, 1 if median standard, and 0 if low standard), and i represents individual dummies and year dummies incremented in years. tyear As childhood obesity is defined as BMI great er or equal to 85th percentile for the same age and gender, we also test if teac her grading standard will affect childrens probability of becoming overwe ight. As the probability of being overweight is a dichotomy, probit or logit model would be the appropriate. Therefore, we begin our empirical work by estimating a simple pooled logit model for the purpose of comparison to the more efficient estimators. Marginal effects are presented in the next section. Various logit regressions are estimated: (3) ] '' [)1Pr(1istst it ist ist istSchool family XGS Y where the dependent variable is a dummy vari able which equals to one if student i is overweight in school s during year t. GS is substituted with two measures described above in different models. If individual effe ct is invariant across time, pooled logit and probit will provide similar and consistent estimates. However, certain unobserved individual-specific characters may bias our estimates because teacher grading standard might not be exogenous. For instance, al though not very likely teachers may give overweight students lower ratings due to certain stereotypes. If this is the case, a more 85

PAGE 86

sensible model is the fixed effects logit model. Chamberlain (1980) shows that a fixed effects logit model can be estimated by conditional maximum likelihood (ML) (conditioning on the fixed effects) consistently. Results Starting by examining some basic descripti ve statistics on the final data set, one can see from Table 3-4 that students under high grading st andards have significantly higher BMI compared to student s from low grading standards and similarly with higher overweight rate. We also observe hi gher percentage AfricanAmerican and Hispanic students in high standard group, implying that race of a particular student may cue the teacher to apply the generaliz ed standards rather than devel oping specific standards tailored to individual students (Baron and Cooper 1985; Lightfoot, 1978). If this is the case, models without considering these fact ors will be biased as Af rican American and Hispanic students tend to have higher BMI and are more likely to be overweight. Moving across columns in Table 3-5, we present results showing the relationship between student BMI and teacher grading standards in the top panel and the probability of being overweight and gr ading standards in the bottom panel, both conditional on increasing number of controls. For the sake of brevity, t he main tables in this paper present only the coefficients or marginal effects associated with the grading standards variables. Starting with simple POLS without any controls, we observe a statistically significant effect of high grading standar ds on student BMI while students under both higher math and reading standards show str onger effects in magnitude. Column 2 presents results including individual contro ls. We can see that all models remain statistically significant with slightly smaller magnitude regardless of which subject is used to evaluate standards. When adding family school and teacher characteristics as 86

PAGE 87

well as school and year fixed effects, a ll models remain statistically significant. Comparing to students with low teacher gradi ng standard, students BMI are higher by 0.17 kg/m2 and are 6 percentage points more likely to be overweight if having high math grading standards, 9 percentage points if having high reading grading standards and 3 percentage points if hav ing both high math and r eading grading standards. There are all sorts of unobserva ble factors that predict the likelihood that a child is overweight. As a result, a nave regressi on of grading standards on overweight may overstate the causal impact if all other re lated factors are not perfectly controlled. Although we have a relatively short panel, T=5, it is important to control individual unobservables. Column 6 in Table 3-5 shows estimates from indivi dual and year fixed effects models. Again, we see a signifi cant impact of gradi ng standards on childhood obesity. With smaller magnit udes, high grading standards increase student BMI by 0.06 kg/m2 and the probability of being overweight by 14 percentage points and 5 percentage points if both math and reading standards are high. Given the childhood obesity rate differs across race, ethnicity and gender, the effects of grading standards will be evaluated separately fo r each of these groups to assess whether grading standards may have had heterogeneous treatment effects on the population using individual an d year fixed effects model. We first stratify data by gender. In the case of female students, student BMI is significantly higher for high standard group versus low standard group for both math and reading, while high math grading standards do not increase the probability of being overweight for female students. However, high reading grading st andards increase the probability of being 87

PAGE 88

overweight for female students by 13 perc entage points and a combined high standards increases this probability by 5.5 percentage points. Associations between teacher grading stand ards and children weight status by race and ethnicity are shown in Table 3-6 co lumns three to six. Interestingly, White students respond to the high grading standards greater than any other race/ethnicity groups. Though all estimates remain positive and are with comparable magnitudes, they are not statistically significant except that African-Americ an and Asian students have higher BMI if they are under both high math and reading standards and for Hispanic students under high reading standards. None of African-American, Hispanic and Asian students present higher probability of bei ng overweight when t hey are treated with higher grading standards. Previous studies in this literature find evi dence of differential effects of school and teacher grading standards, with initially high-performing st udents benefiting more from high grading standards (Betts and Grogger 2000; Figlio and Lucas 2004). Therefore, Table 3-6 column seven and eight shows results testing heterogeneous grading standards effects on childrens health by di viding the sample into two subsamples based on childrens direct assessment scores. Ov erall, we observe consistent results, with stronger effects on hi gh-achieving students vers us their low-achieving counterparts. Studies in Childhood obesity lit erature have shown that obes ity is a significant risk factor for adverse school outcomes (Datar and Sturm 2006; Cawley and Spiess 2008). If teacher rating standard is affected by st udents non-academic characteristics, for instance students with higher BMI may receive a lower rating, the effects of grading 88

PAGE 89

standard will be overestimated. We us e a unique method to define our standard measures in this study to avoid such endoge neity caused by simultaneity. Though we have a rather large sample size, with about 18,176 teachers sampled in the ECLS-K, we should be cautious when gen eralizing our results. Conclusion The existing literature has focused the majo rity of its attention on the treatment effects of grading standards or teachers evaluation standa rds on student test scores. However, given that childhood obesity rate increased dramatically over the last decades and more researchers realized that educati on policies may post unintended effects on childrens health, the purpose of this paper is to complement and extend the existing literature to determine if t eacher level grading standards had any effects on child BMI and the probability of being overweight. More specifically, we study teacher grading standards across two subjects and analyze how they relate to childrens weight outcomes. We find that, conditional on a large set of individuals observable c haracteristics that includes demographic information, family background, school charac teristics, and teacher characteristics, students under higher teacher grading standards across subjec ts have higher BMI and are more likely to be overweight. Moreover, th is effect is stronger for female and White students comparing to their c ounterparts. These results suggest that while high grading standard may help improve student academic per formance, it may at the same time harm childrens health, in this case childrens weight. Throughout the paper, the result s consistently indicate that students health respond uniquely to teacher grading standards. This suggests that school teachers should give careful thought to the grading sta ndards when they are considering how 89

PAGE 90

these standards can be improved to promote academic performances. We certainly do not want to improve students academic perform ances at the expense of their health. As childhood obesity has become on e of the biggest health concerns in this nation, teachers, school and parents shoul d realize that school environment is one of the main factors that may associates with this weight gain. More research is needed in this literature to examine the uni ntended impacts of other education policies on childrens health outcomes. Policy makers should keep in mind that a single incentive policy that is designed to promote academic performance may unintentionally harm childrens health as teachers and schools pursuing higher te st score gains may undertake unwanted actions that cause irreversible impact on children. 90

PAGE 91

Table 3-1. Summary Stat istics of All Variables Variables Mean (Standard errors) Mean (Standard errors) Definition/Questions from ECLS-K ECLS-K Wave 1999 (W2) 2007 (W7) Body Mass Index 16.358 (0.055) 22.843 (0.117) Weight in kilograms divided by height in meters squared Overweight students 0.262 (0.008) 0.374 (0.010) Dichotomous variable that equals 1 if body mass index 85th percentile for children of the same gender and age Male 0.477 (0.013) 0.498 (0.011) Dichotomous variable that equals 1 if respondent is Male Age 6.26 (0.01) 14.288 (0.008) Childs age in years at the time the direct child assessment occurred. White 0.662 (0.013) 0.627 (0.011) Dichotomous variable that equals 1 if respondent is White African American 0.147 (0.011) 0.141 (0.009) Dichotomous variable that equals 1 if respondent is African American Hispanic 0.133 (0.008) 0.164 (0.008) Dichotomous variable that equals 1 if respondent is Hispanic Asian 0.02 (0.002) 0.027 (0.002) Dichotomous variable that equals 1 if respondent is Asian Other race 0.038 (0.004) 0.04 (0.004) Dichotomous variable that equals 1 if respondent is Other race Mom's age 32.716 (0.218) 40.718 (0.209) Age of resident mother, female guardian or mother figure Dad's age 28.351 (0.464) 33.78 (0.483) Age of resident father, male guardian or father figure Mom_White 0.68 (0.013) 0.647 (0.011) Dichotomous variable that equals 1 if mother is White Mom_African American 0.138 (0.011) 0.132 (0.009) Dichotomous variable that equals 1 if mother is African American Mom_Hispanic 0.119 (0.008) 0.142 (0.007) Dichotomous variable that equals 1 if mother is Hispanic Mom_Asian 0.021 (0.003) 0.031 (0.003) Dichotomous variable that equals 1 if mother is Asian Mom_other race 0.023 (0.003) 0.024 (0.003) Dichotomous variable that equals 1 if mother is other race Dad_White 0.592 (0.013) 0.547 (0.011) Dichotomous variable that equals 1 if father is White Dad_Black 0.068 (0.008) 0.069 (0.006) Dichotomous variable that equals 1 if father is African American Dad_Hispanic 0.093 (0.007) 0.112 (0.006) Dichotomous variable that equals 1 if father is Hispanic Dad_Asian 0.018 (0.002) 0.024 (0.002) Dichotomous variable that equals 1 if father is Asian 91

PAGE 92

Table 3-1. Continued. Dad_Other race 0.015 (0.002) 0.016 (0.003) Dichotomous variable that equals 1 if father is other race Mom's Education__highschool 0.306 (0.013) 0.211 (0.009) Dichotomous variable that equals 1 if mother's education is high school Mom's Education__some college 0.265 (0.012) 0.304 (0.011) Dichotomous variable that equals 1 if mother's education is some college Mom's Education__bachelor degree 0.174 (0.01) 0.19 (0.009) Dichotomous variable that equals 1 if mother's education is bachelor degree Mom's Education__graduate degree 0.017 (0.003) 0.025 (0.003) Dichotomous variable that equals 1 if mother's education is some graduate degree Mom's Education_master degree 0.051 (0.006) 0.072 (0.006) Dichotomous variable that equals 1 if mother's education is master degree Mom's Education_doctorate degree 0.018 (0.003) 0.022 (0.003) Dichotomous variable that equals 1 if mother's education is doctorate degree Dad's Education__highschool 0.257 (0.011) 0.19 (0.009) Dichotomous variable that equals 1 if father's education is high school Dad's Education__some college 0.181 (0.01) 0.173 (0.008) Dichotomous variable that equals 1 if father's education is some college Dad's Education__bachelor degree 0.139 (0.009) 0.15 (0.008) Dichotomous variable that equals 1 if father's education is bachelor degree Dad's Education__graduate degree 0.015 (0.004) 0.025 (0.004) Dichotomous variable that equals 1 if father's education is some graduate degree Dad's Education_master degree 0.046 (0.005) 0.062 (0.005) Dichotomous variable that equals 1 if father's education is master degree Dad's Education_doctorate degree 0.035 (0.005) 0.043 (0.004) Dichotomous variable that equals 1 if father's education is doctorate degree Family socioeconomic status (SES) status 0.074 (0.021) -0.039 (0.018) Family Socioeconomic scale Familuy poverty level 1.832 (0.01) 1.829 (0.009) Dichotomous variable that equals 1 if family below poverty threshold Teacher_Male 0.019 (0.003) 0.157 (0.007) Dichotomous variable that equals 1 if respondent is male Teacher_Hispanic 0.033 (0.004) 0.043 (0.004) Dichotomous variable that equals 1 if respondent is Hispanic Teacher_Asian 0.013 (0.003) 0.015 (0.003) Dichotomous variable that equals 1 if respondent is Asian Teacher_African American 0.061 (0.007) 0.078 (0.006) Dichotomous variable that equals 1 if respondent is African American 92

