1 DIFFERENTIAL ASS OCIATION WITH DELINQUENT PEERS: AN INVESTIGATION OF THE CONSEQUENCES OF USING ALTERNATIVE MEASURES OF DELINQUENT PEERS By JOHN H. BOMAN, IV A PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERS ITY OF FLORIDA IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF MASTER OF ARTS UNIVERSITY OF FLORIDA 2010
2 2010 John H. Boman, IV
3 To my mother and my father, for the endless support, love, and encouragement that they have provided me
4 ACKNOWLEDGMENTS I would first like to express my extensive appreciation to my chair, Dr. Chris Gibson, for his time, efforts, and I would like to express the same gratitude to the members of my committee Dr. Ron ald Akers and Dr. Lonn Lanza Kaduce for their assistance and guidanc e in completing this project. In addition, I would like to thank Dr. Marvin Krohn for his guidance support, and interest in my research I would also like to extend my gratitu de to the many research assistants that have helped me with this project. I would like to especially thank Paula Silva for her exhaustive efforts with this project. I extend my gratitude to Alvin Rameau and Giovanni Vidal as well for their high quality, detailed, and timely work. I am further infinitely indebted to my loving fianc e Stephanie Diane Ksionzyk for her assistance with editing, organizing, and providing support during this project. If there are any commas that have been omitted, please blam e her
5 TABLE OF CONTENTS page ACKNOWLEDGMENTS ................................ ................................ ................................ .. 4 LIST OF TABLES ................................ ................................ ................................ ............ 7 LIST OF FIGURES ................................ ................................ ................................ .......... 8 LIST OF ABBREVIATIONS ................................ ................................ ............................. 9 ABSTRACT ................................ ................................ ................................ ................... 10 CHAPTER 1 INTRODUCTION ................................ ................................ ................................ .... 12 Introduction to the Delinquent Peers Construct ................................ ...................... 12 Past Methods of Peer Delinquency Measurement ................................ .................. 15 Current Study ................................ ................................ ................................ .......... 20 2 REVIEW OF RELEVANT LITERATURE ................................ ................................ 25 Background of the Peer Delinquency Debate and Common Measurement Problems ................................ ................................ ................................ ............. 25 Dyadic Studies and Friendship Traits: A Criminological Application ....................... 32 Past Measurement of Peer Delinque ncy: A Tale of Different Trails ........................ 39 Summary of Friendships and Past Measurement of Peer Delinquency .................. 51 Friendships and Social Net works. ................................ ................................ .... 51 Measurement of the Delinquent Peers Construct. ................................ ............ 52 Research Questions and Hypotheses ................................ ................................ ..... 55 Research Question One. ................................ ................................ .................. 55 Research Question Two. ................................ ................................ .................. 56 Research Question Three. ................................ ................................ ............... 57 Research Question Four. ................................ ................................ ................. 58 3 RESEARCH DESIGN AND ANALYTICAL STRATEGY ................................ .......... 64 Researc h Design ................................ ................................ ................................ .... 64 Sample. ................................ ................................ ................................ ............ 64 Procedures. ................................ ................................ ................................ ...... 69 Measures ................................ ................................ ................................ ................ 72 Delinquency. ................................ ................................ ................................ ..... 72 Self reported delinquency ................................ ................................ .......... 73 Perceptual peer delinquency ................................ ................................ ...... 74 Groupy crimes ................................ ................................ ............................ 75 Individualistic crimes ................................ ................................ .................. 76
6 Property crimes ................................ ................................ .......................... 77 Violent crimes ................................ ................................ ............................ 77 Substance use ................................ ................................ ........................... 78 Minor delinquency ................................ ................................ ...................... 78 Friendship Quality ................................ ................................ ............................ 79 Analytic Strategy ................................ ................................ ................................ ..... 81 Research Question One. ................................ ................................ .................. 82 Research Question Two. ................................ ................................ .................. 83 Research Question Three. ................................ ................................ ............... 85 Research Question Four. ................................ ................................ ................. 86 4 RESULTS ................................ ................................ ................................ ............... 92 Research Question One. ................................ ................................ ........................ 92 Research Questi on Two. ................................ ................................ ........................ 93 Research Question Three. ................................ ................................ ...................... 96 Research Question Four. ................................ ................................ ........................ 99 5 DISCUSSION ................................ ................................ ................................ ....... 112 6 CONCLUSION ................................ ................................ ................................ ...... 121 APPENDIX A IRB PROTOCOL, IRB APPROVAL, AND IRB REVISION APPROVAL ................ 128 B INFORMED CONSENT ................................ ................................ ........................ 134 C ORIGINAL SURVEY INSTRUMENT ................................ ................................ .... 136 D DEBRIEFING FORM ................................ ................................ ............................ 151 E LIST OF DELINQUENCY ITEMS, BROKEN DOWN BY SUBCATEGORY .......... 153 F CORRELATIONAL MATRIX OF ALL SCALES USED IN ANALYSES ................. 155 LIST OF REFERENCES ................................ ................................ ............................. 157 BIOGRAPHICAL SKETCH ................................ ................................ .......................... 163
7 LIST OF TABLES Table page 3 1 Demographic characteristics of subject and friend in dyadic sample (n = 485). ................................ ................................ ................................ ................... 88 4 1 delinquency ................................ ................................ ................................ ..... 101 4 2 reported groupy delinquency onto th groupy delinquency. ................................ ................................ ......................... 1 02 4 3 s perception of the ................................ ................................ ... 103 4 4 property delinquency. ................................ ................................ ....................... 104 4 5 violent delinquency. ................................ ................................ .......................... 105 4 6 substance use. ................................ ................................ ................................ 106 4 7 delinquency. ................................ ................................ ................................ ..... 107 4 8 P rinciple Components Analysis of the internal component structure of the reported delinquency measure. ................................ ................................ ...................... 109 4 9 reported delinquency. ................................ ................................ ................................ ..... 110 4 10 Ordinary Le reported moderation effects. ................................ ................................ ........................... 111
8 LIST OF FIGURES Figure page 2 1 Research Question One. ................................ ................................ .................... 60 2 2 Research Question Two. ................................ ................................ .................... 61 2 3 Research Question Three. ................................ ................................ ................. 62 2 4 Research Question Four. ................................ ................................ ................... 63 3 1 Conceptua l model of the scales and subscales of the Friendship Qualities Scale (taken from Bukowski et al., 1994, p. 474). ................................ ............... 90 3 2 T he 23 item Friendship Qualities Scale (taken from Bukowski et al., 199 4, p. 475). ................................ ................................ ................................ ................... 91 4 1 reported delinquency. ................................ ... 100 4 2 reported delinquency. .................... 108
9 LI ST OF ABBREVIATIONS FQS The FQS is the Friendship Quali ties Scale (Bukowski, Hoza and Boivin, 1994). The FQS is theorized to moderate the ability for a reported delinquency. IRB IRB stands for Institutional Review Board. The IRB is the body that overse es original research on campus. PPD PPD stands for perceptual peer delinquency. While not noted this way in the text, the PPD abbreviation is used in several Figures and Tables. SRD SRD stands for self reported delinquency. While not noted this way in t he text, the SRD abbreviation is used in several Figures and Tables.
10 Abstract of Presented to the Graduate School of the University of Florida in Partial Fulfillment of the Requirements for the Degree of Master of Arts DIFFERENTIAL ASSOCIATION WITH DELINQUENT PEERS: AN INVESTIGATION OF THE CONSEQUENCES OF USING ALTERNATIVE MEASURES OF DELINQUENT PEERS By John H. Boman, IV May 2010 Chair: Chris Gibson Major: Criminology, Law and Society The construct of peer delinq uency is of vital importance to the field of criminology. Having delinquent friends is one of the most consistent individual level predictors of over how to measure the cons Most notably, Gottfredson and Hirschi (1987; 1990) have criticized the use of a perceptual delinquent peer measure due to the concern that the measure the subject produces of delinquent peers is merely another measure of self reported delinquency. Past measurement techniques of the peer delinquency construct typically request one research subject to report on the delinquency of an a bstract group of friends. While this practice is quite common, it is largely unclear whether a research subject is capable of perceiving a measure of the delinquency of their peers that is in fact reflective of their The current study seeks to explore how to measure the delinquent peers construct and determine whether past mea surement methods of peer delinquency have actually been tapping the construct intended. Four main findings are sought. The first asks if a
11 perceptual peer delinquency measure produced by a subject is s self reported delinquency. Second, which operationalizations of delinquency are most reported delinquency? The third finding explores whether a reported delinquency is the same as his or her perception of uency. Finally, the last research question inquires as to whether the quality of the friendship between the friends moderates the level to which a subject can Using original, dyadic, best friend data, it is found th at delinquency is reported delinquency under reported deviance measure is strongly related to his/h er measure of perceptual peer deviance but not to the extent that they should be viewed as the same construct. The quality of the friendship does not appear to moderate the level to which a subject can criminality Implications f or criminological theory and measurement of the delinquent peers construct are discussed in detail.
12 CHAPTER 1 INTRODUCTION Introduction to the Delinquent Peers Construct It has long been the goal of science to aggressively pursue knowledge and report n ew findings that build on prior research. In order to properly do this, constructs, theories, and hypotheses must be tested as efficiently and accurately as possible. Measurement, for both physical and social science disciplines, is undoubtedly one of th e most crucial components of all methodological research (Thornberry & Krohn, 1997). Some scientific disciplines have extensively utilized measures of certain constructs that are assumed to be valid but have undergone little empirical analysis. Criminol ogy is no exception to this. Peer delinquency is one particular construct that has been used prolifically in criminological research despite a dearth of knowledge as to what the construct is actually evaluating. Often researchers have asked whether deli nquent peers have an influence on various behavioral outcomes without empirically questioning how the construct is being measured (see Gottfredson & Hirschi, 1987; 1990; Thornberry & Krohn, 1997). This has resulted in many attempts to understand the delin quent peer effect before asking the question of how it should be operationalized. The peer delinquency construct is recognized as one of the first robust findings on delinquency that the field of criminology produced. The concept was first purported to b e a major predictor of deviance in differential association theory (Sutherland, 1939; 1947). Sutherland claims that crime is the result of learning definitions that are consistent with the motives, attitudes, and values of others. Through the peer group it
13 was proposed that one would learn the definitions that are favorable to criminal behavior. Associates with whom one has the highest level of friendship intensity are theorized to affect the behavior of an individual the most. Thus, having close friend s who are delinquent should increase the likelihood of one being delinquent himself, while having non delinquent friends should reduce the likelihood of becoming involved in delinquency. Social learning theory, proposed by Burgess and Akers (1966), incorp orated and deviant behavior. Social learning variables have been found to be consistent and haracteristic of individuals known to criminologists is a better predictor of criminal behavior than the delinquent peers construct produces such a consistent effect that no t including it as a control in predicting delinquency and/or criminal behavior could result in a misspecified model with biased results (see Agnew, 1991). While criminologists seem to have accepted this fact and control for the delinquent peer effect reg ularly, questions have been raised as to the appropriateness of the most common form of measurement of the delinquent peers variable (Agnew, 1991; Asteline, 1995; Gottfredson & Hirschi, 1987; 1990; Haynie, 2001; 2002; Haynie & Osgood, 2005; Kandel, 1996; M cGloin, 2009; McGloin & Shermer, 2009; Thornberry, Lizotte, Krohn, Farnworth, & Jang, 1994). To date, the most standard form of measuring the delinquent peers construct is by asking a person to report on the delinquency of their associates (hundreds of st udies have used this measure, if not
14 more; see, for ex ample, Pratt & Cullen, 2000). While t his produces a perception of their this perception actual delinquency. Such a vast number of studies have q uantified delinquent peers in this form that researchers are seemingly always justified in including such a measure without question. While having been questioned, an empirical evaluation of whether a ng the construct that it claims has not yet been performed. Several researchers have deduced that measuring the construct through a perceptual measure may be problematic (Asteline, 1995; Gottfredson & Hirschi, 1987; 1990; Haynie, 2001; 2002; Haynie & Osgo od, 2005; McGloin, 2009; McGloin & Sherm er, 2009; Weerman & Smeenk, 2005 ). Concern regarding this measure was first proposed by Gottfredson and Hirschi (1987). Gottfredson and Hirschi claim (1987), nquent activities reported for friends are the same delinquent activities previously reported by the respondent for reported peer delinquency is just another measure of self the subject elf reported delinquency and their measure (Asteline, 1995; Weerman & Smeenk, 2005). While no studies have specifically addressed this topic, several have peripherally repo rted findings relevant to the claim of projection. A pair of large studies has
15 reported delinquency are related, the two are distinct constructs (Agnew, 1991; Thornberry et al., 1994), but other studies call these findings into question. Some speculate that the measures are so interrelated that they cannot be distinguished from one another (see Asteline, 1995; Gottfredson & Hirschi, 1987; 1990). To this point, that primary deviance leads to delinquent peer associations, and those associations fuel further delinquency (Agnew, 1991; Warr, 1993; Krohn, Lizotte, Thornberry, Smith, & McDowall, 1996). Other research has concluded that a perceived peer delinquency reported delinquency (Weerman & Smeenk, 2005; see Asteline, 1995). In short, criminologists do not have a clear un derstanding of whether a person self reported deviance. To this point, it is unknown whether certain criminal actions of a friend may be reported less precisely by a subject due to the individual nature of some crimes (e.g., date rape) as opposed to group based deviance (e.g., vandalism) that the subject is more likely to know about. There also exists the possibility that the capacity of a subject to reflectively report on the delinquencies of a peer is heavily depende nt on the intensity of the friendship that the acquaintances share. Empirical research on these issues has been limited in criminology, although they are central to testing the validity of perceived peer delinquency measures. Past Methods of Peer Delinque ncy Measurement The first, most common and potentially most problematic way of measuring
16 the collective behavior of a group of abstract friends. Asteline (1995) puts this problem the major independent variable of peer influence theories involvement with delinquent friends perceptions of their 104, emphasis in original). perception of the quantity of friends that are deviant (and even sometimes the frequency to which they exert that deviance). The Nationa l Youth Survey (see Agnew, 1991; Matsueda & Anderson, 1998; Menard & Elliott, 1990a; 1990b; Warr, 1993), Richmond Youth Study (see Hirschi, 1969), Monitoring the Future (see Bachman et al., 2002; ley, Bachman, & Johnston, 1983 ), Dunedin Multidisciplin ary Health and Development Study (see Bartusch, Lynam, Moffitt, & Silva, 1997) and Boys Town (see Akers, Krohn, Lanza Kaduce, & Radosevich, 1979; Lanza Kaduce, Akers, Krohn, & Radosevich, 1984; Radosevich, Lanza Kaduce, Akers, & Krohn, 1979) studies all me asure peer delinquency in this way, as do a very large number of cross sectional studies that are too numerous to reference. T his is an advantageous method of considerat ions. While this form of measurement is easy and quick to obtain, it has several serious drawbacks that may prevent social scientists from drawing accurate conclusions of how delinquent peers influence behavior. Gottfredson and Hirschi (1987; 1990) discre dit the use of obtaining information reported
17 delinquency twice. If their inclination is correct, t hen using this type of method for Stated differently, if a subject reports on the behavior of a group of friends but unknowingly reproduces a scale that is identical to their self reported delinquency, then the independent and dependent variables become synonymous. Of course, if this is true then results of countless prior studies controlling peer delinquency could be flawed due to the inclusion of a tautology (see Thorn berry & Krohn, 1997). there is no method through which it is possible to determine if that perception is accurate. Thus, the use of this measure relies solely on the assum ption that the common form of measurement of delinquent peers is evaluating what it is purported to. There are several other ways that studies have operationalized this construct. The second way that peer delinquency measures have been obtained is much less common and has been done largely through network analyses and longitudinal research. In these studies, the subject is typically asked to identify a set number of best ure. The difference here is that the subject names the best friends, typically three to five, and provides a general delinquency assessment of those individuals. Often members of that friendship network are also sampled, which allows for the comparison i n measures of peer delinquency measurement can be seen in the method in which the Rochester Youth Development Study implements (see Krohn et al., 1996; Thornberry et al .,
18 1994). 1 It is the purpose of these studies not to measure and evaluate how the peer delinquency construct works in detail, but rather to control for peer delinquency in a more in depth way than is typical in cross sectional research. But again, while t his method of controlling for delinquent peers is clearly more advantageous than the first method in that it can tell us much more about who the friends are and about their behaviors, this genre of peer delinquency measurement finds itself in the same situ ation that the first does. There is no method of determining who is responsible for the reported delinquencies individually when someone reports on the delinquencies of a group of people. As a result, the possibility exists that the results of these stud ies fall identified deviance. While self reported criminality measures in these studies do often exist from others within that peer group, subjects still are asked to report on their specific peer delinquency, thus eliminating the possibility that the measures can be used to determine if the perceptions ctual behavior In order to devise whether a perceptual peer delinquency measure is valid, measures of specific delinquencies of specific friends must be gathered. The most advantageous research design to determine the accuracy of a 1 delinquency with an actual report of delinquency from that individual (Haynie, 2001; 2002; Haynie & O sgood, 2005; McGloin, 2009; McGloin & Shermer, 2009; see also Asteline, 1995, who has similar measures from another dataset). While these studies do not have a perceived delinquency measure, they still provide unique insight into the peer delinquency cons truct and are further elaborated on in the forthcoming chapter.
