1 A COMPARATIVE TEST OF THE SOCIAL STRUCTURE AND SOCIAL LEARNING MODEL OF SUBSTANCE USE AMONG SOUTH KOREAN ADOESCENTS By EUNYOUNG KIM A DISSERTATION PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT OF THE REQUIREMENT FOR THE DEGREE OF DOCTOR OF PHILOSOPHY UNIVERSITY OF FLORIDA 2010
2 @ 2010 Eunyoung Kim
3 I dedicate my dissertation to my wonderful husband, Minwoo, loving parents Jung Nim and Duck H un and my older sister, Soyoung.
4 ACKNOWLEDGMENTS I would like to express appreciation to my major adviser, Dr. Ronald Akers for all his thoughtful advice and wonderful guidance through the years of my Ph.D studies at the University of Florida. I am indebt ed to Dr. Akers for his endless support in all phases of my graduate work. I would like to thank my committee members, Lonn Lanza Kaduce, Chris Gibson, and Larry Winner, for all their great advice and assistance with completing my dissertation. Their insig hts, advice, and assistance were invaluable to my doctoral studies. Finally, I would like to praise my Lord for everything.
5 TABLE OF CONTENTS page ACKNOWLEDGMENTS ................................ ................................ ................................ ............... 4 LIST OF TABLES ................................ ................................ ................................ ........................... 8 LIST OF FIGURES ................................ ................................ ................................ ....................... 10 ABSTRACT ................................ ................................ ................................ ................................ ... 11 CHAPTER 1 INTRODUCTION ................................ ................................ ................................ .................. 13 Purposes of the Present Study and Its Contribution to Knowledge ................................ ........ 13 The Problem of Adolescent Substance U se In the United States and South Korean ............. 16 Gaps in the Research ................................ ................................ ................................ .............. 21 2 LITERATURE REVIEW AND THEORETICAL FRAMEWORK ................................ ...... 25 Social Learning Theory ................................ ................................ ................................ .......... 25 ................................ ................................ ....................... 25 Empirical Validi ty of Social Learning Theory ................................ ................................ 29 Cross Cultural Applicability of Social Learning Theory ................................ ................ 37 South Korean Adolescents and Social Le arning Theory ................................ ................. 39 ................................ ....................... 44 Previous Studies Testing the Social Structure and Soci al Learning Theory .......................... 48 Social Bonding Theory and Self control Theory ................................ ................................ .... 63 ................................ ................................ .................... 63 Empirical Validity of Social Bonding Theory ................................ ................................ 65 Control Theory ................................ .............................. 68 Cross cultural testing of social bonding and self control theories ........................... 73 Relationship between social bonding theory and self control theory ...................... 73 Previous Studies Comparing Social Learning, Social Bonding and Self Control Theory ................................ ................................ ................................ .......................... 75 Cross Cultural Studies Comparing Social Learning, Social Bonding and Self Control Th eory ................................ ................................ ................................ ............. 77 3 METHODS ................................ ................................ ................................ ............................. 83 Individual Level Data ................................ ................................ ................................ ............. 83 Structural Level D ata ................................ ................................ ................................ .............. 85 Measures of Variables ................................ ................................ ................................ ............ 88 Dependent variables: Alcohol, Tobacco, and Depressant Use ................................ ........ 88 Independent Variables: Individual Level ................................ ................................ ........ 90 Independent Variables: Structural Level ................................ ................................ ......... 94 Hypotheses ................................ ................................ ................................ .............................. 97 Analysis Plan ................................ ................................ ................................ .......................... 98
6 Model ing ................................ ................................ ................................ ............................... 102 HLM Models ................................ ................................ ................................ ........................ 103 Unconditional (Random ANOVA) Models ................................ ................................ ... 103 Structural Level M odel s ................................ ................................ ................................ 104 Random Coeff icients Models: Individual Level Predictor Models ............................... 105 Social learning model ................................ ................................ ............................. 106 Social bonding model ................................ ................................ ............................. 107 Self control model ................................ ................................ ................................ .. 108 F ull Model s ................................ ................................ ................................ .................... 108 The SSSL model ................................ ................................ ................................ ..... 108 Social structure and social bonding model ................................ ............................. 109 Social structure and self control model ................................ ................................ .. 110 Com parison Models ................................ ................................ ................................ ....... 110 HGLM Models ................................ ................................ ................................ ...................... 112 Unconditional (Random ANOVA) Model ................................ ................................ .... 113 Structural Level M odel ................................ ................................ ................................ .. 114 Random Coefficients Models: Individual Level Predictor Models ............................... 115 F ull Model s ................................ ................................ ................................ .................... 116 Comparison Model ................................ ................................ ................................ ........ 117 Data L imitations ................................ ................................ ................................ ................... 119 4 RESULTS OF ANALYSES ................................ ................................ ................................ 122 Descriptive Statistics ................................ ................................ ................................ ............ 122 Bivariate Correlations ................................ ................................ ................................ ........... 124 Hierarchical Li near Model Analyses: Alcohol and Depressant Use ................................ .... 125 Alcohol Use ................................ ................................ ................................ .......................... 125 Unconditional (Random ANOVA) Model ................................ ................................ .... 125 Structural Level Model ................................ ................................ ................................ .. 126 Random Coefficient Model ................................ ................................ ........................... 128 Social learning model ................................ ................................ ............................. 128 Social bonding model ................................ ................................ ............................. 129 Self control model ................................ ................................ ................................ .. 130 Models with Structu ral Variables and Variables from the Three Theories ................... 130 The social structure social learning (SSSL) model ................................ ................ 130 Social structure w ith social bonding model ................................ ........................... 132 Social structure with self control model ................................ ................................ 133 Full Comparison Model for Alcohol Use with All Social Structural, Social Learning, Social Bonding, and Self Control variables ................................ .............. 133 Depressant Use ................................ ................................ ................................ ..................... 134 Unconditional (Random ANOVA) Mode l ................................ ................................ .... 134 Structural Level Model for Depressant Use ................................ ................................ .. 135 Random Coefficient Models ................................ ................................ .......................... 136 Social learning model ................................ ................................ ............................. 136 Social bonding model ................................ ................................ ............................. 137 Self control model ................................ ................................ ................................ .. 138 Models with Structural Variables and Variables from the Three Theories ................... 138 The social structure social learning (SSSL) model ................................ ................ 138 Social structure and social bonding variables model ................................ ............. 139
7 Social structure and self control variables model ................................ .................. 139 Full Comparison Model with Social Structural, Social Learning, Social Bonding, and Self Control Variables ................................ ................................ ......... 140 Hierarchical Generalized Linear Model Analyses: Tobacco Use ................................ ......... 141 Unconditional (Random ANOVA) Model ................................ ................................ .... 141 Structural Level Model ................................ ................................ ................................ .. 142 Rando m Coefficient Models ................................ ................................ .......................... 143 The social learning model ................................ ................................ ...................... 143 Social boding model ................................ ................................ ............................... 144 Self control model ................................ ................................ ................................ .. 144 Models with Structural Variables and Variables from the Three Theories ................... 145 The social stru cture social learning (SSSL) model ................................ ................ 145 Social structure and social bonding model ................................ ............................. 146 Social structure and self control model ................................ ................................ .. 147 Full Comparison Model with Structural, Social Learning, Social Bonding, and Self Control Variables ................................ ................................ ............................... 148 5 SUMMARY AND CONCLUSION S ................................ ................................ ................... 174 Variations in Alcohol, Depressant and Tobacco Use across Gus ................................ ......... 174 Social Structural Dimensions of Use of Alcohol, Depressan ts and Tobacco Among Adolescents in Busan ................................ ................................ ................................ ........ 175 Social Learning, Social Bonding and Self Control Variables in Multi Level Analyses ................................ ................................ ................................ ............................ 179 The SSSL Model and Models with Social Structure and Social Psychological Variables ................................ ................................ ................................ ........................... 184 Comparison Models of Social Learning, Social Bonding and Self Control Variables ........ 189 Implications, Contributions, and Suggestions for Future Research ................................ ..... 191 Comparison of Mediation Effects of Social Learning and Other Social Psychol ogical Variables ................................ ................................ ................................ ........................... 196 Limitations of the Current Research and Implications for Future Research ........................ 198 APPENDIX: CORRELATION MATRIX F OR ALCOHOL USE SCALE RELATED EXOGENOUS VARIABLES ................................ ................................ .............................. 202 LIST OF REFERENCES ................................ ................................ ................................ ............. 206 BIOGRAPHICAL SKETCH ................................ ................................ ................................ ....... 230
8 LIST OF TABLES Table page 4 1 Percentages of students who ever used different substances in their lifetime ................. 150 4 2 Descriptive statistics on students and districts ................................ ................................ 151 4 3 Bivariate correlations of the independent variables with alcohol, depressant and tobacco use ................................ ................................ ................................ ....................... 152 4 4 Unconditional r andom ANOVA m odel for variation in a lcohol u se across Gus ............ 153 4 5 Structural level m odel for alcohol use ................................ ................................ ............. 153 4 6 Social learning m odel for alcohol use ................................ ................................ .............. 154 4 7 Social b onding m odel for alcohol use ................................ ................................ .............. 155 4 8 Self c ontrol m odel for alcohol use ................................ ................................ ................... 155 4 9 T he SSSL m odel with social structure and social learning variables for alcohol use ................................ ................................ ................................ ................................ .... 156 4 10 Model with social structure and s ocial b onding variables for alcohol use ...................... 157 4 11 M odel with social structure and s elf control variables for alcohol use ........................... 158 4 12 Full m odel comparing the relative mediating effects of s ocial l earning, s ocial b onding and s elf c ontrol variables for alcohol use ................................ .......................... 159 4 13 Unconditional r a ndom ANOVA m odel for variation in depressant use across Gus ....... 160 4 14 Structural level m odel for depressant use ................................ ................................ ........ 160 4 15 Social l earning m odel for depressant use ................................ ................................ ......... 161 4 16 Social b onding m odel for d epressant u se ................................ ................................ ......... 162 4 17 Self c ontrol m odel for d epressant u se ................................ ................................ .............. 162 4 18 The SSSL m odel, with social structure and social learning variables for depressant use ................................ ................................ ................................ .................. 163 4 19 Model with social stru cture and s ocial b onding variables for depressant use ................. 164 4 20 Model with social structure and s elf control variables for depressant use ...................... 165 4 21 Full m odel comparing the relative mediating effects of s ocial l earning, s ocial b onding and s elf c ontrol variables for depressant use ................................ ..................... 166 4 22 Unconditional r andom ANOVA m odel for variation in tobacco u se across Gus ............ 167
9 4 23 Structural level m odel for tobacco use ................................ ................................ ............. 167 4 24 Social l earning m odel for toba cco use ................................ ................................ ............. 168 4 25 Social b onding m odel for tobacco use ................................ ................................ ............. 169 4 26 Self c ontrol m odel for tobacco use ................................ ................................ .................. 169 4 27 The SSSL Model, with social structure and s ocial l earning variables for tobacco use ................................ ................................ ................................ ................................ .... 170 4 28 Model with social structure and s ocial b onding variables for tobacco use ...................... 171 4 29 M odel with social structure and s elf control variables for tobacco use ........................... 172 4 30 Full m odel comparing t he relative mediating effects of s ocial l earning, s ocial b onding and s elf c ontrol variables for tobacco use ................................ ......................... 173 A 1 Correlation matrix for alcohol use related exogenous and dependent variables ............. 203 A 2 Correlation matrix for depressant use related exogenous and dependent variables ........ 204 A 3 Correlation matrix for tobacco use related exogenous and dependent variables ............. 205
10 LIST OF FIGURES Figure page 2 1 The Social Structure and Social Learning Model ................................ .............................. 82
11 Abstract of Dissertation Presented to the Graduate School of the University of Florida in Partial Fulfillment of the Requirements for the Degree of Doctor of Philosophy A COMPARATIVE TEST OF SOCIAL STRUCTURE AND SOCIAL LEARNING MODEL OF SUBSTANCE USE AMONG SOUTH KOREAN ADOESCENTS By Eunyoung Kim August 2010 Chair: Ronald L. Akers Major: Criminology, Law and Society This purpose of this study is to Learning model (SSSL) model of crime and deviance on substance use behaviors among adolescents. This study attempts to gauge the validity and cross cultural genealizability of the SSSL model by utilizing a sample of 1,021 high school students drawn from a non American cultural setting, South Korea. The data for this dissertation consist of self reported individual level survey data including measures of all of the main explanatory variables found in social learning, social bonding and self control theories. The analysis therefo re includes a comparison of the effects of the social learning variables in the SSSL with the effects of social bonding and self control variables in expanded models. Also this study measures all four social structural components of the SSSL model with mea sures drawn from census data in Busan (formerly Pusan) in South Korea. To examine this multi level data, several Hierarchal Linear Models (HLM) and Hierarchical Generalized Models (HGLM) are used. The chief hypothesis to be tested is that the effects of th e structural variables on adolescent substance use are mediated by social learning variables and more so than they will be mediated by bonding/self control variables. The findings from the analyses provide supportive evidence for the SSSL model. Social lea rning variables substantially mediate the
12 and tobacco). On the other hand, social bonding and self control variables do not mediate the relationship between social structural influence and individual substance use behaviors as much as social learning variables do. Rather, social bonding and self control variables appear to be moderated by the social structural variables. The social learning variables have significant and greater mediation effects compared with the effects of social bonding and self control variables. Overall, this study found support for the validity and generalizability of the SSSL model to a non Western society. The contributions, limitations, and implications of the study for future research are presented.