PAGE 93

93 Table 3-1. Continued Teacher_White 0.923 (0.007) 0.905 (0.007) Dichotomous variable that equals 1 if respondent is White Schooltype_Public 0.831 (0.01) 0.888 (0.007) Dichotomous variable that equals 1 if respondent's school is a Public school Schooltype_Private 0.031 (0.005) 0.019 (0.003) Dichotomous variable that equals 1 if respondent's school is a Private school Schooltype_Catholic 0.078 (0.005) 0.052 (0.004) Dichotomous variable that equals 1 if respondent's school is a Catholic school Schooltype_Others 0.061 (0.007) 0.04 (0.004) Dichotomous variable that equals 1 if respondent's school is other types of schools School_total enrollment 3.258 (0.03) 3.95 (0.024) Total school enrollment School_free lunch 29.061 (0.721) 29.199 (0.561) Percentage of students eligible for free lunch in school School_reduced lunch 8.382 (0.305) 8.446 (0.195) Percent of students eligible for reduced price lunch in school Region_Midwest 0.246 (0.011) 0.261 (0.01) Dichotomous variable that equals 1 if respondent's is from Midwest Region_South 0.415 (0.013) 0.386 (0.011) Dichotomous variable that equals 1 if respondent's is from South Region_West 0.176 (0.01) 0.176 (0.009) Dichotomous variable that equals 1 if respondent's is from West Notes: *Summary statistics and standard erro rs (in parentheses) from authors calculations from the 1999,2000,2002, 2004 and 2007 Early Child hood Longitudinal Study -Kindergarten Class of 1998-1999. Africa Am erican and Hispanics are oversampled in the ECLS-K, results above are weighted means

PAGE 94

Table 3-2. Comparison of BM I and Obesity Across Year and Gender 1999 2000 2002 2004 2007 Gender Female Male Female Male Female Male Female Male Female Male BMI Mean 16.457 (2.268) 16.339 (2.352) 16.928 (2.829) 16.879 (2.912) 18.73 (3.89) 18.595 (3.88) 20.682 (4.782) 20.468 (4.73) 22.783 (5.898) 23.015 (6.653) BMI Median 21.698 22.490 21.480 22.530 17.580 17.570 19.330 19.290 21.180 21.630 BMI 95th Percentile 20.790 20.903 22.580 22.710 26.770 26.230 30.100 29.650 33.530 33.630 Fraction Overweight 26.3% 24.6% 27.6% 25.6% 36.0% 33.3% 42.1% 36.7% 36.7% 34.9% Number of Obs. 9,877 9,444 7,820 7,490 6,925 6,717 5,276 5,245 3,971 3,953 Notes: *Summary statistics and standard errors (in parentheses) from authors ca lculations from the 1999, 2000, 2002, 2004, and 2007 Early Childhood Long itudinal Study-Kinder garten Class of 1998-1999. All t he BMI related calculation excl uded observations with 40
PAGE 95

Table 3-3. Teacher grading standards and observed teacher characteristics Teacher Characteristic High Standards Median Standards Low Standards Female 86.26 88.35 97.53 White 0.925 0.914 0.875 African American 0.044 0.043 0.079 Hispanic 0.040 0.044 0.055 Asian 0.019 0.021 0.022 Teacher has a Bachelors degree 23.59 27.16 20.22 Teacher has a Masters degree 37.81 30.38 40.17 Teacher has regular certification 69.58 48.23 3.39 Teacher has temporary certification 16.69 29.43 37.22 Notes: Each column of the table presents weighted means. Statis tical significance level: *** : p<.001; **:p<.01; :p<.05; Standard errors are cluster-corrected by state. All models include controls for individual race, gender and age. 95

PAGE 96

96 Table 3-4. Means of dependent variabl es and selected student characteristics Variable High Standards Median Standards Low Standards BMI 21.166 18.634 16.883 Overweight 0.381 0.304 0.261 Female 48.11 54.33 51.5 White 56.93 67.75 68.94 African American 12.71 6.83 7.91 Hispanic 19.68 12.84 12.47 Asian 4.84 7.52 5.85 Free lunch 32.445 25.556 22.36 Reduced lunch 8.067 7.593 7.535 Public school 82.72 82.16 75.77 Catholic school 11.38 10.48 14.67 Notes: Each column of the table presents weighted means. Statis tical significance level: ***: p<.001; **:p<.01; *:p<.05; Standard errors are cluster-corrected by state. All models include controls for individual race, gender and age.

PAGE 97

Table 3-5. Estimated effects of t eacher grading standards on students' BMI and the probability of being overweight Dependent Variable: BMI Model (1) (2) (3) (4) (5) (6) High Math grading standards 0.219*** (0.040) 0.3756*** (0.0886) 0.1946 (0.1063) 0.1749 (0.1090) 0.173*** (0.049) 0.0670^ (0.038) High Reading grading standards 0.263*** (0.040) 0.2142*** (0.0606) 0.1561* (0.0701) 0.1752* (0.0764) 0.179*** (0.038) 0.0694^ (0.0371) High Math & Reading gs 0.301*** (0.025) 0.1322*** (0.0335) 0.0889* (0.0386) 0.0802* (0.0400) 0.104*** (0.018) 0.0610** (0.0210) Dependent Variable: 1(Overweight) High Math grading standards 0.2928*** (0.0202) 0.1362** (0.0460) 0.0444 (0.0526) 0.0450 (0.0551) 0.0568* (0.0290) 0.1012 (0.0608) High Reading grading standards 0.1470*** (0.0203) 0.0588 (0.0308) 0.0469 (0.0355) 0.0718 (0.0391) 0.0947*** (0.0222) 0.1405** (0.0433) High Math & Reading gs 0.0973*** (0.0115) 0.0380* (0.0160) 0.0185 (0.0180) 0.0212 (0.0195) 0.0326** (0.0106) 0.0487* (0.0226) Student covariates NO YES YES YES YES YES Family covariates NO NO YES YES YES YES School covariates NO NO NO YES YES YES Teacher covariates NO NO NO YES YES YES School fixed effects NO NO NO NO YES NO Student fixed effects NO NO NO NO NO YES Year fixed effects NO NO NO NO YES YES Sample size 40,663 32,689 22,164 19,385 30,493 7209 97

PAGE 98

98 Table 3-5. Continued Notes: Each column of the table presents c oefficients and standard erro rs in parentheses fr om a different regression. Statisti cal significance level: ***: p< .001; **:p<.01; *:p<.05; ^<0.1 All regression analysis used ECLS-K longitudinal sample weight. Detailed covariates used in the models can be found in Table 1.

PAGE 99

Table 3-6. Estimated differential effects of teacher grading standards on students' BMI and probability of being overweight Dependent Variable: BMI Model Female Male White African American Hispanic Asian High Achievers Low Achievers High Math grading standards 0.1112^ (0.0618) 0.1150 (0.0665) 0.0987^ (0.0606) 0.1964 (0.2631) 0.1452 (0.1594) 0.4420 (0.2319) 0.1800* (0.0737) 0.0225 (0.0631) High Reading grading standards 0.2264*** (0.0475) 0.1035* (0.0490) 0.1541** (0.0493) 0.1637 (0.1608) 0.2975** (0.1071) 0.1923 (0.1740) 0.1867*** (0.0539) 0.1705*** (0.0431) High Math & Reading gs 0.1049*** (0.0219) 0.0480* (0.0225) 0.0806*** (0.0214) 0.1375* (0.0608) 0.0716 (0.0445) 0.2428** (0.0804) 0.0718** (0.0218) 0.1095*** (0.0275) Dependent Variable: 1(Overweight) High Math grading standards 0.0560 (0.0360) 0.0674 (0.0401) 0.0264 (0.0341) 0.1233 (0.1262) 0.0952 (0.0874) 0.2109 (0.1721) 0.0611^ (0.0359) 0.0455 (0.0490) High Reading grading standards 0.1361*** (0.0283) 0.0266 (0.0293) 0.0926*** (0.0277) 0.1137 (0.0773) 0.1129 (0.0579) 0.1453 (0.1299) 0.0869*** (0.0259) 0.0997** (0.0329) High Math & Reading gs 0.0551*** (0.0114) 0.0090 (0.0121) 0.0366** (0.0113) 0.0380 (0.0250) 0.0067 (0.0212) 0.0733 (0.0465) 0.0211* (0.0107) 0.0546** (0.0167) Sample size 14,014 13,400 16,260 1,680 2,856 614 16,446 12,519 Notes: Each column of the table presents coefficients and standard errors in parentheses from a different regression. Statisti cal significance level: ***: p< .001; **:p<.01; *:p<.05; ^<0.1 All r egression analysis used ECLS-K longitudinal sample weight and controlled for student and year fixed effects. Detailed covariates used in the models can be found in Table 1. 99

PAGE 100

16 17 18 19 20 21 22 23 24 19992000200220042007BMIYEARStudent BMI by teacher grading standard Low Standard High Standard Figure 3-1. Student BMI by teacher grading standards 0.2 0.23 0.26 0.29 0.32 0.35 0.38 0.41 0.44 19992000200220042007Percentage overweightYEAROverweight rate by teacher grading standard Low Standard High Standard Figure 3-2. Student overweight rate by teacher grading standards 100

PAGE 101

CHAPTER 4 IS IT WISE TO INVEST IN SEX E DUCATION? ABSTINENCE EDUCATION PROGRAM AND ADOLESCENT RISKY BEHAVIORS Introduction In 1996, the welfare reform law added Titl e V Section 510 (b) of the Social Security Act which established a new funding stream to provide grants to states for abstinence-only 1 programs. The new federal legislation requires states that choose to accept these funds to teach that abstinence is the only certain way to avoid pregnancy and sexually transmitted diseases (STDs), and may not in any way advocate contraceptive use or discuss contraceptive methods except to emphasize their failure rates. Beginning in fiscal year 1998, the Ti tle V abstinence-only pr ogram has allocated $50 million annually to such programs that teach abstinence from sexual activity outside of marriage as the expected standard for sc hool-age children. Under a matching block grant program, states must ma tch this federal funding at 75 percent, resulting in a total of up to $87.5 million annually for Title V, Section 510 abstinenc e education programs (Trenholm et al. 2008). With the official implementation of abs tinence-only programs and the infusion of the federal funds across the nation, a new round of debate over the effectiveness and efficacy of abstinence-only sex education b egins to heat up. A few decades ago, debate over sex education yet focused on whether schools should provide information on sex related matters. Supporters of no school sex education believe that providing such information to adolescents may promote their sexual behaviors, while advocates of sex 1 Abstinence education focuses on teaching young people to abstain from sex until marriage. Comprehensive sex education does not focus either solely or so closely on teaching young people that they should abstain from sex until marriage. Alt hough they do explain to young people the potential benefits of delaying having sex, they also make su re that they are taught about contraceptives. 101