19 behavior. Very few studies have met this criterion (see Kandel, 1978a; 1978b; Weerman & Smeenk, 2005). This me thod of measuring delinquent peers is by far the most time consuming, but provides the best method, to date, for assessing the peer delinquency construct measurement debate. Weerman and Smeenk (2005) used data from a sample of underprivileged Dutch high school students. This study used measures of actual and perceived peer bolste rs my expectation that the delinquencies of a friend may not fully be known by a subject simply because some criminal acts are best committed in an individual context. Although the authors of this study offer no explanation for the underreporting of devia nce on the perceived peer delinquency scale, I expect that individualized deviances will represent the majority of the underreporting of delinquency on a combined deviance scale. Also, their measures of perceived peer delinquency and the re ported delinquency are very similar, which lends support to the claims of projection from Gottfredson and Hirschi (1987; 1990). Although among the most advanced in terms of peer delinquency measurement, this study has several significant limitations that are thematic to this body of research that will be addressed in great detail in the next chapter. In the body of research that includes actual and perceived delinquency measures there are several notable shortcomings, but the most looming one is that ther e lacks a quantitative way to evaluate the intensity of the fri endship. This measure indicates two
20 different things. First, it signifies friendship reciprocity because congruent answers indicating a high level of companionship between the subjects displa y their collective liking for the other. Furthermore, an appropriate measure of friendship intensity should be multidimensional. It should not only measure friendship involvement, for instance, as some studies do (Haynie, 2001; 2002; Haynie & Osgood, 200 5; Kandel, 1978a; 1978b; McGloin 2009; McGloin & Shermer, 2009) but rather evaluate multiple sectors of the relationship. Friendships vary greatly in intensity (Bukowski, Hoza, & Boivin, 1994; Brendgen, Markiewicz, Doyle, & Bukowski, 2001; Gavin & Furma n, 1996; Saferstein, Neimeyer, & Hagans, 2005) and reciprocity (Kandel, 1978a; 1978b), indicating that such a measure should be a central part of a study designed around dyadic relationships and social networks. Current Study To my knowledge, a study has not been performed to date that is designed to self reported delinquency. However, this is not to say that such a study has not been desired by the academic community. Ra ther, researchers have attempted to answer this question by utilizing datasets that simply are not suitable to satisfy such a query. A leading social network and dyadic researcher has noted that in order to properly test this research question a study mus t be developed specifically for this purpose (see Kandel, 1996). To this end, the importance of the current study is two fold. First, this study empirically delinquency is a va lid measure of peer delinquency. Many studies have utilized a
21 peers, and as a field we have allowed this to go empirically unquestioned while largely assuming that this sc ale is in fact performing as we think. To advance our understanding of this measure, a much more rigorous test of the delinquent peers construct will be performed for the current study. I will examine which types of delinquency (e.g., property, violence, etc.) that a friend can identify efficiently as well as what demographic and friendship circumstances result in more concise reporting of peer delinquency. I hypothesize 1) that individuals in friendships higher in reciprocal intensity will show an incre delinquencies. measure of a reported delinquency (the projection hypothesis) will also be evaluated. The projection argument is a major criticism of the use of perceived peer delinquency scales and has not been prope rly evaluated to date. Being that this study strives to determine the validity of a perceived peer delinquency measure, it is crucial that the possibility of projection occurring be addressed. Not testing this hypothesis leaves an obvious gap in the resu lts as this claim is consistently acknowledged in literature (Agnew, 1991; Asteline, 1995; Haynie, 2001; 2002; Haynie & Osgood, 2005; Kandel, 1996; McGloin, 2009; McGloin & Shermer, 2009; Thornberry, Lizotte, Krohn, Farnworth, & Jang, 1991). Prior studies that have reported on this (with mixed results and clearly no conclusion) have used insufficient data to address this issue. The data for the current study were specifically collected to answer research questions related to the accuracy of a
22 commonly use d perceived peer delinquency measure. In order to thoroughly evaluate the delinquent peers construct, it is necessary to correct for exclusions and shortcomings of past investigations. Existing research on delinquent peer measurement has excluded several crucial constructs. First, past studies have included measures of delinquency that do not evaluate a wide range of deviant behaviors. It is surprising that they have done so because of the fact that deviance normally serves as the dependent variable in o ur discipline. To this end, this thesis will seek to advance existing knowledge of the delinquent peers construct not only b y assessing whether someone can evaluating a wide range of both individual and group bas ed delinquencies. It can reasonably be hypothesized that a person may not know the full extent of the deviance of a best friend due to the fact that certain behaviors, such as date rape, are committed individually. Including such a wide range of items on the deviance scales allows for exploration as to which delinquencies of a friend a respondent can (and cannot) provide reported delinquency. Developing an understanding of which items a subject can re port for a friend will provide direction as well as justification for researchers that they are measuring the delinquent peers variable as efficiently as possible. A second shortcoming of prior research on the delinquent peers construct is that it fails to include a seemingly obvious variable a quantitative way to evaluate the intensity of friendship between the subjects. I have included a measure of sorts (the Friendship Qualities Scale; see Bukowski et al., 1994) in this research. The current study will determine if a subject is more capable of producing a measure of perceptual
23 reported delinquency under varying levels of friendship intensity. It is hypothesized that members of a friendship wit h elevated intensity will show perceptual reports that are more reflective self reported behaviors than those with low friendship intensity The inclusion of this measure is important because to my knowledge this is the first time in crimi nological research that a detailed measure of the intensity of a friendship has been included. Due to the uniqueness of the inclusion of this measure, the results have the possibility of producing an original and noteworthy contribution to peer delinquenc y measurement. This study used primary data collection and sampled a large number of self identified, best friend college student pairs. Potential respondents were asked to bring bers of the dyad were placed, at the same time, in separate areas and were administered identical paper based survey instruments. The dyad responded to a self report delinquency scale as well as an identical scale indicative of their perception of their f over the past year. Additionally, both members completed a scale indicating the level of intensity of their friendship. Participants were compensated with extra credit in any of more than twenty classes they may have be en enrolled in acr oss the university for the spring semester of 2009. Not all subjects were directly compensated due to some not being enrolled in any selected classes instead they came to the study to help their friend receive credit. Due to Institutional Review Board restrictions, only college students 18 years and older were sampled. The following chapter provides a summary and critique of past dyadic and peer network studies in criminology with specific attention given to the measurement of the
24 delinquent peers con research design and analytical plan will be presented, as well as chapters on results and discussion of them.
25 CHAPTER 2 REVIEW OF RELEVANT LITERATU RE Background of the Peer Delinquency Deba te and Common Measurement Problems Despite a fairly sizeable body of literature that addresses friendship pairs, networks, and influences, it remains unclear whether a su bject is capable of providing a valid es (see Agnew, 1991; Asteline, 1995; Gottfredson and Hirschi, 1987; 1990; Haynie, 2001; 2002; Thornberry and Krohn, 1997; Thornberry et al., 1994; Warr, 1993; Weerman and Smeenk, 2005). Past studies have had common methodological design problems that have hindered their abilities to completely test whether a perceived peer delinquency measure is in fact a valid reported delinquency. Additionally, no prior literature has sought to determine which types of peer delinquen cy a subject can most accurately identify and no research has addressed how varying levels of friendship intensity may /her although by no means all of them, end up repo footnote (see Agnew, 1991; Thornberry et al., 1994; Warr, 1993). Determining how a perceived delinquency measure performs in regards to an reported delinquency is both a significant empirical and theoretical issue. The peer delinquency construct was originally proposed in the theory of differential association (Cressey, 1955; Sutherland, 1937; 1940; 1947; Sutherland and Cressey, 1978) and is utilized as a major construct of social learning theory (Burgess and Akers, 1966). Several studies have called into question the ability of a research
26 1995; Gottfredson and Hirschi, 1987; 1990; Weerman and Smeenk, 2005), but it is the work of Gottfredson and Hirschi that has put this argument to the forefront. When incorporating a perceived delinquency measure of a friend, the concern is that the measu re could itself be reported behavior are the same as reported peer delinquency is just another measure of s elf p. 597). The authors provide several different speculations as to why it could be expected that a perceived peer delinquency measure could be the same as a self report ed delinquency measure: questions about the delinquencies of their friends. Several possibilities come to mind: (1) the respondent may have been at the scene, himself engaging in the activi ty; (2) the respondent may impute his own qualities to his friends; (3) the respondent may impute friendship to people like delinquencies he did not himself witness; and (5) the respondent may h itnessed (or heard about) them. ( p. 598) Gottfredson and Hirschi (1990) elaborate on the notion that the two apparently distinct measures could be the same thing in an attempt to deny the feasibi lity of differential Akers, 1966; see Akers et al., 1979) abilities to explain deviance: a creati on of faulty measurement and the tendency of people to seek the co mpany of others like themselves ( p. 156) differential association and social learning theories in the p ast would have suffered from
27 misspecified models and would themselves have been measuring a tautology in that reported delinquency is typically regressed against a perception of their their own delinquency. The argument of the perceived peer delinquency measure being circular has been acknowledged as a potential problem in other literature (see Agnew, 1991; Asteline, 1995; Thornberry, 1987; Thornberry et al., 1994). Notably, Astelin e (1995) echoes similar concerns over the possibility that an theories involveme nt with delinquent friends perceptions p. 104; emphasis in original). This assertion is correct. Social learning theory has been measured in a fairly similar pattern, al beit with exceptions, for most o f its history since the original development in 1966 by Burgess and Akers. A potential outcome of implementing such a measurement strategy of the delinquent peers construct could produce tentatively overestimated effects in a social learning or differential association based model: behavior and thus are clearly biased toward confirmation of differential association and subcu lture theories (Asteline, 1995, p. 104) While such an emphatic statement may or may not be correct, research designed to test whether or not a subject is c behavior reported behavior is something of great importance to criminologists, especially considering that a vast majority of studies have succumbed
28 to the notion tha because this variable may have a causal effect on both delinquency a (Agnew, 1991, p. then ire body of research assessing the relationship between association with deviant peers and self reported deviance would be useless because the measures are not independe p. 222). It should be remembered, though, that despit e the projection hypothesis being fairly well known it is only one piece of the bigger picture of peer delinquency measurement. It is altogether unclear as self reported behavior, across which circumstances a correct perception is most likely, and which crimes are hypothesis for an initial finding to the bigger question of how, when, and which measures of perceive d peer delinquency are accurate. Measuring delinquent peers with a perceptual measure is extremely common in the criminological literature. Valid and efficient use of proven measures should be a to have assumed that a construct. Denise Kandel (1996, p. 297 298), a leading dyadic and network researcher, points to the fact that to measure peer delinquency accurate ly, a study must be specifically designed with this purpose in mind: Existing data are truly inadequate to estimate the extent to which attribution and projection distort and inflate the estimates of peer influence. Optimally, identical information should be available for all three reported behavior. To the best of my knowledge, such data are rarely collected
29 Although fairly few r esearchers have followed this advice, the literature does include several studies that have used specific reports of actual and perceived peer delinquency (Bauman and Fisher, 1986; Weerman and Smeenk, 2005), but only one of which evaluates the perceived me asure validity. Bauman and Fisher (1986) measured cigarette smoking and alcohol consumption for minors for both members of the friendship. This study concludes that the ability for close accomplices to accurately s dependent on a number of unknown factors. Unfortunately, this study did not take into account any of these potential unknown factors such as friendship intensity from the subject, reciprocal friendship intensity from ures of delinquency and deviance were the most accurately reported. Another design limitation is that the Bauman and Fisher (1986) study administered take home surveys in which a subject was asked to identify one of their three best friends and answer ques tions about each specific one. Being that the respondent was (most likely) at least decent friends with the person they answered questions about, the friendship pair could have communicated either in person, over the telephone, or at school the next day a bout the ir responses before the instrument was handed in to the researchers. This could obviously give rise to the potential for data contamination between the friendship pairs. Thus, the results from their study should be interpreted cautiously as they could be overstated or biased based on a number of characteristics that could include socio geographical location of the friends in relation to each other, or any number of other unknown demographic or character based traits that are contained within the stochastic
30 error term. In sum, no dyadic based or network analysis to date has been able to provide a definitive answer to w hether a subject can provide a representation of his/her delinquent behaviors t reported behavior due to having one or more typical problems. The most common problems are that a study will measure (1) a perception of the delinquency of an abstract group of non specific friends as a whole (Akers et al., 1979; Agnew, 1991; Krohn et al., 1996; Lanza Kaduce et al., 1984; Radosevich et al., 1979; Thornberry et al., 1994; Warr, 1993) or of a group of specific friends (Weerman and Smeenk, 2005); (2) the study measures an actual report of a specific frie vior behavior (Asteline, 1995; Haynie, 2001; 2002; Haynie and Osgood, 2005; McGloin, 2009; McGloin and Shermer, 2009; Kandel, 1978a; 1978b; Kandel and Davies, 1991); (3) the study fails to measure the intensity of the friendship (Asteline, 1995; Bauman and Fisher, 1986; Kandel, 1978a; 1978b; Kandel and Davies, 1991; Haynie, 2001; 2002; Haynie and Osgood, 2005; Krohn et al., 1996; McGloin, 2009; McGloin and Shermer, 2009; Thornberry et al. 1994; Weerman and Smeenk, 2005); (4) the sample is too homogenous and lacks diversity nearly completely; (Asteline, 1995); (5) the study treats non reciprocal friendships as reciprocal friendships (Asteline, 1995; Bauman and Fisher, 1986; Haynie, 2001; 2 002; Haynie and Osgood, 2005; Krohn et al., 1996; McGloin, 2009; McGloin and Shermer, 2009; Thornberry et al., 1994; Weerman and Smeenk, 2005), or (6) the data collected between the friendship pairs could be contaminated due to having been collected at eit her different times or in the same physical space (e.g., a classroom) (Asteline, 1995; Bauman and Fisher, 1986; Haynie,
31 2001; 2002; Haynie and Osgood, 2005; Kandel, 1978a; 1978b; Kandel and Davies, 1991; Krohn et al., 1996; McGloin, 2009; McGloin and Sherm er, 2009; Thornberry et al., 1994; Weerman and Smeenk, 2005). To properly understand the full nature of differing forms of peer delinquency measurement, this chapter seeks to cover findings from prior studies of characteristics of friendships, such as simi larities and differences in substance use, demographics, minor delinquency, and more serious delinquency. Past measurement strategies of how actual peer delinquency, perceived peer delinquency, self reported delinquency and friendship intensity will also be summarized and evaluated. valid is not only an important empirical finding for controlling peer delinquency in future studies, but also a central theoretical issue with i mplications for social learning and differential association theories. There is no contention about whether peer delinquency has an effect on criminal behavior. However, there is contention on how peer delinquency has been measured and whether the effect s observed have been spurious or even tautological based on improper measurement strategies. To this point, it is unknown under what circumstances of friendship intensity a perceived measure of a most rep orted deviance as well as on which crimes a subject can more precisely report, both of which are major exclusions from prior literature. It is ominous to think how the results of past studies of (or controlling for) peer delinquency could be altered based on the employment of a measure that is potentially not measuring what it claims The section that follows addresses characteristics of friendships from past dyadic and network based studies.
32 Dyadic Studies and Friendship Traits: A Criminological Applica tion While dyadic data is somewhat rare in a criminological context, Denise Kandel, a professor in the Mailman School for Public Health at Columbia University, has authored several studies based on dyadic data that have proven to be of importance to the cu rrent study (Kandel, 1978a; 1978b; 1996; Kandel and Davies, 1991; Kandel and Lesser, 1972). Specifically, one study should be mentioned and elaborated on. Kandel and Davies (1991) compared two theories of criminal behavior using a large sample of adolesc ent dyads that were originally sampled in New York high schools in 1971 and continued for three waves across ten years. Social control theory (Hirschi, 1969) was compared with social learning theory (Burgess and Akers, 1966) to determine which theory bett er predicted how drug users interacted socially. Social control theory would propose that drug users are not bonded well with society and that their friendship networks are either non existent or display very low levels of homophily. Alternatively, socia l learning theory proposes that drug users have friendship networks that are just as intimate as nonusers and that their friendship networks should be highly interconnected. Kandel and Davies (1991) used a three wave longitudinal sample of students colle cted originally in high school and followed to the age of 30 with a 75% retention rate. Friendship similarity, the finding of interest, was measured across seven categories. However, the authors may have been limited in their measure of friendship. In o rder to identify which participants were friends with other participants, respondents were only asked to write down their three best friends at wave one not waves two or three. The underlying assumption is that the relationships that were identified at the first sampling period remained intact for nearly 15 years. This is not the best way to
33 measure friendships given that some evidence has more recently shown that friendships are quite fluid (Haynie, 2001; 2002; Haynie and Osgood, 2005; McGloin and Sher mer, 2009), thus emphasizing the need for a more detailed measure of fondness between research subjects. The authors established that social control theory was not an adequate representation of the friendship networks that drug users have. Drug user net works were found to be the same size as nonuser networks and there was no significant difference in intimacy between user and nonuser women. Men, however, tended to show higher levels of relationship intimacy among user groups than they did nonuser groups important and statistically significant determinant of intimacy in all three types of p. 457 459). Women were found to have higher levels of intimacy overall, too, which echoes other findings in the field regarding intimacy (see Bukowski et al., 1994; Giord ano, Cernkovich, and Pugh, 1986 ). il legal status of [drug use] may further reinforce the formation of bonds among drug p. 460). Other studies of friendship networks have found that social learning and cultural deviance theories are the best predictors of behavior (Fraser and Hawkins 1984a; 1984b; Giordano et al., 1986 ; Kandel, 1973; 1978a; 1978b; Krohn, 1986; Krohn, Massey, and Zielinski, 1988). While her research is not specifically designed to address whether or not a perceptual delinquency measure of a peer is accurate, her findings and measurement are of
34 thesis uses the Bukowski et al. Friendship Qualities Scale (19 94), a tested and reliable multidimensional indicator of friendship intensity. Akin to several studies that measure friendship involvement (a unidimensional quality; see Haynie, 2001; 2002; Haynie and des a very similar unidimensional scale that includes five items measuring the amount of time spent with best friends (see Kandel, 1978a, p. 309). It should be noted that a unidimensional measure should not be misclassified as a measure of friendship inte nsity because of the multidimensional nature of friendships (see Bukowski et al., 1994). Instead, a properly developed, tested, and proven measure should be used to represent this particular construct. To the immediate relevance of this thesis, no exist ing studies within criminology include a multidimensional measure of friendship intensity, which makes efficiently Several other studies have examined friendship dyads and how friends relate to one another. Kandel (1978a) measured the level of similarity among adolescent friendship dyads (91% were same sex friends). Using a sample of 1900 high school best friend dyads (defin identified best friend and the subject himself), it was found that four demographic factors showed very high similarity school progression (grade level), sex, race and age, respectively. The findings, however, may have been over stated due to the similarity of age and school progression that is inherent in the high school culture. Despite this, besides demographic characteristics marijuana use and minor delinquency were very similar among the fri endship pairs. For
35 the purpose of this thesis, it is expected that a subject will be able to report on certain peer b ehaviors more reflectively due to them being group activities (such as minor delinquency and recreational marijuana use). Students that used marijuana needed of the peer group for their initiation as well as th 1978a, p. 311). Marijuana usage was found to be very similar between close friendship dyads in several other studies as well (Giordano et al., 1986; Kandel 1978b; Kandel and Davi es, 1991; Kandel and Lesser, 1972). that the measures of friendships are self defined. In other words, a subject identifies a best friend through a fill in the blank line bu t only in 38% of their overall sample did the friendships showed some bias. Individuals with unreciprocated best friendships had heavier minority status, lower grades higher drug use, and females outnumbered males significantly, all of which are potentially serious limitations. Where Kandel advances literature, though, is that she treats reciprocal friendships as quantitatively different from non reciprocal friendshi ps. If a subject identified a best friend and the best friend identified the subject reciprocally as best friends, they were included in the analysis. But, if one identified a different best friend that pair was excluded. Kandel recognizes that treating reciprocal and non reciprocal friendships as quantitatively the same is not a safe assumption, as findings from some studies indicate that friendships are often very studies have, however, committed the error of treating non reciprocal and reciprocal friendships the same (Asteline, 1995; Bauman and Fisher, 1986; Haynie, 2001; 2002;
36 Haynie and Osgood, 2005; Krohn et al., 1996; McGloin, 2009; McGloin and Shermer, 2009; Thornberr (1978a; 1978b; Kandel and Davies, 1991). However, the method through which respondents were asked to identify a best friend was insufficient. They were simply asked to list a best friend, which is a potential shortcoming due to the inherent assumption that all reciprocal friendships are high in intensity. This point brings up the need of an actual, proven measure of friendship intensity that the subject completes for each of his identified best friends in future longitudinal network research. When discussing friendship networks, homophily is undoubtedly a major factor that should be taken into account. Homophily, a term originally conceived by Lazarsfeld and Merton (1954), can be laconic ally described as the tendency of individuals to associate and bond with others that share common characteristics with them the very high across best friend dyads and s atisfied intimate relationships, especially in regards to age and race (Hu, Davies, and Kandel, 2006; Kandel, 1978a; 1978b; Kandel and Davies, 1991; McPherson, Smith Lovin, and Cook, 2001). Homophily has been found to vary in certain sectors of the population. Friendships that are newer are typically lower in homophily than friendships that have been stable over an extended time period (Kandel, 1978b). Typically, friend ships that dissolve are very low in homophily; accordingly, high levels of homophily are statistically significant predictors of a steady, strong relationship. Notably, levels of homophily can fluctuate based on behaviors, specifically drug use. In the c ase of an imbalance of attitudes or
37 order to maintain the level of homophily within the dyad (Kandel, 1978b). Thus, the concept of similarity, or homophily, within a friendship is a vital part of the relationship and is positively related to both people being satisfied. Expanding on homophily in relationships, Warr (1993) found t are a commonplace in an annually measured, longitudinal panel study. Sticky friends refers to the phenomenon that once delinquent friends are acquired they are typically not lost in the immediate future. However, to this point very r ecent research suggests that the concept of sticky friends can be largely explained by other factors, and particularly genetics (Beaver et al., 2009). This is not to say, however, that a person will forever have delinquent friends once acquired. Rather, a very small proportion of people maintain delinquent behavior throug h out their life course (Moffitt, 1993), thus bringing into relevance recent work showing that behavior over the life course continually tends to increase in similarity between close frien ds (Warr, 2002; McGloin, 2009). In contrast to past dyadic research, Haynie (2002) found that friendship networks p. 125). This study proposes that three types of adolescent friendship networks exist those that ar e all delinquent, those that are partially delinquent and partially non delinquent, and those that are completely non
38 delinquent. The overwhelmingly most common network is mixed with people that are delinquent and those who are not, indicating that friend s may be more different than prior studies reported (e.g., Kandel, 1973; 1978a; 1978b; 1996; Kandel and Davies, 1991, and Krohn et al., 1988). Interestingly, Haynie concludes that the proportion of delinquent friends is more important than the absolute de linquency level for the network in predicting the delinquency of a particular respondent within the network. While dyadic studies are more common in some fields, mainly medical and psychiatric sciences, they are quite rare in criminology. However, among the studies that do exist, members of friendship networks share common demographic characteristics (Hu et al., 2006; Kandel, 1973; 1978a; 1978b; Kandel and Davies, 1991; Kandel and Lesser, 1972; McPherson et al., 2001; Newcomb, 1961; 1963), drug use habit s (Agnew, 1991; Asteline, 1995; Fraser and Hawkins, 1984a; 1984b; Hawkins and Fraser, 1985; Kandel, 1973; 1978a; 1978b; Kandel and Davies, 1991; Marsden, 1988), minor delinquency tendencies (Agnew, 1991; Asteline, 1995; Kandel, 1978a; 1978b; Kandel and Dav ies, 1991), and cigarette smoking behaviors (Krohn, 1986; Krohn et al., 1988). Setting any measurement debates temporarily aside, there appears to be evidence that differential association is correct and that both primary (i.e., initial) and secondary (i. e., repetitive) adolescent delinquency needs the support of the friendship group. To this point, social learning theory has been consistently found to be the most stable theoretical explanation for the aforementioned findings (see Warr, 2002). It should be emphasized that none of the aforementioned studies were specifically designed to measure whether a subject could report the behavior of a group of friends or a specific friend. To this point, there can be several possible solutions as
39 to why this is t he case. First, in the case of earlier work (Bauman and Fisher, 1986; Kandel, 1978a; 1978b), nearly no attention had been paid to the peer delinquency measurement debate as scholars simply had not postulated that the measure could be potentially problemat ic. Second, these studies were not methodologically designed to address whether a perceived peer delinquency measure is reported deviance and across what conditions it becomes more reflective As such, the larger body of net work research (e.g., Kandel, 1978a; Kandel and Davies, 1991; Haynie, 2001; 2002; Haynie and Osgood, 2005; McGloin, 2009; Warr, 1993) is not developed with the current question in mind, although authors acknowledge that the perceptual delinquency measure ma y not be performing as optimally as possible (Agnew, 1991; Kandel and Davies 1991; Haynie, 2001; 2002; McGloin, 2009; Thornberry et al., 1994; Warr, 1993). In the forthcoming section, it is my intention to evaluate some of the strengths and weaknesses of how studies in the past have Past M easurement of Peer Delinquency: A Tale of Different Trails The most common form of all peer delinquency measurements is asking a subject to report the delinquencies of either a friend, or most frequently their group of friends, either represented by a proportion of the friends that are deviant or an actual count of delinquent acts they commit. Employing such a method of measurement is time saving, inexpensive and very frequent ly utilized. Indeed, the studies that have employed such measures are so numerous that it is unreasonable to attempt to quantify them. The most obvious reason for such a high volume of studies employing this measure is due to the consistently significant explanatory power that the peer
40 2009). Further, researchers have suggested that peer delinquency be statistically controlled in studies when possible to decrease the chance s of misspecified models (see Agnew, 1991). For a concrete example of how peer delinquency is typically measured, let us briefly examine a well known and frequently referenced study. Pratt and Cullen (2000) conducted a meta analysis on the 21 different studies (with 17 separate datasets) at the time that had assessed the general theory of crime (Gottfredson and Hirschi, 1990). In an attempt to avoid a misspecified model, every study that was included in the meta analysis had controlled for the construc t of delinquent peers and had done so through a extremely common nature of peer delinquency measurement of this form in criminological applications. 2 Because so much li terature has implemented the measurement strategy of asking a subject to produce a perceived delinquency measure for the friend, it is fruitless to examine more of those studies in detail. Rather, it is appropriate to focus on studies that have utilized d iffering forms of peer delinquency measurement. These 1978a; 1978b; Kandel and Davies, 1991; Haynie, 2001; 2002; Haynie and Osgood, 2005; McGloin, 2009; McGloin and Shermer, delinquency (Krohn et al., 1996; Thornberry et al., 1994; Weerman and Smeenk, 2005), 2 All of the self control (Gottfredson and Hirschi, 1990) studies used in this analysis that measure vior, or some variation very akin to this. There are currently no studies of self control that incorporate an actual peer delinquency measure in place of a perceived peer delinquency measure except for a very recent study by McGloin and Shermer (2009). F analysis, see Pratt and Cullen (2000), p. 939.