13 CHAPTER 1 INTRODUCTION Purposes of the Present Study and Its Contribution to Knowledge In the field of criminology, there are relatively few comprehensive and integrated theoretical frameworks that can be used to investigate the relationship between structural Structure and Social Learning (SSSL) model proposes the social learning process as the principal pro cess by which social structure, including society, community and neighborhoods contexts and characteristics as well as sociodemographic and other variables have an effect on crime, deviance, and substance use. The SSSL model is an elaboration of social lea rning theory which integrates macro level factors with individual level factors in accounting for behaviors. As a cross level theoretical explanation, the SSSL is designed to guide ecological research exploring social structural influence on deviant and cr iminal behaviors, including mediate the effects of structural variables on delinquency and crime. One application of this general hypothesis is that adolescents who reside in socially disorganized neighborhoods are more likely to engage in delinquent and substance use behavior because they more often associate with substance using peers, are more exposed to deviant role models, and more likely to experience reinfo rcing consequences of their deviant and substance use behaviors. In substance use. Since the emergence of the SSSL model, a relatively small but growing number of studies have attempted to examine the main propositions of the theory in the United States (Bellair, Roscigno, & Velez, 2003; Haynie, Silver & Teasdale, 2006; Holland Davis, 2 006; Lanza Kaduce & Capece, 2003; 2006; Lee, Akers, & Borg, 2004; Verrill, 2008;
14 Wu, Eschbach, & Grady, 2008). But most of these studies failed to include all of the predictors that the theory required. That is, perhaps, partially due to lack of comprehens ive data that include all of social learning variables at the individual level, as well as various type of structural level predictors as indicators presented in the SSSL model. Moreover, none of these studies have used samples drawn from different societi es with different social and cultural contexts from American society. Therefore, there is no study examining whether the behaviors in different cultural contexts. As stated by Akers, the SSSL model is not culture or society bounded. Therefore, if the model is a general one, then it should be supported not only by research in America but in other societies as well. The research reported in this dissertation makes a con tribution to our knowledge by adding additional specification and test of the SSSL model to the small number of studies thus far and by being the first to test the model in a non western cultural context. Moreover, the current study contributes by using e mpirical measures of all of the social structural variables specified in the SSSL (differential social organization, differential location in the social structure, theoretically defined structural variable and differential location in primary and secondary groups) and empirical measures of all four of the major social learning concepts (differential association, definition, imitation, and differential reinforcement). As noted above, this has been done in only a very few of the previous tests of the model. This study also goes beyond all of the previous studies by measuring other social psychological variables drawn from other major criminological theories, social bonding and self control theory, to test the extent to which they offer processes mediating mac ro level factors on individual substance use as alternatives to the social learning processes that are proposed in the SSSL theory. That is, none of the previous research testing the SSSL model
15 has provided findings on not only the social learning variable s but other social psychological variables which potentially could mediate structural effects. It would be beneficial in testing and perhaps improving the SSSL model by examining if alternative variables have additional and perhaps stronger mediating effec t than the social earning variables. In sum, this study not only tests the major proposition of the SSSL model, testing wether the social learning variables mediate structural influences on adolescents substance use, but also incorporates social bonding an d self control variables to compare the relative mediation effects of the three different sets of variables among youth in the non western society of South Korea. Such multilevel study is also particularly critical to advancement of the literature examini ng substance use and delinquency in South Korea in general as well. To the best of my knowledge, no study has been conducted which considers multilevel factors, both neighborhood and individual influence on adolescence substance use behavior with South Kor ean samples. Additionally, the structural level data used in this study are drawn from South Korean census data (15 districts in Busan). One of the strengths of these data is that the census districts are exactly matched with the districts where the partic ipants who attended the schools in which data were collect resided in Busan, South Korea. It overcomes the methodological problems commonly found in multilevel studies using structural variables from U.S. census data, which often have unmatched residential contexts information and the school districts where surveys are generally conducted. Finally, this study utilizes Hierarchical Linear Modeling (HLM) to analyze nested data appropriately. Recent advancements by Raudenbush and Bryk (2002) consider not only the variation of contextual variables.
16 The Problem of Adolescent Substance Use In the United States and South Korean Substance use by adolescents has been a l ong time public concerns as one of the pervasive and costly social challenges in most countries. The detrimental effects and problems associated with drug and substance use have been well documented by a large body of research. The major concern is the wel l known harmful effects of substance use on being and its relationship to delinquency and crime. In the United States, it has resulted in tens of thousands of deaths for youths annually, according to the 2000 National Instituti on on Drug Abuse report (Lazarou, Pmeranz, & Corey, 1998). Substance use is also very expensive for the entire society. The estimated annual costs for drug related problems has continued to increase (Perl, 2003). As a result of it, the projected annual ex penditure of the United States government on drug control policy for Fiscal Year (FY) of 2009 was $14.1 billion. It represents an increase of $459 million from the $13.7 billion FY 2008 budget (Office of National Drug Control Policy, 2009). Besides the ex particularly problematic because it entails other seriously harmful problems. A large body of research reports that illicit (i.e., marijuana, stimulant, and inhalants) and licit drug (i.e., a lcohol, tobacco use) use among adolescents represents an important risk factor for the development of psychological, biological, interpersonal and social problems (Office of Applied Studies, 2009a; Lindstrm 2008; Sakai, Hall, Mikulich Gilbertson, & Crowl ey, 2004; Valente, Hoffman, Ritt Olson, Lichtman, & Jonson, 2003; Valente, Gallaher, & Mouttapa, 2004). Furthermore, adolescent substance and drug use is linked to various lifetime problems, such as the greater risk of developing abusive or substance depen dent behaviors and the higher risk of later delinquency and criminal behaviors (Akers, 1985; Barnes, Barns & Patton, 2005; Brook, Pahl, Balka, & Fei, 2004; Fergusson & Horwood, 1996; Hawkins et al., 2000; Huizinga, Loeber, & Thronberry, 1995; Reed & Roundt ree,
17 1997; ONDCP, 2009; Ellickson, Hay, & Bell, 1992;Golub & Jonhson, 2001; Kandel, 1984, 2002; Lynskey, Coffey, Carline, & Patton, 2003; Lynskey, Vink, Boomsma, 2006; Green and Ensminger, 2006; Brook, Gordon, Brook, & Brook, 1989a; Donovan, 1996; Wagner & Anthony 2002; Wilcox, Wagner & Anthony, 2002; Brook, Whiteman, Finch, & Coen,1996; Greenwood, 1992; Sealock, Gottfredson, & Gallagher, 1997). Studies have linked drug use behaviors to failure in academic achievements, such that students who engage in subs tance use have had bad school performance and lowered achievement motivation and school persistence (Barns, Welte, & Hoffman, 2002; Jeynes, 2002). In a study examining the relationship between substance use and youth delinquency, Barnes and his colleges ( 2 002 ) from a sample of New York high school students. Initial alcohol use in early age is correlated with developing problematic drinking habits and poor coping mech anisms (Warner & White, 2003). Adolescents who are at high risk of committing serious crime and incarceration in juvenile facilities have a high level of propensity to alcohol and drug abuse problems in their childhood (Hawkins et al, 2000; Guo, Elder, Cai & Hamilton, 2008). Alerted to these detrimental impacts of illicit and licit drug use among adolescents, American society exerted great effort to curb early substance use. Consequently, general substance and illicit drug use among adolescents began to de cline in the 1980s, and after a period of little change or some increase has declined again since 2002 in the United States ( 2008a). However, the overall proportion of adolescent who use substances such as alcoho l, tobacco and illicit drugs is still alarming ( Ritt Olson, Unger, Valente, Nezami, Chou, &Trinidad, D, 2005; Bachman, & Schulenberg 2008b). For example, alcohol use is still prevalent among U.S adolescents (Cleark & Loheac, 2007). Th e trend in alcohol use is parallel with that of illegal drug use in the U.S (Johnston et
18 al., 2007). Alcohol and tobacco are the most widely used substance among adolescents than any other drugs, is continuously consumed among adolescents. According to a 2 007 national survey, the proportion of any alcohol and tobacco use among 12 th graders in their life time are 72.3 percent and the proportion of tobacco use among 12 th graders is 43.6 percent respectively (Johnston, et al., 2008a). The prevalence and entwi limited to the United States. Substance use among adolescents has also been considered as a critical social issue in South Korea for decades and the trends of substance use among South Korean adolescen ts shows somewhat similar trends to those in the United States. First, many South Korean youths use substances as much as American adolescents do. Especially, legal substances, such as alcohol and tobacco use, as well as the abuse of pharmaceutical such as stimulants and depressants are serious problems. Second, substance use among adolescents, particularly, is considered as one of the major probable causes that are linked to various types of deviant behaviors for South Korean youths (National Youth Commiss ion, 200 4 ; Son, Kim, illicit and licit drug use seriously. The South Korean government also spends large sums of money each year on substance use control policy for adolescents, such as education and prevention programs in schools and anti substance use advertisement thorough mass media. Despite such tremendous efforts made by the South Korean government and society, the prevalence of substance use among South K orean youths is still high. Furthermore, similar to the declining trends of adolescent alcohol tobacco and other illicit drug use in the United States, the recent trends of adolescents substance use in South Korea is getting lower but the magnitude of sub stance use, particularly, alcohol use, among adolescents is as high as the statistics in Western nations (National Youth Commission, 200 4 ; Son, et al., 2002; Park, Kim, Kim, & Sung, 2007; Je, He, Kim, & Lee, 2004). In 2004, The National Youth
19 Commission a government funded research institution, in South Korea interviewed 2,990 South Korean youths aged 12 to 18 to investigate the prevalence of alcohol use among South Korean adolescents. The National Youth Commission (NYC) reports that approximately 74.4 per cents of youths aged 1 2 to 1 8 reported that they have experienced alcohol in their lifetime. Alcohol use among youths has increased from 60.2 percent in 1999, and 70.4 percent in 2002 survey. Also, the proportion of binge drinking during the past month amo ng youths is 70.1 percent for high school students attending industrial schools and 55.4 percent for students attending liberal high schools, according to the 2004 NYC report. Notably, the prevalence of a significant level. The proportion of frequency of binge drinking among female high school students is slightly higher than the frequency of male students. The age of onset of alcohol use is significantly reduced. Smoking is also a common trend seen in adolescence in South Korea. Particularly, it was prevalen t among youth in the early 2000s. Recently, the proportion of smoking adolescents is getting lower; yet, the overall prevalence of cigarettes smoking is still high. In 2007, National Youth Commission investigated the tobacco use among adolescents and found that 32 percent of high school students who attend industrial school and, 11.2 percent of high school s tudents who attend liberal school reported that they had smoke during the last year. Furthermore, the age of beginning smoking is getting younger and the prevalence of female h Institution, 2006). Contrary to the high prevalence of alcohol and tobacco use, the general trends of illicit drugs use among adolescents have been substantially deceased in South Korea. For example, in 1987 to 2.8 percent in 2002, according to the South Korean Supreme Court report in 2002. However, some licit and illicit
20 drugs are also a serious threat for South Korean youths. For instance, inhalant use by adolescents constituted about 80 percent of the total inhalant users in South Korea; it has decreased somewhat, but it is still high (South Korean Supreme Court Drug Report, 2002 ) Furthermore, about 20 percent of adolescents reported that they have been involved in licit drug abuse, such as depress ants use (Choi, 2003). Studies suggest that the decreased illicit drugs use among adolescents is partially due to the difficult accessibility of illicit drugs such as marijuana and cocaine (Choi, 2003;Sakai, Hall, Muijlich Gulbertson, & Crowley, 2004). How ever, other over the counter drugs, such as depressants can be obtained from many drugs stores without prescriptions just like other commercial products. Since the depressants are cheap and readily available (often in the home) and legal to buy and possess they are commonly preferred by young adolescents. As a result, depressants are particularly preferred by those teens experiencing significantly more abuse and neglect (South Korean Supreme Court Drug Report, 2002 ; Choi, 2003). Such depressant abuse shoul d not be taken lightly. That is because studies found that most drugs have a strong impact on adolescence behavior and their brain activity and may result in impaired perception and thought process, and even damaging and killing brain cells (Santrock, 2007 ). South Korea. Just as in the United States research on the causes and correlates of adolescent use is needed to inform public policy to control, prevent, or treat adolesce nt substance use and abuse in Korea. However, until today, only little research has been conducted to examine limited number of studies investigating substance use amo ng Korean youth has focused mainly on the trends and the prevalence of substance use. Just a few of these studies attempt behavior (Choi, 2003; National Youth Commission 2004; Son, et al., 2002; Je, et al., 2004;