PAGE 102

education at the time argue t hat the probability of an adolescent becoming sexually active will not be affected solely due to the sexual information provided to her/him at school. Therefore, it is necessary for school s to provide accurate sex education which will help to prepare these willbe-sexually-active young adults to acquire the necessary knowledge about pregnancy and STD and HIV/AIDS prevention. According to recent studies, in 2007 close to half of (47.8%) all teenagers 15 to 19 years old in the United Stat es have had sexual intercourse at least once and 7.1% of them initially had sexual intercourse before age 132. Coincidentally, the highest agespecific rates of reported Chlamydia and Gonorrhea, the most commonly-reported sexually transmitted disease (STD) in the United States, for women in 2007 were among those 15 to 19 years old of age (3004.7 and 647.9 cases per 100,000 females, respectively), which increased by 12.4% and 2.6% compared to 2003. While the number for adolescent males is substant ially lower, both Chlamydia and Gonorrhea case rates among them have increased steadily since 20033. Both of these diseases have also been associated with increased HIV transmission. Estimated numbers of cases of HIV/AIDS for the same age gr oup increased 20 percent from 1010 cases per 100,000 in 2001 to 1213 in 20054. In addition, the birth rate for U.S. teenagers 15 to 19 years rose 3% to 41.9 births per 1000 fema les in 2006, the first increase reported since 19915. Given current patterns of adolescent sexual behaviors and the alarming 2 Centers for Disease Control and Prevention, Youth risk Behavior Surveillance Summaries, June 2008. MMWR 2008; 57. 3 Centers for Disease Control and Prevention, Sexually transmitted Diseases Surveillance, 2007 4 Centers for Disease Control and Prevention. HIV/AIDS Surveillance Report, 2005. Vol.17. 5 National Vital Statistics Report, 2007. Volume 56, Number 7 102

PAGE 103

numbers of STDs and HIV incidences as well as teenage pregnancy rates, it is safe to believe that preventing teenagers from being sexually active is at best unrealistic a is urgent to develop effective education programs to prevent the HIV/AIDS and othe STDs epidemic. nd it r The ongoing debate over sex ed ucation is now between advocates for abstinenceonly-until-marriage programs, focusing on teac hing young people to abstain from sex until marriage, and supporters of comprehen sive sex education, which includes contraceptives information besides abstinence, in the nations schools. With the incessant increase of the number of STD/HIV infections among adolescents and millions of dollars floodi ng into abstinence-only programs, the unavoidable question is whether the money is well-spent. Studies in this literature have provided frui tful evidence showing that sex education in general does not promote st udents sexual behaviors in any way. Social scientists and lately economists have been longing fo r discovering further evidence of the potential differential effects, if there is any, of abstinence-only and comprehensive sex education have upon adolescent se xual behaviors. Credible estimates of such analysis, however, are limited. Specifica lly, most of the empirical wo rk can be divided into two categories based on their research methods : quasi-experimental or pre-post approach and a randomized controlled evaluation technique. Much of the studies that adopted quasi-experimental approaches have used cr oss-sectional data (Lerner 2004; Doniger Riley, Utter, and Adams 2001; Ku, Sonenstein, and Pleck 1992, 1993). Though several studies found that exposure to abstinence-on ly sex education has positive effects on delay of first intercourse and increasing cont raceptive use, failing to control for the 103

PAGE 104

unobservables between program s participants and comparison groups make these results susceptible (Ku, Sonenstein, and Pl eck 1992, 1993). Another stream of studies, including the most recent study conduc ted by Trenholm et al. (2008), have used randomized experimental design to test the effects of a specific school program on students sexual behaviors. They examined the impacts of four abstinence-only education programs on adole scent sexual activity and risks of pregnancy and STDs using survey data collected in 2005 and ear ly 2006 from more than 2,000 teens who had been randomly assigned to either a program group that was eligible to participate in one of the four programs or a control group that was not. The findings show no significant impact on teen sexual activity, no di fferences in rates of unprotected sex, and some impacts on knowledge of STDs and perceived effectiveness of condoms and birth control pills. So far, all the studies on sex education have been focusing on program-evaluation at school levels, although major sex educ ation policies had been made at the state level. Therefore, our paper serves to f ill this gap in the sex education literature and provides further insights on the current issue from at least three im portant angles. First, we collected detailed data on state sex education laws from various sources regarding when states have sex educati on mandates and if they do, whether they require comprehensive or abstinence-only sex educat ion. Subsequently, employing a rich micro-level data set, we use various panel data techniques to identify the impact of state sex education programs has upon their students. We believe a study of outcomes at state level may paint a clearer picture of the benefits of co mprehensive sexuality versus abstinence-only programs. Howeve r, critics of such method may argue that state sex 104

PAGE 105

education is endogenous due to the 1996 reform which may have altered some states sex education focuses and hence bias our esti mates in unknown directions. Therefore, we first estimate the heterogeneous impacts of the 1996 welfare reform on state sex education legislatives. Based on each states policy change affected by the Title V 510, this is the first paper to employ the Interrupt ed Time-Series design to identify the causal effects of state sex education on students sexual behaviors. Although all 50 states applied for the abstinence-only program funds in 1997, there are significant variability and creativity in the program planning and implementations. Without proper control for the observable and unobservabl e program variations acro ss states and the potential pre-existing state trends, we will not be able to consistently estimate the impacts of 1996 reform through a simple pre-post mean diffe rences analysis. In order to overcome the problem and to try to identify causal e ffects, we requested access to a detailed state-by-state survey conducted by Sexualit y Information and Education Council of the United States (SIECUS) in winter 1997. SIE CUS focus its research on the first-year implementation of the abstinence-only-until-ma rriage program by conducting a survey of state abstinence program administr ators about a variety of iss ues related to the content and implementation of their Se ction 510(b) program. Especially, according to the report state plans vary in the age of target audi ence, media campaigns, in-school curricula, after-school curricula and et c. We then construct an i ndex score for each state to evaluate the state-specific program implementation and t hen divide them into high implementation states and low implementations stat es based on their scores. This time series contrast with greater statistical power allow us to better identify the effects of Section 510(b). 105

PAGE 106

Last, most previous cross sectional studies used relatively older data set or use only one or two-period data, while we employed Youth Risk Behavior Survey Surveillance System of years from 1993 to 2005, which not only allows us to control for state by year trend but also provides us in formation about the most current pattern of adolescent sexual behaviors. Additionally, th is unique data set enables us to examine a variety of different outcomes with the most effi cient estimators as well as to investigate the heterogeneity of effects across gender, racial groups and grades. The remainder of this paper is organized as follows. Section 2 provides some background on the abstinence-only programs. Section 3 discusses the theoretical model followed by empirical strategy and Section 5 describes the data. Section 6 presents the main findings and Section 7 concludes. A Brief Background on Abstinence-only programs Federal funding for teen pregnancy prevent ion in the form of abstinence-only education began in 1981 with the passage of the Adolescent Family Life Act (AFLA). Ever since then, around 4 million dollars annually are invested in abstinence-only programs across the nation until 1996 when t he 1996 welfare reform law added Title V, Section 510 (b) of the Social Security Ac t which established a new funding stream to provide grants to states for abstinence-onl y programs. Under the Title V abstinence-only program, states that choose to accept these funds may not in any way advocate contraceptive use or discuss contraceptive methods except to emphasize their failure rates. Every state has at one time accepted Titl e V funds, though, in 2009, nearly half the states no longer participate in this program. Under Title V, states decisively coalesced around youth 10 to 14 years of age-in upper elementary and middle schools-as the 106

PAGE 107

intended audience for their efforts because students in this age group most likely have not begun to explore sexual relationships. If the intended or say the treated group are students aged between10 and 14, the first tr eated group observed from YRBSS are students in 9th grade in 1999, and the second treated group is students in 9th to 11th grades in 2001. Therefore, to include the entire data when analyzing the effects of 1996 reform will result in inconsistency of the es timates. We thus carefully define treated and control groups based on these information. Deta ils are discussed in the method section. In October 2000, the federal government created another fu nding stream to support abstinence-only programs, Community-Based Ab stinence Education (CBAE). Different from the 1996 reform, CBAE is managed directly by Family and Youth Services Bureau (FYSB), Administration for Children and Fam ilies, Department of Health and Human Services rather than providing funds to states directly to gain cont rol over the programs and hence to adhere to the tenets of the welfare law more clos ely. In addition, majority of the funding is provided to public and private institutions for community-based abstinence education projects ra ther than school based program s. The objective of the grants is to reduce teen pregnancy rate and ST D by teaching abstinence to adolescents ages 9 through 18 and by creating an environ ment within communities that support decisions to postpone sexual activity. Al though a great amount of funds flooded into community abstinence programs, they did not alter the focu s of state sex education or programs implemented at school level. We furt her control the implementation of CBAE, in case of any unobserved bias. Between 1996 and 2008, over $1.5 billion dollars were spent to promote abstinence-only programs through these three main tunnels and for fiscal year 2009 alone, the total amount allocated was just over $160 million. 107

PAGE 108

Theoretical Framework In this section, we provide the theoret ical basis for linking state sex education mandate to sexual outcomes. The conceptual m odel of sexual decision making in this work is built on the model of sexual activity and pregnancy proposed by Oettinger (1999) and Tremblay and Ling (2005). We sh ow that the relationship varies with individual preferences and with the type of sex education (riskaltering, risk-revealing, and utility-altering). Rational individuals become se xually active if the expected utility of intercourse exceeds the cost, for instance, becoming pregnant or contracting STD or HIV/AIDS. As we do not have data regarding STD or HIV/AI DS (only available in 2005 YRBSS) infection, we assume the only co st is the risk of becoming pregnant. Assume the perceived probability of becoming pregnant P where 0 P 1 depends on the optimal use of condom (k*). The term U(y|k*) is defined as the net utility of sex in the present period plus the present value of ex pected future net utilit y. Net utility varies across pregnancy y, where y=p if the individual is pre gnant from an incident of intercourse and y=np if the individual is not pregnant. An adolescent chooses to engage in sexual intercourse rather than abstinence ( V ) if and only if: VkyUkyPkyUkyP )](1)][(1[)()(* Let )(1)(* *kyUknU we have VknUkyPkyUkyP )()](1[)()(* or )]()(/[])([)(* *kyUknUVknUkyP 108

PAGE 109

Equivalently, with )]()(/[])([* *kyUknUVknUZ the individual chooses to be sexually active or not if and only if ZkyP )()(*. Similarly, a sexually active teen will choose to use a condom if and only if the expected benefit of using a condom exceeds the cost, i.e. becoming pregnant or causing someone pregnant: )](1)][(1[)()()](1)][(1[)()( ncyUncyPncyUncyPcyUcyPcyUcyP Rearranging them, we have )]()(/[)]()([)()( ncyUcyUncnyUcnyUncyPcyP Sex education may influence an individuals ultimate decision regarding sexual activity by revealing or altering the probability of pregnancy and by affecting the relative utility of abstinence. Risk-altering sex education includes courses that provide information of alternative contraceptive met hods which may allow sexually active teens to alter the risk of pregnancy ( P ). Comprehensive se x education including contraceptives would fall in this category. In our model, if co mprehensive sex education provides information regarding condom use wh ich in turn affects teens perception on the risk of pregnancy, the risk altering sex education will increase the probability of becoming sexually active conditional on the usage of condom, but has ambiguous effects on the probability becoming sexually ac tive conditional on if condom is not used or the probability of abstinence conditional on the optimal condom usage. The probability of condom usage wi ll increase if schools provide students information about safe sex and the benefits of c ondom usage. It is also likel y that the probability will decrease if other forms of contraceptiv es are introduced to the students. Define K=other as if student chooses any other type of c ontraceptives besides condom, the probability 109