41 and perceived delinquency measures that are specifically reported on for one friend (Bauman and Fisher, 1986; Weerman and Smeenk, 2005). The previously mentioned works are unique in that they vary from the standard peer delinquency measurement process. Recall the Kandel and Davies (1991) study that evaluated social control and social learning theories using dyadic data. This stu dy was part of a longer, longitudinal study of New York high school students that transitioned into adults across a three wave data collection over a bit more than a decade. This study is advantageous in that an actual delinquency measure for the friends was used. Additionally, the authors treated reciprocal friendships as quantitatively different from non reciprocal friendships. However, the Kandel (1973; 1978a; 1978b; Kandel and Davies, 1991; Kandel and Lesser, 1972) sample could potentially have been obtained in a problematic way. This study chose to go to the classroom of high school students and administer a paper based survey to all the students present on that particular day. Clearly, the most significant drawback to this sort of collection is t hat there is no guarantee that the best friends that are identifying each other will not be in the same classroom. Due to this possibility, a potential problem can arise if the teacher has an open seating policy that allows for best friend dyads to sit ne xt to each other. This could increase the likelihood that the two could share answers, thus potentially contaminating the data. To correct for this methodological shortcoming, members of the dyad must be sampled in different places at the same time. Oth er studies have suffered from similar collection problems (Asteline, 1995; Bauman and Fisher, 1986; Haynie, 2001; 2002; Haynie and Osgood, 2005; Kandel, 1978a; 1978b; Kandel and Davies, 1991; Kandel and Lesser, 1972;
42 Krohn et al., 1996; McGloin, 2009; McGl oin and Shermer, 2009; Thornberry et al., 1994; Weerman and Smeenk, 2005). Part of the design of the current research was that the self identified, best friend dyads were placed into separate physical spaces and administered the survey at the same time. the National Longitudinal Study of Adolescent Health (Add Health) data. Similar to schools class rooms for wave one. Each student was asked to identify up to five of their closest female and up to five closest male friends. The Add Health data have measures reported delinquency, as researchers are able to link friendship pairs together through the identification of each other participants who had completed the study at wave one were selected to complete an in home surv problematic due to potential contamination effects because contact between friends is very likely to have occurred due to there being a significant time lag in between the first sampling period when the students were at school and the in home follow ups (see above paragraph). A measure of friendship involvement was also included in the Add Health measures. An involvement measure, while better than having no measure in its place, is not as complete as a friendship intensity measure instead, it is but one piece of the friendship intensity construct (see Bukowski et al., 1994). Haynie (2002) found that friendship groups were not frequently homogenous. Instead, networks had many members that identified each other as friends non
43 reciprocally and the most common form of friendship web included delinquent and non delinquent friends. The findings of this study may be more accurate than past analyses of other networks and dyads due to the fri endship measure not being about an abstract group of friends but rather the actual group of friends that the respondent identifies. Using the Monitoring the Future dataset, Asteline (1995) tested social learning theory ( Burgess and Akers 1966) against s ocial control theory (Hirschi, 1969) to determine which better predicted delinquency. Similarly to the Add Health, Monitoring reported delinquency, but excludes a perceptual measure of delinquency. H owever, Asteline recognizes that a perceived peer delinquency measure may not be an appropriate scale to use as he involvement with delinquent friends has been derived from the respo perceptions p. 104; emphasis in original). To this point, despite the lack of a perceived n self reported delinquency than a perceptual peer delinquency measure. This is a rather unique finding as it provides confirmatory speculation to the previously most visible part of the larger question of the accuracy of a perceived peer delinquency meas ure reported delinquency is The Asteline (1995) study ha s several limitations. A three wave longitudinal design is implemented in the Monitoring the Future study, but participants are not asked to identify their friends until wave three. Since high school students are the targeted
44 population, and friendships are constantly changing at this poi nt in the life course (see Agnew, 1991; Kandel, 1978a; 1978b), wave three was not the most amenable time to every wave to assess how they changed over time through the p recarious adolescent back to the initial study wave and thus would not be significant predictors of 110). Another major drawback is that this study has no reported delinquency is included in the instrument, which draws into questio n his conclusion that self reported delinquency is a better measure to use. Warr (2002) has echoed One additional and quite i mportant methodological problem with the Asteline (1995) study is that it treats unreciprocated best friends as best friends. Stated differently, if person A said that person B was their best friend, but person B did not identify person A as a friend, thi s was still treated as a best friend dyad for the statistical analysis. This is a significant drawback to this study due to the fact that only 40% of the sample had reciprocally identified friendships. This shortcoming has only been corrected for by a fe w studies (Kandel, 1978a; 1978b; Kandel and Davies, 1991). Despite limitations, this study makes some very strong theoretical assertions and provides a unique test of social learning and social control theories. Thornberry and colleagues (1994), in a test of interactional theory, used data
45 from the Rochester Youth Development Study (RYDS) to explain juvenile delinquency across four waves of data. While the RYDS does use a perceived peer delinquency measure of an abstract group of friends, as did the studi meta perceived delinquency measure and self reported delinquency could possibly be the same construct (see Gottfredson and Hirschi, 1987; 1990; also see Thornberry and Krohn, 1997). To address whether the perceptual peer delinquency measure was identical to reported delinquency, Thornberry and colleagues (1994) statistically compared the two measures. The researchers concluded th items loaded on one factor, and the self reported delinquency items either loaded on a second factor or did not meet the criterion level (.40) on either fa 1994, p. 62). This finding is quite significa nt for the current study for two reasons. First, 1990) claims of projection within the perce ptual peer delinquency measure. Second, this finding is important because measure is not the same thing as his self reported delinquency, despite being in opposition to related results (Asteline, 1995; Gottfredson and Hirschi 1987; 1990; Weerman and Smeenk 2005). While there are no significant advances in whether the use of a perceived peer delinquency measure reported delinquency or under what conditions it does so the most this finding is quite significant as its contradictory evidence to prior lite rature provides us with further questions about what a perceptual measure of peer delinquency is in fact measuring.
46 Akin to the Thornberry and colleagues (1994) study and using data from the National Youth Survey (NYS), Agnew (1991) tested Gottfredson and 1990) hypothesis of projection and compared two delinquency measures to evaluate if they were the same thing. The NYS uses the same measure of delinquency that the Rochester Study does by asking the respondent to provide a perception of t he proportion of their friends that are delinquent. After comparing the perceptual peer reported delinquency, Agnew also concluded that while related, the two measures were not the same thing. This finding provi des further evidence against the projection hypothesis, although the two measures are highly correlated in that they load on each other at .70 to .76 across three waves of data. e argued that the perceived peer delinquency and self reported delinquency measures from the NYS data are indeed measuring the same thing due to them being very highly correlated. However, Agnew (1991) addressed this concern and removed the delinquent pee r measure in one of his analyses (not reported) and found that the results from the model were unaltered. However, Gottfredson and Hirschi (1987) are correct that this is a very high correlation between two measures that are supposedly unique constructs. Due to the measures being so highly correlated in his analysis, Agnew (1991) does acknowledge that multicollinearity may have inflated his standard errors, thus potentially decreasing the t values artificially. However, it is noteworthy to remember that peer delinquency measure are but one solution to one piece of the peer delinquency
47 measurement debate. Instead, research should focus on how reflective a perceived peer delinquency m easure is under what conditions it becomes more reflective and what behaviors a subject can most precisely report. Several studies have moved towards this goal and have measures of peer reported behavior, perceived peer behav ior, and self reported behavior the three measures that are implemented in this thesis. Bauman and Fisher (1986) analyzed adolescent beer drinking, liquor consumption, and tob acco use from two similar three wave, longitudinal studies. For the alcohol s tudy, which represented one dataset, middle school aged adolescents were sampled, while the tobacco study sampled freshman in high school that were on average two years older than the middle schoolers. This study asked participants to list their three b est friends in order of closest best friend first and provide a specific perceived behavior report for each person. Siblings were not allowed to be listed as a best friend. Being that this study sampled students only in selected schools, a small portion of the sample could not be linked to individuals in the study. Researchers used the three listed best friends in order to create a best friend dyad, attempting to match the closest best friend first and then working their way down the list. This is a ver y efficient way of collecting data on specific individuals, as 95% of the sample was able to be linked to a best friend that was also in the study. Running only correlational analyses between the measures of delinquency, the research found that the meas
48 but only in cross sectional designs. When a longitudinal analysis was conducted, neither the actual fri interesting, Bauman and Fisher (1986) note that they were not altogether certain in how to interpret the res ults found. To this point, longitudinal analyses indicated that the delinquency when prior behavior was not controlled. The authors noted that their sample did have s ome notable drawbacks. The two longitudinal studies utilized were not completely ideal to be compared due to different cigarette smoking study that led responses (self rep ort and perceived) to be very different from the alcohol study. Furthermore, students in schools were given self administered, take home surveys to complete, indicating that data contamination may have occurred in this study as there could easily have bee n contact between the best friends. There was also no indication of friendship intensity. Despite its limitations and lack of a powerful analysis of comparison of the measures of perceived and self reported delinquency for the subject and friends, this s tudy is unique and of use to this thesis due to its rare use of self reported delinquency for a subject, a best friend, and a perceived behavior report for specific individuals. a with who m the subject spends time and is amenable towards but is not very closely bonded with) delinquency
49 unde rprivileged frame of Dutch high school students, allowed subjects to select up to two best friends. Two perceived peer delinquency measures were gathered. One was method of asking a subject to report the proportion of his friends that are delinquent, and the other measure was a perceived peer delinquency measure for the best friend(s) indicated. If a person selected more than one best friend, the perceptual report of both f riends was summed and divided by two for the analysis. friend matters beyond the delinquency of regular friends in predicting criminality in a subject. As such, the best frien each other. If a respondent had higher delinquency than the best friend, the del inquency level would increase in a sense the two met halfway. It was also found that non delinquent respondents tend to have non delinquent friends over half the time. In a similar finding to what Bauman and Fisher (1986) reported, this research concluded that the respondents own behavior was more strongly correlated with their perceptions of their peer s s actual behavior. However, Weerman and Smeenk (2005) include no measure of friendship intensity, and students were sampled in school at different times, indica ting that there may be contamination in the data. While these drawbacks are common, there are several limitations in this particular study that extend beyond criticisms of prior work. First, other studies have treated non reciprocated friendships (where one subject thinks another is a friend, but that person does not list the subject as a friend) as reciprocal
50 (Asteline, 1995; Bauman and Fisher, 1986; Haynie, 2001; 2002; Haynie and Osgood, 2005; Krohn et al., 1996; McGloin, 2009; McGloin and Shermer, 2009 ; Thornberry et al., 1994), which this study does also. B eing that Weerman and Smeenk focus so heavily on the relationship of best friends, failing to treat reciprocated best friendships and non reciprocated best friendships as statistically different thi ngs (as Kandel, 1978a; 1978b; have such a strong effect on the delinquency of the acquaintances. A second limitation iminality was probably also captured in the general perceived peer delinquency scale, indicating that the measures may not be independent of each other. To bolster this criticism, the perceptual measures of a best reported delinquency, was dichotomized so that reporting only one delinquent behavior classified the respondent into the deviant group. Considering there were ten delinquency items for respondents to report at school and thirteen away from school, dichotomizing this measure could have potentially produced a large loss of information which may have significantly affected the findings. Despite drawbacks, the Weerman and Smeenk (2005) and Bauman and Fisher (1986) studies are very relevant to the current work because of nearly identical identified deviance, and a self reported measure of delinquency from the subject. While no definitive solutions have come out of this pair of studies that suggest whether a perceptual delinquency measure is accurate or across what circumstances it is more accurate, two findings ar e
51 of immediate relevance. First, both studies provided results that suggested that a general and change in criminality in a subject (Weerman and Smeenk, 2005). This suggests that utilizing a best friend for the unit of analysis in a study attempting to determine if a perceptual peer delinquency scale is measuring the construct it claims is most likely a fruitful path to take. Summary of Friendships and Past Measurement of Peer Delinquency Friendships and Social Networks. Friends have been found to be similar on many characteristics, especially on demographics (Hu et al., 2006; Kandel, 1973; 1978a; 1978b; Kandel and Davies, 1991; Kandel and Lesser, 1972; McPherson et al., 2001; Newcomb, 1961; 1963), age, and share d marijuana and drug use habits (Agnew, 1991; Asteline, 1995; Fraser and Hawkins, 1984a; 1984b; Hawkins and Fraser, 1985; Kandel, 1973; 1978a; 1978b; Kandel and Davies, 1991; Marsden, 1988) as well as minor (Agnew, 1991; Asteline, 1995; Kandel, 1978a; 1978b; Kandel and Davies, 1991), moderate, and index crime behaviors (Agnew, 1991; Kandel, 1978a; 1978b; Kandel and Davies, 1991). To this point, it should be emphasized that best friends and regular friends have differing impacts on delinquency. Warr (2002), the first to make this claim, has sup portive evidence that this hypothesis is accurate. McGloin (2009) found that best friends will not terminate a friendship that is imbalanced in delinquency as some researchers have found (Kandel, 1978b; Kandel and Davies, 1991). Rather over time they wi ll tend to balance out delinquency indiscretions with the other so that a friend who is more
52 delinquent will decrease his deviance while the other friend increases his deviance. This finding is of importance to this thesis, which only focuses on best frie nds, indicating that the subjects of focus for this study should have a more direct effect on the delinquency of each other than if they were just simply amenable acquaintances. Friendship networks are not as homogenous as some have previously thought (H aynie, 2002), and some research even suggests that a respondent will report higher deviance homogeneity in a group than actually exists (Asteline, 1995). Rather, a typical friendship group is composed of both delinquent and non delinquent peers. This spe reported delinquency because a general perceived group based delinquency measure most likely covers delinquent and non delinquent individuals. Furt hermore, research demonstrates that networks do not have completely reciprocal friendships between the people in them (Haynie, 2001; 2002). This finding emphasizes the need of a measure of friendship intensity in order to divulge which research subjects a ctually can classify as best friends. This measure has been excluded in past studies. Measurement of the Delinquent Peer s Construct. The predominant reason why a measure of delinquent peers is included as a statistical control so frequently is because research demonstrates that deviant friends 2009; Akers and Sellers, 2009). It has been recommended that researchers use deviant friendships as a control variable whenever po ssible (see Agnew, 1991). The most common form of measuring deviant peer relationships is by asking the subject to
53 This measure has been called into question in research (see Asteline, 1995; Gottfredson and Hirschi, 1987; 1990; Weerman and Smeenk, 2005). It has been b own self reported delinquency. Clearly, the use of a perceived peer delinquency measure is commonplace in the criminological literature, as it is often assumed that th is measure accurately captures the peer delinquency construct. However, several authors have recognized that this may not be a safe assumption, and have addressed this measure in more rigorous detail (see Agnew, 1991; Asteline, 1995; Bauman and Fisher, 19 86; Thornberry and Krohn, 1997; Thornberry et al., 1994; Warr, 1993; 2002). delinquency is the same as their own self report ed delinquency. Two studies measured a perceived peer that the two measures were moderately correlated, but did in fact load on separate factors (Agnew, 1991; Thornberry et al., 1994). Asteline (1995) addressed this issue as well, but instead Another study ( his/her own self reported crime. The two measures were extremely highly related, but were treated as separate constructs, although the a uthor noted this could have caused multicollinearity in his dataset.