21 Hwang & Akers, 2006 ; Kim, Kwak, & Yun, 2010 ). But, generally, studies attempting to substance use are still largely limited in explaining what mechanism and factors might In the United States, on the contrary, researchers from various disciplines have actively conducted research to address the causes and correlates of adolescent substa nce use. Interdisciplinary studies investigating predictors of adolescent illicit and licit drug use consistently have shown that various factors such as peer influence (i.e., peer use and delinquent peer association), family (i.e., parental substance use and ineffective parenting), and individual personality traits (i.e., low conventionality, and other individual level (Barnes, Barnes, & Patton, 2005; Newcomb & Bentler, 1988; Swadi, 1999; Warr, 1993a; 1993b). Gaps in the Research In addition to the contribution from numerous studies which extended current knowledge about the etiology of adolescent substance use, there is a small body of studies that attempts to extend the focus to inc lude both individual level factors and the role and impact of neighborhood on adolescent drug (Jang & Johnson, 2001; Sunder, Grady, & Wu, 2007; Winstanley, Steinwachs, Ensminger, Latkin, Stizer, & Olsen, 2008). These attempts may reflect recent recognition s among scholars that the predictors of substance use must be measured not only at the individual level, but also at the community level. The neighborhood provides the context in which individual behavior occurs. Therefore, it is reasonable to assume that this context may change the way in which individual protective and risk factors operate (Browing, 2008). However, a recent review of these previous studies examining fi ndings for some relationship. For example, while some studies found that the higher level of
22 illicit use, other studies found that the greater level of neighborhood poverty is positively associated with increased adolescence substance and illicit drug use (Allison, et. al, 1999; Ennett, Flewelling, Lindrooth, & Norton, 1997; Luthar & Cushing, 1999; Ford & Beveridge, 2006; Saxe, Kadushin, Beveridge, Liverty, Tighe, Ri ndskopf, Ford, & Brodsky, 2001; Sunder et al., 2007; Wright, Bobashev, & Folsom, 2007; Winstanley et al., 2008) Saxe et al. (2001) tested the relationship between substance use and neighborhood disadvantages and found that people in poorer neighborhoods o nly slightly used more substance than those residing in better neighborhoods. In a study using the same data, Ford et al, (2006) reported that use, but instead merel y addresses the more overt or visible drug problems (e.g., drug dealing). This was contradicted by Winstanley et al., (2008), who reported that neighborhood disadvantages and social capital were significantly associated with adolescent drug and alcohol use and dependence, even after controlling for individual and family variables. Furthermore there are some other social structural factors believed to have disruptive Along with neighborhood poverty, n e ighborhood level of residential mobility and population density are key pieces of these social structural factors These neighborhood characteristics have been identified as possible risk factors by social disorganization theory not only for adolescents su bstance and drug use but also for crime and delinquency (Ennett, et al., 1997, p.56). However, studies found mixed evidence regarding the positive relationships between the effects of residential mobility and bstance use. In terms of residential mobility, a s it is well known, social disorganization theory and some studies testing its contention found that higher level of residential mobility as one neighborhood characteristic, is associated with higher level o f adolescent substance use (Brook, Whiteman, Gordon, A.S.,
23 Nomura, & Brook, 1986; Clark & Loheac, 2007; Dewit, 1998 ; Ennett, et al., 1997; Hwakins, Catalano, & Miller, 1992; Sampson & Groves, 1989). Clark and Loheac (2007) found that adolescents who move f requently are more likely to be susceptible to peer group pressure for using drug and substance use such as marijuana and cocaine use. DeWit (1998) found that alco hol and drug use. Generally, studies found that residential mobility negatively affects However, Ennett and her colleges (1997) found that e alcohol use, the opposite relationship posited by the social disorganization theory literature. Moreover, they found that population mobility and density are partially mediated by some school characteristics, such as substance use norms in schools. Stud ies examining the effects of h igher levels of population density also generated Some studies found positive relationships between population density and individual substances and drug use. Sundq u ist and Frank ( 2004) examined if the level of urbanization is associated with hospital admissions for alcohol and drug abuse with a sample drawn from Sweden. They defined the level of urbanization by population density. They found that both women and men who live in the most densely populated communities had significantly greater levels of increased risk of being hospitalized for alcohol and drug abuse compared to women and men living in the least populated communities. The po sitive relationship between population density and hospital admission rates for alcohol and drug abuse is significant even after controlling for demographic characteristics, such as marital status, education, age, and immigrant status. However, Ennett et a l (1997) found that high population density is The rates of lifetime alcohol use and cigarette use are higher in schools in low density
24 lley et al (1988) analyzed Monitoring the Future data and found associated with population density. Overall, this review of previous studies examining the impact of neighborhood differences in the relationships between neighborhood characteristics and adolescent substance use. Also, some studies found that many of the finding s are not predicted by social disorganization theory. These contradicting findings among studies must be interpreted carefully, however. Perhaps, these differences could be explained by the different outcome operationalizations examined in different studie s (different types of substance and substance use vs. including other risk behaviors) or the difference in age of populations in studies. More fundamentally, it might be possible to assume that such differences from these neighborhood context studies are d ue to the lack of comprehensive and integrated theories to develop appropriate analysis models. Therefore, although there is an increased interest in examining the relationship and interactions between structural level factors and individual level factors the complexity of the causes and consequences of substance use is still a challenge in both theoretical conceptualization of neighborhood effects and developing appropriate methodological models. Consequently, to date, there is still a limited number of studies that have actually considered the mechanisms or processes by which neighborhood or community factors influence adolescent behavioral outcomes in what way (Akers, 1998; Crane 1991; Elliott, Wilson, Huizinga, Sampson, Elliott, & Rankin, 1996; Dembo, Blount, Schmeidler, & Burgos, 1986; Jang & Johnson, 2001). Therefore, there are still many gaps in the knowledge of relating neighborhood contexts to adolescents substance and drug use and this gap suggests the need for further investigation.
25 CHAPTER 2 LITERATURE REVIEW AN D THEORETICAL FRAMEW ORK Social Learning Theory Kaduce & al association theory in 1947(Sutherland, 1939; Akers, 1998; Sutherland & Cressey, 1966). Sutherland first proposed his theory, in his book titled Principles of Criminology in 1939. Later, he revised his first version of differential association theory in 1947 and proposed nine statement explanations referring to the process by which a particular person comes to engage in criminal behavior. learned through the proce ss of social interaction in intimate groups or networks which supply favorable and unfavorable definition of crime. Recognizing the omission of the specified mechanism of learning, Burgess and Akers (1966) took an initiative to address the task of specif ying the learning process left implicit by Sutherland (1947). This revision of differential association theory by Burgess and Akers was p.146 ociation Later, Akers (Akers, 1973, 1985, 1998) further elaborated and enhanced the conceptualization and presentation of the theory which he came more and more to refer to as focuses on four major distinctive concepts: differential association, differential reinforc ement, definition and imitation (Akers, 1998 p. 50; Akers, et al. 1979). In his most
26 recent statement of the theory, he describes the central proposition of the social learning theory of criminal and deviant behavior as follows; The probability that p ersons will engage in criminal and deviant behavior is increased and the probability of their conforming to the norm is decreased when they differentially associate with other who commit criminal behavior and espouse definitions favorable to it, are relati vely more exposed in person or symbolically to salient criminal/deviant models, define it as desirable or justified in a situation discriminative for the behavior, and have received in the past and anticipate in the current or future situation relatively g reater reward than punishment for the behavior. (Akers 1998, p.50) As such, it proposes that individual conformity and deviant behavior are products of a learning process, operating in a context of social structure, interactions with significant others, and situation. The difference between conformity and deviance is the direction of the process in which theses mechanisms operate. These learning mechanisms have balance in al association, definitions, differential reinforcement, and imitation. Differential association behaviors through direct and indirect, verbal, and nonverbal communication, interaction, and ide ntification with others. These significant others can be both conforming and deviating others who are in their primary (or secondary, reference or symbolic) groups (i.e., family, peers, school, church groups and co workers). These persons can contain the i major sources of reinforcement, behavioral models, and effective definitions which has a 53). The composition of the groups also varies over the life course; family is the most salient grou p in childhood; the importance of peers and school are peak in adolescence; and spouse and colleagues are the most significant others in adulthood (Akers, 1998 ,pp. 60 66). The influence of differential association varies depends on the modalities of associ ation: frequency, duration, priority, and intensity of association with those significant others and groups. Therefore, it can be expected
27 that conforming or deviating behaviors are more likely, when associations occur earlier (priority), last longer and o (frequency), and involve others with whom one has the more important or closer relationship (intensity) (Akers & Sellers, 2009, p.90) Differential reinforcement refers to the net balance of the pa st, present, and anticipated future costs and rewards, which relates to a given behavior, conformity and crime and delinquency. Reinforcers can be positive (i.e., obtaining approval, money or pleasant feelings) or negative (i.e., avoid or escape aversive o r unpleasant events) and punishers can be direct (i.e., painful, unpleasant consequences) or indirect (i.e., reward or pleasant be non social (i.e., positive or neg ative effects of psychological and physical stimuli) and social (i.e., positive and negative social consequences including reactions and relationship with parents and friends) which influence on individual behaviors contingent on previously learned experie nces of that person (Akers & Sellers, 2009, p.92). Social reinforcement is performed, but also the whole range of tangible and intangible rewards valued in society and modalities and the most significant rewards are social (most of reward and punishment are social). However, recent studies reveal that psychological and neurological effe cts of intrinsic reinforcers that are influenced by social expectations (e.g., past experience, influence by definition) also affect social reinforces, while interacting with individual characteristics (i.e., seeking trills, impulsivity) (Akers, 1998; Bren zina & Piquero, 2003; Wood, Gove, & Cochran, 1994; Wood, Cohran, & Pfefferbaum, 1995). Definitions unfavorable toward deviant and conforming behavior. The concepts of definitions are bot h
28 general and specific. General beliefs mean that religious, moral and other conventional values and norms that are favorable to conforming behaviors are also generally unfavorable to deviant behavior. Specific definitions related to specific acts (Akers, 1998, p.78). Definitions then are positive, neutralizing and negative (Akers, 1998, pp.77 87). Definitions can generally or specifically favor deviance, oppose deviance or justify or excuse deviancy under certain conditions (Verill, 2008, p.39). Definition s serve as cues that make one more willing to commit an act given anticipated reinforcement or punishment for certain behavior. The dimensions of the concept of definitions include beliefs, attitude, and rationalization/neutralization (Akers, 1998). Imita tion able outcomes from adopting the behavior. Whether an individual will imitate the observed behavior or not is depends on the characteristics of the models (e.g., admired or not) the behavior observed, and the observed consequences (i.e., rewarded or punishe d) (Akers, 1998, p.65). Imitation is more likely to have an effect in initiation of delinquent and criminal behaviors (Akers, 1998). In sum, with those four concepts, social learning theory proposed separate testable hypothesis that when an individual is m ore likely to be engaged in deviant or criminal acts: 1. He or she differentially associated with others who commit, model, and support violations of social and legal norms. 2. The violative behavior is differentially reinforced over behavior in conformity to t he norm. 3. He or she is more exposed to and observes more deviant than conforming models. 4. His or her own learned definitions are favorable toward committing the deviant acts. (Akers, 1998, p.51).