PAGE 110

of choosing other contraceptives conditional on sexually active and no condom usage increases ),1 ( ncaotherKP if other method introduced to the students. Next, if a student is already sexually active, we might be also interested in their other sexual behaviors, for instance, number of sexual partners. Under riskaltering sex education, if )(*kpPdecreases, we would observe an increase in the number of se xual partners. Utility altering sex education changes the teens per ceived utilities (U or V) throughout their adolescence. Abstinence se x education that teach strategies for resisting sex (thereby increasing V), t hat highlight the costs of teen parenthood (presumably reducing U), or that provide information about abortion (likely increasing U) is more likely to be in this category. T herefore the expected e ffects depend on the relative change of utility of abstinence (V) and the utility of intercourse. Furthermore, if utility altering sex education affects the re lative utility of having intercourse and abstinence but does not affect the probabili ty of becoming pregnant with or without using condoms, the probability of using condom or any ot her contraceptives will not change. If the sex education increases the ut ility of having sex, we might also observe an increase in the number of sexual partn ers among sexually active young adults. Finally, risk-revealing sex education provides accurate information to teens that initially may misjudge pregnancy risks. Instruction on contraceptive methods would be purely risk-revealing which again coincides with the focus of many comprehensive sex education programs. Having such sex educat ion or not alters the probability of pregnancy conditional on the optim al condom usage. Additiona lly, if the probability of becoming pregnant increases due to the sex education, we would expect an increase of condom or other contraceptives use and decrease otherwise. 110

PAGE 111

The expected signs of the impacts of these types of sex education on the probability of engaging in intercourse given the optimal condom choice are listed in Table 4-1. We test these hypot heses in the following sections. Empirical Framework The theoretical model portra ys a simplified model without taking into account of the potential heterogeneity in individual preferences, the differential state policies and the frequency of sexual activity. In this section, we develop regression models incorporating these issues. To test for the existence of differential effects associated with state sex education programs, we estimate the marg inal propensities of becoming sexually active and using condoms, respecti vely. As the probability of conducting sexual intercourse and the probability of using condoms given they are sexually active are dichotomies, probit or logit model would be t he appropriate. We start our analysis with a standard nonlinear panel model: (1) ), (),1Pr(2 1st ist st ist istSexEd xf SexEdxy where yist.represents a series of student sexual behaviors for student i residing in state s during year t for instance if a student is sexually active or not, xist is a set of individual covariates, and SexEdst is our main independent variable of interest which measures different state level sex education policies. In addition, we assume the state effect is invariant across time and if this is the ca se, pooled logit and probit will provide similar and consistent estimates. However, certai n unobserved state-specif ic characters may bias our estimates because xit may contain lagged yit. For instance, states with higher percentage of students who are sexually active in year 1999 might expect a similar outcome the following year. Si milarly, states sex education may have lagged effects on 111

PAGE 112

students sexual behaviors or their effect might not become pr esent until the next period. If this is the case, a more sensible model is the unobserved effects logit model. Model (1) thus becomes: (2) ), (),1Pr(2 1st ist s sist istSexEd xfxy where s is the unobserved state-specific effe ct. If we further assume that: (3) is independent once conditional on )(,...,1 itiyy),,(sist istSexEdx and (4) st istsSexEdx ,~ Normal ),0(2The random effects model wi ll produce consistent esti mates. Nonetheless, the assumption thats and are independent may not hold. For example, states with higher teen pregnancy rates or higher STD infection cases are more likely to implement stricter sex educat ion programs. If we relax this assumption by allowing st istSexEdx, s and to be correlated, we might consistently estimatest istSexEdx,s using fixed effects logit. Before we characterize our model, one thing should be po inted out that we obtained information regarding state level se x education policies for years 1995, 1999, 2001, 2003, and 2005. Given this short panel, the only legitimate model that can consistently estimate the impact of sex educati on at state level is the fixed effects logit model (Chamberlain 1980). In particular, we consider the following underlying latent model: (5) istst ist sistSexEd x y 2 1* where is a continuous but unobserved index of utility of being sexually active of individual i residing in state s during year t xist is a vector of individual characteristics, SexEdst represents the type of sex educ ation implemented in state s in year t which isty* 112

PAGE 113

equals to 1 if state requires sex educati on mandate with contraceptive, 2 if state requires abstinence sex educat ion, with states without se x education mandates as the baseline, s is an idiosyncratic fixed effect which a ccount for state-specific time invariant unobservables, and ist is the stochastic error term. Rather than observing we observe: isty*(6) if y0 0 1istotherwise yist*and hence the logit model is: (7) ) exp(1 ) exp( ),i|1Pr(2 1 2 1 ist ist ist ist istSexEd x SexEd x x isty Chamberlain (1980) shows that such a fi xed effects logit model can be estimated by conditional maximum likelihood (ML) (conditi oning on the fixed effects) consistently. For T=5 we consider the set si =, then 4,...,2,1 isty(8) iDd t t st ist ist t st ist ist i i ii id SexEd SexEd2 2x y x y SexEd SexEd x y ]) exp[( ]) exp[( ),, ,...., ,| Pr(5 1 1 5 1 1 15 11 15 1115 11 x ,...., y ,..., where di is the set of all possible combinations of si ones and 5-si zeros, is independent of state-specific effect. The primary concern with identification of2 in fixed effects logit is that the state may not change their sex education frequently and hence will be cancelled out when differencing out the state fixed effects. If th at is the case, SexEd will not be specified. On the other hand, if state does vary their sex educations from year to year, this could be endogenous due to the pre-existing state trend or due to certain state-specific 113

PAGE 114

characters, for instance high teen pregnanc y rate/high percentage of teenage had sex. Last, the 1996 welfare reform may also bias the estimate because it might influence state sex education legislativ e differentially. In addition, one might be worried that other contemporaneous programs or certain state so cioeconomic characters may also affect students sexual behaviors which will again bias the estimate. To overcome these concerns to the mo st extent, we use the SIECUS data, including information for 46 states that parti cipated in the survey, to construct an index score for each state: (9) ss sZ Index where Zs is a set of variables representing stat e-specific policy differentials collected from the SIECUS. We categorize these variables into five s ubgroups. First, we evaluate if a state make-up an advisory panels6 which may help plan and develop the state programs. Second, we document if states specify intended audience for their programs. We notice that 45 out of the 46 states have a specific age-intended audience. Third, we include data regarding states media campaigns; especially we specify type of method in the campaign, the initiate year and etc. F ourth, majority of the states provide funding to community-based organizations and hence we further classify these programs based on their nature, for example if they prov ide education progra ms or recreational programs or other programs. Also, we include the absolute number of grants awarded for each state. Last, we collect data on if states provide grants to education agencies. According to SIECUS, 3 states provi ded grants to do abstinence programs at elementary school level, 7 states at junior high and 5 states at high school levels. A 6 26 states and the District of Columbia had advisory panels of some sort (SIECUS 1999) 114

PAGE 115

detailed list of variables included in the model is shown in Appendix 1. In total, we include 48 variables to estimate an index sco re for each state. With the handful of data, we, however, do not know the pattern of how these numbers are related. To discover the unknown trends in the data, we apply the Principal Component Analysis (PCA) to identify the pattern of the data and reduce its dimensions. Specifically, we first standardize the data by subt racting the mean and dividi ng by the standard deviation and then perform the component analysis which bas ically tells us the most significant relationship between the data dimensions. We used the calculated components constructing two different scores for each st ate. As component 1, in our analysis, accounted for around 30% of the variance and 45 components were needed to explain all of the variance. Ther efore, we use component 1 and the summation of all 45 components as our index score respectively We then define stat es with index score higher than the median as hi gh implementation st ates and low implem entations states otherwise, as showed below. (10) otherwise DoseLow IndexTot Index Indexif DoseHigh groupmedian median s s/1Using aggregated state level data, for instance percentage of students who had intercourse, we employ the interrupted time-series design to estimate the average treatment effects of 1996 welfare reform. Fo r this model to consistently estimate the average treatment effects, we will need info rmation regarding the pre-trend for each state. The data employed in this study allo w us to go back to 1991 which leaves us 3wave data prior to the introduction of 1996 re form. We use these data to calculate the potential pre-existing sexual behaviors trends and thus look for a sharp decrease/increase, if we assume Title V has negative effects on the probability of 115

PAGE 116

students engaging in sexual activities, in sexual behaviors/performing safer sex, increased condom usage for instance. Furthe rmore, to account for the potential bias from time-varying state-specif ic characteristics, we include a set of variables and statefixed effects to capture such effects. Us ing state level data on students risky behaviors, we estimate the following equation: (11) stss s t s t s t t stState group trend group trend group trend trend Y 6 2 5 1 4 322110) ( ) ( where takes on value of for year and equals to for year (respectivelyttrend1,4,2,0,0, )6,6,6,6,6,6,4,2,0( )2007,...,1991 )2007,...,1991(stgroupttrend2 4)10,8,60,0(7, is a dummy variable which equals to one if the state is a high implementer and zero otherwise, captures the difference in average growth rate between groups prior to 1996 Reform and 5 indicates how the comparison groups differ in the difference in growth rate from before to after the reform. This equation enables us to break up the potential nonlinear growth trajectories for high and low implementers into separate linear components. Specifically, we will observe the heterogeneous growth ra tes for high and low implementers during before and after 1996 reform respectively. Essentia lly, this is a difference-in-differencesin-differences (DDD) analysis where the firs t difference is a within high dose and low dose states change over time, the second di fference is a within state pre-post reform change, and the last difference is between the high dose and low dose states pre and 7 Another coding scheme is let equal to for year and equal to The only difference would be the interpretation. In this case,ttrend1)10,8,6 )16,14,12,10,8,6,4,2,0( )2007,...,1991(ttrend24,4,2,0,0,0,0( indicates the different growth rates betwe en groups without the reform treatment and5 captures the heterogeneous increment (or decrement) to growth rates among comparison groups post reform treatment. Regression results, not shown here, presen t little difference and are available upon request from the author. 116

PAGE 117

post reform. The advantage of this model over DDD is not only we observe the average treatment effects of 1996 on high implementers relative to low implement ers; we will also observe the pre and post growth ra tes between these two types of states. Data Our dependent variables are drawn from the Youth Risk Behavior Surveillance System (YRBSS) for the years 1993 to 2005 which has been administered biennially since 1991 by the Centers for Disease C ontrol and Prevention (CDC). The YRBSS system monitors six categories of priority health-risk behaviors among youth, including those relating to student sexual behaviors, information on HIV/AIDs education, and other risky behaviors. It includes national, st ate, and local school-based surveys of nationally representative samples of 9t h through 12th grade students. Our main outcome variables include education programs delay the onset of sex, reduce the frequency of sex, reduce the number of sexual partners among teens, or increase the use of contraception. From Table 4-2 we see that, around 47% st udents in our data are sexually active and the average age of firs t time sex is 13.6. For students who had sexual intercourse, 59% of them chose to use a condom and on average they have 1.5 sexual partners. Other indi vidual control variables are quite standard, for instance gender, race and grade. Results are shown in T able 4-2. Additionally, to control for school characteristics, we add two variables: if the respondent feels safe to be at school or if she/he carries weapons to school. We also include a few other individual control variables which may measure the respondents lifestyle, for example number of days drank alcohol/number of time s used marijuana. Last, to cont rol for time-varying state characteristics, we incorporat e a set of state covariates to capture such differential effects, including state-level public school dropout data and student to teacher ratio in 117