54 However, it is important to remember that while the projection hypothesis has been piece of the larger mystery that surrounds the peer delinquency measurement construct. Longitudinal studies represent 100% of the research that addresses how friendship dyads and networks interact with one another. The most prominent motivation for this is that none of the se studies are designed specifically to evaluate how the perceived peer delinquency measure a respondent provides interacts with the actual report of the friend. Although not designed specifically for this reason, one study (Bauman and Fisher, 1986) indic reported delinquency are separate constructs. Unfortunately, this particular test is marked by data collection shortcomings and methodological exclusions. Being that there are n o studies to date that are specifically designed to test w hether a subject can provide a precise series of common errors occur in existing studies. These errors include possible data contamination (Asteline, 19 95; Bauman and Fisher, 1986; Haynie, 2001; 2002; Haynie and Osgood, 2005; Kandel, 1978a; 1978b; Kandel and Davies, 1991; Kandel and Lesser, 1972; Krohn et al., 1996; McGloin, 2009; McGloin and Shermer, 2009; Thornberry et al., 1994; Weerman and Smeenk, 200 5), a lack of proper measures of friendship intensity (Asteline, 1995; Bauman and Fisher, 1986; Haynie, 2001; 2002; Haynie and Osgood, 2005; Kandel, 1978a; 1978b; Kandel and Davies, 1991; Krohn et al., 1996; McGloin, 2009; McGloin and Shermer, 2009; Thornb erry et al., 1994;
55 Weerman and Smeenk, 2005), failing to include any measure of peer delinquency other than a perception of the delinquency of an abstract group of non specific friends (Akers et al., 1979; Agnew, 1991; Krohn et al., 1996; Lanza Kaduce et a l., 1984; Radosevich et al., 1979; Thornberry et al., 1994; Warr, 1993) or of a group of specific fri ends (Weerman and Smeenk, 2005), not having a perceptual measure of delinquency (Asteline, 1995; Haynie, 2001; 2002; Haynie and Osgood, 2005; McGloin, 2009 ; McGloin and Shermer, 2009), poor sample qualities (Asteline, 1995), and failing to treat reciprocal friendships and non reciprocal friendships as statistically different things (Asteline, 1995; Bauman and Fisher, 1986; Haynie, 2001; 2002; Haynie and Osgo od, 2005; Krohn et al., 1996; McGloin, 2009; McGloin and Shermer, 2009; Thornberry et al., 1994; Weerman and Smeenk, 2005). Despite errors, researchers are clearly interested in this topic as it has both theoretical and empirical implications for research (see the debate from Gottfredson and Hirschi, 1987; 1990; also the discussion from Agnew, 1991; Asteline, 1995; Kandel, 1996; Thornberry and Krohn, 1997). For instance, if a perceived peer delinquency measure does in fact prove to be efficient and reflect ive only for certain delinquencies or decades may have included a measure that might have caused a misspecified model. It is indeed surprising that no study to date has been designed to examine the several peer delinquency measurement issues that are of relevance to this research. Research Questions and Hypotheses Research Question One.
56 practice, little is known about the validity of this measure and whether it is in fact measuring delinquency reflective of that reported delinquency (see Figure 2 1)? To date, research has been unable to empirically address this question due to a surprising lack of app ropriate measures in currently available datasets (e.g., Agnew, 1991; Asteline, 1995; Gottfredson and Hirschi, 1987; 1990; Haynie, 2001; 2002; Thornberry and Krohn, 1997; Thornberry et al., 1994; Warr, 1993; 2002; Weerman and Smeenk, 2005). Research Ques tion Two. Warr (2002) discusses, among other issues relating to peers and delinquency, the nature of group based and individual based crimes. Individual crimes, which are nearly non existent in early adolescence, surge around the time of late adolescenc e to adulthood. Warr, and also Reiss (1986), conjure that the reason behind this is that people tend to grow out of the need for a supportive peer group to commit deviance as they mature. To expand on this notion, Warr notes that while certain types of c rimes tend to be committed individually, the majority of criminal behaviors are still committed delinquencies typically committed in a group setting, are typically property o ffenses, public order offenses, and substance use. Adversely, sexual battery, assault, threatening assault, and shoplifting tend to be behaviors that people carry out in private. This brings rise to the second research question of this thesis. Which pee r delinquencies can a subject report through a perceived peer delinquency scale reported delinquency ?
57 assault, threatening assault, sexual misconduct, and shoplifting are peer behaviors that a subject will not be able to identify because of their tendency to be individualistic behaviors. However, it is expected that crimes pertaining to property violations, public order offe nses, and substance use will be more precisely reported by a subject because groupy literature aside, I will further explore which genres of perceptual peer crime (property vio lations, substance use, viol ence, and minor delinquency ) tend to be reported most correctly by a subject. More detailed information will be provided in the subsequent chapter on which crimes are theorized to be individualistic and groupy an d into what de linquency categories the behaviors are classified. Figure 2 2 is a visual representation of the reported delinquency measures are represented by Research Question Three. Gottfredson and Hirschi (1987; 1990) have been avid critics of the use of to a perceptual peer reported peer delinquency is just another measure of s elf 597). It is identified delinquency. In order for the two measures to be found similar they must have the same predictive ability on the desired
58 outcome To this the same as his or her own delinquency, then do they have the same predictive ability reported delinquency ? Prior literature has not been capable of answering e ither of these questions, thus making this an exploratory research question. Figure 2 reported delinquency represent any simil arity that there may between the measures while the dotted lines measure. Research Question Four. Prior criminological research has not included a detailed measure of fri endship intensity when examining the peer delinquency construct. When a measure of friendship intensity (the Friendship Qualities Scale; see Bukowski et al., 1994) is included, is the level of friendship closeness capable of decreasing and/or intensifying members of friendships that are very intense will show a high ability to perceive each lf reported behavior The consistency of the perceptual delinquency report to the self reported delinquency can be expected to decrease incrementally when friendships with lower levels of closeness are accounted for, resulting in the expectation that the reflectiveness friendship intensity. I further seek to determine which of the five subsections the FQS is comprised of (play, conflict, help, security, and closeness) appear to be most important
59 Figure 2 4).
60 Figure 2 1 Research Question One. Self Reported Delinquency Perception of Delinquency Subject Friend D V IV
61 Figure 2 2 Research Question Two. Subject Friend Perception of Delinquency PPD Groupy PPD Individualistic SRD Individualistic SRD Groupy Perception of Delinqu ency PPD Property PPD Substance PPD Violence SRD Violence SRD Property SRD Substance Perception of Delinquency PPD Minor Delinquency SRD Minor Delinquency D V IV D V IV D V IV D V IV D V IV D V IV
62 Figure 2 3 Research Question Three. Self Reported Delinquency Self Reported Delin quency Subject Friend Perception of Delinquency
63 Figure 2 4. Research Question Four. Self Reported Delinquency Perception of Delinquency Friendship Qualities Scale Play Conflict Security Help Closenes s Self Reported Delinquency Perception of Delinquency Sub ject Friend Play Conflict Security Help Closenes s
64 CHAPTER 3 RESEARCH DESIGN AND ANALYTICAL STRATEGY Research Design Samp le. Primary data were collected during the spring semester of 2009 at a large southeastern university. Participants were sought through a list of the largest classes that the university offered for the spring semester of 2009. This list was obtained thro ugh a public information request at the Office of Institutional Planning and Research. Instructors with large classes were contacted in order to see if they would be interested in offering an extra credit opportunity for participation in original research Smaller criminology classes were also solicited. The sampling method employed was a convenience sample. A convenience sample was implemented primarily because this research lacked a source of funding, thus making it difficult to gain a sample that was representative of the university population. Individuals nested within dyads are the unit the subject plus one of his/her ined as the person person that comes with them. If both students were participating in the study for extra credit in one of the selected courses, one was randomly ass igned to be the subject and one the friend. The same survey instrument, coded with a pre designated random three or four digit code to identify the dyad as a linked pair, was given to both the subject and the friend (for example, 123 0 is the subject, and 123 1 is the friend). The final sampling frame included 22 undergraduate classes in the fields of psychology, pre med, wildlife ecology, criminology, communication studies, African
65 American studies, conservation, and economics. There were six very larg e, non criminology classes that offered the study for extra credit. The size of the large classes ranged from 299 (a general communications class) to 1,521 students (a microeconomics class). The other four large classes that were in the sampling frame in cluded a course on wildlife issues (993 students), an introduction to psychology class (631 students), a biodiversity class (439 students), and an introduction to health class (333 students). There were 4,216 students that were solicited from these six cl asses. Fourteen criminology courses also offered participation in this study for extra credit. The total number of students taking these classes was 568. However, this number is inflated by roughly 30% due to criminology majors overlapping in multiple co urses. Two other classes in African American studies combining for 52 students with heavy minority enrollment (96% for each) were also included to oversample black populatio n characteristics of the campus. Without counting for students overlapping in courses, the final sampling frame included 4,836 students that could have taken the study for extra credit. However, this number is more likely to be in the vicinity of 4,000 s tudents once undergraduates in multiple classes are accounted for. Roughly 2,200 students completed the survey instrument (or approximately 1,100 dyads). However, due to the survey being paper based and taking a very long time to enter into a computer, the initial sample size for this thesis before performing missing data analyses was 1,552 individuals (776 dyads). This represents roughly 70% of the entire sampled population. Out of the 1,552 surveys that were completed, 253
66 (16.3%) of them were checke d systematically for errors. To further ensure accuracy in nonsensical responses caused by data coding errors. When an error was found, the original survey was reexamined and the proper response was reentered into the dataset. A listwise deletion was implemented on the initial sample size of 776 pairs, 3 Overall, the individual sample (n = 970) was pr edominantly female (66.5%), non Hispanic (82.1%), white (63.4%), employed in the past year (66.5%), 19 years old (39.9%), a freshman (47.5%), not in a relationship (57.4%), had married parents (70.8%), were not in a 3 All variables that were included for the analysis were examined for missingness. Missing values in the dataset appeared to be missing at random. There were no signs that there was systematic missingness among responses. The demographics section showed very low levels of missing responses. Out of the 14 items in the demographics, only eight showed any missing values. Of these items, the range of missing responses ranged from 0.13% (school class rank and the frequency of attending a religious service) to 1.16% (grade point average). Of the items that showed any missingness, the average amount was 0.45%. All items on the self report de linquency scale showed missingness, although it was a minor issue. The range of missingness was 0.13% (vandalized property of friends, neighbors or roommates, whether the subject had cheated on school tests, and whether the subject had sold hard drugs) to 0.58% (vandalized other property that is not yours). Of the 26 items on the scale, the average percent missing was 0.32%. The perceptual peer delinquency scale showed a higher level of missing responses, but still the overall percent missing was very low. All items showed missingness which ranged from 0.26% (six different items) to 1.03% (how often the friend drank alcohol). The mean percentage missing for the perceived peer delinquency scale was 0.44% per item. The Friendship Quality Scale (FQS) sho wed the highest level of missing data of any scale on the instrument. All items showed missing values that ranged from 0.13% (item 12) to 2.32% (item 17). Of the 23 items on the scale, the mean missing percentage was 0.86% per item. While this is slight ly higher than the other scales on the instrument, it is not indicative of any significant problem with missing data. Data missingness did not appear to contain any sort of systematic patterns. Due to a relatively large dyadic sample size (n=776), a listw ise deletion was implemented on the sample to determine if the responses from both members of the dyad on a single line of data. As a result, a cas e being eliminated indicates that at least one of the two members of the dyad failed to respond to at least one item on the survey. The listwise deletion eliminated 291 dyads, leaving a final sample size of 485 student pairs. Listwise deletion, when impl emented without seriously affecting sample power, is the preferred method of dealing with missing data (see Allison, 2000). Even when employed with advanced techniques, imputing values is still an educated guess as to what a respondent would have answered As such, for the purposes of the analysis a conservative approach to missing data was implemented, providing us with a final sample size of 485 dyads.
67 fraternity or sorority (80.7%), never at tended a religious service (41.9%), wanted to obtain a doctoral or professional degree (41.6%), had a grade point average of 3.5 to 4.0 (48.5%), and were sexually active (57.3%). Being that this is a convenience sample, it is fruitful to examine the char acteristics of the sampled students to the population they are supposed to represent. The sample for this study was demographically very similar to the overall 2009 school year, 10 .3% black students, 8.7% Asian and Pacific Islander students, 15.5% Hispanic students, 60.7% white students, and 3.1% other race stu dents. The classification of diffe rent from the overall university population. To this point, the university treated the for this study is roughly 15% higher, meaning that 18% of the participants in the sample When summed, population. Roughly 19% of the sample for this study was a member of a fraternity or sorority ulation. However, the semesters, at which time the fraternities and sororities are having recruitment. The potential members, thus, are not included in the number o f 15.3%. Considering that this
68 s ample was collected during the s pring of 2009, pledges of the fraternities and sororities are included as they would then have become active members of the Greek organizations. In sum, it is very likely that the percentage of Greeks to take this study is in actuality very similar to the overall population of the university. For summary statistics on those individuals who served as the subject and those who served as the friend, consult Table 3 = 485) characteristics are also quite informative. The student pairs consisted of 54.2% female x female dyads, 21.2% male x male dyads, and 24.6% mixed se x dyads. In 71.3% of the cases neither member of the student pairs were Hispanic, in 7.2% of the case s both students were Hispanic, and in 21.5% of the cases there was one student who was Hispanic. Slightly over half (51.1%) of all dyads were white x white, 9.5% were black x black, 1.1% were Asian x Asian, and roughly 33% of the dyads were of mixed races More often than not, members of the student friendship dyad shared the common class rank. Freshman x freshman dyads were most common (37.1%), sophomore x sophomore dyads were next common (15.1%), junior x junior dyads were somewhat common (8.4%), and s enior x senior dyads comprised the smallest amount of the sample (4.3%). Approximately 35% of the sample consisted of friends that were not in the same collegiate class. 12% of the sample was student pairs who were both in a fraternity or sorority, while 73.4% of the sample was students who were not Greek affiliated. Roughly 15 % of the sample of best friend pairs contained one member who was affiliated with a social Greek organization. 46% of the dyads contained students who were both employed in the pa st year, 13% of the pairs had neither been employed, and 31% of the dyads had one member who was employed and one who was not.