29 Social learning theory posits that the social learning pr ocess involves reciprocal and which an opportunity for a crime is present (Akers & Sellers, 2009, p.93). These social learning concepts are applicable to initiation repetition, maintenance, and desistence (1998 p. that it can explain all types of crime and deviance. Empirical Validity of Social Learning Theory Social learning theory has consistent ly received positive empirical support (strong to moderate) by a large body of empirical research (for a review see, Akers, 2006; Agnew 1991a, 1993, 1994, ; Akers, 1985; 1998; Akers & Lee, 1996; Akers & Sellers, 2009; Bahr et al., 2005; Burkett & Jensen, 1 975; Burkett & Warren, 1987; Cao, 2004; Conger, 1976; Conger & Simons, 1995; Dabney, 1995; Dembo et al., 1986; Elliott, Huizinga, Suzanne, & Ageton, 1985; Hwang & Akers, 2006; Inciardi, Horwitz, & Potttiger, 1993 ; Jensoen, 1972, 2003; Johnson, Marcos, & Ba hr, 1987; Kandel & Davies, 1991; Krohn, et al., 1985; LaGrange & White, 1985; Matsueda & Heimer 1987; Meier, Burkett, & Hickman, 1984; Patterson & Dishion, 1985; Pratt, Cullen, Sellers, Winfree, Madenson, Diago, Fearm, & Gau, 2009 ;Orcutt, 1987; Reed & Rount ree, 1997; Sellers et al., 2005; Warr, 1993a; 1993b; 1996; Warr & Stafford, 1991; White, Pandina, & LaGrange,1987 ; Winfree & Griffiths, 1983; Winfree, Sellers, & Clason, 1993; Wood, Cohran, & Pfefferbaum, 1995). Sometimes, social learning theory is tested in comparison with other theories or as integrated or combined models with other variables from other theories and social learning concepts usually have the strongest net effects on the dependent variables than concepts from other theories (Akers & Sellers 2009; Benda, DiBlasio, 1991; Burton, Gullen, Evans, & Dunaway, 1994; Catalnon et al., 1996; Elliott, Huizinga, & Ageton, 1985; Huang et al., 2001; Hwang & Akes, 2003; 2006; Jang, 2002; Kaplan, Jonson & Barley, 1987; Kaplan, 1996; McGee, 1992; Neff & Wait e,
30 2007; Rebellon, 2002; Thonberry, Lizotte, Krohn, Franworth, & Jang, 1994; White, Johnson, & Horowitz, 1986). Agnew ( 1991a, 1993, 1994, 200) Agnew ( 1991a, 1993 ) acknowledges the empirical validity of the social learning theory through his own reviews on empirical and substance use behaviors Particularly, studies testing the effects of the four measures of 78). Kubrin and his colleges (2009) also s tate that the accumulated research evidences from have earned an important Recently, Pratt et al (200 9 ) conducted a meta analysis, a standard approach in review of empirical studies that can examine the absolute and relative influence of theoretical components of on outco me behaviors. This meta bias and specify the criteria for review (Pratt, et al 2009, p.8; Rosenthal & DiMatteo, 2001 ). In their analysis they reviewed about 133 studies that tested the social learning theory dir ectly and studies using social learning theory concepts. In particular, the sample studies that they used for their meta analysis were collected in a systematic way; they used studies which included social learning variables that were published in the lead ing criminal justice/criminology journals from 1974 to 2003 (p. 9). In the analysis, they found that etc) is the measure that have strongest overall effect size (M z = .225, p < .001). Among the (M z =
31 .406, p < .001). Specifically, the and alcohol abuse, which have the range of effect size of 0.361 and 0.338 (M z = .361 and .338, respectively, p < .05). Also, the index of definition variable has great influence for general popu M z = .658, p < .001) (p.23). Overall, they found consistent supportive evidence from those studies. In conclusion, they stated that the empirical validity of social learning theory has been well supported by the me ta analysis. analysis was incomplete in that it did not include findings on different ial reinforcement and imitation. Prior to the meta analysis of Pratt et al (200 9 ) other meta analysis have been conducted ( Andrews & Bonta, 2006; Gendreau, Goggin, & Law, 1997; Gendreau, Little, & Goggin, 1996; Lipsey & Derzon, 1998). Andrews and Bonta (20 06) conducted a meta analysis for risk and need predictors of criminal recidivism. Particularly, the major risk and/or criminogenic need factors that used in their meta and/or cognitive behavioral influence strateg ies are readily identified within general p.10). Their findings reveal consistent findings to the propositions of social learning theory. That is, beliefs favorable for law violations and delinquent peer association significantly predict recidivism. Other evaluation research on correction and rehabilitation programs with criminal offenders are another source for supporting the empirical validity of social learning t heory (Pratt et al., 200 9 ). That is because correctional or school based prevention programs are based on underlying criminological theories of crime which select certain risk factors that are believed to cause certain criminal behaviors for change of offe nders conduct. Therefore, the reduction of risk factors and misbehaviors can be seen as supporting evidence for the selected
32 theory ( Pratt et al., 2006). In this context, although program evaluation does not provide direct empirical evidence on a theory, e xperimental and quasi experimental studies can provide evidence regarding the validity of criminological theories (Cullen, Wright, Gendreau, & Andrews, 2003). Recently, evaluation studies on treatment and prevention programs, especially cognitive behaviora l interventions consistently found that the major propositions and the four major concepts of social learning theory are the most effective factors in Cullen et al., 20 03; Lipsey, Chapman, & Landenberger, 2001; MacKenzie, 2006). Therefore, these findings provide experimental and quasi experimental evidence consistent with social learning theory (Cullen et al., 2003; Pratt et al., 200 9 ). Social learning theory is a genera l theory that claims to explain various types of deviant behaviors and offending. While research has focused more on delinquency and adolescent deviance, social learning research provides support for the claim of application to a wide range of behavior, fr om serious to minor form of behaviors (see Akers & Sellers, 2009; Warr & Stafford, 1991), including adolescent and elderly substance use (Akers et al., 1979, Akers et al., 1989; Akers & La Greca 1991; Akers, 1992 ; Akers & Lee, 1996; Hwang & Akers, 2003, 20 06; Sellers, Cochran, & Branch, 2005; Spear & Akers, 1988), and gang membership and activities (see Akers, 1998 pp. 110 117; Battin, Hill, Abbott, Catalano, & Hawkins, 1998 ; Currey, Decker, & Egley 2002; Esbensen & Deschenes, 1998; Winfree Mays, & Vigil Backstrom, 1994b; Winfree Vigil Backstrom, Mays, 1994a ). Social learning principles have also been tested on other forms of criminal and deviant behaviors, including computer crime (Skinner & Fream, 1997), premarital sex (DiBlasio & Benda, 1990), courtshi p violence (Sellers, Cochran, & Winfree, 2003), sexual coercion among fraternity men (Boeringer, Shehan, & Akers, 1991), college students cheating (Lanza Kaduce & Klug, 1986), felony offending of both gender (Alarid, Burton, & Cullen, 2000), terrorist viol ence (Akers &
33 Silverman, 2004), homicide rates of various societies (Batton & Ogle, 2003; Jenson & Akers, 2003). The principles have also been found to be operative in evaluation of more successful correctional rehabilitation programs (Cullen & Gendreau, 2 000; Cullen, Wright, Gendreau, & Andrews, 2003). While still more studies assessing serious types of criminal behaviors, such as homicide and violent criminal behaviors, are necessary, those recent empirical efforts expanded the scope of the theory and hav e supported the generalizability of social learning theory (Sellers et al., 2003). Studies conducted by Akers and his colleagues have employed measures of the four main constructs proposed by the theory (Akers, 1992; Akers, LaGreca, Cochran & Sellers, 198 9; Akers & Lee, 1996; Akers et al., 1979; Boeringer, Shehan & Akers, 1991; Hwang & Akers, 2003, 2006; Krohn, Skinner, Massey & Akers, 1985) and have included cross sectional land longitudinal data (Krohn et al., 1985; Akers & Lee, 1996). Those studies used all of the four major concepts proposed by social learning theory to test its propositions in cross sectional and in longitudinal ways (Krohn et al, 1985; Akers & Lee, 1996). Some of ocial bonding, self control theory, and general strain theory) (Akers & Cochran, 1985; Hwang & Akers, 2003). cross cultural generaliziablity (Hwang & Akers, 2006). Using an adolescent sample of approximately 3,000 in grades 7 through 12 in eight communities in Midwest, Akers and his This data set has been called Boys Town dat a. In this study, they found that the model with measures of each of four concepts is moderately to strongly associated with substance use in the expected direction. These variables accounted for 55 % of the variance in alcohol use and 68 % of the variance in marijuana use of the sample. Generally, the power of explanation of the social learning variables was very high compared to variables from other theories (e.g.,
34 social bonding). Other subsequent studies that tested social learning theory with this data set garnered moderate to strong empirical supports for the theory (Krohn et al., 1982 1985 ; Lanza Kaduce et al., 1984; Akers & Cocran, 1985; Akers, 1992). Krohn et al. (1985) conducted a longitudinal analysis to examine the influence of social learning v ariables on initiation and maintenance of adolescents smoking. They used a five year longitudinal data of smoking among junior and senior high school students in Muscatine, Iowa. This study provided some support and structured a causal map describing a pro cess of initiation and maintenance of smoking behaviors. They found that imitation and definition has importance in the initiation while differential association and differential reinforcement (including both peer and nonsocial reinforcement) are important in maintenance of smoking behaviors. propositions ( Lauer, Akers, Massey, & Clarke, 1982 Akers, Skinner, Krohn, & Lauer, 1987; Spear & Akers, 1988). In the Iowa smoking stu dy, Spear and Akers (1988) conducted tests from the first year survey. They found that each of the four social learning variables is moderately to strongly correlated with smoking frequency of the participants. All of the social learning variables combined explain 54% of smoking behaviors among adolescent. data from a panel of 454 adolescents in Iowa, and examined the causality of the process posited by the theory. They found some support for its reciprocal and feedback effects among variables over time with greater effects of social learning variables on smoking than the effects of smoking on the social learning variables. Social learning theory also provides a sou nd theoretical framework for elder alcohol use. Akers and his colleges (1989; Akers & LaGreca, 1991; Akers, 1992) studied the influence of social learning variables on elderly alcohol use with a sample of 1,410 adults aged sixty and over from four communit ies in Florida and New Jersey. Akers and his
35 drinking behaviors in the past year and 52% of the variance in the quantity of drinking of the participants. Although all of social learning concepts received substantial empirical support for their validity, the effects of the four concepts on outcome behaviors are not equal. Th e existing research, including studies fully or partially (e.g., used one or two social learning variables) testing social leaning concepts, suggests that the strongest predictor of criminal involvement typically is differential association. In particular, social learning variables operative in peer association has received abundant evidence to demonstrate the significant impact on delinquency and crime of learning, especially during adolescence (Akers, 1998). Also, it is the mostly commonly used variable i review see, Akers & S ellers, 2009; also see Warr, 2002:40). A recent study by Haynie (2002) focused on influence of contextual factors on the relationship between friendship networks and delinquency. She found that the proportion of delinquent peer associations (that is differ ential peer association) is the most important factor in peer networks in explaining delinquency. Furthermore, Sellers et al. (2003) found that peer association is also the strongest predictor of aggression and dating violence even when peers are typically absent during the commission of the aggressive act. Furthermore, empirical studies present evidence supporting the social learning argument that the learning process operating in the family setting is a strong predictor for either conforming or deviant be havior. Family, a key primary group provides conforming and deviant modeling, definition and reinforcement through child parent interactions (Akers & Sellers, 2009; Patterson, 1995).
36 With regard to the validity of the other social learning concepts, empir ical studies have also generated significant supports for differential reinforcement and definition, while imitation has received relatively less support. Akers argues that imitation is an important Akers et al., 1979; Krohn et al., 1985). Specifically, differential reinforcement can be considered as the central causal mechan ism in the social learning model, since research evidence suggest that differential deviance in relation to a process of differential reinforcement. Triplett & Payn e (2004) 1,725 youths from the National Youth Survey data. This study employed three social learning measures, imitation, differential association, and differential r einforcement for analysis. In particular, they employed two types of reinforcement measures that are specified social (i.e., reactions from peers) and nonsocial reinforcement measures (i.e., drug use as problem solving) separately. Their findings provide s upportive evidence for the propositions of social leaning theory. They found that the three measures were significant for adolescents in their sample, in particular, for frequent drug users. Also, they found that reinforcement measures mediate the relation ship between differential association and drug use. Further, their study revealed that youths tended to use drug as a way to solve problems and confirmed the effects of nonsocial reinforcement on adolescent drug use as social learning theory predicted. Woo d and his colleges (1995) also found some evidence of psychological or emotional arousal (e.g., internal sensations of excitement or thrill) while committing or anticipating deviant behaviors can be played as intrinsic rewards to an individual. Their findi ngs suggest that some types of taking
37 and thrill seeking behavior provide sensory or physiological stimulation which is highly 1995, p.174). In terms of social reinforcement, research found that it is a very important variable in cessation/continuation of substance and drug use. Studies found that social reinforcement contributing significantly to the amount of variance explained by the model. For example, friends and family were important in (Akers et al, 1979; Krohn et al, 1985; Winfree, Sellers, & Clason, 1993). In spite of years of theor etical specification, the importance of all of social learning concepts and several examples provided by Akers himself, researchers still rarely include good measures of differential reinforcement and imitation. Therefore, it is necessary to develop more s ophisticated measures to examine the differential reinforcement process, such as the role of nonsocial reinforcers In terms of the imitation concept, in fact, research has revealed a relationship between exposure to deviance and violence on television and other media outlets and behavior problems in early childhood (Akers, 1998, p.77; Comstock & Rubinstein, 1972; Murray et al., 1972; Pearl et al., 1982; Donnerstein & Linz, 1995). Also, imitative effects on pro social behaviors have been reported as well (R ushton, 1980, 1982). Cross Cultural Applicability of Social Learning Theory The applicability of social learning theory is not limited only to American samples, but also to sample from various countries confirmed the genealizability of the theory. As Aker s clearly states, the social learning theory is not a culture bounded or society specific explanation of deviance. Accordingly, cross cultural studies have been conducted in South Korea, Sweden, New Zealnd, Israel, Ireland, Europe, Taiwan, & China (Ferguss on, Swain Campbell, & Horwood, 200 2 ; Fergusson & Horwood, 199 9 ; Hwang & Akers, 2003, 2006; Kandel & Adler, 1982; Lee, 1989; Junger Tas, 1992; Bruinsma, 1992; Miller, Jennings, Alvarez
38 Zhang & Messner, 199 5). Rumpold et al. (2006) used a sample of Austrian adolescents to exa mine the effects of various predictors on adolescent substance and drug use. They conducted multivariate structural equation model analyses and found that peer group influence is particularly strong predictors associated with adolescent substance and drug use family atmosphere and school difficulties, including other individual level risk factors. Fergusson and his colleges (200 2 ) examined the deviant peer influence substance use as well as deviant behaviors among New Zealand adolescents. They used twenty one years of longitudinal data in a birth cohort of 1,265 children who were born in Christchurch, urban area, in New Zealand in mid 1977. The cohort has been followed by Christchurch health and Development Study program at birth, 4 month, 1 year and at annual interval to age 16, 18 and 21 years old. With such a carefully designed study, they found deviant behaviors, such as violent crime, property crime, alcohol and other drug abuse. On the basis of their findings, researchers concluded that deviant peer associations are correlated with increased rates of a range of problem behaviors in adolescence and young adulthood, and the influence of deviant peer association is the most influential at younger age. Miller et al. (2008) also specifically assessed the cross cultural efficacy of social learning theory with a sample of Puerto Rican high school adole scents, who are attending private and public schools. This study focused on comparison of the relative influences of personal and peer definitions (differential association) on substance use (tobacco, alcohol, and marijuana use ). They found statistical dif ferences between public and private school students across all three substance use behaviors and the two aspects of social learning theory, definitions and differential association. In this study, peer definitions are the strongest predictors on adolescent substance use after controlling for demographic factors, age, and gender. That is, students
39 who perceive more peer approval of substance use are at greater risk to be involved in lifetime tobacco, alcohol, and marijuana use than students who perceive le ss peer approval of substance use, regardless of their own personal definitions of these substance use behavior. South Korean Adolescents and Social Learning Theory These findings are consistent with the results of studies that have used South Korean samp les and studies conducted in South Korea context. First, Hwang (Hwang, 2000; Hwang &Akers 2003; 2006) extended testing social learning theory to in a sample of South Korean youths. He collected data from a cross sectional sample of 1,012 high school studen ts in a metropolitan city, Busan (formerly Pusan), in South Korea. In their study, specifically, all of four social learning concepts are employed. The findings of the study reveal that the concept of differential peer association, which was measured by as sociation with friends using That is, the more students reported that they are associated with substance using friends, the more likely it was that they used su bstances and drugs. Besides differential association with peers definitions, differential reinforcement, and imitation were also significant predictors for (2010) used a n ational sample of 3,188 junior high school students drawn from South Korea to examine comparative predictability and generalizabilty between social learning theory and social bonding theory. This study compared the relative importance between peer influenc e drawn from social learning theory and parental influence reflecting social bonding attachment measure on adolescent drinking and smoking behaviors. This study found support for both social learning theory and social bonding theory, suggesting substantial peer and parental for influence in predicting the risks of adolescent substance use. Other studies conducted by South Korean researchers also support the social learning propositions in explaining adolescents substance and drug use behaviors. National Yo uth
40 Commission (2005) conducted research investigating patterns of alcohol abuse among South Korean youths by using a self reported data from randomly selected sample across South Korea. Researchers in this project collected a national representative sampl e of 1,615 adolescents which is composed with 361 elementary school students (22.4%), 469 junior high school students (29%) and 785 high school students (48.5%). Specifically, the high school participants consist of 395 students attending liberal school an d 390 students attending industrial schools. Although they used randomized data collected nationwide, the researchers only conducted descriptive analysis, cross tab analysis and t test for comparing means between groups, and correlation analysis, including some narrative analysis with several interview data. Therefore, although they attempt to collect data and conduct empirical analysis their efforts are limited in understanding correlations between risk and protective behavior. However, this study provides the overall patterns of alcohol use among adolescents in South Korea and general information regarding important factors influencing on their alcohol use behavior. They found that approximately 65 percent of particip ants have ever used alcohol. Although the overall proportion using alcohol among adolescents has decreased slightly from the proportion of the previous years, 74percent, the pattern of alcohol use among youths is still high and comparative to the rates of western countries. Also, the proportion of binge drinking among students increased from 44.9 percent in 2004 to 56.9 percent in 2005. In addition, this study found that there is no significant gender difference in the life time alcohol use (lifetime alcoho l use for male students is 63.4% versus for female students is 62. 2%). Furthermore, the proportion of alcohol use during the past year is greater for female students than male students (40.3% for males and 43.0% for females respectively), indicating incre ased trends of alcohol and other substance use among female students in South Korea.