PAGE 118

public elementary/secondary schools collected from Common Core of Data (CCD). These variables meant to control for the school quality which may in turn reflect students tendency of behaving riskily. We al so include data on percentage of people below poverty for each state (i t is the below 100% poverty by weighted person count for all ages (in thousands)), total crime which eq uals to the sum of violent and property crimes both obtained from Census, and divorc e rate obtained from CDCs National Vital Statistics based on provisional counts of divorces by state of occurrence (rates are per 1,000 total population residing in area) to cont rol for state socioeconomic status which might influence teen behaviors. Our data on state sex education is mainly collected from three sources: State Policies in Brief-Sex and STD/HIV E ducation for years 1995-2005 from Alan Guttmacher Institute (AGI), School Heal th Policies and Programs Study (SHPPS) conducted and maintained by CDC and Between the Lines: States Implementation of the Federal Governments Se ction 510(b) Abstinence Educat ion Program in Fiscal Year 1998 from collected by SIECUS. Firstly, we requested access to the surv eys data collected by AGI and then handcoded data which subsequently are augment ed with YRBSS using Fips code. AGI collects data annually on state sex education legislative since 1988, however, data are not coded in a consistent way until 1995 and there is no survey conducted in 1997. Therefore, we obtai ned 5 years data from years 1995, 1999, 2001, 2003, and 2005. AGI report documented, in detail, the type of sex education mandat e in each state including if the state has comprehens ive sex education, abstinence sex education, or no sex education mandate, if state has sex education mandate, HIV mandate, both sex 118

PAGE 119

education and HIV mandate or onl y sex education or HIV mandate. As shown in Table 4-2, around 36% of students reside in stat es where comprehensive sex education is provided while 13% of students live in st ates where abstinence sex education is required. However, information might not be available for every state in each year. Based on the nature of the survey data, we construct two main independent variables: if state has comprehensive sex education or abs tinence-only sex educat ion, if state has sex education or HIV m andate, or both of them or one of them or none of them and then analyze each policy variable respectively. As mentioned in the previous section, SIECUS conducted a thorough survey in 1998 concerning the impact of 1996 refo rm had upon state sex education programs. According to their report, California and New Hampshire did not conduct any abstinence programs and withdrew their application for the funding in fiscal year 1998, South Dakota and South Carolina chose not to participate in the survey and Louisiana was still in the development state in the survey y ear and hence chose not to participate which left 46 states in the survey. Last, we make use of the School Health Policies and Programs Study (SHPPS), a national survey periodically conducted to assess school health policies and practices at the state, district, school, and classroom levels, to evaluate their effects on adolescent sexual behaviors conducted by CDC. Specif ically, we employ state-level data which were collected by computer-assisted telepho ne interviews or self-administered mail questionnaires completed by designated res pondents in state education agencies in all 50 states and the District of Columbia. Survey was conducted in 1994, 1996, and 2006. For the purposes of the current study, t hese data are merged wit h the database with a 119

PAGE 120

one year lag (i.e. 2006 is merged with year 2005) under the assumption that state will not change its health policy significantly within one year. Especially, we include information on if providing pregnancy prevent ion programs, STD pr evention programs and sex education is a state mandate. Results Nonlinear Panel Data Analysis We begin our empirical work by estima ting a simple pooled Logit model for the purpose of comparison to the more effici ent estimators. All models employed YRBSS survey weights and conditional on gender, race and age. Result presented in Table 4-3 column 1 suggests that states with abstinenc e-only sex education is associated with a 12 percentage points decrease in the probability of becoming sexually active while states with comprehensive sex education programs in creases the probability of becoming sexually active by around 11 percentage points. We further include a set of school covariates to control for differ ential impacts due to school environment on students risky behaviors. As showed in Tabl e 4-3 column 2, the negative effects of abstinence-only sex education disappeared while the effects of comprehensive sex education doubled. When further c ontrolling for state-specific time-varying covariates which might affect respondents behaviors in an unobserved fashion, we see that abstinence-only sex education again present negative sign on the probability of being sexually active, however it failed to present any statistical significance. On the other hand, comprehensive sex education presents consistent positive and statistically significant effects. However, without contro lling for state fixed effects, the estimates might be overestimated due to the endogeneity of state sex education policy. It is likely that states with high er teen pregnancy rate or HIV/AI DS rates are more likely to 120

PAGE 121

implement sex education earlier or in a more aggressive way. We thus include year and state fixed effects in the re st of the models presented in Table 4-3. Abstinence-only sex education presents negative and significant effe cts on the probability of a student being sexually active, however the effect became insignificant when conditioning on a set of state covariates. Similar pattern exists for comprehensive sex education but with positive sign. Overall, abstinence-only (comprehensive) sex education suggests a negative (positive) impact on stud ents decision of being sexua lly active or not, however this effect might be due to unobserved stat e characteristics or other unknown state policies. Consistent with our theoretical model, comprehensive sex education serves as risk-altering and risk-revealing types of programs reduces the probability of becoming pregnant due to sexual intercourse which in tu rn increases the pro bability of adolescent being sexually active. On the other hand, if abstinence-only educat ion promotes the utility of abstinence relative to the utility of having sexual intercourse, we will observe a negative effect on the probability of being sexually active. Although, the empirical model fails to present any statistically significant effect, it does show a negative sign. We employed the same models substitu ting the abstinence/comprehensive sex education measures with if a state has sex education mandat e or HIV mandate or have both. From Table 4-1, we see that states with both sex and HIV education mandates suggest a lower probability of being sexually ac tive in fixed effects models, however this effect became insignificant once controlling for state time-varying characteristics. We next test the hypothesis that if risk revealing and risk altering sex education, here comprehensive, reduces the probabili ty of becoming pregnant by providing information on safer sex, we should see an increase in condom usage. From the results 121

PAGE 122

showed in Table 4-3, one notices that comp rehensive sex education indicates positive effects in all three fixed effects models while abstinence-only education significantly reduces the probability of using condoms. When we use sex and HIV education mandates as independent variables, we do not observe any statistically significant causal effects of these programs. As mentioned previously, we also obt ained data from SHP PS. These policies variables are collected directly from school administrators, which might control for the variation of programs school levels. Top panel of Table 4-5 presents results for intercourse outcome across three models. Programs on pregnancy indicates a negative and statistically significant e ffect on the probability of having sex without controlling for state covariates while program s on STD significantly lower such probability. Because state policies on STD program do not vary across years we fail to estimate its effects in fixed effects models. Bottom panel of Tabl e 4-5 shows results for condom usage. We find a significant positive effect of STD programs on the probability of using condoms, which is consistent with our t heoretical model as it serves as a risk revealing program. To measure the impact of sex educati on on the usage of other birth control methods we began by examining the differ ential effects of abstinence-only and comprehensive programs. Using no birth control method as the omitted group, we find that audiences of abstinence-only education ar e less likely to choose birth control pill and use condoms. Although comprehensive sex education fails to present any statistically significant effects on any type of birth control methods once conditioning on state time-varying variables and state fixed effects, we find positive signs on using 122

PAGE 123

condom and Depo-Provera8. Similarly, sex and HIV mandates fail to show any differential statistically significant effect s on choosing different birth control method using fixed effects model with st ate and individual covariates. Another indicator of safe sex is t he number of partners a respondent had during her/his life time, therefore we further test the effects of sex educ ation on the number of sexual partners a respondent had with results showed in Table 4-8. As can be seen in column 3, only abstinence-only sex educatio n indicates negative and statistically significant effects on number of partners for male students. Specifically, male students residing in states with abstinence-only sex education mandate on av erage have 0.5 less sexual partners in their life time. Last, we present results testing the effects of sex education on the deferring of first time sex in Table 4-9. When both Sex and HIV education are required at state level, the policies significantly decrease the age of initial sex for female students by nearly 1.7 y ears and when only HIV is required the magnitude decrease slightly to 1.1 years. However, we do not observe any significant effects on male students. However, because these analyses did not ta ke into account of the 1996 reform as well as their potential differential impacts on states policy choice s, it is difficult to trust the credibility of the results and interpret these results as effects of state sex education policies. Interrupted Time Series Model As Congress allocated $50 million in federal funds for the abstinence-only program starting from federal fiscal year 1998 (October 1,1997 Sept ember 30,1998), we 8 Depo Provera (also known as DMPA or Depot Medro xyprogesterone Acetate) is a hormone injection that lasts for 3 months to prevent pregnancy. 123

PAGE 124

therefore use 1998 as the policy intervention poi nt to estimate the di fferential effects it has upon state policies and then the policy e ffects on their audiences. Before using a regression framework, it is worthwhile to take a look at the di fferential trends between high and low implementers in figures. Using year 1998 as the cutoff point, figure 4-1-5 present the state level percentage st udents who had sex, per centage students who used a condom last time had sex and perc entage of students who had sex before 13 years of age before and after the cutoff point for high and low implementers which is decided based on the state index scores. Looking first at figur e 4-1 where we plot state level percentage of students who had sex agai nst year, we see some evidence of a negative effects of the treat ment on both high and low im plementers defined using component one as state index score. T he slopes for both low and high implementers become more positive post 1998 implying the negative effects of 1996 reform on students risky behaviors. In addition, the di fference in trends between high and low implementers before 1998 is much bigger than post 1998 and converges toward 2007. We then aggregate state level dat a by years to get a clearer look at the changes in trends. Figure 4-2 shows same pattern as the pa ttern in figure 4-1, th erefore in the rest of the figures we use aggregated level data. In figure 4-3 we use all components calculated from the PCA to identify high and low implementers, we observe slight changes in the shape of trends while the ge neral pattern is the same. Figure 4-4 focuses on the differential trends in t he percentage of condom usage between two groups. In the first of these two figures, we used component 1 only to identify two groups and one can see that t he before trends are much wi der relative to the post trends between two groups. In fact, t he percentage of students who used condoms 124

PAGE 125

increases more for low implementers suggesting that abstinence-only education might reduce the probability of using condoms. The last two figures shows the differential trends in the percentage of students who had se x before 13 years of age. As was the case before, we see slopes become more po sitive for both low and high implementers. When using the total components to identif y high and low implementers, the slope for high group in fact shifts up. The advantage of using interrupted time seri es model is that we can break up the curvilinear growth trends into separate li near components and then estimate the trends difference. Table 4-10 provides results for the same model using different dependent variables. We see negative and significant e ffects on percentage of students who had sexual intercourse as well as the perc entage of students who had sex before 13% for the high group post 1998, relative to the low group. Specifically, the percentage of students who had sex significantly decreased 0.5% and the percentage of students who had sex before 13 decreased ar ound 0.5%, both at 5% signi ficance level, for highimplement states after 1998, re lative to low implementers. One should also notice that ITS design c ontrols for baseline level and trend when estimating expected changes due to the inte rvention. The regression method assumes linear trends over time, and the adolescent risky behavior data, in particular, had a poor fit, resulting in large standard errors in t he post-intervention per iod. Therefore, the negative impacts of sex education policies post 1996 reform we observe could be because of the linearity and nonnegativity assumptions. In addi tion, to conduct the ITS analysis, we aggregated data to state level with three data points before and after the intervention, respectively. The limited number of observations may significantly affect 125

PAGE 126

the power of ITS design which may be t he reason why we could not detect any statistical significant effects for some of our analysis. Last, time series is strongest when the intervention produces an immediate effect. However, in our case, the sex education reform may present delayed effects and henc e cannot be captured by our model which again may underestimate the treatment effects. Conclusion Our study provides evidence on the c ausal links between 1996 Title V 510 and students sexual behaviors. However, we di d not detect any statistically significant difference between state level compr ehensive sex education and abstinence sex education mandates. One might be worried that school policies vary considerably and therefore we will not identify the effects of state sex educat ion policies. However, we employ a rich dataset which allows us to control for individual characteristics, school characteristics and state covariates to the most extent and then id entify the effects of sex mandate at state level. Overall, we do not find that abstinence-only and comprehensive sex education decrease the probability of bei ng sexually active or increase the likelihood of performing safe sex. Instead, we find that abstinence-only lower the probability of using condoms and birt h control pills relative to not using any birth control method. We then use a unique identificat ion strategy that allows us to create two groups based on how strict states implement thei r sex education policies and then test the effects of the differential trends in students risky behaviors due to 1996 reform. We find that the trend in percent age of students who had sex (per centage of students who had sex before 13) decreases by 0.5% (0.5%) in high-implement states relative to lowimplement states post 1996 reform. 126