69 Those dyads in which one member never attended a religious service had friends that also never did (23.1%), went only once a mon th (21.2%), went two to three times a month (6.4%), went once a week (8%), or went more than once a week (1.9%). 7.4% of the pairs contained students that both went to a religious service once a month, while only 3.1% indicated that they both went to chur ch once a week. The majority of the student pairs contained at least one member whose grade point average was between a 3.5 and a 4.0 (70.9%), and nearly all dyads had at least one student with a GPA of over a 3.0 (96.5%). Procedures. Data collection was set at recurring times on Wednesday and Friday afternoons for two months, with occasional alternate times on Monday and Thursday evenings. Students were requi red to come to the research study together (as a dyad) to be able to participate in the study. Upon entering the building where the study was being conducted, the participants were directed down a short flight of stairs to a sign in station. A statement of informed consent was read and signed by each student stating that they understood that participation was voluntary and that the answers were confidential. Upon signing the informed consent (see Appendix B), a copy was given to the students to take hom e with them, as the document discussed their rights as participants and was important for their own documentation. Students in the dyad were both required to present their student identification cards in order to participate in the study. No one was allo wed to participate multiple times. While students read and signed the informed consent, their identification
70 numbers, last names and first names were recorded in a spreadsheet. Included on the spreadsheet was an allocation of spaces to record which appli cable class, or classes, each student was enrolled in for extra credit. Participants were asked to verbally indicate which classes they were in off a printed list of the 22 selected courses. There was a special effort made by researchers to assign as muc h extra credit as possible due to students being unable to participate in the study twice. Instructors were contacted near the end of the semester with a list of those who had participated in the study for their classes. The study consisted of both studen ts from each dyad being administered identical, 14 page surveys that were pre coded with a unique numerical identifier to identify the students as one dyad (the survey instrument can be seen in Appendix C). Surveys were administered in a large, split room Each side of the room was where the sign in station was. Upon entering the room, the subject was asked to stay and take the survey on the side the pair entered on (d eemed Room 3). In order to create separation in the friendship dyads, the friend was asked to walk to the end of the first room in order to enter the second room (deemed Room 3A). Rooms 3 and 3A were separated by a full concrete wall that had an open doo rjamb at the opposite end of the room from where pairs entered. There were a number of desks that were set up in both parts of the room where students could fill out their surveys. Upon completion, students were instructed to bring their completed surve ys to a research assistant that was stationed next to the doorjamb that separated the two rooms. When the survey was handed back a debriefing form was issued to the
71 participant (see Appendix D). The subject (in room 3) was asked to leave out the door th ey came in and the friend (in room 3A) was asked to leave out another door that was at the end of their room. Being that both parts of the room had their own exit, this eliminated the possibility that the subject and the friend could come into contact wit h each other while the instrument was being administered. Due to high volumes of participants that showed up to participate in the study, surveys were administered with two research assistants and the principal investigator the majority of the time; occa sionally, subjects were administered surveys with only the while the job w as to issue the debriefing form. T he third party either helped with sign ins, answered questions f rom participants, pre coded surveys or assisted with other logistical work. It should also be noted that several students were given alternative assignments their being a recent transfer student or a non traditional student. Additionally, several participants were given alternative assignments on the grounds that they were under the ag e of 18 and were not allowed to participate in the study due to IRB restrictions. Most participants were compensated directly by being given extra credit for the class, or classes, they were in that offered this study for extra credit. Since a significant portion of the sampling frame was large, survey level classes with hundreds of students in them, about 20% of participants were compensated with extra credit in more than one class. Approximately the same number of students came to the study only as a fr iend,
72 and thus were not compensated beyond helping their friend get extra credit. Roughly 80% of the people that showed up for the study were compensated with extra credit. request ed extra credit of 1% / 100% increase in overall grade will be asked for; however, the researchers cannot make this decision for the instructors, who will have majority of in structors did utilize their discretion in deciding the amount of extra credit to assign for the study. The two most frequently used extra credit values that were given were five to ten extra credit points on a test or one percentage point overall increase in the grade, respectively. Typically, those instructors who offered five to ten points on an exam had two to three examinations in the class. The minimum extra credit value offered came from two criminology class instructors who gave point extra cred it to the overall grade. Measur es Delinquency. The measurement of delinquency was an essential aspect to this research project. To assess the construct, delinquency was measured using scales administered to each member of the friendship dyad. All delinqu ency items were adapted from the National Youth Survey (NYS; see Matsueda & Anderson, 1998; Menard & Elliott, 1990a; 1990b) Being that the NYS was originally constructed for high school populations, several of the delinquency items were modified in order to be more relevant to a college money or other things from your
73 delinquency scale were adjusted slightly to make better grammatical sense when asking times in the last year have you sexual relations with someone against their will? measured using a nine point Likert based response scale (see Appendix C, parts V and Both members of the dyad were asked to respond to 26 items rega rding their own self delinquency (see Appendix C, parts V and VI, respectively). This produces two delinquency measures for each member of the dyad a self report delinquency mea measures, and their subscales, were coded so that higher scores indicate higher levels of delinquency. Self reported delinquency Self reported delinquency was measured using a 26 item measure that inquired how many times the reporter had committed a wide range of delinquencies in the past 26 item self reported delinquency measure served as the dependent variabl e in both the first, fourth and part of the third researc h questions. This measure 4 ). The third research question, which explores 4
74 reported delinquency as an independent variable. The subject reported reported delinquency scales, the responses for the delinquency items were summed and divided reported delinquency ( = 1.48, SD = 0.43) was reported delinquency ( = 1.49, SD = 0.43). Higher scores indicate higher delinquency, while scores of one indicate no delinquent actions. Perceptual peer delinquency Each member of every dyad was also asked to specifically report on the delinquencies of their friend. Note that this is not a perceptual friendship group delinquency measure, as has been used by many network studies (e.g., Akers et al., 1979; Agnew, 1991; Krohn et al., 1996; Lanza Kaduce et al. 1984; Radosevich et al., 1979; Thornberry et al., 1994; Warr, 1993), but rather this measure inquires about the reported delinquency 26 item index which implemented as an independent variable in both the first and fourth research questions, and serves as an independent and dependent variable in different parts of the third together and dividing by 26 to create an average ( = 1.42, SD = 0.44). Scores of one would indicate the subject perceives that their friend is completely non delinquent, while higher scores correspond with higher delinquency. The self reported and perceptual peer delinquency measures were designed to
75 produce unique opport unities for the use of alternate forms of both dependent and independent variables. The two measures were designed to cover a breadth of different types of crime as well. To this point, a series of subscales representing two operationalizations of delinq uency were created for the second research question, which asks what types of crimes a subject can most validly report for a friend. The first (see Warr, 2002) are better predicted by a subject. Subscales for both groupy and individualistic crimes were formed from each overall delinquency measure. The second way in which delinquency was operationalized was by disaggregating the overall delinquency indices into trad itional crime categories property crime, violence, substance use and minor delinquency. Groupy crimes The second research question disaggregates the overall delinquency measures in two ways across typical categorical crime operationalizations Warr (2002) theorizes that certain types of crimes are committed in groups and certain types of crimes are committed alone. Those delinquencies committed with accomplices typically are vandalism, subs tance reported groupy delinquency, were created to measure groupy delinquency. reported groupy indices contain 17 items (see Appendix E for which items are groupy and individualistic). These items oupy offenses because they directly inquire as to a
76 substance use, public order offenses and acts of vandalism. as an independen dependent variable in a bivariate OLS regression model. These scales were created by summing the 17 groupy items and dividing by 17, thus producing an average groupy delinquency. The subjects perceived ( = 1.54, SD = 0.56) slightly less groupy crimes than the friends reported ( = 1.62, SD = 0.56). Higher scores on these scales indicate more perceptual or self reported groupy delinquency, while a score of exactly one indicates no gr oupy delinquency. Individualistic crimes Individualistic crimes, or those crimes that are typically committed alone, are theorized to be shoplifting and theft, violence, threatening violence, and sexual assault (see Warr, 2002). Similar to the groupy crim e model, individualistic crimes were disaggregated from the overall delinquency scales. Two individualistic crime scales rep orted individualistic crime. The perceptual and self report individualistic crime scales both contain nine items rizing, the items included in these scales are justifiably individualistic crimes as they strictly are measuring a form of lf
77 = 0.57) both show moderate internal consistency. These scales were created by adding the items together and dividing by nine, with higher scores indicating higher amounts of individualistic delinquency and scores of one representing no individualistic delinquency. Subjects ( = 1.18, SD = 0.34) tended to perceive slightly less individualistic delinquency than the friends reported themselves ( = 1.24, SD = 0.32). Property crime s The second form of operationalizin g delinquency measurement for this thesis was by measuring categorical crimes. To see which crimes fall into which categories, see Appendix E. Two property crime scales were created from the 26 item, overall delinquency scale, each containing nine items many times in the last year have you stolen [or tried to steal] things worth between $5 reported property property crime subscales, the items were ad ded and divided by the total number of items. Higher scores indicate higher levels of property crime, while scores of one indicate no property crime. The subject perceived ( = 1.10, SD = 0.31) that the friend committed slightly less property crimes than the friend typically reported ( = 1.17, SD = 0.29). Violent crime s reported violence. The two violent crime indices contained four
78 w many times in the last year have you hit [or threatened to hit] report scale most likely contributed to the low reliability statistics that were generated. The violent crime subscales were created by adding the violent crime items together and dividing by f our. As such, higher scores indicate more violence. The subjects mean perception ( = 1.16, SD = 0.31) was fairly similar. Substance use A nine item scale was used to measure substance use. In a ccordance with the reported substance use measure was also generated. Each scale contained ve you used together and dividing by nine, so that higher scores represented more subst ance use perception ( reported substance use ( = 1.93, SD = 0.83). Minor del inquency reported minor
79 delinquency, were also generated for the second research question. Each minor delinquency that o the minor delinquency items were summed and divided by four. Higher scores = 1.50, SD = 0.73) mean percep tion was very similar to the mean the friend reported ( = 1.53, SD = 0.68) for minor delinquency items. It was my intention to measure delinquency as completely as possible given the goals of this thesis. As such, there are a large amount of delinquency subscales that were created. In order to make these scales as clear as possible, consult Appendix E to see detailed information o n which items were considered property crime, substance use, violent behavior, minor delinquency, groupy crimes and individual istic crimes. Not surprisingly, many of the delinquency scales that are being used for this thesis are highly correlated with one another. To see the correlations between the individual scales used, consult Appendix F. Friendship Quality Past studie s of social networks and best friend dyads have utilized unidimensional measures of friendship involvement (Haynie, 2001; 2002; Haynie and Osgood, 2005; Kandel, 1978a; 1978b; Kandel and Davies, 1991; McGloin, 2 009) and several of these studies have incorre ctly labeled the construct measured as friendship intimacy (Kandel, 1978a; 1978b; Kandel and Davies, 1991). Friendships are marked by
80 multiple ties of closeness (Bukowski et al., 1994; Furman, 1996; Gavin and Furman, 1996), indicating that a correct measu re of friendship quality should not only indicate the amount of time friends spend together but also address different sectors of the relationship. To this point, this thesis extends prior criminological literature due to the inclusion of a multifaceted m easure of friendship intensity that has become common in the psychological sciences. Friendship strength within dyads was measured by the Friendship Qualities Scale (FQS), a well tested measure developed by Bukowski, Hoza and Boivin (1994). The FQS is a 23 item instrument originally intended for use on adolescents, but research has successfully applied this measure to older populations (Brendgen et al. 2001; Furman, 1996; Gavin and Furman, 1996; Saferstein, et al., 2005). When administered to adult su bjects, the measure shows high reliability and is very simple for participants to complete (see Saferstein et al., 2005). This measure is central to the fourth research question which attempts to understand how level of friendship quality alters the accur The FQS is a multidimensional measure of friendship intensity that is disaggregated into five sub scales measuring companionship, conflict, help (formed from two categories aid and protectio n from victimization), security (formed from two categories transcending problems and trust), and closeness (formed from two categories reflected appraisal and an affective bond) (see Figure 3 1). Since the scale does include subsections, the order of the questions on the instrument was randomized in order to have participants not remain on any one sub category for multiple questions (see Appendix C, part III). All scale items were measured on a five
81 point Likert scale so that higher scores indicated higher levels of friendship intensity. Examples of items on the FQS instrument include : someth ing, m 2 Reliabilities for all five of internal consistency = 4.04, SD = 0.61), and closen = 3.96, SD = 0.52). The overall FQS scale and the subscales were created by adding the items together and dividing by the total number of items for that measure. Al l scales were coded so that scores more near five indicated higher levels of friendship quality and scores nearer to one represented lower friendship quality. Further detail on the deployment of these measures is included in the analytical plan section. Analytic Strategy It should be noted that nearly all dependent variables exhibited some degree of skewness (as is normally the case with delinquency and crime measures in the general population). As such, regression models were estimated two ways. The f irst approach assumed normality and did not correct for skewness. Each regression model was then re estimated using the natural log of the dependent variable. Taking the natural log of
82 the dependent variable adjusts or corrects for skewness, thus not vio lating an assumption of the Ordinary Least Squares (OLS) regression model. Despite taking the extra precaution, the OLS outcomes were nearly unchanged. Because of this, non logged models are reported. All models were also checked for outliers 5 heterosk edasticity 6 and collinearity 7 Research Question One. that friend actually self reported for his/her own delinquency. To properly test this, I report. A bivariate OLS regression is estimated that regresses reported b 5 Outliers were checked for in three different ways. First, studentized residuals were created for all independent variables. Second, leverage statistics associated with the independent variables were difference in the beta coefficient if an observation were to be removed from the an alysis. To this point, if an observation for any of the models proved to be very askew from the rest of the observations, it was removed. Thus, although all regression models began with a sample size of 485, they have differing sample sizes depending on the severity of the outliers in the model. 6 Each reported model was also checked for heteroskedasticity. Heteroskedasticity is a tendency for the error terms to have a non constant variance around the predicted regression line. Thus, having heteroske dasticity in a regression model affects the t value and is a violation of the Gauss Markov assumptions in that the distance between the error terms do not have a systematic deviation from the regression line (see White, 1980). Many independent variables d id show a significant amount of heteroskedasticity. To detect heteroskedasticity, the Breusch & Pagan test (1979) for heteroskedasticity was run on all models. When the Breusch & Pagan test statistic was significant, the model was run with robust standar values. 7 Collinearity was also examined for those regression models that contained more than one p redictor be a major concern for the modeling except for the third research question. This particular research question showed signs of severe c ollinearity between its two independent variables. To correct for this, the third research question.
83 test or model, this particular regression is crucial to understanding the remainder of results. Research Question Two. Research question two asks what types of crime a subject could predict for a friend. This question utilized a series of six bivariate OLS regression models that use a nding reported delinquency. Two operationalizations were implemented in measuring delinquency in this question. The first seeks to determine whether a subject is better at predicting groupy or non groupy delinquency of his/her groupy offenses than their individualistic offenses. The second operationalization of delinquency seeks to determine which genres of crime (property, violence, su bstance use, and minor delinquency) a subject can best predict for a friend. For the latter of the operationalizations of delinquency four sets of indices were created from the overall delinquency scales measuring property offenses, violence, substance u se and minor delinquency. The first perspective on how perceptual delinquency should relate to actual The first OLS regression model (model one) regresses what grou py crimes a friend crimes that his or her friend commits (independent variable). I hypothesized that a ld better predict the that the friend commits. The second model is another bivariate OLS regression model
84 reported individualistic del inquency (dependent variable) variable). The second theoretical perspective on perceptual and actual peer delinquency measures comes about by addressing which genre s of criminal behavior a subject can best predict for a friend. This perspective takes four common criminological crime categories property crime, substance use, violence, and minor delinquency and seeks to determine with what accuracy a subject can p reported delinquency. Model three seeks to determine the extent to which someone is capable reported property crime as a dependent variable reported violent behavior ity (independent variable). This will allow us to see if a subject is capable of estimating his reported violence The fifth bivariate OLS regression model seeks to determi ne with what accuracy reported substance individualistic crime, here it should be noted that since substance use is predominantly a group activity all the substance items are also classified as groupy behaviors.
85 minor delinquency (independent variable) and uses it to determine how i ndicative it is of reported minor delinquency (dependent variable). Research Question Three. asure are justified. The projection hypothesis is a major criticism of the use of a perceived peer delinquency i dentified delinquency, reported peer delinquency is just another measure of se lf 597). Despite some literature on the topic (see, for example, Thornberry and Krohn reported delinquency is the same as question, two approaches were implemented. First, a factor analysis was model ed to examine the internal structure validity of the two measures to determine if they were in fact justifiably the same measure. Second, a bivariate OLS regression was utilized that reported delinquency as the independent vari able and that in order to test the projection hypothesis in past research (e.g., Agnew, 1991; Thornberry et al., 1994) measures of only one person were used. Implementin g a similar model will enable me to examine whether the perceived and actual delinquency measures of the same person are in fact justifiably identical. However, Gottfredson and Hirschi (1987; 1990) suggest that for projection to be more thoroughly examine d another step must be taken. Theoretically, since they claim
86 that the self reported and perceptual peer delinquency measures are the same, they measure self reported delinquency. In order to determine this, the reported delinquency measure were both included in a multivariate OLS regression model that reporte d delinquency (dependent variable). This model, reported estimate the effects of the two in the fac e o (1997) w as implemented 8 Research Question Four. The fourth research question asks if a multidimensional measure of friendship quality is capable of moderating the relationship between a subj does not disaggregate the delinquency measures into subscales, but rather uses the full, 26 item perceptual peer delinquency measure as an indepen dent variable and the report delinquency scale as the dependent variable. To empirically address the moderating effect of friendship quality, the 8 This technique was developed by Dennis Roncek but never published in literature, although it was presented at a conference It is only effective on models with two independent variables. The goal of residualization is to determine two different effects for each independent variable a maximum and minimum effect. In order to residualize the regression equation, one independent variable (X1) is regressed on the other independent variable (X2). From this regression equation, a residual term (R1) is 1 residual term is then included in a regression model as an independent variable with X2 that predicts the dependent variable (Y). This model, which no longer shows symptoms of severe collinearity, will indicate the maximum effect of X2 on Y and the mini mum effect of X1 on Y (by interpreting the R1 residual term). The inverse process is also performed. X2 is regressed on X1 and a residual term (R2) is created. The R2 residual is then included as an independent variable in a model that regresses Y onto X1 and R2. The R2 residual is indicative of the minimum effect of the X2 variable on Y, while the X1 coefficient in this equation serves as the maximum effect of X1 on Y.
87 friend ship quality indicator) were used to create a multiplicative interaction term. In creating a multiplicative interaction it is first necessary to mean center the two independent variables so that problems associated with multicollinearity are reduced when including an interaction and its two component terms into a regression model (Bond and Fox, 2007). To create the multiplicative interaction term, the mean centered perceptual peer delinquency (PPD) and the mean centered FQS measure were multiplied. When a model is run with a multiplicative interaction term, the coefficients of uninterpretable (Baron and Kenny, 1986; Bond and Fox, 2007). As such, if the coefficient for the perceive d peer delinquency FQS interaction is significant, that would be indicative that the friendship quality is moderating the relationship between perceptual and actual delinquency measures.
88 Table 3 1. Demographic characteristics of subject and friend i n dyadic sample (n = 485). Subject Demographics Friend Demographics Mean & standard deviation Minimum value Maximum value Mean & standard deviation Minimum value Maximum value Sex 0.318 0.466 0 1 0.353 0.478 0 1 Hispanic 0.177 0.382 0 1 0. 181 0.386 0 1 Employed in past year 0.645 0.479 0 1 0.685 0.465 0 1 Sexually active 0.581 0.494 0 1 0.565 0.496 0 1 Social Greek 0.194 0.396 0 1 0.192 0.394 0 1 Age 19.260 1.323 18 29 19.320 1.253 18 26 White 0.631 0.483 0 1 0.637 0.481 0 1 Black 0.136 0.343 0 1 0.138 0.345 0 1 Asian 0.080 0.272 0 1 0.076 0.266 0 1 Pacific Islander 0.014 0.119 0 1 0.010 0.101 0 1 Other Race 0.138 0.345 0 1 0.138 0.345 0 1 Freshman 0.493 0.500 0 1 0.458 0.499 0 1 Sophomore 0.287 0.453 0 1 0.282 0.451 0 1 Juni or 0.153 0.360 0 1 0.173 0.379 0 1 Senior 0.068 0.252 0 1 0.087 0.282 0 1 Single 0.557 0.497 0 1 0.592 0.492 0 1
89 Table 3 1. Continued Subject Demographics Friend Demographics Mean & standard deviation Minimum value Maximum value Mean & standard d eviation Minimum value Maximum value Casually Dating 0.124 0.330 0 1 0.103 0.304 0 1 In committed relationship 0.320 0.467 0 1 0.303 0.460 0 1 Parents married 0.707 0.456 0 1 0.709 0.455 0 1 Parents widowed, divorced, or separated 0.293 0.456 0 1 0.29 1 0.455 0 1 Never attend religious service 0.429 0.495 0 1 0.408 0.492 0 1 Religious service once a month 0.256 0.437 0 1 0.254 0.436 0 1 Religious service 2 3 times a month 0.122 0.327 0 1 0.111 0.315 0 1 Religious service once a week 0.151 0.358 0 1 0.163 0.370 0 1 Religious service more once a week 0.043 0.204 0 1 0.064 0.245 0 1 Format: Mean Std. Deviation Coding: Sex (1 = male, 0 = female); Hispanic (1 = yes, 0 = no); Employed in past year (1 = yes; 0 = no); Sexually activ e (1 = yes; 0 = no); Social Greek (1 = yes; 0 = no); Age (write in value); Race: White (1 = white; 0 = not); Black (1 = black; 0 = not); Asian (1 = Asian; 0 = not); Pacific Islander (1 = Pacific Islander; 0 = not); Other race (1 = Other race; 0 = not); Cla ss rank: Freshman (1 = yes; 0 = no); Sophomore (1 = yes; 0 = no); Junior (1 = yes; 0 = no); Senior (1 = yes; 0 = no); Current relationship: Single (1 = yes; 0 = no); Casually Dating (1 = yes; 0 = no); In committed relationship (1 = yes; 0 = no); Parental m arital status: Married (1 = yes; 0 = no); Widowed, divorced, or separated (1 = yes; 0 = no); Religious service attendance: Never (1 = never; 0 = attends a religious service); Once a month (1 = once a month; 0 = other attendance); Two or three times a month (1 = two or three times; 0 = other attendance); Once a week (1 = once a week; 0 = other attendance); More than once a week (1 = more than once a week; 0 = other attendance).
90 Figure 3 1. Conc eptual model of the scales and subscales of the Friendship Qualities Scale (taken from Bukowski et al., 1994, p. 474). Friendship Companionship Companionship Conflict Help Security Closeness Conflict Aid Protection from Victimization Transcending Problems Trust Reflected Appraisal Affective Bond
91 Friendship Qualities Scale Items organized by Friendship Subscale Scale Subscale Item Companionship My friend and I spend all our free time together. My friend thinks of fun things for us to do together. My friend and I go to and on weekends. Sometimes my friend and I just sit ar ound and talk about things like school, spo rts, and things we like. Conflict I can get into fights with my friend. My friend can bug me or ann oy me even though I ask him/her not to. My friend and I can argue a lot. My friend and I disagree about many things. Help Aid If I fo rgot my lunch or needed a little money, my friend would loan it to me. My friend helps me when I am having trouble with something. My friend would help me if I needed it. Protection If other kids were bothering me, my friend would help me. My friend would stick up for me if another kid was causing me trouble. Security Trust If I have a problem at school or at home, I can talk to my friend about it. If there is something bothering me I can tell my friend about it even if it is something I cannot tell to other people. Transcending If I said I was sorry after I had a fight wit h my friend, problems he/she would still stay mad at me. If my friend or I do something that bothers the other one of us, we c an make up easily. If my friend and I have a fight or argument, we can Closeness Affective If my friend had to move away, I would miss him/her. bond I feel happy when I am with my friend. I think about my friend even when my friend is not around. Reflected When I do a good job at something, my friend is appraisal happy for me. Sometimes my friend does things for me, or makes me feel special. F igure 3 2. The 23 ite m Friendship Qualities Scale (taken from Bukowski et al., 1994, p. 475).
92 CHAPTER 4 RESULTS Research Question One. The following pages present answers to the research questions that were laid out at the end of chapter two, using the methodological plan t hat was laid out in chapter reported delinquency. Figure 4 1 is a visual representation of t he inaccuracies in the perceptual peer delinquency scale. Notice how the distribution is a traditional, bell shaped curve. This indicates delinquency roughly equivalently. It is also quite noteworthy that responses were completely accurate more times than they were skewed in any one direction. This can be seen by examining the zero value on the x axis. Responses that are on zero indicate subject perceptions of peer delinquency that w ere totally accurate. Table 4 9 10 This s well as the findings in 9 It should be noted that the findings presented in this chapter were also run rec iprocally. That is, models research question one, the self reported delinquency was regressed on the perception of that delinquency; this is the reported finding. However, the inverse model was also run in which the perceptual delinquency measure was the independent variable and the self reported delinquency was the dependent variable This was done for all reported research findings, and the results were extremely similar in all cases. As such, these findings were not reported. However, knowing that all reported findings were confirmed using inverse delinquency measures allows for a more certain picture that the results presented herein are as accurate as the sampling strategy allows. 10 The research questions in this thesis require directional effects to be tested. That is, they use one should be noted that this is not a causal relationship that is being tested but rather the models in this ion of his/her reported delinquency.