41 They also found some differences in the frequency of alcohol use per month among students by type of schools they attend. For students attending industrial schools, the m ean frequency of alcohol use per month is 3.45 times per month, while the average alcohol use frequency for students attending industrial schools is 2.03 times per months. In terms of binge drinking among youths, students attending industrial schools repor ted that they drunk three times per month on average, while students attending industrial schools reported that they have been involved in binge drinking less than five times per year. That indicates students attending industrial schools more frequently us e alcohol per month and were at greater risk of binge drinking than students attending liberal type of schools. This study explicitly employed predictors for adolescents alcohol use drawn from social learning, social bonding and strain theory. They found t alcohol use. In particular, the influence of friends using alcohol has great impact on both the initiation and maintaining of alcohol use behavior among the participants. The correlation coefficient o f the peer variable is 0.20 which is the greatest one among variables used in this study. Again, the variables measured by the proportion of close friends who use alcohol and their frequency of alcohol use, which are also other explicit peer association me asures, had wanted to maintain good relationship with friends who using alcohol and that is one of the main reasons they initiate and continue to use alcohol while the y associate with friends. Interestingly, approximately 21 percent of student who ever used alcohol reported that they used alcohol because other adults encouraged drinking. This suggests that differential association with other adults who use alcohol has a compared to the students who did not hav e binge drinking problems. The difference is
42 differences between th al cohol use. Imitation through exposure to alcohol using advertisement appeared to have supervision and parental attachment are also significantly associated with the part substance use. Yun and her colleges (1999) used differential reinforcement measures from social a list of index of 21 questions asking students about t heir expectance from alcohol use. In analysis they found that question items reflecting good or bad effects of substance use, which indicate both social and non social reinforcement measures, are significantly and strongly and substance abuse behaviors. This study concluded that students who have greater expectation of substance use on the basis of measures of the effects of substance use, such as rewards and costs of substance use, are more likely to use substance. This fi nding supports the concept of differential reinforcement in social learning theory. Choi (2003) investigated correlates associated with drugs use among a sample of incarcerated delinquents in six juvenile confinement facilities in South Korea. In her fina l sample, she surveyed 508 delinquents which is 27.9 percent out of the total juvenile delinquents incarcerated in any confinement facilities in South Korea (n=1,822 a nationwide sample ) in 2003. She also studied the trends of substance and drug use among non delinquent and delinquents in confinement facilities from 1987 to 2003 by using data from other studies
43 of adolescents and delinquents substance and drug use (Kim, Chio, & Chin, 1990; National Youth Commission, 200 2). In her analysis, first she found that illicit drug use among non delinquent adolescents subsequently decreased compared to a decade earlier. Choi noted two important predictors for substance and illicit drug use among non delinquents and delinquents are 1) the availability of illicit and substances and 2) peer influences. Although this study does not explicitly use social learning theory as the theoretical framework for analysis, the findings support the proposition of peer influences in the theory. However, this study did not utilize rigo rous statistical methods; rather it used simple descriptive, content and narrative analyses. Studies investigating smoking behaviors among South Korean adolescents also support major concepts of social learning theory. Specifically, peer and parental inf luences were found that the most important causal factor for initiation of cigarette use is peer influences, particularly for female adolescents. Kang and Kim (2005) investigated factors associated with male students is also peer influence. P articularly, the variable measuring proportion of close friends using cigarette, a standard variable reflecting the differential peer association concept of social learning theory, was the strongest predictor in his study. Another study by Kim and Park (2 009) smoking behaviors followed by the differential peer association with smoking friends. In summary, studies conducted in South Korea generally support the propositions a nd the core concepts of social learning theory. Although these studies provide valuable information regarding patterns and behaviors of adolescent substance and drug use, only a few of the studies employed rigorous statistical methods and variables from ma jor
44 criminological theories (Hwang & Akers, 2003, 2006; Kim, K wa k, &Yun, 2010). Therefore, there are still great knowledge gaps in explaining and investigating the predictors related with all, unlike the studies conducted in the United States and other Western countries, none of the studies, which used samples of South Korean, have systematically used neighborhood or community level factors, which are considered as important predictors in u and drug use behaviors. level ing theory that specifies the process and mechanism by which structural level variables affect individual deviant and criminal behaviors (Akers, 1998, p.330). The key proposition of the SSSL theory is that, the social learning process mediates social struc tural influences on individual criminal and deviant behaviors that make up the macro level rate of crime and deviance. More specifically, the cognitive/behavioral process specified in social learning theory process is hypothesized to be the primary mechani sm linking social structural variables (meso or macro level structural factors) to individual behaviors (Akers, 1998, p.329). Therefore, level and meso generate difference in the operation of social learning variables (Akers, 1998, p.322). Akers vari ables remain in these models while the net effects of socio demographic variables are reduced, typically to statistical non Social Learning and Social Structure, he commented about the SSSL theory th at,
45 Its basic assumption is that social learning is the primary process linking social structure to individual behaviors. Its main proposition is that variations in the social structure, culture, and locations of individuals and groups in the social syste m explain variations in crime rates, principally through their influence on differences among individuals on the social learning variables mainly, differential association, differential reinforcement, imitation, and definitions favorable and unfavorable an d other discriminative stimuli for crime. The social structural variables are indicators of the primary distal macro level and meso level causes of crime, while the social learning variables reflect the primary proximate causes of criminal behavior by indi viduals that mediate the relationship between social structure and c rime rates (Akers, 1998, p.322). In SSSL, Akers maintains that four main dimensions (differential association, differential reinforcement, definitions, and imitation) of the social learni ng process are the key mechanisms that mediate the effects of structural conditions Akers identifies four major dimensions of social structure that provide the various contexts within which the social learning process is assumed to operate: differential s ocial organization (society, community culture), differential location in social structure (age, gender, class, race), meso level of social location (primary, secondary, and reference groups), and theoretically defined structural variables (social disorgan ization). Figure 2 1 illustrates the conceptual model of the SSSL theory. social structural correlates: differential social organization level characteristics of cultural, societal and geographical differences in crime rates. Akers has little emphasis on established, with or without specification of the causative structural or c ultural characteristics of these systems. From this point of view, it makes little difference what the specific theoretical explanations of the variations are But the measure could be tapping some unspecified combination of the features of the social orga nization, culture, or social concept include empirical correlates that have been used as statistical controls in prior macro
46 level studies, such as population size, de nsity and other regional, geographic or economic social system (Akers & Sellers, 2009; Lee et al., 2004). socio demographic/socioeconomic correlates: differential location (p.333). It refl level social organization or larger groupings and other dimensions of differentiations in societies and communities (p.333). This concept is commonly co nceptualized as direct (p.333). These indicators are gender, race, marital statu s, occupation, religion, age, class and socioeconomic status and other individual characteristics that characterize groups in society. In empirical studies, these variables can be used in analysis of models of crime and deviance either by measuring the cha racteristics for individuals or by aggregating these measures into proportions of individuals with these characteristics. For example, a measure of race and gender proportions of the population in a community can be incorporated into an empirical model as an indicator of differential location in the social structure. Third theoretically defined structural causes refers to crime causative structural deviance that prop ose elevated rates in those societies, or segments of societies, that are hypothesized to have higher levels of some abstractly defined condition like anomie, conflict, lates of crime such as inequality, social class, poverty, population instability or others may be taken as indicators of these theoretical constructs or other measures taken at the group or individual level may be defined as indicators of the theoretical c onstructs. According to Akers, the most
47 relevant explanatory concepts are theoretical variables drawn from social disorganization theory and anomie theory (Akers, 1998, pp. 330 334). The fourth structural dimension in the SSSL theory is d ifferential soci al location in primary, secondary, and reference groups (p.334) It refers to the meso level or more immediate social context than to more macro level and distal contexts. These consist of primary or secondary groups and individual networks such as fami ly, work groups, peer groups, church groups, reference groups and others to which one is affiliated or belongs (p.335). exposure to different levels of the social learning variables. In turn, this different level of maintainance, and cessation of criminal and deviant behaviors, and ultimately crime rates among a population in which the crim es of these individuals are counted (p.335). The expected relationships between social structural dimensions and social learning variables and social learning variables and individual criminal and deviant behaviors are as follows: The strongest expectatio n is that the variations and stabilities in the behavioral and cognitive variables in the social learning process account for all variations and stabilities in criminal behavior and thereby mediate all of the significant relationship between the structural variables and crime. The more realistic statement is that variations and stabilities in the behavioral and cognitive variables specified in the social learning process account for a substantial portion of individual variations and stabilities in crime and deviance and mediate a substantial portion of the relationship between most of the structural variables in the model and crime. A weak statement of the theory is that the social learning process accounts for some portion of the variation and stability in criminal behavior and mediate some portion of the relationship b etween the correlates and crime (Akers, 1998, p.340). Specifically, Akers claims that the SSSL model rests on the expectation that the social tructural effects. The purpose of this SSSL theory is not to explain why there are social structural variations in social disorganization,
48 anomies, as well as age and gender structures. Rather, he is more interested in explaining how social structural diff erences generate the difference in individual criminal behaviors though the effects on the social learning process and thus to the differences in crime rates (p.336). Previous Studies Testing the Social Structure and Social Learning Theory Although Akers proposed the social structure and social learning theory and invited empirical research for this model more than a decade ago, complete research testing the SSSL model which incorporates all four of social structural concepts, as well as the four major so cial learning concepts, is still rare. However, even the limited number of studies that have been conducted, generally have found supportive evidence for the major proposition of the theory, the mediating role of social learning measures between structura l conditions and various deviant outcome behaviors (i.e., delinquency, crime and substance and drug use) (Akers & Lee, 1999; Bellair, Roscigon & McNulty, 2003; Gib so n, Poles, & Akers, 2010; Haynie, Silver, & Teasdale, 2006; Holland Davis, 2006; Jensen, 200 3; Krohn, Lanaza Kaduce & Akers, 1984; Lanza Kaduce & Capece, 2003; Lonza Kaduce, Capece, & Alden, 2006; Lee, Akers, & Borg, 2004; Page, 1998; Verrill, 2008; Wu, Eschbach, & Gardy, 2008). Lee et al (2004) tested the propositions of SSSL model with Boy s Town data to examine whether social learning variables substantially mediate structural variables on adolescent alcohol and marijuana use behaviors. The study used three social structural variables; differential locations in social structure (gender, cla ss, and age), differential social location in primary and secondary groups (family structure), and differential social organization (community size) as well as all four social learning variables, differential association, imitation, reinforcement, and defi nition favorable and unfavorable to substance use. Utilizing structural equation modeling, they found supportive evidence of the model: Three social learning variables mediated substantial amounts and sometimes virtually all of
49 the effects of gender, soci o economic status, age, family structure, and community size on the adolescents alcohol drinking and marijuana smoking behaviors. Notably, the net effects of gender, although substantially reduced, remained statistically significant in the models. Although empirically supported set of concepts and principles for understanding how social influences of social structural measures, particularly, predictors that garnered substantial empirical supp orts, such as labor market opportunity, and concentrated community (fighting). They used multilevel data from a longitudinal survey data of 25,000 middle school students a nd various type of structural characteristics drawn from U.S. census data. The structural variables used in this study included poverty, unemployed, percent of urbanization, percent of professional occupation, percent of service job, and family headed hous ehold etc. Also, this study utilized various types of individual sociodemographic variables measured at the individual level such as race, employment, sports involvement, school attachment, exposure to violence, family mobility during the past two years, e tc. However, this study used less ideal social learning variables in testing the SSSL theory. What they expected in their analysis is that the employment structure of community may influence social learning processes as well as family well being in the co mmunity directly. (p.199). Using hierarchical linear models analysis, they found significant effects of structural variables on the dependent variable and when social learning variables are entered in the models, the effects of social structural variables, particularly concentrated disadvantage, on adolescent violence were substantially reduced. Additionally, they also found that family