PAGE 127

Searching for an effective sex education policy which will reduce the incidence and prevalence of STD and HIV/AID and teen pr egnancy rate has been the main goal for many policy makers, school administrators and parents for decades. However, credible studies on such issue are limited. As the firs t study evaluating the differential effects of 1996 reform at state level, our results i ndicate that a heterogeneous trend exists between high and low groups. With the impl ementation of 1996 reform, however, the decreasing/increasing trends in the likelihood of having sex (performing safer sex) slows down implying that such policies may resu lt in negative effects on adolescent risky behaviors overall. 127

PAGE 128

Table 4-1. Predicted Effects of Different types of Sex Education on the Probability of becoming sexually acti ve and Condom Usage Risk-Altering* Utility-Altering** RiskRevealing*** Utility Ordering Impact on (Comprehensive) (Abstinenceonly) (Comprehensive) Case 1 Case 2 Case 1 Case 2 (1) (2) (3) (4) (5) (6) V>U(np|k*) P(a=1|k=c) + 0 or + 0 or ? ? P(a=1|k=nc) 0 or + 0 or + P(a=0|k*) 0 0 or 0 ? ? V1|k*,a=1) + + Risk-altering sex education reduces the P(p|k*) **Case 1 utility-altering sex education in creases U(np|k*) relative to V; case 2 utility-altering sex education decreases U(np|k*) relative to V. ***Case 1 risk-revealing sex education increases P(p|k*) case 2 risk-revealing sex education decreases P(p|k*) 128

PAGE 129

Table 4-2. Summary Stat istics of All Variables Variables Mean (Standard errors) Definition/Questions from YRBSS If had sex 0.466 (0.007) Dichotomous variable that equals 1 if had sexual intercourse If used condom 0.591 (0.004) Dichotomous variable that equals 1 if used a condom the last time had sex. Age of first time sex 14.187 (0.020) How old were you when you had sexual intercourse for the first time? Number of partners 1.43 (0.023) During your life, with how many people have you had sexual intercourse? Female 0.485 (0.004) Dichotomous variable that equals 1 if respondent is Female Asian 0.032 (0.003) Dichotomous variable that equals 1 if respondent is Asian African American 0.138 (0.006) Dichotomous variable that equals 1 if respondent is African American Hispanic 0.100 (0.005) Dichotomous variable that equals 1 if respondent is Hispanic White 0.643 (0.010) Dichotomous variable that equals 1 if respondent is White Grade 9 0.241 (0.428) Dichotomous variable that equals 1 if respondent is in Grade 9 Grade 10 0.246 (0.431) Dichotomous variable that equals 1 if respondent is in Grade 10 Grade 11 0.252 (0.434) Dichotomous variable that equals 1 if respondent is in Grade 11 Grade 12 0.258 (0.437) Dichotomous variable that equals 1 if respondent is in Grade 12 # Hours Watching TV per day 4.06 (0.027) Number of hours a respondent watches TV on an average school day Smoke regularly 0.214 (0.004) Dichotomous variable that equals 1 if respondent smoked cigarettes regularly # days drink of alcohol 1.74 (1.076) Number of days a respondent had at least one drink of alcohol # of times used marijuana 1.709 (0.022) Number of times a respondent used marijuana # Days have PE Class 3.170 (0.048) Number of times a respondent exercises or participates in physical education class # of times carry a weapon 0.559 (0.011) Number of times a respondent carry a weapon such as a gun, knife. 129

PAGE 130

130 Table 4-2. Continued # Days unsafe at school 1.099 (0.004) Number of days a respondent did not go to school because you felt you would be unsafe at school or on your way to or from school State Level Variables % had comprehensive sex education 0.362 (0.036) Percentage of students who had comprehensive sex education classes. % had abstinence sex education 0.131 (0.025) Percentage of students who had abstinence sex education classes. % have no sex mandate 0.246 (0.033) Percentage of students residing in states with no sex education mandate. % had both HIV and SEX education 0.348 (0.038) Percentage of students residing in states with both sex education and HIV education mandate. % only had HIV education 0.398 (0.038) Percentage of students residing in states with HIV education mandate. % only had SEX education 0.007 (0.006) Percentage of students residing in states with sex education mandate. State high school Dropout rate 15.23 (7.07) High school dropout rate in respondents state of residence Total Crimes 4.042 (1.74) Sum of violent and property crimes per 1,000 total population residing in the area. %Adult with Bachelor Degree 24.712 (0.272) Percentage of adults with bachelor degree in respondents state of residence % Family below poverty 12.53 (0.011) Percentage of family below poverty line in respondents state of residence State Divorce rate 72.9 (0.188) State divorce rate in respondents state of residence No. of Obs. 105,724 Notes: *Summary statistics and standard dev iation (in parentheses) from authors calcul ations from the 1991-2005 Youth Risk Behavior Surveillance System. Africa American and Hispanics are oversamp led in the YRBSS, results above are weighted means non-weighted means are similar for all variables except fo r the proportion of blacks and Hispanics

PAGE 131

Table 4-3. Estimated Effects of State Sex Ed ucation on adolescent P r(sexual intercourse) Dependent Variable: 1(Sexual Intercourse) Model (1) (2) (3) (4) (5) (6) Panel A Abstinence SexEd -0.1229** (0.0375) -0.0717 (0.0716) 0.0821 (0.0867) -0.1751*** (0.0382) -0.1903*** (0.0397) -0.0380 (0.1822) Comprehensive SexEd 0.1162*** (0.0274) 0.2267*** (0.0371) 0.1451* (0.0691) 0.1699*** (0.0292) 0.1551*** (0.0302) 0.1158 (0.2224) Panel B Sex Education and HIVMandate -0.0178 (0.0448) -0.0460 (0.0457) 0.1709* (0.0756) -0.1379** (0.0507) -0.1486** (0.0530) -0.6029 (0.4933) HIV mandate only -0.0765 (0.0564) -0.1023 (0.0582) 0.0296 (0.0844) -0.0535 (0.0650) -0.0305 (0.0705) 0.4364 (0.3015) Sex Education Mandate only -0.1157 (0.1707) -0.1439 (0.1768) 0.0326 (0.2026) -0.1417 (0.1733) -0.1767 (0.1801) -0.7200 (0.5594) School covariates NO YES YES NO YES YES State covariates NO NO YES NO NO YES State fixed effects NO NO NO YES YES YES Year fixed effects NO NO YES YES YES YES Sample size 61354 40023 27184 61354 59523 27184 Notes: Each column of the table presents coefficients and standard errors in parentheses from a different regression. Statis tical significance level: ***: p<.001; **:p<.01; *:p<.05; Standard erro rs are cluster-corrected by state. All models include controls for individual race, gender and age. School covariates include number of times a respondent carry a weapon such as a gun, knife, or club an d number of days a respondent did not go to school because you felt you would be unsafe at school or on your way to or from school State covariates include state dropout rate, percentage of adults with bachelor degree, Percentage of family below poverty line and state divorce rate 131

PAGE 132

Table 4-4. Estimated Effects of State Se x Education on Performing Safe sex (using condoms) Dependent Variable: 1(Used a Condom) Model (1) (2) (3) (4) (5) (6) Abstinence SexEd 0.0888* (0.0438) -0.0686 (0.0448) -0.0490 (0.0921) 0.0136 (0.0523) 0.0379 (0.0535) 0.5133* (0.2470) Comprehensive SexEd 0.0618 (0.0336) 0.0606 (0.0343) -0.0744 (0.0701) 0.0135 (0.0393) 0.0131 (0.0401) 0.3096 (0.3142) Sex Education and HIV Mandate -0.0698 (0.0500) -0.0433 (0.0496) 0.1389 (0.0783) -0.0087 (0.0678) 0.0209 (0.0692) 0.3298 (0.3140) HIV mandate only -0.0938 (0.0587) -0.0640 (0.0585) 0.0608 (0.0882) -0.1308 (0.0922) -0.1197 (0.0939) 0.3288 (0.5233) Sex Education Mandate only -0.2124 (0.2329) -0.1784 (0.2355) -0.1829 (0.2544) -0.1031 (0.2480) -0.0552 (0.2510) School covariates NO YES YES NO YES YES State covariates NO YES YES NO YES YES State fixed effects NO NO No NO NO YES Year fixed effects NO NO YES YES YES YES Sample size 32341 31207 14371 32341 31207 14371 Notes: Each column of the table pres ents coefficients and st andard errors in parentheses from a different regression Statis tical significance level: ***: p<.001; **:p<.01; *:p<.05; Standard erro rs are cluster-corrected by state. All models include c ontrols for individual race, gender and age. A dditional controls are listed in the notes to Table3 132

PAGE 133

Table 4-5. Estimated Effects of Stat e Policies on Pr(Sexual intercourse) and Pr(Condom Usage) Dependent Variable: 1(Sexual intercourse) Model (4) (5) (6) Program on Pregnancy -0.4849*** (0.0875) -0.4849** (0.1498) -0.1809 (0.1815) Program on STD 0.7263*** (0.0790) 0.7264** (0.2474) Program on Sex education -0.1220 (0.1049) -0.1221 (0.1529) 0.2939 (0.2530) Dependent Variable: 1(Used a Condom) Program on Pregnancy -0.2223 (0.1236) -0.2223 (0.1928) -0.1688 (0.2484) Program on STD 0.5362** (0.1955) 0.5362 (0.3208) Program on Sex education -0.2675 (0.1902) -0.2675 (0.2192) -0.2428 (0.3709) School covariates NO YES YES State covariates NO YES YES State fixed effects NO NO YES Year fixed effects YES YES YES Notes: Each column of the tabl e presents coefficients and standard errors in parentheses from a different regression. Statis tical significance level: ***: p<.001; **:p<.01; *:p<.05; Standard errors are cluster-corrected by state. All models include c ontrols for individual race, gender and age. Addi tional controls are listed in the notes to Table 3 133

PAGE 134

Table 4-6. Estimated Effects of State Sex Education on Birth Control Method Dependent Variable: Categorical variable-Birth control method Model (4) (5) (6) Birth Control Pill Abstinence 0.1198 (0.1002) 0.1526 (0.1020) -1.2025* (0.5947) Comprehensive 0.1375 (0.0740) 0.1345 (0.0752) -0.5001 (0.5403) Condom Abstinence 0.0479 (0.0696) 0.0925 (0.0714) -0.8606* (0.3851) Comprehensive 0.0152 (0.0531) 0.0213 (0.0543) 0.0683 (0.4981) Depo-Provera Abstinence 0.6503*** (0.1219) 0.6827*** (0.1240) -0.1732 (0.5651) Comprehensive 0.4872*** (0.1052) 0.4804*** (0.1066) 0.6120 (0.8600) Withdrawal Abstinence -0.2909* (0.1145) -0.2490* (0.1158) -0.3874 (0.5393) Comprehensive -0.1211 (0.0791) -0.1271 (0.0803) -0.4412 (0.8900) School covariates NO YES YES State covariates NO YES YES State fixed effects NO NO YES Year fixed effects YES YES YES Notes: Each column of the tabl e presents coefficients and standard errors in par entheses from a different regression. Statis tical significance level: ***: p<.001; **:p<.01; *:p<.05; Standard errors are cluster-corrected by state. All models include c ontrols for individual race, gender and age. Addi tional controls are listed in the notes to Table 3 134