93 understanding the other results that are presented. A bivariate OLS regression was ncy was a robust The evidence suggests that a research subject is capable of predicting the delinquency of a friend, albeit not perfectly. The overall model is also significant (F = 91.53, p < .001 ). The model explains nearly 30% of the variance (R = 0.294), which is reported delinquency, indicating that of that delinquency also increases at a statistically significant moderate level (b = 0.467) 11 Research Question Two. The second research question inquires as to how well differe nt self reported delinquency. Two sets of operationalizations of delinquency are explored. Second, property crime, violent crime, substance use, and minor delinquency were each examined. Results of these regressions are found in Tables 4 2 through 4 7. Table 4 perception of his fri 11 Coefficients are the only report ed findings in models with only delinquency scales in them. Beta weights are not necessary in these models because all of the delinquency items were measured using the same Likert based response. As such, coefficients can be interpreted as beta weights i n these type of models. Beta weights were only used in the fourth research question because of the inclusion of the Friendship Qualities Scale.
94 Al groupy delinquency (b = 0.551, F = 128.74, p < .001 R = 0.367). Interestingly, the 26 item delinquency measure that was displayed in the baseline model. The level of explained variance als o increases from roughly 30% to roughly 37%, indicating that a groupy perceptual peer delinquency reported groupy delinquency than a reported full delinqu ency measure. Table 4 3 displays results for individualistic offenses. To explore the predictive reported indi vidualistic offending on The overall model remains significant, but the F stat drops over tenfold (F = 9.00, p > .01; b = 0.112, p > .01) in comparison to the groupy model. The explaine d variance in the groupy delinquency model is nearly 20 times higher than the explained variance self reported individualistic offending, it is not nearly as robust of a predictor as the actual individualistic delinquency is
95 perceptual individualistic coefficient is five times lower than that of the respective coefficient in the groupy model. Tabl es 4 4 through 4 self reported property crime, violence, substance use, and minor delinquency onto the Table 4 4 shows results from an OLS model repo rted property delinquency (F = 1.01, non significant). Furthermore, the model accounts for nearly no explained variance (R = .002) and leaves over 99% of the variance in the stochastic error term. Based on these not at all reflective of what the friend self reports. Table 4 delinquency. Again, as with the previous property crime model, the model is non significant (F = 1.51, non statistically significant levels (b = 0.055), and this model expl ains the same amount of variance as its predecessor (R = .003). To this point, the evidence suggests that a delinquency whatsoever. Table 4 6 shows OLS regression result self
96 The model operates v ery efficiently (F = 175.08, p < .001 ) and has quite a high explained variance (R = 0.461), thus mini ghly significant (t = 13.23, p < .001 ) and has a large coe fficient as well (b = 0.658, p < .001 ). The model has a strongly significant positive relation reported substance t s use is a robust and reflective predictor of what substances that friend reports using. Table 4 The over all model is significant (F = 19.56, p < .001 gnificant effect (b = 0.194, p < .001 ). Evidence suggests that a research participant is capable of explainin g the explained variance in this model is low, meaning that minor forms of delinquency perhaps are not the best method of operationalizing a perceptual peer delinquency m easure. Research Question Three. The third research question that this thesis addresses is the projection hypothesis (see Gottfredson and Hirschi, 1987; 1990). It should be noted that the self reported delinquency measures are cor related rather highly (0.71, p < .001 ). This correlation is slightly higher than those found by Thornberry and colleagues (1994), who reported a
97 correlation of .50 to .59 between the two measures across three wav es of data. Agnew (1991) reported that his perceptual and self reported delinquency measures were correlated at .70 and .76 for two waves of data, which suggests that the correlation found between the measures in this study is comparable with prior litera ture. This is a p. 144). To ensure that the two measures were not the same, a principle components analysis (P CA) was performed. Two interpretations of the results were used. The first interpretation of the factor analysis, presented in Figure 4 2, examined from appearing roughly parallel with the y axis to roughly parallel with the x axis (hence could suggest that the two delinquency scales have the same number of internal factors, potentially servi ng as evidence that the measures are quite similar. However, reported delinquency broke after the fifth component. This suggests that the measures are indeed two separate constructs. To confirm this finding, a second interpretation of the PCA was examined. Table 4 8 shows the second interpretation of the PCA that evaluated the eigenvalues of the individual components of the two scales. Eigenvalue s were interpreted so that values over one indicated a component being retained. The reported delinquency measure retained nine
98 internal factors. The fact that the two scales did not retain the same number of principle components in either interpretation of the factor analysis is indicative that they are not measuring the same thing. Table 4 9 shows a bivariate OLS regression t 26 item self 26 item perceptual peer delinquency measure. The model is statistically significant (F = 281.18, p < .001 ) and explains nearly half of the overall variance ( reported delinquency (b = 0.615, p < .001 ). However, the explained variance is too low to his or her self reported delinquency are interrelated, they are two separate cons tructs. In addition to the prior methods, I attempted to delineate the joint effects of the reported delinquency and perceptual peer delinquency measures when reported delinquency was regressed onto them. In order to do thi s, (1997) which was designed for instances when there are two highly collinear independent variables in a regression model, was when there are two hi ghly collinear independent variables in a regression model After running the analysis, t he results from t he modeling were inconclusive. A weakness to the residualization technique is that there are instances where results are uninterpretable. Although the results from this model could not be interpreted, the principle components
99 reported and perceptual peer delinquency measures are separate constructs. Research Question Four. The final research questio n is designed to address how friendship quality may reported delinquency. Friendship quality, measured by the Friendship Quality Scale (FQS; see Bukowski et al ., 1994), is theorized to be a significant reported reported delinquency. Table 4 10 shows results from an OLS regression that assessed th e interaction reported delinquency. To statistically approach this question a multivariate OLS regression was implemented. The full, 26 item delinquency scale and the full, 23 show that the interaction was non significant 12 indicating that quality of friendship did not impact the predictive ability of the perceptual peer delinquency measure. T hat is, those subjects who have better quality friendships with their friend were no more likely to 12 In a regression model that has a multiplicative interaction, the only interpretation that can be made is of 9. We can see that the interaction term not only fails to reach significance, but the coefficient is nearly zero (b = 0.033).
100 Figure 4 reported delinquency.
101 Table 4 1. Ordinary Least Squares regression mod reported delinquency n = 47 0 b RSE t value Confidence Interval Lower Upper 0.467*** 0.049 9.57 0.371 0.562 Constant 0.799*** 0.065 12.30 0.671 0.926 F Statistic = 91.53*** R = 0.294 *** denotes significance at p < .001 Self Reported Delinquency Perception of Delinquency Subject Friend b = 0.467*** D V IV
102 Table 4 2. Ordin groupy delinquency. n = 469 b RSE t value Confidence Interval Lower Upper Subject PPD 0.551*** 0.049 11.35 0.455 0.646 Constant 0.739*** 0.069 10.72 0.604 0.875 F Statistic = 128.74*** R = 0.367 *** denotes significance at p < .001 Self Reported Groupy Delinquency Perception of G roupy Delinquency Subject Friend b = 0.551*** D V IV
103 Table 4 3 Ordinary Least Squares regression mod el regressing the frien reported individualistic d elinquency n = 473 b SE t value Confidence Interval Lower Upper Individualistic PP D 0.112** 0.037 3.00 0.039 0.185 Constant 1.081*** 0.045 24.04 0.993 1.169 F Statistic = 9.00** R = 0. 019 ** denotes significance at p < .01 *** de notes significance at p < .001 Self Reported Individualistic Delinquency Perception of Individualistic Delinquency Subject Friend b = 0.112** D V IV
104 Table 4 4. Ordinary Least Squares regression mod el regres reported property delinquency property delinquency. n = 477 b SE t value Confidence Interval Lower Upper PPD 0.044 0.044 1.01 0.042 0.130 Constant 1.104*** 0.049 22.57 1.008 1.200 F Statistic = 1.01 R = 0. 002 *** denotes significance at p < .001 Self Reported Property Delinquency Perception of Proper ty Delinquency Subject Friend b = 0.044 D V IV
105 Table 4 5 Ordinary Least Squares regression mod reported violent delinquency onto th violent delinquency. n = 475 b SE t value Confidence Interval Lo wer Upper PPD 0.055 0.045 1.23 0.033 0.143 Constant 1.080*** 0.051 21.09 0.980 1. 181 F Statistic = 1.51 R = 0. 003 *** denotes significance at p < .001 Self Reported Violent Delinquency Perception of Violent Delinquency Subject Friend b = 0.055 D V IV
106 Table 4 6 reported substance use substance use n = 470 b RSE t value Confidence Interval Lo wer Upper substance use 0.658*** 0.050 13.23 0.561 0.756 Constant 0.697*** 0.083 8.44 0.535 0.860 F Statistic = 175.08*** R = 0. 461 *** d enotes significance at p < .001 Self Reported Substance Use Perception of Substance Use Subject Friend b = 0.658*** D V IV
107 Table 4 7 reported minor delinquency minor delinquency. n = 469 b RSE t value Conf idence Interval Lower Upper PPD 0.194*** 0.044 4.42 0.108 0.281 Constant 1.185*** 0.064 18.39 1.059 1.312 F Statistic = 19.56*** R = 0. 055 *** denotes significance at p < .001 Self Reported Minor Delinquency Perception of Delinquency Subject Friend b = 0.194*** D V IV
108 Figure 4 2. Scre reported delinquency. Elbow Break Elbow Break
109 Table 4 8 Principle Components Analysis of the internal component structure of the ency measure and his/her self reported delinquency measure. Perceptual Peer Delinquency Self Reported Delinquency Eigenvalue Proportion Cum. Proportion Eigenvalue Proportion Cum. Proportion Comp. 1 7.823 0.301 0.301 5.202 0.200 0.200 Comp. 2 2.982 0.1 15 0.416 2.291 0.088 0.288 Comp. 3 1.632 0.063 0.479 2.027 0.078 0.366 Comp. 4 1.409 0.054 0.533 1.776 0.068 0.435 Comp. 5 1.327 0.051 0.584 1.342 0.052 0.486 Comp. 6 1.144 0.044 0.628 1.243 0.048 0.534 Comp. 7 0.969 0.037 0.665 1.182 0.046 0.579 Com p. 8 0.938 0.036 0.701 1.105 0.043 0.622 Comp. 9 0.848 0.033 0.734 1.006 0.039 0.661 Comp. 10 0.788 0.030 0.764 0.876 0.034 0.694 Comp. 11 0.700 0.027 0.791 0.785 0.030 0.724 Comp. 12 0.640 0.025 0.816 0.779 0.030 0.754 Comp. 13 0.569 0.022 0.838 0.72 8 0.028 0.782 Comp. 14 0.545 0.021 0.859 0.685 0.026 0.809 Comp. 15 0.512 0.020 0.878 0.670 0.026 0.835 Comp. 16 0.474 0.018 0.896 0.581 0.022 0.857 Comp. 17 0.459 0.018 0.914 0.530 0.020 0.877 Comp. 18 0.366 0.014 0.928 0.518 0.020 0.897 Comp. 19 0. 330 0.013 0.941 0.492 0.019 0.916 Comp. 20 0.301 0.012 0.953 0.467 0.018 0.934 Comp. 21 0.263 0.010 0.963 0.426 0.016 0.950 Comp. 22 0.257 0.010 0.973 0.396 0.015 0.966 Comp. 23 0.218 0.008 0.981 0.294 0.011 0.977 Comp. 24 0.193 0.007 0.988 0.246 0.01 0 0.986 Comp. 25 0.162 0.006 0.995 0.204 0.008 0.994 Comp. 26 0.141 0.005 1.000 0.150 0.006 1.000
110 Table 4 9 Ordinary Least Squares regression model regressing the reported deli nquency n = 470 b RSE t value Confidence Interval Lower Upper SRD 0.615*** 0.037 16.77 0.543 0.687 Constant 0.482*** 0.052 9.34 0.381 0.583 F Statistic = 281.18 *** R = 0.447 *** denotes sign ificance at p < .001 Self Reported Delinquency Subject Perception of Delinquency b = 0.615*** D V IV
111 Table 4 10 Ordinary Least Squares regression model regr self reported delinquency onto pe rception Friendship Quality indicator) to test for moderation effects. n = 46 0 b RSE t value 0.495*** 0.044 11.26 0.565 Score 0.123*** 0.030 4.11 0.171 FQS PPD Interaction 0.033 0.078 0.42 0.017 Constan t 1.248*** 0.141 8.84 -F Statistic = 54.70*** R = 0. 358 *** denotes significance at p < .001 Self Reported Delinquency Perception of Delinquency Subject Friend = 0.565*** FQS PPD Interaction Term = 0.017 IV IV D V
112 CHAPTER 5 DISCUSSION Past research has not reached a definitive conclusion on how to measure peer delinquency (see Thornberry and Krohn, 1997). Whil e many researchers have utilized erformance (Agnew, 1991; Kandel and Davies, 1991; Haynie, 2001; 2002; McGloi n, 2009; Thornberry et al., 1994; Warr, 1993) and others have claimed that implementing such a measure is highly problematic (Asteline, 1995; Gottfredson and Hirschi, 1987; 1990). Based on the evidence garnered in this study it appears that a research s ubject is capable, although not perfectly, of providing a measure of perceptual peer reported delinquency. This finding is of importance to studies that have implemented perceptual peer delinquency measure s in the past, needless to say which are numerous. Despite this, it would be inaccurate to produces an equally valid peer delinquency indicator. It does appear th at a general, wide range perceptual peer delinquency scale (such as the 26 item scale utilized here) reported delinquency construct somewhat. However, even implementing that sort of measure may be potentially problematic. Evidence from the second research question clearly shows that a subject is reported delinquency. Groupy crimes are delinquent actions that are typically committed in the presence of other s namely, they are substance use, property crimes (excluding shoplifting), and public order violations (see Warr, 2002). Rather, individualistic crimes
113 known to subje cts. Thus, perceptual peer delinquency measures may not work well for individualistic types of crime. To this point, the overall delinquency scale that serves as both the independent and dependent variables in the baseline model (see Table 4 1) consists of 26 items, 17 which are groupy crimes and nine which are individualistic. However, when compared to the perceptual groupy delinquency model (see Table 4 2), the baseline model shows s coefficient, and a lower ANOVA statistic as well. This is quite convincing evidence that the individualistic delinquency items are underperforming and altering the more valid measures of perceptual peer groupy delinquency in the overall delinquency inde x. As such, a word of warning should be issued regarding the inclusion of too many individualistic crimes in a perceptual peer delinquency measure. It seems rather useless, and in fact harmful, to include individualistic items in a perceptual peer delinq uency measure because a subject simply cannot predict them in a manner report. delinquency cannot be overstated. The perceptual groupy peer delinque ncy measure outperformed nearly all other perceptual delinquency measures in explanatory ability. reported delinquency, although substance use and minor delinquency were. Substance use of a friend was highly predicted by a research subject. Interestingly, Warr (2002) contends
114 that substance use is almost completely a groupy activity. Perhap s the reason why peer substance use was so precisely perceived by a subject is because all nine items When alcohol and drug behaviors are compared to the reflectiveness of the erent stories are told. The violence subscale consisted of four items, half of which were individualistic delinquencies and half of which were groupy delinquencies. Warr (2002) states that most violent acts are non groupy delinquencies, and there appears to be violence did not predict more than chance could alone in the model. A similar story can be told for the property crime subscale. But first, notice how the phra are qualitatively very different from each other. For instance, shoplifting is quite different from destroying property that belongs to a school. Stealing from your roommate s is very different from breaking their things. Warr (2002) recognizes this, and operationalizes this category so that shoplifting and theft are individualistic activities but vandalism is not. As such, the property crime scale contains nine items, only four of which are classified as groupy crimes. Once again, the property crime subscale contains several individualistic crimes that a subject simply cannot perceive in a manner report. Three implications for future resea rch, then, can be made about the nature of
115 measuring peer delinquency through a perceptual measure (which is quite common in our field). First, when requesting that a research subject provide a perception of his or the researcher should be wary. The scale that reported including sexual violence and threatening violence, and items about shoplifting and theft. Second, many authors have used a large number of items to create a general delinquency scale in the past. Such scales typically encompass a moderate to high numb However, it has been shown in the prior chapter that including individualistic items in an overall delinquency measure reduces the strength of the reflectiveness of the and reduces the explained variance of the overall model as well. The implication is that by including any individualistic crime items in a general perceptua l peer delinquency index a researcher can underestimate the effects of a perceptual peer delinquency variable in a statistical model. h subject cannot predict his or her self reported individualistic delinquency, then the subject is unlikely to produce an indicative representation of their more abstract, individualistic crimes. Although the following hypothesi s needs further investigation, it is possible that the inaccuracy of these reports across multiple friends could amplify the error variance in perceptual peer
116 group measures. Network studies that employ perceptual peer group delinquency measures (e.g., Ake rs et al., 1979; Agnew, 1991; Krohn et al., 1996; Lanza Kaduce et al., 1984; Radosevich et al., 1979; Thornberry et al., 1994; Warr, 1993) should be wary of the potential for these inconsistencies to potentially alter results in statistical modeling. Seve ral reasons may exist for why groupy crimes are highly predicted by a research subject. First, the subject could have heard about the actions of his or her friend from a mutual accomplice, being that the act was most likely committed in a group setting. Second, the subject could have been present at the time of the deviance themselves, thus attributing direct knowledge of the event to the reporter. Both scenarios seem feasible, and several points should be emphasized regarding these two possibilities and First, if a subject was merely present while a peer vandalized a car, took ecstasy, or used a fake identification to get into a bar does not necessarily mean that the subject is also committing that crime. This seems to be a logically unsubstantiated assumption the subject would have been present while the friend committed those delinquencies. This would lead to them report measure the subject would have no reason to indicate committing those crimes themselves. However, the subject could very likely still have committed other acts (with or without a specifi c friend) that make the two measures correlate to a moderate level. For instance, they probably would have been consuming alcohol in the bar with their friend, could have partook in marijuana use with them and without them, or could have
117 vandalized a buil ding by throwing rocks at the windows. Who is to say, then, that the subject did not commit crime with other friends at some point during the measurement time frame? Would this not create the same correlation between the two measures that Gottfredson and Hirschi speak of? In a standard survey design, delineating whether the incredibly difficult. reported de linquency and perceptual peer delinquency are correlated it produces a situation where the measures are not independent of each other, thus meaning a tautology is being measured. How can this be eliminated? The most glaring answer to this question is to use peer crimes that a subject cannot predict delinquency. Thus, a self reported individualistic delinquency measure would be most likely not be indicative of that self reported behavior. As such, the measures most likely would not be highly correlated (which would be ideal in a regression model). However, while this is an attempt to make peer delinquency measurement more reliable it is not m aking it more valid and in fact is decreasing validity. So, we must ask ourselves why it makes theoretical reported delinquency should not be correlated to their perc eptual peer delinquency measure? This brings up the second point that should be emphasized about the projection hypothesis. Let us control theory. Using postulations from the general theory of crime (Gottfredson and Hirschi,
118 1990) measures of peer and self reported delinquency are identical. The reason for this is s pending time together and partaking in similar activities, despite the postulation that friendships are hard to form when low self control is a factor. Ironically, this point seems very Gottfredson and Hirschi, 1990). The central claim of this argument is that individuals with low self control will seek out other individuals with low self control to spend time with. The authors also claim that low self control is the latent cau se of criminality. If this is in fact true, then theoretically individuals with low self control would intentionally seek out others with low self control to associate with. Quickly, those relationships would become synonymous with crime because both mem bers of the friendship would possess the latent cause of criminality low self control. Thus, since both members of the newly formed friendship are both committing crime it can be speculated that that they will very likely be committing similar crimes, whether it be together or separate. At very least the friends should be aware of the crimes that their peer commits considering their self control levels caused each other to seek out a relationship in the first place. From this discussion, we are left with only one logical theoretical conclusion from reported peer delinquency measure and a perceptual peer delinquency measure, the two measures theoretically should be cor related for two reasons; 1) the individuals had sought each other out to form a relationship and 2) they both possess the latent cause
119 of criminality, low self control. If the measures were not correlated, that would be indicative of a theoretical flaw in that the birds of a feather did not flock together. As reported and perceived peer delinquency are the same and that it is a bad thing is counterintuitive to the general B ut we see, though, that while the measures of perceptual peer delinquency and self reported delinquency are highly related they are in fact separate constructs. Thus, Gottfredson and Hirschi (1987; 1990) were correct in saying that the measures are very similar, but they were incorrect in the postulation that they are same. If a subject did indeed project themselves into the perceptual delinquency measure of a peer, the measur reported delinquency and perceptual peer delinquency would have been the same, thus meaning that the same construct, self reported delinquency, was measured twice in many studies (Thornberry and Krohn, 1997). Knowing that the two constructs contain separate internal components is a useful finding to scholars because we can be sure that obtaining a perceptual peer delinquency report from only one research subject is not an invalid method of obtaining a measure of peer delinquency. Surprisingly, friendship quality did not play a significant role in the accuracy of the the relationship is reciprocal should be conducted. That is, this study o nly implemented would be interesting to see if the friend produced a friendship intensity measure
120 ip quality by no means needs to be necessarily shared by the friend; rather, it could be totally different (see, for example, Kandel, 1978a). It could be that the moderating effect that was hypothesized but not observed could only appear when reciprocal f riendships are measured. reported delinquency. Thus, the underlying broad intention was to find a method that can ensure th at the perception of report as possible when only surveying one person. Based on the evidence presented here, it seems of little use to include a non reciprocal measure of friendship quality into a model that uses a perceptual peer delinquency measure. The friendship quality indicator shows no moderating effect with a perceptual delinquency measure, indicating that a perceived peer delinquency scale is not conditioned by friendship quality.