50 processes partially mediate the structural measures. On the basis of the findings, they concluded that local labor market study provided partial support for the SSSL model. Lanza Kaduce and Capece (2003) conducted a partial test of the SSSL theory examining if social learning process mediates the influences of socia l structural variables on different universities, which is drawn from the Core Alcohol and Drug Survey which conducted survey among various types of colleges in the Uni ted States. The final sample only includes college students who are full time, not married, and of traditional college age from 17 to 23 years old to avoid extra social structural influence beyond the social structural measures included in the model. In th eir SSSL model, they utilized all four social structural measures: differential social organization (diverse university), differential location in social structure (gender and race), differential social location in meso level groups (Greek organization ext racurricular involvement), and the two structural theoretical variables, one is groups: male students, female students, faculty/staff, alumni, and athletes). With regard to social learning measures, their study include two major concepts, differential reinforcement (anticipated rewards of alcohol use and anticipated punishments by fr iends for binge drinking) and definitions, while neither peer association nor imitation measures are included. They found that the coefficients of most of the structural variables (Greek Participation, gender, race, and campus climate regarding alcohol use ) are substantially attenuated as social learning variables are included in the model. Therefore, they suggest that their study is supportive of the SSSL theory in that, social learning processes mediates substantial proportion of the relationship between most
51 alcohol use (p.194). However, they note that studies including the better operationlized social structural variables are more likely to increase the mediation effects of social learning variables that th e SSSL model predicted (p.194). Later, Lanza Kaduce, Capece, and Alden (2006) explored the effects of gender on and within the feminist theory perspective. That is, th is study examined if the social learning variables may substantially mediate the effects of structural variables, such as gender, rather than modulate or moderate them. Second, it also examined feminist theory arguments that not all group dynamics, particu larly gender, to be mediated by social psychological processes (Morash, 1999). Ultimately, the authors test whether gender, as a structural indicator, would hold in predicting drinking behaviors as suggested by feminist theory or whether the effect of gend er would be mediated by social learning. Using a sample of 688 White single college students (18 years of age or older) from Core Alcohol and Drug Survey data in the U.S., they developed research models by including three structural variables (gender, camp us Greek system involvement, and grades) and two social learning variables from one differential reinforcement concept (anticipated risk of harm and anticipated positive consequences). In the analysis, the authors created a grade and gender interaction ter m (GradeGender) in the SSSL model to test if the social learning variables mediate the interactions between structural lts suggested that social learning variables did mediate the direct effect of gender; however, a significant statistical interaction between gender and grades was not mediated by social learning. They interpreted this finding after interaction between gend er and other structural components that may act independently the social learning variables do not mediate. Particularly, the authors suggest that gender may interact with other structural variables independently of the mediation of social learning
52 process However, as the authors noted, there was no differential association variable which variables, some mediation would have been expected. Without a differentia l association research, this study concluded that the findings are still tentative and additional research tural relationships are substantially mediated, which are not mediated, which interact and which are population Holland micro level data and developing appropriate m ultilevel analysis models. In this dissertation, she tested the mediation effects of social learning variables on structural variables for study used individual self report survey data from the Boys Town study and structural variables from the U.S census data. Utilizing Hierarchical Linear Modeling, this study includes all four social structural dimensions (age, SES, sex, population density, poverty, ethnic heterogene ity, residential mobility, and religiosity) and all four social learning variables (differential association, definition, differential reinforcement, and imitation) in the analysis models. This study provides strong support for the SSSL mediation propositi ons for marijuana and illicit drugs use and moderate support for alcohol use among adolescents. Overall, the findings revealed that most of the social structural influences are substantially mediated by each of the four social learning processes, with a f ew exceptions. Most of the social structure indicators (e.g., population density, residential mobility, poverty, age, and SES) became insignificant and their coefficients reduced toward zero when the social learning variables are entered into the each of t he three substance and drug use models. Among the social learning variables, differential association, definitions, and differential
53 reinforcement, are strongest predictors and mediators on drug use behaviors in some social contexts than others. Notably, h owever, this study found that to some extent social structure The coefficients of gender in each drug and substance model were reduced, but remained significant on alcohol, an d marijuana and while not initially significant became significant for illicit drug use once social learning variables were included in the models. Her findings were consistent with the previous tests for the mediation effects of the social learning proces s, which mediated but did not make gender effects insignificant in the SSSL models (Lanza Kaduce & Capece, 2003; Lanza Kaduce, et al. 2006; Lee et al., 2004). The researcher, however, interpreted this gender effects as not completely inconsistent with the SSSL predictions because it suggests that the social learning variables interact in different ways to influence boys and girls differently (p.111). Also, this study found that the effects of poverty on alcohol use, which had no significant direct effects, increased and became significant when social learning variables included in the model. The researcher suggests that this interaction effect of social learning variables on poverty may also imply the moderating effects of social context on social learning p rocess. Another recent empirical study by Verill (2008) specifically tested the idea of mediation versus moderation effects of the social learning process on the relationship between social structural influences and delinquent behavior. Jumping to the co nclusion, Verill reported that social learning variables more likely moderates the social structural effects on delinquency because his study found some moderated mediation effects of the ill criticized some theory with structural concepts does not specify th e linking proposition clearly, accordingly
54 complete mediation (p.129). Therefore, he argues that empirical studies need to test the theory with more scrutiny to rule out moderation effects to verify mediation effects of the social learning process than the previous studies afforded. He used a survey data of 1,674 Largo, Flor ida high school and middle school students with 1998 U.S. census data. This study utilized various types of social structural measures (population density, race, sex, age, near poverty, racial composition, family disruption, poverty, residential mobility, socioeconomical status, and ethnic heterogeneity) that reflecting the differential social organization and theoretically defined structural concepts of social structural in the SSSL theory. He also used three elements of social learning variables (differen tial association, definition, & differential reinforcement. First, he tested moderation effects of the social learning process, by running twelve sets of different Ordinal Least Square models which included interaction terms of each social learning variabl es and each social structural variables. He found some evidence of moderation effects of social learning population density, age composition, and sex on log delin quency; definitions moderate rather than mediate the effects of population density, individual sex, socioeconomical status, and log ethnic heterogeneity on the delinquency measures; and costs moderate rather than mediate the effects of log race composition 133). In a set of Structural Equation Models (SEM) estimating the mediation effects of the social learning process between three structural influences on log delinquency, he reported that the findings do not
55 represent the differential social organization and theoretically defined structural cause dimensi (p.142). However, he suggests that this finding of moderation effects does not invalidate the riables are related to the social structural variables and their impact on delinquency. Furthermore, he noted that in a series of analyses examining moderation effects, some social learning variables were correlated with social structural variables as well as outcome variables. This correlation invalidates the moderation interpretation of the relationship (Baron & Kenny, all social learning variables, nor did one soc ial learning variable statistically interact with all macro social learning as both a moderator and a mediator of social structural variables (p.143). Further he interprete than social learning mediating the social structural effects on delinquency, distal macro social correlates of crime may influence criminal behavior through their interaction with the s ocial learning process, whereas more proximate meso level crime correlates may provide the context social learning mediates. This explanation accounts for both the moderation effects observed in the present research and for the mediation effects noted in t However, his study has several limitations: it does not include social structural dimensions of differential locations in the social structure and the differential social location in primary, secondary groups, representing more meso level structures. Also, the sample used in this study was collected from only one high school and middle school in Florida. Further, as the author reported, there is possibility of some misinterpretations of the study findings by using strict model fit cr iteria of structural equation models. Since the study utilized a sample drawn from only one school area, the strict criteria accepting supportive evidence of the
56 mediation effects models might be rejected with errors. In addition, the author was concerned that the non normality issues of data which might have generated error in the moderation One of the most recent study by Gibson et al. (2010) provides a partial test of the SSSL th eoretical model by examining whether the social learning process mediates social structural factors, such as theoretically defined variable (i.e., social disorganization), differential t behaviors. This study used longitudinal data from the 9, 12 and 15 years old cohorts from Project on Human Development in Chicago Neighborhoods (PHDCN) data (n=1,214) and 1990 U.S. census data (343 neighborhood clusters). This study used two social struc tural dimensions of the SSSL theory, theoretically defined constructs (concentrated disadvantage as an indicator of social disorganization), and differential location in the social structure (age, race, gender, and socioeconomic status), and one social lea rning variable, differential association with delinquent peers. Particularly, concentrated disadvantage measure represents the structural conditions of neighborhoods in social disorganization theory and was operationalized by six items drawn from 1990 U.S. census data; percentage of people below the poverty line, percentage of people on welfare, percentage of female headed households, percentage unemployed, percentage less than 18 years of age, and percentage Black (p.141). They hypothesized that if the SS SL proposition is correct, 1) children and adolescents reside in more disadvantaged neighborhood will have more delinquent peers, 2) children and adolescents reside in more disadvantaged neighborhood will engage in delinquent behaviors; and 3) delinquent p delinquency and will mediate substantially the impact of concentrated disadvantage on Childre n and adolescents residing in highly disadvantaged neighborhoods associated with
57 more delinquent peers and are more likely to engage in delinquency. Specifically, as anticipated by the SSSL model, the impact of concentrated disadvantage has been substantia lly reduced once the differential peer association is entered into the model while neighborhood concentrated disadvantage measures remained statistically signific ant and became insignificant only after differential location in the social structure variables were included in the model (age, race, gender, and socioeconomic status). Also, most of the effects of differential location in the social structure, such as ag e, race, gender, and socioeconomic status, have been also substantially reduced after differential association with delinquent peers is taken into account. On the basis of the findings, the researchers conclude that the effects of social disorganization an d demographic characteristics are partially mediated by differential association with delinquent peers and the SSSL model is partially supported by the results from their study. It indicates that the differential association in fact shows moderate effects at best and did not fully explain the effect of neighborhood disadvantage on state that full mediation effects are expected (p.146). Although this study utilized many neighborhoods in combination with individual data on children and adolescents and used a good theoretically defined measure (i.e.., concentrated disadvantage), this study still has a major limitation in that it includes only one social learning measure. Al though differential association is the most empirically influential elements of social learning theory, the lack of the other key social learning constructs may affect the results of this study. There are some studies that did not explicitly claim that th ey are testing the mediation proposition of the SSSL, but they found partial support for the theory (Haynie et al., 2006; Wu, et al, 2008). Haynie and colleges (2006) examined whether delinquent peer exposure mediates the neighborhood characteristics on ad
58 this study is not an explicit test of the SSSL theoretical proposition, it provides partial support for the mediation effects of the differential peer association. Using the national longitudinal study of Adoles cent Health (Add Health) data, and 1990 U.S census data, they developed Hierarchical Linear regression models for analysis. The findings reveals that adolescents reside in more socioeconomically disadvantage neighborhoods are more likely to engage in viole nce net of individual demographic and family characteristics and neighborhood selection measures. Importantly, once the peer network measure is taken into account in the model, they substantially reduced the effect of neighborhood disadvantage and other st ructural level factors toward zero and the two factors became insignificant. In addition, neighborhood disadvantage is associated with exposure to violent peers, and peer exposure mediates part of the neighborhood disadvantage and violence association. The authors concluded that the findings suggest that neighborhood disadvantage influences adolescent violence indirectly by increasing opportunities for youth to become involved in violent peer networks and that this association is in part mediated by exposur e to violent and academically unmotivated peers (p.163). or tests the SSSL model, but this study also provides partial test of the SSSL model. They examined contextual risk for marijuana and other hard drugs of young women and found that socially and economically disadvantaged neighborhoods influence their drug use, which is mediated by personal networks. The measures used for personal network variable include number of Overall, previous studies testing the mediation proposition of social learning proc ess posited by the SSSL model provide at least partial support. However, there are still some limitations regarding the previous studies. First, a complete test of the SSSL theory is still