PAGE 135

Table 4-7. Estimated Effects of State Sex Education on Birth Control Method Dependent Variable: Categorical variable-Birth control method Model (4) (5) (6) Birth Control Pill Sex Education and HIV Mandate -0.0973 (0.1287) -0.0584 (0.1310) -0.4646 (0.5401) HIV mandate only -0.3824* (0.1714) -0.3927* (0.1741) -0.0965 (0.7140) Sex Education Mandate only 0.3823 (0.4287) 0.3937 (0.4296) Condom Sex Education and HIV Mandate -0.0621 (0.0887) -0.0222 (0.0906) 0.1037 (0.4979) HIV mandate only -0.2960* (0.1301) -0.2992* (0.1325) 0.1310 (0.5835) Sex Education Mandate only 0.0941 (0.4049) 0.1226 (0.4073) Depo-Provera Sex Education and HIV Mandate 0.5869*** (0.1493) 0.6125*** (0.1520) 0.5962 (0.8589) HIV mandate only 0.8257*** (0.2236) 0.8015*** (0.2270) 18.4641 (298.5175) Sex Education Mandate only 0.1130 (0.7237) 0.1040 (0.7239) Withdrawal Sex Education and HIV Mandate 0.7041*** (0.1550) 0.6880*** (0.1571) -0.4240 (0.8897) HIV mandate only -0.4810* (0.2006) -0.5181* (0.2032) -0.2508 (0.7699) Sex Education Mandate only -0.0840 (0.6298) -0.3033 (0.6686) School covariates NO YES YES State covariates NO YES YES State fixed effects NO NO YES Year fixed effects YES YES YES Notes: Each column of the table pr esents coefficients and standard errors in parentheses from a different regression Statis tical significance level: ***: p<.001; **:p<.01; *:p<.05; Standard erro rs are cluster-corrected by state. All models include controls for individual race, gender and age. A dditional controls are listed in the notes to Table 3 135

PAGE 136

Table 4-8. Estimated Effects of State Sex Education on num ber of sex partners Dependent Variable: number of sex partners in respondent's life time Model Full Sample Female Male Abstinence -0.1803 (0.1560) 0.0334 (0.2136) -0.4957* (0.2311) Comprehensive 0.0168 (0.2013) -0.2276 (0.2897) 0.2639 (0.2811) Sex Education and HIVMandate -0.6395 (0.4318) -0.8171 (0.6255) 0.2827 (0.2810) HIV mandate only 0.3005 (0.2642) 0.2323 (0.3947) 0.0812 (0.6448) School covariates YES YES YES State covariates YES YES YES State fixed effects YES YES YES Year fixed effects YES YES YES Notes: Each column of the table presents coefficients and standard errors in parent heses from a different regression St atistical significance level: ***: p<.001; **:p<.01; *:p<.05; Standard erro rs are cluster-corrected by state. All models in clude controls for individual race, gender and age. Addi tional controls are listed in the notes to Table 3. 136

PAGE 137

Table 4-9. Estimated Effects of State Se x Education on the age of first time had sex Dependent Variable: age of first time sex Model Full Sample Female Male Abstinence 0.0057 (0.1936) -0.0422 (0.2736) 0.0892 (0.2771) Comprehensive 0.0263 (0.2664) -0.3267 (0.3811) 0.0673 (0.3697) Sex Education and HIVMandate 0.0260 (0.2663) -1.7415* (0.8469) 0.0632 (0.3695) HIV mandate only 0.7592 (0.4386) -1.1090* (0.5258) 0.5134 (0.8204) School covariates YES YES YES State covariates YES YES YES State fixed effects YES YES YES Year fixed effects YES YES YES Notes: Each column of the table presents coefficients and standard errors in parentheses from a different regression. St atistical significance level: ***: p<.001; **:p<.01; *:p<.05; Standard erro rs are cluster-corrected by state. All models in clude controls for individual race, gender and age. A dditional controls are listed in the notes to Table 3 137

PAGE 138

Table 4-10. Estimated Effects of St ate Sex Education using ITS design Dependent Variable: %had sex %had sex before 13 Model %used a condom #of sex partners Pre-Trend (Before Reform) -1.2982** (-0.4672) 0.541 (-0.5573) -0.5023 (-0.4137) -0.9128*** (-0.2321) Post-Trend (After Reform) 0.1509 (-0.4863) 1.2014*** (-0.2666) 0.2265 (-0.3955) -0.2128 (-0.1744) High Implementers -10.2151 (-7.1281) -16.0338 (-8.261) -1.5822 (-7.114) 3.7087 (-2.6642) Pre-Trend*High Implementers 0.1292 (-0.5326) 0.4061 (-0.6917) -0.2596 (-0.5485) -0.0992 (-0.4196) Post-Trend*High Implementers -0.5054* (-0.2678) -0.2574 (-0.3441) -0.4693* (-0.2532) -0.2817 (-0.2341) R 2 0.929 0.87 0.911 0.89 Sample size 131 137 137 138 Notes: Each column of the table pres ents coefficients and st andard errors in parentheses from a different regression. Statisti cal significance level: ****: p<. 001; ***:p<.01; **:p<.05; *:P<0.1 Standard errors are cluster-corrected by state. All models include controls for individual race, gender and age. A dditional controls are listed in the notes to Table 3 138

PAGE 139

40 50 60 70 80 %students who had sex 1991 1993 1995 1997 1999 2001 2003 2005 2007 year Low Implementers Low Implementers High Implementers High ImplementersAmong High and Low ImplementersEvidence of Differential Trends in %students who Had Sex Define High Implementation> Indexcomp1 Figure 4-1A. Percentage of students who had sexual intercourse Pre-Post 1996 Reform (State level) 139

PAGE 140

High Low 45 50 55 60 65 %students who had sex 1993 1995 1997 1999 2001 2003 2005 2007 yearUsing only component 1Aggregated Level Among High and Low ImplementersEvidence of Differential Trends in %students who Had Sex Define High Implementation> Indexcomp1 Figure 4-2 Percentage of students who had se xual intercourse Pre-Post 1996 Reform (Aggregate level) High Low 45 50 55 60 65 70 %students who had sex 1993 1995 1997 1999 2001 2003 2005 2007 yearUsing all ComponentsAggregated level Among High and Low ImplementersEvidence of Differential Trends in %students who Had Sex Define High Implementation> IndexcompTotal Figure 4-3 Percentage of students who had se xual intercourse Pre-Post 1996 Reform (Aggregate level) 140

PAGE 141

High Low 50 55 60 65 %students used a condom last time had sex 1993 1995 1997 1999 2001 2003 2005 2007 yearUsing component 1Among High and Low ImplementersEvidence of Differential Trends in Condom Usage Define High Implementation> Indexcomp1 Figure 4-4 Percentage of students used condoms last time had sex Pre-Post 1996 Reform (Aggregate level). High Low 50 55 60 65 %students used a condom last time had sex 1993 1995 1997 1999 2001 2003 2005 2007 yearUsing all componentsAmong High and Low ImplementersEvidence of Differential Trends in Condom Usage Define High Implementation> IndexcompTotal Figure 4-5 Percentage of students used condoms last time had sex Pre-Post 1996 Reform (Aggregate level). 141

PAGE 142

142 High Low 6 8 10 12 14 %students who had sex before 13 years of age 1993 1995 1997 1999 2001 2003 2005 2007 yearUsing component 1Among High and Low ImplementersEvidence of Differential Trends in %students who Had Sex Before 13 Define High Implementation> Indexcomp1 Figure 4-6 Percentage of students who had se x before 13 years of age Pre-Post 1996 Reform (Aggregate level) High Low 5 10 15 20 %students who had sex before 13 years of age 1993 1995 1997 1999 2001 2003 2005 2007 yearUsing all componentsAmong High and Low ImplementersEvidence of Differential Trends in %students Who Had Sex before 13 Define High Implementation> IndexcompTotal Figure 4-7 Percentage of students who had se x before 13 years of age Pre-Post 1996 Reform (Aggregate level).

PAGE 143

CHAPTER 5 CONCLUSION The first two essays in this dissertati on examine the causal relation between education policies and childhood obesity. The first study contributes to the sparse literature on adolescent obesity by identifying school account ability system as a strong factor that may contribute to higher adolescent BMI and the probability of being overweight. We construct three differ ent measures to ev aluate state school accountability over the years between 1999 and 2005. According to our estimates, all measures present statistically significant and positive effects on student BMI and the probability of being overweight. Specifically, an additional y ear of exposure to school accountability systems will lead to an increase in BMI by around 0.06 kg/m2 and this effect is bigger among females, and among As ian and White students, as compared to their counterparts. One more ye ar exposure to school accountability also increases the probability of being overweight by around 0.5 percentage points and this effect is slightly bigger among males and Hispani c students. Second, our empi rical results show that school accountability systems present signific ant lag effects on adolescent BMI and the probability of being overweight. It appears that students need an adj ustment period before the full response to changes of school environmental factor. Last, school accountability significantly decreases the number of times female students participate in PE classes. The purpose of the second study is to complement and extend the childhood obesity literature to determine if teacher leve l grading standards had any effects on child BMI and the probability of being overweight. More specifically, we study teacher grading standards across two subjects, math and r eading, and analyze how these relate to 143

PAGE 144

childrens weight outcomes. We find that, co nditional on a large set of individuals observable characteristics including dem ographic information, family background, school characteristics, and teacher charac teristics, students under higher teacher grading standards across subjects have higher BMI and are more likely to be overweight. Moreover, this effect is str onger for female and White students. These results suggest that while high grading stan dard may help improve student academic performance, it may at the same time harm childrens health, in this case childrens weight status. Throughout the paper, results consistently indicate that students health respond uniquely to teacher grading standards. This suggests that school teachers should carefully give thought to the gradi ng standards when they are considering how these standards can be improved to promote academic performances. We certainly do not want to improve students academic perform ances at the expense of their health. As childhood obesity has become on e of the biggest health concerns in this nation, teachers, school and parents shoul d realize that school environment is one of the main factors that may associate with this weight gain. More research is needed in this literature to examine the uni ntended impacts of other education policies on childrens health outcomes. Policy makers should keep in mind that a single incentive policy that is designed to promote academic performance may unintentionally harm childrens health as teachers and schools pursuing higher te st score gains may undertake unwanted actions that post irreversible impact on children. The last study provides evidence on t he causal links between 1996 Title V 510 and students sexual behaviors. However, we di d not detect any statistically significant difference between state level compr ehensive sex education and abstinence sex 144

PAGE 145

education mandates. One might be worried that school policies vary considerably and therefore we will not identify the effects of state sex educat ion policies. However, we employ a rich dataset which allows us to control for individual characteristics, school characteristics and state covariates to the most extent and then id entify the effects of sex mandate at state level. Overall, we do not find that abstinence-only and comprehensive sex education decrease the probability of bei ng sexually active or increase the likelihood of performing safe sex. Instead, we find that abstinence-only lower the probability of using condoms and birt h control pills relative to not using any birth control method. We then use a unique identificat ion strategy that allows us to create two groups based on how strict states implement thei r sex education policies and then test the effects of the differential trends in student s risky behaviors due to 1996 reform. We find that percentage of students who had sex (p ercentage of students who had sex before 13) decreases by 0.5% (0.5%) in high-implement states relative to low-implement states post 1996 reform. Searching for an effective sex education policy which will reduce the incidence and prevalence of STD and HIV/AID and teen pr egnancy rate has been the main goal for many policy makers, school administrators and parents for decades. However, credible studies on such issue are limited. As the firs t study evaluating the differential effects of 1996 reform at state level, our results indi cate that heterogeneous effects exist between high and low groups. With the implement ation of 1996 reform, however, the decreasing/increasing trends in the likelihood of having sex (performing safer sex) slows 145

PAGE 146

146 down implying that such policies may resu lt in negative effects on adolescent risky behaviors overall.