121 CH APTER 6 C ONCLUSION Since the advent of criminology as a discipline, and especially since the construct has been of central importance to scholars. A very large body of literature in criminology has included the construct as a measure. Given this fact, it is rather surprising that the method through which peer delinquency is measured has not garnered more scrutiny than it has. Despite the surprising lack of knowle dge on the measurement of the delinquent peers construct, prior forms of operationalizing the variable have yielded powerful is a better predictor of criminal behav ior than the number of delinquent friends an necessary to control for association with delinquent peers because this variable may have a causal effect on both delinq p. 137). The most common form of measuring delinquent peers has been through a perceptual delinquency measure which entails requesting that one respondent report on overwhelmingly common (see Thornberry and Krohn, 1997), questions about this form of measurement have been raised by scholars. The best recognized of these criticisms reported delinquency is delinquency (Gottfredson and Hirschi, 1987; 1990; see also Asteline, 1995). If this is association with deviant peers and self re ported deviance would be useless because
122 the measures are not independe 222). construct and divulge 1) if a perceptual peer delinquency measure is reflective of what delinquencies the peer reports, 2) what behaviors the peer perceptual measure is most reflective for and 3) under what conditions the perceptual peer delinquency measure is nravel the peer delinquency Evidence suggests that a subject is capable of providing a fairly representative perception of what crimes his/her friend is committing. However, the strength of the accuracy of that perception varies based on which peer crimes a subject is asked to predict. The overall perceptual delinquency scale, which was not disaggregated into subscales of specific crimes, was not as reflective of the perceptual groupy crime delinquency scale. This study suggests that groupy crimes (see Warr, 2002) are the most precise form of peer delinquency that can be perceived by a single research subject. To this point, my predic delinquency subscale was s ignificantly predicted by the subject it s predictive power was five times less than the perceptual groupy crime measure It is recommended that perceptual individualistic crime items such as theft, violence, and sexual battery be excluded from perceptual peer delinquency measures because these behaviors are not validly reported by a research participant.
123 (1987; 1990) projection hypothesis was not supported. Furthermo re, the projection hypothesis does not seem altogether logical based on the claims of the general theory of crime (1990). The authors claim that individuals with low self control will seek out other individuals with low self control to spend time with (th argument). As such, these individuals both possess the theorized latent trait of criminality low self control. Because these individuals marked by low self control are spending time with each other, they should be able to predict accurately. Their new relationship most likely signifies that the friends are committing similar criminal acts and very possibly could be committing those delinquencies together. The measures then should be moderately to h ighly correlated because the criminal individuals sought each other out to form a relationship centered around crime. delinquency would be more precise across higher levels o f friendship was not supported. Very clear evidence shows that friendship intensity has no relation to whether the subject can more validly represent his or her pee wa s somewhat unexpected Upon conceptualizing the procedure s of gathering data, I study. I thought that by asking this I would get friendships that are very close, moderately close, and then some individuals would come that onl y peripherally knew each other. Based on looking at the graphic distributions of the Friendship Qualities Scale, this did in fact happen. Thus, while the FQS measure was distributed the way I expected, it can be definitively said that based on this data the friendship intensity
124 measure does very little. If indeed friendship intensity does not alter the effectiveness to which a subject of measurement error, an impl ication for measurement can be made. When requesting that one participant report on the delinquency of his/her best friend, the researcher may not need to inquire how intense that friendship is. Instead, the delinquency of a best friend, regardless of th e intensity of that friendship, could have a stagnant effect on an But, i f one is asking about a group of friends, this is not the case. Research shows that the delinquency of a best friend matte rs beyond the delinquency of a more measures of friendship intensity will be of importance. While the findings of this study are interesting, several limitations should be acknowledged. First, as with all student samples, the results are most likely not generalizable to a general sample of the population or even a more representative sample of studen ts. Another potential limitation is that this study restricted the participants to bringing one of their five best friends to the study that was also a student at the university. It is possible that a participant could have defined one of their five best university friends differently than one of their five best non university friends. To this point, though, because this was the first time that the research questions proposed here had been tested I feel this method of sampling is appropriate. Furthermor e, many studies in criminology and other related social science fields regularly use college
125 student samples. Another drawback to the research design was that it was cross sectional. In the past, dyadic researchers have warned that relationships are flu id and are constantly changing, and due to this fact longitudinal data is much more advantageous than cross sectional designs (Kandel, 1978b). Several studies suggest that cross sectional on substance use and minor delinquency tendencies (Kandel, 1978b; 1996). Also, friendship groups are fluctuant and members of friendship groups identify different members as close friends at different times (Haynie and Osgood, 2005). Thus, lacking the a bility to observe the stability of the friendships that were measured here is a limitation of the design of this thesis. Another possible limitation is that since this study was offered for extra credit, a selection bias may have occurred. Since participa tion was voluntary, it is possible that this sample could be weighted towards having more students that were concerned for their grade than those who were not. Past research has showed that students with lower grade point averages tend to have higher numb ers of delinquent friends (Haynie, 2002). Thus, this sample could be weighted towards having less delinquency than would actually be observed in the true university student body. Despite the limitations though, this thesis presents findings that are of importance and relevance to criminology. Knowing that a research subject is largely incapable of reporting on individualistic types of delinquency offers several relevant implications for improving future measurement of the delinquent peers construct. Fi desire is to capture the true delinquency of a peer through a perceptual peer
126 delinquency measure, asking groupy crime items will give a fairly indicative measure of the true delinquencies of the friend. However, asking individualis tic crime items will produce a perceptual measure that is completely unrelated to the crime of the peer. Second, a large amount of prior literature has focused on perceptual peer delinquency measures by either condemning their use (e.g., Asteline, 1995; G ottfredson and Hirschi, 1987; 1990; Weerman and Smeenk, 2005) or by noting that perceptual measures are in fact useful (e.g., Akers, 1991; Agnew, 1991; Thornberry et al., 1994, Warr, 2002). To this point, it is quite beneficial to criminologists to learn that a perceptual measure of reported delinquency. The unfortunate implication garnered from this evidence is that results for past studies that have included perceptual indivi dualistic types of peer delinquency measures may have suffered the same effects that were observed here. Indices that contain more than several perceptual individualistic peer crime items were marked by low F statistics, drastically lowered coefficients f or the perceptual peer delinquency variable, and lowered explained variance in the model. Clearly, there is reason to expect that prior studies that have included these items may have underestimated the one misspecified variable obviously has implications for the rest of the variables that were measured as well. Future studies should seek to replicate these findings in other populations, including adolescent samples longitudinal samples, and most importantly nationally representative studies. It would also be interesting to investigate if individuals in
127 criminal (e.g., incarcerated, probational, etc.) populations are capable of predicting the delinquencies of their friends. Furthermore it may be beneficial to perform a study akin to this one with multiple friends instead of merely a dyad. Criminologists, and especially those interested in network analyses, are in need of further knowledge on the performance of pe rceptual peer group delinquency measures. A comprehensive understanding of how peer delinquency should be measured is crucial to the field of criminology. Peer delinquency is one of the most (if not the most) consistent predictive variables of individua l level criminality of which we are aware. This fact only increases the urgency for the measurement of this construct to be completely understood. Measurement is the most vital aspect to science. Constructs, theories and hypotheses must all be validly t ested and measured if unbiased conclusions are to be made. The use of a traditionally untested measure, such as perceptual peer delinquency, calls past findings into question. Thus, criminologists should further pursue how to measure the peer delinquency construct instead of assuming a priori that past measurement precedents were valid.
128 APPENDIX A IRB PROTOCOL, IRB AP PROVAL, AND IRB REVISION APPROVA L This appendix contains the original Institutional Review Board (IRB) Protocol for the current study, the subsequent approval that the IRB submitted, and the IRB Revision that was submitted that requested an increase in sample size for the study. The following pages contain these documents
129 UFIRB 02 Social and Behavioral Research Protocol Submission Title of Protocol: Delinquent Peers, Self Control, and Deviance Principal Investigator: John Boman UFID #: 1942 1161 Degree / Title: Graduate Student Department: Department of Sociology and Criminology & Law Mailing Address: John Boman 3219 Turl ington Hall PO Box 117330 Gainesville, FL 32611 7330 Email Address & Telephone Number: email@example.com 392 1025 x 212 office Co Investigator(s): UFID#: Supervisor: Dr. Chris Gibson UFID#: Degree / Title: Assist ant Professor of Criminology Department: Department of Sociology and Criminology & Law Mailing Address: 3219 Turlington Hall PO Box 117330 Gainesville, FL 32611 7330 Email Address & Telephone Number: firstname.lastname@example.org u 392 5065 x 206 office Date of Proposed Research: January 1, 2009 through December 31, 2009 Source of Funding (A copy of the grant proposal must be submitted with this protocol if funding is involved): None Scientific Purpose of the Study: The pur delinquency of self control? How does delinquen Describe the Research Methodology in Non Technical Language: ( Explain what will be done with or to the research participant. ) This study involves the use of student pairs. The subject (a n undergraduate student) and one of their space of the Division of Criminology, Law & Society inside the Department of Sociology and Criminology & L aw (Room 3, Walker Hall Basement). The pair of students will be asked which one is taking the survey for credit and which one is the friend. The subject (the student taking the survey for credit) will be given a paper survey printed on both sides of a sm all packet. This survey will be pre coded with a random three digit number that matches the survey that their friend will be given. The use of the code is to ensure
130 anonymity of the participants. The friend will be escorted onto the other side of the conc rete wall into an adjoining room and will be allowed to read the informed consent and take the survey. Likewise, the subject will be allowed to do the same in the other part of the split room. The duo will both complete their surveys and will be both give n a debriefing form when they hand the survey back to the person in charge of administering the survey. The large, split room will be adequate for multiple student dyads to take the survey at once while in complete confidentiality of each other. The data Describe Potential Benefits and Anticipated Risks: ( If risk of physical, psychological or economic harm may be involved, describe the steps taken to protect partic ipant.) The anticipated risks of participating in this study do not exceed any risks encountered in the normal day. The study will be conducted during normal school operating hours which will allow for the student pair. This could help lessen transportat ion risks. There is minimal physical, psychological, or economic harm that will be created that is not already present in everyday life. The only risk associated with this survey method of the student friends is that they will chat with each other later about their responses. We have attempted to control for this the best way possible by including a statement in both the informed consent and the debriefing form that prohibits the students from discussing their responses among themselves, and stressing th at having confidential responses is the golden standard of social science research. The student participants could benefit from taking a study like this. First, since the students are already very good friends, traveling to and from the study will allow friends some potential time to spend together. Second, t of and each of the students could potentially realize that they need to curb their risky behavior. This could indirectly lead to a healthy change in lifestyle habits that promotes the participants own well being. Describe How Participant(s) Will Be R ecruited, the Number and AGE of the Participants, and Proposed Compensation: Participants will be recruited from undergraduate courses at the University of Florida. Students in the class will be informed that they will receive extra credit for participati ng. The potential participants will be asked to come to the split room with one of their five best undergraduate friends at the University of Florida. A requested extra credit of 1% / 100% increase in overall grade will be asked for; however, the researc hers cannot make this decision for the instructors, who will have the final decision. While presenting the study in the classroom, the potential participants will be told that they will need to bring their UF student ID card with them to the study (this i s to make sure that no one other than a UF student will be sampled). Since the subject and the friend are very good friends already, having someone that is not associated with the class should not be a hindrance to either party. Inversely, many undergradu ate students take classes with their best friends, which would enable both parties to be able to receive compensation. The space that this research will take place in is very large. Room 3 of Walker Hall is approximately 12 feet wide by 30 feet long; ther e are two sides that share similar dimensions. There is only one entrance, although there are two exits (one for each room half). The rooms are joined by an empty doorjamb at the far end of the room. Room 3 used to be a very long, split bathroom in a re sidence hall, and half of it was the showers and half of it was the toilet facilities. Participants will be instructed to sit at any of a large number of desks in the room that were put there for use by Criminology graduate students. Thus, the minimum sp ace between participants will be several feet per participant more than a normal seminar classroom at the University of Florida. The participants will then be asked to leave out the exit on the side of the room that they are on in order to not see their f Upon completion, participants will be instructed to place surveys into a sealed and undated manila
131 on the outside in t he very rare chance it is stolen or lost. This envelope will be sealed upon completion next work week in order to ensure privacy of the students. Not da ting the envelope with the surveys inside will ensure that the participants from any given day will not be able to be identified by researchers. A sample size of 800 student pairs (1600 participants) is necessary in order for the research to be able to rea ch the level of statistical power necessary. Only undergraduate students will be allowed to participate in this study. Thus, the age range will be that of a typical random sample of college students. However, no undergraduate student under 18 will be al lowed to participate in this study. Instructors will be asked to provide an alternative extra credit assignment for those students that are under 18 that desire to participate. The participating student dyads will be compensated with extra credit for t he subject. The friend will not be directly compensated. In order to receive compensation, the name of any student seeking extra credit will be recorded on a separate Microsoft Excel Document and their instructors will be notified. A student ID card wil l be shown to the person administering the survey in order to ensure that both of the participants are in fact students. If the friend is also in the class, both students will be recorded for compensation. Describe the Informed Consent Process. Includ e a Copy of the Informed Consent Document: Before either of the participants will be allowed to complete the study, they must both sign the Informed Consent. This document will be given out to both students the subject and the friend. The informed consen t will be labeled with the same three digit code that the pair of surveys is labeled with in order to ensure that they are not mixed up. The students will be given the informed consent document on a piece of paper. They will be asked to carefully read i t, and then sign it and date it. After signing it, they will turn in the original copy, which will be kept by the researchers. A copy of the informed consent will be given to each student for their own records. The informed consent will stress the impo rtance of confidentiality. It will stress that the students not share information about the study with each other, and explain that social science research is held to a high standard of strict confidentiality. Principal Investigator(s) Signature: S upervisor Signature: Department Chair/Center Director Signature: Date:
134 APPENDIX B INFORMED CONSENT This is the Informed Consent document that was signed by all participants i n order to participate in this study. A copy of the Informed Consent was also taken home with participants as it instructs them who to contact with questions. The following page shows the full informed consent document. Informed Consent Delinquent Peers Self Control, and Deviance Please read this consent document carefully before you decide to participate in this study. Purpose of the research study: The purposes of this study are t o 1) evaluate factors affecting crime and delinquency and 2) assess s everal scales that are theorized to measure criminological concepts. This is a paper based survey. What you will be asked to do in the study: Please complete this survey in private. You will be asked to answer a series of questions by circling appropri ate responses and filling in blank lines. Once completed, please return this survey to the researcher in the room. Some questions in this survey ask about self evaluated personal behaviors, including criminal topics such as drug use, theft and violence. Please note that your responses are confidential and cannot be traced back to you. Time required: Most participants will finish in half an hour or less, although some may take up to one hour. Risks and Benefits: class. If you are the subject, the benefit for you participating in this is that you will receive extra credit in the class that you are here for. If you are the friend, the benefit is that you a re helping your f riend by completing this survey. However, if both of you are in the same class, you may both receive extra credit for this. The one potential r isk is that the answers could be discussed between you and your friend at the end of the surve y. It is imperative to stress that you do not share your answers with your friend when the survey is over Confidentiality of responses is the ultimate goal in social science research. Thus, you must not share your answers with anyone. Compensation: There is no compen sation for your participation other than granting extra credit for the subject in the class that they are here for. If both the subject and the friend are in the same class together, they will both be given extra credit. Confidenti ality: Some questions ask you to report personal behaviors, including criminal behaviors such as drug use, theft and violence. However, this survey does not ask you to give information that can link you to a particular event. You will remain completely anonymous in taking this survey. There is no way that anyone can track your personal responses through this paper based survey The only way your names will be recorded is in a document that will be used to communicate to your instructor that you have c ompleted extra credit. In our efforts to protect you, y our responses will be kept confidential to the extent provided by law. They are not traceable. Voluntary participation: Your participation in this study is completely voluntary. There is n o pe nalty for not participating. You will not have any points removed for not participating in this study Right to withdraw from the study: You can stop answering questions at any time and you may skip any question you do not wish to answer. Nothing bad will happen to you if you stop answering the questions or if you skip questions. As long as you complete the survey, you will be given credit. Whom to contact if you have questions about the study:
135 Principal Researcher : John Boman, Graduate St udent, Dep artment of Sociology and Criminology & Law, University of Florida, P.O. Box 117330, Gainesville, FL 32611 7730. P hone: 352 392 1025 ext. 212, email: jboman@u fl.edu Research Supervisor : Chris Gibson Ph.D., Assistant Professor Department of Sociology and Criminology & Law, University of Florida, P.O. Box 117330 Gainesville, FL 32611 7330. P hone: 352 392 0265 ext. 206 email: email@example.com Whom to contact about your rights as a research participant in the study: UFIRB Office, P.O. Box 112250, Univers ity of Florida, Gainesville, FL 32611 2250, phone: 352 392 0433, email: firstname.lastname@example.org Agreement: I have read the procedure described above. I voluntarily agree to participate and I understand that I should keep a copy of this informed consent document f or my records You will be provided a copy of this document upon agreeing to complete this study. Please sign below if you wish to take this survey. If you do not wish to participate, no penalty will be assessed against you or your friend. If you agr ee to this statement, please sign and date on the lines below. _______________________________________________________ ____________________________ Date Approved by University of Florida Institutional Review Board 02 Protocol # 2009 U 0008 For Use Through 01/16/2010
136 APPENDIX C ORIGINAL SURVEY INST RUMENT This survey instrument was given to each participant. The subject and the friend completed identical survey instruments. Each instrument was 14 pages in length and contained a wide range of measures that will be used in manuscripts beyond merely this study.