59 rarely conducted (one exception is Holland Davis ). Instead, m ost of previous research built up hypotheses drawn from one or two of primary propositions of the SSSL model; differential association or definitions mediate structural variables on crime and delinquency or used less than ideal social structural measures and soc ial learning variables. Also, the analytical models frequently failed to address measurement errors when using nested data in geographical units or groups. Since the SSSL theory is a macro and micro level theory, testing its theoretical proposition requir es developing multi level models. Yet, some of previous studies utilized traditional Ordinary Least Square (OLS) regression models that are not appropriate for examining multi level data. Moreover, some of previous studies reported not only mediation, but also some moderation effects of social learning variables. That is, social structural concepts may interact with the social learning process to increase or to decrease individual deviant or criminal acts. Therefore, more studies are necessary to test the mediation proposition of the SSSL model to investigate whether the social learning process intervenes in the relationship between social structure and outcome behaviors by mediation only or both mediation or moderation simultaneously. Furthermore, in term s of the study evidence suggesting some moderation effects of the social learning process, some (for example, Sampson, 1999) may say that there are perhaps other linking process between the impact of social structural conditions and individual behaviors, i nstead of or in addition to the social learning process proposed by Akers (1998) such as family wellbeing (Bellaire et al., 2003), informal social control process (Gibson et al., 2010) and other social psychological processes (Lee et al., 2004). Although the findings of moderation effects of social learning process do not invalidate the SSSL model, this unexpected finding needs to be addressed and further tested by rigorous empirical studies
60 with diverse data sets to suggest modification of the SSSL model if any modification is necessary. Consequently, identifying alternative processes linking structural influences and individual deviant and criminal acts is also a critical consideration in testing the SSSL model. That is, researchers need to investigate whether alternative or other social psychological variables, besides social learning variables, may mediate the impacts of social structural aimed to test the SS SL model must consider including variables drawn from major social control theories such as social bonding and self control theory. In fact, Akers (Akers & Lee, 1999; Lee, et al., 2004) emphasizes several times the importance of examining mediation effects of other social psychological theories, such as social bonding and self control theory, in testing of the SSSL model along with the social learning variables. For example, in a study that examined the relationship of crime and deviance to age with social learning and social bonding theories, Akers and Lee (1999) suggests that they basically agree with the statement (p.2). They suggest that the causes of crime and deviance are principally the social psychological causative variables specified in social learning theory as well as social bonding theory. They clearly mention that Hirschi himself did not state it explicit ly, but it is at least implicit in social bonding theory. Therefore, they hypothesized that both social learning and social bonding elements mediate the age effects on adolescents marijuana use and found supportive evidence of mediation for both theories. A similar study had earlier been conducted by Krohn, Lanza Kaduce and Akers (1984) and tested relative mediation effects of social learning and social bonding variables on the relationship between community contexts (rural or urban) and drugs use. For bot h of the
61 previous studies, social learning found stronger mediation effects compared to the social bonding theory. More recently, Akers and his colleges (Lee et al., 2004) clearly state that earning variables, good measures of other potential mediating processual or micro level variables. The most obvious control (Gottfredson & Hirsch, 1990), or other social psychological or personality v Moreover, it has been known that many scholars noted that there are a lot of similarity between social learning theory and social bonding theory. More specifically, Akers and Cochran (1985) (and Akers, 1997, 1998) suggested that social learning and social bonding theory has the greatest compatibility. In their arguments about conceptual integration between the conceptual o verlap between social learning and social bonding theories to illustrate the (1985) pointed out that there is considerable conceptual overlap of the belief concept in bo nding theory and the definitions concept in social learning theory. Further, they argued that negative and neutralizing attitudes and verbalizations which are (p.339). When beliefs are measured as the proximal beliefs specific to smoking or drug use, they are empirically the same as definitions favorable and unfavorable. Later, Akers (1998) distinguished between the two concepts when c research: He allocated studies that used more generalized beliefs as studies testing bonding theory while studies used more specific beliefs to delinquency and substance use as research testing learning theory (p.339) Therefore, although the belief terminology comes directly from social bonding theory it does not need to be seen as separate from and can be integrated into the general overall concept of definitions. However, in empirical studies, researchers di d
62 not make the connection or distinction between the two concepts, proximal beliefs of social bonding theory and definitions in social learning theory. With regard to the other conceptual overlaps between social bonding and social leaning theory, Akers an d Cochran (1985) argue that social bonding variables are subsumable under some concepts of social learning variables; the rational components in conformity, the cost of deviant behavior, of social bonding theory is subsumable under one side of the concept of differential reinforcement, negative punishment; the concept of attachment overlaps with the intensity of differential associations or other social leaning variables such as the concept of definitions, self reinforcement in and imitation social learnin g theory (pp.339 340). This is another reason for emphasizing the importance of including other social psychological theories, such as social bonding and self control variables to compare the relative mediation effects of the three theories. Despite the importance of comparing and considering other social psychological variables such as social bonding and self control variables, none of the studies has fully included other social psychological variables in models testing the SSSL theory, except the limite d efforts made by Akers and Lee (1999) and Krohn, Lanza Kaduce and Akers (1984). Therefore, by considering the importance of comparing and examining other social psychological variables with social learning variables, this study makes a contribution to the existing literature and development of the SSSL model. Other contributions are that this study is a compete test of the SSSL model that utilizes the full set of social structural dimensions as well as the four major concepts of social learning variables, as well as including all of the major concepts of social bonding and self control variable. Also, using HLM and HGLM analysis models on three different substance use dependent variables, this study attempts to estimate not only mediation effects of the soc ial learning variables but also attempts to find
63 moderation effects, if there is any. Finally, by using a sample drawn from Korea, this study attempts to extend the generalizability of the SSSL theory into another social cultural context. Social Bonding T heory and Self control Theory control are now considered to be the major control theory within criminology and have received tremendous research interests and are endorsed by a high proportion of criminologists (Akers ogically coherent, and parsimonious theory that is applicable to any type of criminal or deviant p. 129) Social bonding theory effectively incorporates key elements from all previous control theories a nd provides new insights to explain delinquent behaviors and drug and substance use. In his book, Causes of Delinquency Hirschi (1969) posits that delinquency and socie ty (1969, p.16). In other words, the more closely an adolescent bonds to his or her family school, and friends, the less likely he or she is to act in deviant ways. Hirschi identified four major elements referring to social bonding to society. The four m ain concepts in social bonding theory include attachment, commitment, involvement, and belief. According to Hirschi (1969), the four major elements are highly intercorrelated each other, indicating that the weakening or strengthening of one element will be accompanied by the weakening or strengthening of another element. First, Hirschi argues that attachment to others, regardless of whom they are, will make a person to be constrained from committing deviant behavior to make the person adhere to conventional societal rule and standard. That is, the more one has close emotional ties to others, admire and honor them and the person
64 cares about their expectations the less likely he or she is to be involved in delinquency and substance use. Second, commitment ref in law breaking behaviors or an investment of time or energy for conventional activities, such as academic achievement or occupational aspiration (Akers & Seller, p.130) It suggests that i f one has greater commitment, the more the individual risks jeopardizing what he or she has by committing delinquent and substance use behaviors. Third, the concept of involvement studying, spending time with friends and family, and participation in extra curricular activities. It assumes that if a person is heavily engaged in conventional activities, this person becomes too busy or too pre occupied to engage in delinquency and sub stance use. Fourth, with regard to the concept of involvement, some researchers have continuously argued that there is conceptual and empirical overlap between involvement and commitment (Akers & Cochran, 1985; Conger, 1976; Hwakins & Weis, 1985; Krohn & Massey, 1980). Therefore, measuring time spent in conventional activities is perhaps a better measure for the concept of commitment as the temporal dimension of commitment (Hwang, 2000). Finally, the fourth element is belief. Hirschi defined the concept o f belief belief in the general conventional values and norms, and endorsement of moral validity of societal rules and laws that keeps the individual from violating those rules and law. It assumes that if an individual believes he or she should obe y the rules and laws, the more likely he or she is to act in a manner consistent with these rules and laws (Akers & Sellers, 2009, p.131). Although Hirschi described the concept of belief as generalized beliefs, research on adolescent alcohol and cigarette smoking has used primarily specific positive and negative beliefs about substance use behaviors (Krosnick & Judd, 1982). Furthermore, some
65 smoking initiation while gen eral belies are important only to the extent that they impact more proximal beliefs (Hwang, 2000, p.32; Fishbein & Ajzen, 1975; Chassine et al., 1981). Therefore, the social bonding theory predicts that a dolescents are more likely to refrain from deviant and substance use behaviors: 1) if they are more strongly attached to others ( attachment ); 2) if they have greater stakes in conformity (commitment); 3) if the majority of their time is spent involved in conventional activities (involvement), and 4) if the y accept conventional moral law and conventional belief s (belief). Empirical Validity of Social Bonding Theory Previous studies generally have supported the contentions of social bonding theory and the major concepts, except for the concept of involvement The strength of the support from empirical studies is weak to moderate (Akers, & Seller, 2009). Among the social bonding concepts, attachment seems to be the strongest predictor for delinquency (Lackey & Williams, 1995; Nagine & Paternoster, 1994; Wiatro wski, Griswold, & Roberts, 1981; Mak, 1991; Junger & Marshall, 1997) and for substance and drug use (Jessor & Jessor, 1975; Brook, Whiteman, Gordon, & Brook, 1981; Burkett, 1977; Jessor, Chase, & Donovan, 1980; Krohn, Massey, Skinner, & Lauer, 1983; Flewel ling & Bauman, 1990; Farrell, Barnes, & Banerjee, 1995; Jang, 2002). P arental influence, attachment to teachers, and attachment to peers a re associated with adolescent deviance and substance use. Parenting style and parental control (as indicators of atta chment) are also important determinates of delinquency (Coombs & Landsverk, 1988; Messner & Krohn, 1990). Lack of attachment to parents, regardless of whether an individual comes from a broken or intact family, was a strong predictor of g, alcohol, marijuana and other drugs (Brook, Whiteman, Gordon, & Brook, 1984 ; used a sample of 882 adolescents who took part in an initial survey when they were12 15
66 and 18 y ear old, and followed by the researchers three years later again to collect data. This study found that greater proportion of adolescents who reported a lower level of perceived parental affection and love were involved in the most serious levels of alcoho l and other drug use than adolescents who reported greater levels of perceived parental attraction. In some studies, however, the relationship between attachment and substance use were rather weak (Agnew 1991a; Bahr, Macos, & Maughan, 1995 ; Krohn et al., 1983). A longitudinal study conducted by Krohn, et al., (1983) found that belief and commitment are associated with smoking abstinence among adolescents, but attachment to friends and behavior contrary to the theory prediction. Agnew (1991a) also conducted a longitudinal study and found that three social bonding variables are weakly related with de linquency. Bahr et al (1995) found but these measures operated through peer association. That is, adolescents who have strong bonds to family are less likely to ass ociate with friends using drugs. Studies also found some support, but weak, for the predictions of the effects of commitment, involvement and belief on delinquency and drug use behaviors ( Bahr, et al. 1995; Brownfield & Sorenson, 1993; Hagan, 1991; Jan g & Jonshon, 2001; Lyerly & Skipper, 1981 ). Using a sample of 1,505 adolescents from the 1977 National Youth Survey data, Jang and Johnson (2001) found that adolescents who were more committed to both school and educational work used drugs less than those who are less committed to these conventional works. Bahr, et al., (1995) also found that commitment was negatively associated with committed to educational work t end to drink less frequently and to drink less when they do
67 drink than those adolescents who are less committed to education. As such, some studies found some support for the concept of commitment to grades, educational and occupational expectations on del inquency, substance and drug use, but the overall relationship was weak (Agnew, 1985; Kandel, Kessler, & Margulies, 1978; Paternoster & Iovanni, 1986; Liska & Reed, 1985). Sometimes there are findings opposite to expectations from the social bonding theory (Krohn, et al., 1983). The belief concept received some mixed support ranging from moderate to weak Cochran, 1985; Krohn & M a ssey, 1980; Kaplan, Martin, & Robbins, 1982; Krohn et al., 1983; Paternoster & Iovanni, 1986). In a longitudinal study, Krohn et al (1983) found that belief and commitment are the strongest predictors among social bonding variables in so found some supportive evidence that adolescents who endorse conventional and normative beliefs less are more likely to use drugs. Akeres and Cochran (1985) found that all of the four major concepts of social bonding theory were significantly associated use. Contrary to the theory prediction, however, some findings contradicted to the social bonding theory. For example, Burton et al. (1995) found evidence non supportive of the belief concept in a study of 263 high school students. They found a significant and positive, With regard to the concept of involvement, previous studies generally found that substance use behaviors (Agnew, 1991b; Ginsberg & Greenly, 1978; Jenkins, 1997; Wiatrowski & Anderson, 1987). For exampl e, Ginsberg and Greenly (1978) used a sample of