PAGE 147

APPENDIX VARIABLES USED IN THE PRIN CIPAL COMPONENT ANALYSIS Variable Variable Who is the Primary authorit y that control over the program Provided Grants to community-based organizations If the state has a Advisory panel Grants_ education programs Number of grants granted to the community-based programs Grants_recreational programs Intended audience Grants_mentoring programs Less than 10 years old Grants_life option or career planning programs 10 to 14 years old Grants_motivational speaker 15 to 17 years old Grants_less than 14 18 to 19 years old If provide grants to education agencies If state conduct media campaign Education_state funding classroom Status of campai gn: continue an old program/develope d new program Education_Continued existing abstinence program Media campaign initiated year Education_introduc ed new abstinence program Media_using radio programs Education_parents optout choice Media_using billboard statefunds_provide afterschool programs Media_using newspaperads Afterschool_education programs Media_using TV PSAs Afterschool_recreational programs Media_using posters Afterschool_tutoring remedial edu Media_using pamphlets Afterschool_community service Media_using Paid TV Radio Afterschool_mentoring programs Organization that responsib le for media campaign: Programs in Elementary schools Health organization Programs in Junior High Education organization Programs in High school Social service abstinence coordinated with state govt. initiatives Faith-based with private efforts 147

PAGE 148

LIST OF REFERENCES Anderson, P.M, Butcher, K.F., Levine, P.B. 2003. Maternal employment and overweight children. Journal of Health Economics 22(3), 477-504 Anderson, P.M., Butcher, K.F. 2005. Reading, Writing, an d Raisinettes: Are school finances contributing to childrens obes ity? NBER Working Paper No. 11177 Anderson, P.M., Butcher, K.F., Schanzenbac h, D.W. 2009. The effect of school accountability on childrens health. Working Paper. Bagues, M.,Labini, M.S., Zinovyeva, N. 2008. Differential Grading Standards and University Funding: Evidence from Italy, CESifo Economic Studies 54(2), 149-176. Bar-Or, O., Foreyt, J., Bouchard, C., Brownell, K. D., Dietz, W. H., Ravussin, E., Salbe, A. D., Schwenger, S., St. Jeor, S., Torun, B. 1998. Physical activity, genetic, and nutritional considerations in childhood weight management. Medicine & Science in Sports & Exercise 30(1), 2-10. Becker, W., Rosen, S. 1990. The Learning Effe ct of Assessment of Evaluation in High School. Discussion paper 90-7, Economics Research Center, NORC. Betts, J. 1995. Do Grading Standards A ffect the Incentive to Learn? Working paper, University of California-San Diego. Betts, J. 1998. The Impact of Educational Standards on the Leve l and Distribution of Earnings. American Econom ic Review 266-275. Betts, J., Grogger, J. 2000. The Impact of Grading Standards on Student Achievement, Educational Attainment, and Entry-Level Earnings. NBER working paper 7875, September. Bokhari, F.A.S. Scheneider, H. 2009. Schoo l accountability laws and the consumption of psychostimulants. Working Paper, wp2009_03_02, Repec. Brener N.D., McManus T., Galuska D.A., Lowr y, R., Wechsler, H. 2003. Reliability and validity of self-reported height and weight among high sc hool students. Journal of Adolescent Health 32(4), 281. Cawley, J., Meyerhoefer, C., Newhouse, D. 2006. The impact of state physical education requirements on youth physical activity and overweight. Health Economics 16(12), 1287-1301. Cawley, J., Meyerhoefer, C.D., Newhouse, D. L. 2007. The correlation of youth physical activity with state policies. Contemporary Economic Policy 25(4), 506-517. 148

PAGE 149

Centers for Disease Control and Prevention, Na tional Center for Health Statistics. CDC growth charts: United States. http:// www.cdc.gov/growthcharts/ May 30, 2000 Chakrabarti, R. 2006. Do public schools fa cing vouchers behave strategically? Evidence from Florida. Nashville, TN: National Re search and Development Center on School Choice. Chou, S.Y., Grossman, M., Saffer, H. 2004. An economic analysis of adult obesity: results from the Behavioral Risk Factor Surveillance System. Journal of Health Economics 23(3), 565-587. Costrell, R. 1994. A Simple Model of Educational Standards. American Economic Review 956-971. Cutler, E.L., Glaeser Shapiro, J. M. 2003. Why have Americans become more obese? NBER Working Paper No. 9446. Dietz ,W.H. 1994. Critical peri ods in childhood for the developm ent of obesity. Am J Clin Nutr. 59(5), 955-959. Doniger, A. S., Riley, J. S., U tter, C. A., Adams, E. 2001. Im pact evaluation of the "Not Me, Not Now" abstinence-orient ed, adolescent pregnancy Prevention Communications Program, M onroe County, NY. Journal of Health Communication 6(1), 45-60. Eisenmann, J.C., Bartee, R.T., Wang, M.Q. 2002. Physical Activity, TV Viewing, and Weight in U.S. Youth 1999 Youth Risk Beha vior Survey. Obesity Research 10(5), 379-385. Figlio, D. N., Rouse, C.E. 2006. Do acc ountability and voucher threats improve lowperforming schools? Journal of P ublic Economics 90(1-2), 239-255. Figlio, D., Lucas, M. 2004. Do High Grading Standards Affect Student Performance? Journal of Public Ec onomics 88(9-10), 1815-1834. Figlio, D.N., Winicki, J. 2005. Food for thought: the effects of school accountability plans on school nutrition. Journal of Public Economics 89(2-3), 381-394. Finkelstein, E. A., Trogdon, J. G., Cohen, J. W., Die tz, W. 2009. Annual medical spending attributable to obesity: Payer-and service-specific estimates. Health Affairs 28(5), w822-w831. Freedman, D.S., Khan, L.K., Dietz, W.H., Srinivasan, S.R., Be renson, G.S. 2001. Relationship of Childhood Obesity to Co ronary Heart Disease Risk Factors in Adulthood: The Bogalusa Hear t Study. Pediatrics 108(3), 712-718. 149

PAGE 150

Friedman, M.A., Brownell, K.D. 1995. Psychological correlat es of obesity: moving to the next research generation. Psyc hological Bulletin 117, 3. Goodman, E., Whitaker, R.C. 2002. Prospective study of the ro le of depression in the development and persistence of adolesce nt obesity. Pediatrics 109, 497-504. Guo, S.S., Wu, W., Chumlea, W.C., Roche, A.F. 2002. Predicting overweight and obesity in adulthood from body mass index values in childhood and adolescence. American Journal of Clinic al Nutrition 76(3), 653-658. Haney, W. 2000. The myth of the Texas miracl e in education. Education Policy Analysis Archives 8(41). Retrieved December 20, 2007, from http://epaa.asu.edu/epaa/v8n41 Haney, W. 2002. Ensuring failure: How a states achievement test may be designed to do just that. Education Week 21(42), 546-58. Jussim, L. 1989. Teacher Expectations: Self -fulfilling Prophecies, perceptual biases, and accuracy. Journal of Personality and Social Psychology 57(3), 469-480. Jussim, L. 1991. Social percept ion and social reality: A re flection-construction model. Psychological Review 98, 54. Katzmarzyk, P.T., Malina, R.M. Bouchard, C. 1999. Physical activity, physical fitness, and coronary heart disease risk factors in youth: The Qubec Family Study. Preventive Medicine 29, 555-562. Koretz, D., Barron, S. I. 1998. The Validity of Gains on the Kentucky Instructional Results Information System (KIRIS). MR-1014-EDU, Sant a Monica: RAND. Ku, L., Sonenstein F.L., Pleck, J.H. 1993. Young men's risk behaviors for HIV infection and sexually transmitted diseases, 1988 through 1991. American Journal of Public Health 83(11), 1609-1615. Ku, L., Sonenstein F.L., Pleck, J.H. 1992. Factors influencing first intercourse for teenage men. Public Health Report, 108, 680-694 Lakdawalla, D., Philipson, T. 2002. The Grow th of Obesity and Technological Change: A Theoretical and Empirical Examinati on. NBER Working Papers No. 8946. Lakka, H.M., Laaksonen, D.E., Lakka, T.A., Niska nen, L.K., Kumpusalo, E., Tuomilehto, J., Salonen, J.T. 2002. The metabolic syndrome and total and cardiovascular disease mortality in middle-aged men. The Journal of the American Medicine Association 288(21), 2709-2716. 150

PAGE 151

Lerner, R. 2004. Can Abstinence Work? An A nalysis of the Best Friends Program. Adolescent and Family Health 3(4), 185-192. Lillard, D.R., DeCicca, P. 2001. Higher St andards, More Dropouts? Evidence Within and Across Time, Economics of E ducation Review 20(5), 459-473. Locard, E., Mamelle, N,, Billette A., Miginiac, M., Munoz, F., Rey, S, 1992. Risk factors of obesity in a five year old population. Pa rental versus environmental factors. Int J Obes Relat Metab Disord 16(10), 721-729. Mokdad, A.H., Ford, E.S., Bowman, B.A., Dietz W.H., Vinicor, F., Bales, V.S., Marks, J.S. 2003. Prevalence of obes ity, diabetes, and obesity-rela ted health risk factors." Journal of the American Medi cal Association 289, 76-79. Must, A., Spadano, J., Coakley, E.H., Field, A.E., Colditz, G., Dietz, W.H. 1999. The disease burden associated with overweight and obesit y. The Journal of the American Medical Associ ation 282(16), 1523-1529. Oettinger, G.S. 1999. The effects of sex educ ation on teen sexual activity and teen pregnancy. Journal of Politic al Economy 107(3), 606-644. Rouse, C.E., Hannaway, J., Goldhaber, D., Figlio, D.N. 2007. Feeling the Heat: How Low Performing Schools Respond to V oucher and Accountability Pressure. NBER Working Paper No. 13681. Schanzenbach, D.W. Forthcoming. Does t he Federal School Lunch Program Contribute to Childhood Obesity? Journal of Human Resources. Stunkard, A.J., Harris, J.R., Pedersen, N.L., McClearn, G.E. 1990. The bodymass index of twins who have been reared apart. T he New England Journal of Medicine 322(21), 1483-1487. Tremblay, C.H., Ling, D.C. 2005. AIDS education, condom dem and, and the sexual activity of American youth. Health Economics 14(8), 851-867 Trenholm C, Devaney B, Fortson K, Clark, M. Quay, L., Wheeler, J.. 2008. Impacts of abstinence education on teen sexual activity, risk of pregnancy, and risk of sexually transmitted diseases. Journal of Policy Analysis and Management 27 (2), 255-276. West, M. R., Peterson, P. E. 2006. The efficacy of choice threats within accountability systems: Results from legislatively i nduced experiments. The Economic Journal 116(510), c46c62. Woolston, J.L. 1987. Obesit y in infancy and early child hood. Journal of American Academy of Child & Adolesce nce Psychiatry 26, 123-126. 151

PAGE 152

152 Yin, Lu. 2009. Are School Accountability Syst ems Contributing to Adolescent Obesity? Working paper, November.

PAGE 153

BIOGRAPHICAL SKETCH After earning her bachelors degree in finance from Beijing Capital University of Economics and Business in China, Lu worked as an associate auditor in Pricewaterhouse Coopers for one and half years. In 2005, Lu started graduate school in economics at University of Florida. Her fields of specializat ion are economics of education, health economics and applied econometrics. She received her Ph.D. from the University of Florida in the summer of 2010. 153