151 APPENDIX D DEBRIEFING FORM This was the Debriefing Form that was given out to students upon completion of the survey instrument. They took this form and the Informed Consent form home with them. The Debriefing Form was approved by the IRB t o be administered to students following the instrument. Debriefing Form Delinquent Peers, Self Control, and Deviance control has address this question, we asked friend both took the same survey in order to allow researchers, for the first time ever, to determine how delinquent peers are viewed by people of differing levels of self control. This study has profound implications for the two leading criminological theories; Self Control Theory (Gottfredson and Hirschi, 1990) and Social Learning Theory ( Burgess and Akers 1966). Even though you b oth took the same survey, we must stress to not share your answers with anyone, including your friend. Confidentiality of surveys is the gold standard in the social sciences, and your answers should only be known by you. Likewise, we will have no idea wh o answered what as there was no name asked for on the survey you took. The first questions you answered were demographics, which prove very useful in data gathering. Then you started taking circled response, grid type questions. Included in these were s everal scales that measured your level of self control. The first was called the Grasmick et al. Scale, d esigned in 1993 (Grasmick, Tittle, Bursik, Jr., and Arneklev, 1993) self control l evel. The second self control scale you took, which asked you to answer questions from the time you were ages 8 13, 14 18 and 18+ was called the Retrospective Behavioral Self Control Scale (RBS), and is a new measurement scale that was developed by Dr. Be rnd Marcus in Germany in 2003 and 2004 Having both scales in the survey instrument allows for researchers to answer which one is better control variable to self control. You also took a scale that measured your closeness to your friend. This is called the Friendship Qualities Scale, and was developed by Bukowski et al. in 1994. The scale was originally intended for use on a young adolescent population, but literature ha s showed that this scale also works very well for college student samples. This scale control affects their view
152 The hypot hesis that this research has is that the lower the level of self control of the subject, the higher they perceive the delinquency of their friend to be. This can be accurately measured for the first time due to the fact that you and your friend both came delinquency as well as your own actual delinquency. Having both of these measures in the survey will allow the researchers to measure whether or not your own level of s elf control affects whether you think your friend is more delinquent than they really are. Having both of you come allows us to measure your perceptions of their delinquency against what they actually reported themselves. The section on potential conseq uences, which multiple sections at the end of the survey asked about, leads into a debate on whether or not a redefinition of self control theory (proposed by Hirschi in 2004) is acceptable. There has not been much literature on it, but the general premis e of the theory is that Social Control Theory (Hirschi, 1969) is the same as self self control. The thesis of this theory says that the higher someone is bonded to conventional norms the mor e they will consider consequences. Thus, consequence measures of specific crimes were included. questions about this study, you may contact the Principal Investigator, J ohn Boman, at email@example.com firstname.lastname@example.org Full contact information is available on the Informed Consent Document.
153 APPENDIX E LIST OF DELINQUENCY ITEMS, BROKEN DOWN B Y SUBCATEGORY Delinque ncy Index crime (I) or Property (Prop.), Substance (Subs.), Violent (Viol.), Other minor delinquency (Mdel.) P urposely damaged or destroyed property belonging to your friends, neighbors, or roommates G Prop. Purposely damaged or destroyed property belonging to a school G Prop. Purposely damaged or destroyed other property that did not belong to you (not countin g friends or school property) G Prop. Stolen (or tried to steal) things worth $5 or less I Prop. Stolen money or other things from your friends, neighbors, or roommates I Prop. Stolen (or tried to steal) things worth between $5 and $50 I Prop. Stole n (or tried to steal) something worth more than $50 I Prop. Avoided paying for such things as movies, clothing, and food I Prop. Broken into a building or vehicle (or tried to break in) to steal something or just look around G Prop. Drank alcohol G S ubs. Drank more than five alcoholic drinks at once G Subs. Bought or provided liquor for a minor G Subs. G Subs. G Subs. Sold hard drugs such as heroin, cocaine, and LSD G Subs. Used hard drugs such as heroin, cocaine, and LSD G Subs. Used Salvia (salvia divinorum) G Subs. Been drunk in a public place G Subs. Thrown objects (such as rocks, bottles, etc.) at cars or people G Viol. Been involved in a group fight G Viol. Hit (or threatened to hit) other people I Viol.
154 Had (or tried to have) sexual relations with someone against their will I Viol. Lied about your age to gain entrance or to purchase something; for example, lying about your age to buy liquor G Mdel. Had sexual relations with a person other than your significant other I Mdel. Cheated on school tests I Mdel. Been loud, rowdy, or unruly in a public place (disorderly conduct) G Mdel.
155 APPENDIX F CORRELATIONAL MATRI X OF ALL SCALES USED IN ANALYSES On the following page is a correlational matrix of all the scales used in the analyses presented in this thesis.
156 Correlation matrix reporting correlations and p values for all delinquency scales. Subject PPD Full Subject SRD Full Friend SRD Full Subject PPD Groupy Friend SRD Groupy Subject PPD Ind. Friend SRD Ind. Subject PPD Prop. Friend SRD Prop. Subject PPD Violent Friend SRD Violent Subject PPD Subs. Friend SRD Subs. Subject PPD Minor Friend SRD Minor Sub ject PPD Full 1.000 Subject SRD Full 0.714 0.000 1.000 Friend SRD Full. 0.602 0.000 0.461 0.000 1.000 Subject PPD Groupy 0.976 0.000 0.732 0.000 0.639 0.000 1.000 Friend SRD Groupy 0.611 0.000 0 .475 0.000 0.975 0.000 0.659 0.000 1.000 Subject PPD Ind. 0.740 0.000 0.419 0.000 0.290 0.000 0.574 0.000 0.257 0.000 1.000 Friend SRD Ind. 0.332 0.000 0.228 0.000 0.679 0.000 0.312 0.000 0.499 0.000 0.282 0.000 1.000 Subjec t PPD Prop. 0.673 0.000 0.372 0.000 0.241 0.000 0.532 0.000 0.213 0.000 0.877 0.000 0.236 0.000 1.000 Friend SRD Prop. 0.310 0.000 0.200 0.000 0.656 0.000 0.306 0.000 0.517 0.000 0.218 0.000 0.852 0.000 0.197 0.000 1.000 Subject PPD Violen t 0.539 0.000 0.367 0.000 0.216 0.000 0.466 0.000 0.204 0.000 0.582 0.000 0.170 0.000 0.484 0.000 0.125 0.001 1.000 Friend SRD Violent 0.210 0.000 0.147 0.001 0.399 0.000 0.200 0.000 0.330 0.000 0.169 0.000 0.467 0.000 0.158 0.001 0.323 0.000 0.212 0.000 1.000 Subject PPD Subs. 0.915 0.000 0.710 0.000 0.661 0.000 0.967 0.000 0.691 0.000 0.452 0.000 0.292 0.000 0.387 0.000 0.292 0.000 0.334 0.000 0.168 0.000 1.000 Friend SRD Subs. 0.609 0.000 0.480 0.000 0.919 0.000 0.663 0.000 0.962 0.000 0.242 0.000 0.404 0.000 0.186 0.000 0.390 0.000 0.184 0.000 0.200 0.000 0.715 0.000 1.000 Subject PPD Minor 0.836 0.000 0.570 0.000 0.439 0.000 0.779 0.000 0.428 0.000 0.732 0.000 0.299 0.000 0.549 0.000 0.267 0.000 0.415 0.000 0.175 0.000 0.654 0.000 0.400 0.000 1.000 Friend SRD Minor 0.412 0.000 0.320 0.000 0.779 0.000 0.422 0.000 0.723 0.000 0.242 0.000 0.650 0.000 0.218 0.000 0.519 0.000 0.167 0.000 0.322 0.000 0.397 0.000 0.568 0.000 0.373 0 .000 1.000 Format: Correlational Value P valu e of correlation All scales are significantly correlated at p < .001 or p = .000.
157 LIST OF REFERENCES Agnew, R. (1991). A longitudinal test of social control theory and delinquency. Journal of Research in Crime and Delinquency, 28, 126 156. Akers, R. L. (1991). Self control as a general theory of crime. Journal of Quantitative Criminology, 7, 201 211. Akers, R. L. (2009). Social Learning and Social Structure: A General Theory of Crime and Deviance New Brunswick, NJ: Transaction. Ak ers, R. L., Krohn, M. D., Lanza Kaduce, L. & Radosevich, M. (1979). Social learning and deviant behavior: A specific test of a general theory. American Sociological Review, 44, 636 655. Akers, R. L., & Sellers, C. S. (2009). Criminological Theories: Intro duction, Evaluation, and Application Los Angeles, CA: Roxbury Publishing Company. Allison, P. (2000). Multiple imputation for missing data: A cautionary tale. Sociological Methods Research, 28, 301 309. Asteline, Jr., R. H. (1995). A reconsideration of parental and peer influences on adolescent deviance. Journal of Health and Social Behavior, 36, 103 121. Merline, A. C. (2002). The Decline of Substance Use in Young A dulthood: Changes in Social Activities, Roles, and Beliefs. Mahwah, NJ: Lawrence Erlbaum Associates. Baron, R. M., & Kenny, D. A. (1986). The moderator mediator variable distinction in social psychological research: Conceptual, strategic, and statistical considerations. Journal of Personality and Social Psychology, 51, 1173 1182. Bartusch, D. R. J., Lynam, D. R., Moffitt, T. E., & Silva, P. A. (1997). Is age important? Testing a general versus a developmental theory of antisocial behavior. Criminology, 35 13 48. Bauman, K. E., & Fisher, L. A. (1986). On the measurement of friend behavior in research on friend influence and selection: Findings from longitudinal studies of adolescent smoking and drinking. Journal of Youth and Adolescence, 15, 345 353. Bea ver, K. M., Gibson, C. L., Turner, M. G., Delisi, M., Vaughn, M. G., & Holand, A. sticky friends hypothesis. Crime and Delinquency, forthcoming.
158 Bond, T. G., & Fox, C. M. (2 007). Applying the Rasch Model: Fundamental M easurement i n the Human S ciences Mahwah, NJ: Lawrence Erlbaum Associates. Brendgen, M., Markiewicz, D., Doyle, A. B., & Bukowski, W. M. (2001). The relations between friendship quality, ranked friendship prefe behavior with their friends. Merrill Palmer Quarterly, 47, 395 415. Breusch, T. S., & Pagan, A. R. (1979). A simple test for heteroscedasticity and random coefficient variation. Econometrica, 47, 1287 1294. Bukowski, W. M., Hoza, B., & Boivin, M. (1994). Measuring friendship quality during pre and early adolescence: The development and psychometric properties of the Friendship Qualities Scale. Journal of Social and Personal Relationships, 11, 471 484. Burgess, R. L., & Akers, R. L. (1966). A differential association reinforcement theory of criminal behavior. Social Problems, 14, 128 147. Cortina, J. M. (1993). What i s coefficient Alpha? An examinat ion of theory and applications. Journal of Applied Psychology 78(1): 98 104. Cres sey, D. R. (1955). Changing criminals: The application of the theory of differential association. American Journal of Sociology, 61, 116 120. Psychometrika, 12, 1 16. Cronbach, L. J. (1951). Coefficient alpha and the internal structure of tests. Psychometrika, 16, 297 334. Fraser, M., & Hawkins, J. D. (1984a). Social network analysis and drug misuse. Social Service Review, 58, 81 97. Fraser, M., & Hawkins, J. D. (1984b). The soci al networks of opioid abusers. The International Journal of the Addictions, 19, 903 917. Furman, W. (1996). The measurement of friendship perceptions: Conceptual and methodological issues. In Bukowski, W. M., Newcomb, A. F., & Hartup, W. W. (Eds.), The Co mpany They Keep: Friendship in Childhood and Adolescence. New York: Cambridge University Press. friends. Child Development, 67, 375 386.
159 Giordano, P. C., Cernkovich S. A., & Pugh, M. D. (1986). Friendships and delinquency. American Journal of Sociology, 91, 1170 1202. Gottfredson, M. R., & Hirschi, T. (1987). The methodological adequacy of longitudinal research on crime. Criminology, 25, 581 614. Gottfredson, M. R ., & Hirschi, T. (1990). A General Theory of C rime. Stanford, CA: Stanford University Press. Grasmick, H. G., Tittle, C. R., Bursik, Jr., R. J., & Arneklev, B. J. (1993). Testing the core empirical implications of Gottfredson and Hirschi's general theory of crime. Journal of Research in Crime and Delinquency, 30, 5 29. Hawkins, D. J., & Fraser, W. M. (1985). The social networks of drug users: A comparison of competing propositions of control and cultural deviance theories. Social Work Research and Abstrac ts, 21, 3 12. Haynie, D. L. (2001). Delinquent peers revisited: Does network structure matter? The American Journal of Sociology, 106, 1013 1057. Haynie, D. L. (2002). Friendship networks and delinquency: The relative nature of peer delinquency. Journal of Quantitative Criminology, 18, 99 134. Haynie, D. L., & Osgood, D. W. (2005). Reconsidering peers and delinquency: How do peers matter? Social Forces, 84, 1110 1130. Hirschi, T. (1969). Causes of Delinquency Berkeley: University of California Press. Hirschi, T. (2004). Self control and crime. In Baumeister, R. F., & Vohs, K. D. (Eds.), Handbook on Self Regulation: Research, Theory, and Applications New York: Guilford Press. Hu, M., Davies, M., & Kandel, D. B., (2006). Epidemiology and correlates o f daily smoking and nicotine dependence among young adults in the United States. American Journal of Public Health, 96, 299 308. Kandel, D. (1973). Adolescent marihuana use: Role of parents and peers. Science, 181, 1067 1070. Kandel, D. (1978a). Similari ty in real life adolescent friendship pairs. Journal of Personality and Social Psychology, 36, 306 312. Kandel, D. (1978b). Homophily, selection, and socialization in adolescent friendships. American Journal of Sociology, 84, 427 436.
160 Kandel, D. (1996). The parental and peer contexts of adolescent deviance: An algebra of interpersonal influences. Journal of Drug Issues, 26, 289 315. Kandel, D., & Davies, M. (1991). Friendship networks, intimacy, and illicit drug use in young adulthood: A comparison of tw o competing theories. Criminology, 29, 441 469. Kandel, D., & Lesser, G. (1972). Youth in Two Worlds. San Francisco: Josey Bass. Krohn, M. D. (1986). The web of conformity: A network approach to the explanation of delinquent behavior. Social Problems, 33 S81 S93. Krohn, M. D., Lizotte, A. J., Thornberry, T. P., Smith, C., & McDowall, D. (1996). Reciprocal causal relationships among drug use, peers, and beliefs: A five wave panel model. Journal of Drug Issues, 26, 405 428. Krohn, M. D., Massey, J. L., & Zielinski, M. (1988). Role overlap, network multiplexity, and adolescent deviant behavior. Social Psychology Quarterly, 51, 346 356. Lanza Kaduce, L., Akers, R. L., Krohn, M. D., & Radosevich, M. (1984). Cessation of alcohol and drug use among adolescent s: A social learning model. Deviant Behavior, 5, 321 341. Lazarsfeld, A. J., & Merton, R. K. (1954). Friendship as a social process. In Berger, M., Abel, T., & Page, C. H. (Eds.), Freedom and Control in Modern Society Princeton, NJ: Van Nostrand. Marcus B. (2003). An empirical examination of the construct validity of two alternative self control measures. Educational and Psychological Measurement, 63, 674 706. Marcus, B. (2004). Self control in the general theory of crime: Theoretical implications of a measurement problem. Theoretical Criminology, 8, 33 55. Marsden, P. V. (1988). Homogeneity in confiding relations. Social Networks, 10, 57 76. Matsueda, R. L., & Anderson, K. (1998). The dynamics of delinquent peers and delinquent behavior. Criminology, 36, 269 308. McGloin, J. M. (2009). Delinquency balance: Revisiting peer influence. Criminology, 47, 439 477. McGloin, J. M., & Shermer, L. O. (2009). Self control and deviant peer network structure. Journal of Research in Crime and Delinquency, 46, 35 7 2. McPherson, M., Smith Lovin, L., & Cook, J. L. (2001). Birds of a feather: Homophily in social networks. Annual Review of Sociology, 27, 415 444.
161 Menard, S., & Elliott, D. S. (1990a). Longitudinal and cross sectional data collection and analysis in the study of crime and delinquency. Justice Quarterly, 7, 11 55. Menard, S., & Elliott, D. S. (1990b). Self reported offending, maturational reform, and the Easterlin hypothesis. Journal of Quantitative Criminology, 6, 237 267. behavior: A developmental taxonomy. Psychological Review, 100, 674 701. Newcomb, T. M. (1961). The Acquaintance Process New York: Holt, Rinehart, & Winston. Newcomb, T. M. (1963). Stabilities under lying changes in interpersonal attraction. Journal of Abnormal and Social Psychology, 66, 376 386. alley, P. M., Bachman, J. G., & Johnston, L. D. (1983). Reliability and consistency in self reports of drug use. Substance Use & Misuse, 18, 805 824. Pr att, T. C., & Cullen, F. T. (2000). The empirical status of Gottfredson and Hirschi's general theory of crime: A meta analysis. Criminology, 38, 931 964. Radosevich, M., Lanza Kaduce, L., Akers, R. L., & Krohn, M. D., (1979). Sociology of adolescent drug and drinking behavior: A review of the state of the field, part one. Deviant Behavior, 1, 15 35. Reiss, Jr., A. J. (1986). Why are communities important in understanding crime? Crime and Justice, 8, 1 33. Roncek, D. W. (1997). A regression based strateg y for coping with multicollinearity Paper presented at the 1997 Midwest Sociological Society, Des Moines, IA. Saferstein, J. A., Neimeyer, G. J., & Hagans, C. L. (2005). Attachment as a predictor of friendship qualities in college youth. Social Behavior and Personality, 33, 767 776. Sutherland, E. H. (1937). The Professional Thief Chicago: University of Chicago Press. Sutherland, E. H. (1940). White collar criminality. American Sociological Review, 5, 1 12. Sutherland, E. H. (1947). Principles of Cri minology. Philadelphia: Lippincott. Sutherland, E. D., & Cressey, D. R. (1978). Criminology: Tenth Edition Philadelphia: J. B. Lippincott. Thornberry, T. P. (1987). Towards an interactional theory of delinquency. Criminology, 25, 863 891.
162 Thornberry T. P., & Krohn, M. D. (1997). Peer, drug use, and delinquency. In Stoff, D., Breiling, J., & Maser, J. (Eds.), Handbook of Antisocial Behavior New York: Wiley. Thornberry, T. P., Lizotte, A. J., Krohn, M. D., Farnworth, M., & Jang, S. J. (1991). Testin g interactional theory: An examination of reciprocal causal relationships among family, school, and delinquency. Journal of Criminal Law and Criminology, 82, 3 35. Thornberry, T. P., Lizotte, A. J., Krohn, M. D., Farnworth, M., & Jang, S. J. (1994). Delin quent peers, beliefs, and delinquent behavior: A longitudinal test of interactional theory. Criminology, 32, 47 83. Warr, M. (1993). Age, peers, and delinquency. Criminology, 36, 17 40. Warr, M. (2002). Companions in Crime: The Social Aspects of Criminal Conduct Cambridge: Cambridge University Press. Weerman, F., & Smeenk, W. H. (2005). Peer similarity in delinquency for different types of friends: A comparison using two measurement methods. Criminology, 43, 499 524. White, H. (1980). A heteroskedastic ity consistent covariance matrix estimator and a direct test for heteroskedasticity. Econometrica, 48, 817 838.
163 BIOGRAPHICAL SKETCH John Boman was born in 1984 in Atlanta, Georgia, where he was raised. After graduating from The Westminster Schools in 2003, he attended college at Ohio University. He received Bachelor of Arts d egrees in sociology / criminology and in h istory from Ohio University in 2007. Currently, he is a graduate student of the Depart ment of Sociology, Criminology and Law in the d ivision of Criminology, Law & Society at the University of Florida where he plans to continue his pursuit of a PhD His research interests include criminological theory, measurement and operationalization of criminological constructs, biological correlate s to crime, and delinquency in friendships.