68 Overall, previous research testing social bonding theory found th at bonding factors frequently reported that social bonding measures had weaker effects if social learning measures were included in their models. While some studies reported greater support for social bonding theory when it is used as a single theoretical model, the effects of bonding disappear in studies including some measure of social learning variables while social learning variables retain their strength. Gott Control Theory In their book, A General Theory of Crime Gottfredson and Hirschi (1990) claim that this theory is suited to explaining all types of crime and deviance, at all times and places, all age, regardless of social class, gender, and race, etc., focusing on one unidimential trait; low self beh avior are due largely to individual differences in the personality trait they call low self Ameklev, Grasmick, Tittle, & Bursik, 1993, p.225). That is, the propensity to engage in criminal and delinquent behavior is caused by low self control (Ak ers & Sellers, 2009). Low self control is formulated as the outcome of ineffective parental socialization t occurs (Gottfredosn & control, once established in early childhood between aged 8 to 10 years old, remains stable over the life course and are relatively unaffected by othe r fact ors (1990, pp.107 108). That is, those individuals with low self control will be substantially more likely to engage in criminal acts
69 at all periods of life while people with high self control are less likely to commit crime, according to self contr ol theory (Akers, 1997, p.91). Also, individuals with low self control will have a greater and stable tendency to commit crime and deviance across all social circumstances (e.g., school, work, and marriage) at all stage of life after childhood (Akers, 1997 p.93). Although Gottfredosn and Hirschi do not explicitly spell out the elements of the low self control concept, they suggest the attributes that characterize people as constituting low self control in their original work (Arnekev et al., 1993; Grasmi ck, Tittle, & Bursik, 1993). According to the theory, individuals with low self physical (as opposed to mental), risk Hirschi, 1990, p.90). These individuals ar a desire for immediate and simple gratification while people with high self control are able to defer gratification and recognize that involvement in crime and delinquency provide only Also, individuals who have low self control tend to be adventuresome and engage in risky and exciting activities rather than be cautious (risk seeking) (1990, p. 89). These people with low self c ontrol also are deemed to be self centered, indifferent, and insensitive to the sufferings of others (self centered) (1990, p.89). The other characteristic of individuals with low self consol includes a tendency to be (1990, p.89). Also, these low self control people tend to be more physical and less verbal, rather tha Lastly, low self control characterized by a temper, which refers to tendency to have low level of tolerance for frustration and little ability to respond to conflict through verbal rather
70 than Hirschi believe to comprise self control: impulsivity, preference for simple tasks, risk seeking, physical activity, self centeredness, and temper (Grasmick et al., 1993). The refore, Gottfredosn and Hirshi (1990) propose that individuals possessing low self control, who are impulsive and self centered, and oriented to simple task, risk seeking, physical activity, and have low tolerance for frustration are more likely to engaged in crime and deviant behaviors, including analogous acts (Hwang, 2000, p.37). With regard to the empirical validity of self control theory, it has received moderate support by a number of studies on criminal behaviors among adults (Tittle, Ward, & Grasmi ck, 200 3 ), adolescents (Cauffman, Steinberg, & Piquero, 2005), adolescents in various Western and non Western countries (Vazonyi et al., 2001, 2004), inmates or offenders (Piquero, MacDonald, Dobrin, Daigle, & Cullen, 2005), street kids (Baron, 2003), and university students (Higgins, 2004). Also, it has received moderate support in explaining adolescent delinquent behaviors as well as drug and substance use behavior (Akers & Sellers, 2009; Arneklev et al., 1993, 1999; Brownfield & Sorenson, 1993; Burton, Cullen, Evans, & Dunaway, 1994 ; Costello, 2000; Deng & Roosa, 2007; Evans et al., 1997; Gibbs & Giever, 1995; Grasmick et al, 1993; Keane, Maxim, & Teevan, 1993; Nagin & Paternoster, 1993; Pratt & C ullen, 2000; Vazsony et al, 2001, 2004; Wood et al., 1993 ) Pratt and Cullen (2000) conducted a meta analysis with 21 cross sectional and longitudinal studies directly examining the relationship between low self control and crime, delinquency and substance use to examine the empirical power of self control measur e. In their analysis, they found that self control is a significant and strong predictor of crime, delinquency and drug use among analysis reported here furnishes fairly impressive empirical support for Gottfredson et al. (1998) tested the proposition of self control theory on imprudent and criminal behaviors
71 using self reported survey data. They found supportive evidence for the self control theory and con Gibbs and Giever (1995) found strong relationship between low self control and alcohol use and a minor type of misbehavior, class cutting. Unnever, Cull en and between self control and academic dishonesty. College students with lower levels of self control were significantly more prone to cheat on an exam. Evans et al. (1997) examined a significant positive relationship between low self control and criminal behaviors, as well as behaviors analogous to crime. They found that self control was a more significant predictor of crime and delinquency than were measures of social control. Furthermore, they found that low self control was also associated with various social outcomes such as quality of marriage and employment. The findings indicate that persons with low self control make decisions in their lives that lead to negative social consequences, including crime or delinquency. However some studies found mixed support for the self control theory (Arnek l ev et al., 1993; Brownfield & Sorenson, 1993; Perrone, Sullivan, Pratt, & Margaryan, 2004 ; Wood et al., 1993). Contrary to the theory expectation, however, Wood et al (1993) found that some demographic factors, such as age and gender had significant effects on self control. Their fin dings reveal that males have a greater likelihood to engage in illegal substance abuse, and vandalism compared to that of women. Arnek l e v et al. (1993) also examined the relationship between low self behaviors. In analysis, they found some mixed effects of low self control on such misbehaviors. While the low self control scale predicts drinking and gambling behaviors, it does not explain smoking behaviors. Grasmick et al. (1993) also found mixed result s for the low self control and crime relationship. While low self control had a strong positive effect on fraud, it had no significant
72 effects on force. Moreover, they found significant interactive effects between low self control and opportunity on offend ing (both fraud and force). Patternoster and Brame (1998) presented another study with mixed support for the theory. They found significant positive low self control effects on criminal and analogues behaviors among children. Yet, the correlation between c riminal behaviors and analogous behaviors remained high after holding self control constant, suggesting other possible forces might be operating on the relationship between criminal behaviors and behaviors analogous to crime. This finding contradicts Gottf control has direct effects on crime and deviance, while they do not consider any mediation effects between self control and criminal behavior. In sum, vast amount of previous research testing self control theory fo und that low self including substance and drug use. However, there is still a tautology issue in testing the low self control theory because much empirical research a ssumes low self control from the commission of certain acts, behavior measures (i.e., drinking and driving). Therefore there is control. In sum, therefore, over the past two decades, a large body of studies tested the theory, but evidence suggesting that the theory is empirically valid and error free have yet surfaced Additionally, t here are studies that at all t Beaver, Wright, & Delisi, 2007 ; Burt, Simons, & Simons, 2006; Grasmick et al., 1993; Hay & Forest, 200 6 ; Wright et al., 1999; Pratt & Cullen, 2000; Patternoster & Brame, 1998 ; Perrone, Sullivan, Pratt, & Margary an, 2004).
73 Cross cultural testing of social bonding and self control theories Relationship between social bonding theory and self control theory Although the social bonding and self control theory have been frequently tested individually and these two t heories once have been considered as different theory, now the two theories are considered as theories that share the same core elements of explanatory v ariables. As it is well known, Hirschi is the author of social bonding theory (1969) as well as the coa uthor of self control theory (1999) along with Gottfredson. Although he is the author of the two theories, the propositions of these two theories appeared to contradict each other and the relationship between the two theories was not specified by the autho rs until recently. In fact, when Hirschi first posited social bonding theory, he criticized earlier control or self control Hirschi suggested that attachment is a better concept than self control, because it avoids the tautology problem in measures of self control and because self control can be subsumed under the concept of attachment (Akers & control theory (1990) later emphasized the causal significance of self c ontrol while keeping silent on the key elements of social bonding theory, including attachment. This is was a rather abrupt and surprising theoretical transition without much clarifying explanations (Akers & Sellers, 2009). In the 1990 self control theory Hirschi and Gottfredson refrained from detailing precisely how the two perspectives (social control and self control) converged and diverged. That indicates that Hirschi once thought that social bonds were the main determinant of crime. Later, however, h e (and Gottfredson) came to believe that self control is the sole individual level causal factor in criminal involvement. In 2004, Hirschi (2004) made a more concerted effort to clarif y the relationship between social control theory and self control theo ry. Low self control and the propensity to crime were defined in the same way Gottfredson and Hirsch (1990) asserted that the
74 propensity of crime was viewed as the tendency to be impulsive, insensitive, physical, risk taking, short sighted, and nonverbal. Also, Gottfredson and Hirschi argued that it could be best measured by the very criminal and deviant behavior it was supposed to explain. This of course presented both a conceptual and a measurement tautology issue (Gottfredson & Hirschi, 1990). In the n ew er concept, Hirschi (2004) redefined self the full range of potential costs of a particular act that moves the focus from the long term .534) and asserted that social control and self control are the same thing. Interestingly, this new concept seem ed to be exactly opposite from his original social bonding theory suggested in 1969 (Hirschi, 1969). As noted above, once he (1969) subsumed sel f control under the concept of attachment. However, he (2004) now equates all elements of the social bond with, and perhaps subsumes them under, a new concept of self control (Akers, 2005). In other words, notion social bonding theory is now merged under the umbrella concept of self control. These measures of the f our elements of social bonds (attachment, commitment, involvement, and belief) now become measures of self control concept (Akers, 2005). Akers (2005) pointed out this tautological problem in the 1990 original version of self control theory, because the new concept of self control is not synonymous with criminal propensity or criminal behavior and the new measures of it s uggested by Hirschi are not indicators of the dependent variable converted to measures of the independent variable. self control concept presente d by Piquero and Bouffard (2007). They collected data from college students to measure the newly defined and reconceptualized self control concept and to compare its
75 control, which is th e most commonly used to measure the self self control was significantly related with criminal outcome measures, while the effects of control reduced to i nsignificant. They concluded with the suggestion that the new self characteristic, one that may not necessarily be stable within persons, over time, and across 304). Such research finding indicates that redefined self control concept ensures further empirical research. However, Hirschi has not proposed any new measure of self control concept by himself in his modification. Rather, on the basis of his newly equat ed self control and major elements of social bonding concepts, he simply used the same measures that had been used in the past (Hirschi, 1969) for measuring social bonding concepts. This current xamine the reconceptualized self measures that captures the six dimensions of the self control concept that Gottfredson and Hirschi originally proposed in 1990. Previous Studies Comparing Social Learning, Social Bonding and Self Control Theory Social leaning, social bonding theory, self control theory have been most frequently also in combined models. In general, these three theories have received substantial empirical supports, individually and together. Also, sometimes the three theories have been compared with each other as competing theories in explaining crime, delinquency and substance use. When r esearch compares social leaning to social bonding and/or self control theories, it provides more support for the propositions of social learning perspectives, as well as the relative greater importance of peer association and del inquent peer influence than variables
76 from social bonding theory and self control theory (Akers & Cochran, 1985; Bahr, et al., 2005; Brown, et al, 1993; Cooper, May, Soderstrom, & Jarjoura, 2009; Dillon, et al., 2008; Dishion et al., 1991 ; Erikckson, et. a l., 2000; Ghanizadeh 2005; Hoffman, 1993; Hwang & Akers, 2003, 2006; Krohn et al., 1984; Neff & Waite, 2007; Triplett & Payne, 2004; Warr, 1993b; Zhang & Messner, 1995). Krohn et al (1984) found social learning variables accounted more than social bonding variables for community variances in adolescent substance use. Akers and Cochran (1985) marijuana use. They found that social leaning theory is strongly supported while social bonding theory is moderately supported by the research. The social bonding variables such as attachment, commitment, and beliefs are significantly related to marijuana use, but the effects of those variables are much weaker than peer associat ion, reinforcement and attitudes toward marijuana smoking. McGee (1992) examined the relative importance of parental versus peer influences on adolescents illicit drug use. The study used data extracted from the Monitoring the Future data in 1985. Using m easures from social bonding, strain and social learning theories, the other hand, the effect of parental influence measure based on social bonding theory does not h ave significant impact on adolescents drug use. Cooper et al (2009) compared multiple criminological theories to examine their respective strength in predicating substance use behaviors among incarcerated juvenile ng a sample of approximately 800 delinquents drawn from incarcerated population in Midwestern states, they examined the relationship In analysis, they used nonsoc ial reinforcement theory, social learning theory, social control
77 theory, and strain theory. Their findings suggest that nonsocial reinforcement and social learning theory demonstrate greater predictability for both preference for and use of illegal substan ces among juvenile delinquents compared to social control and strain theory. Finally, although analysis reported that low self control is a significant and strong predictor of offending, they also noted that variables drawn from social learning theory, in particular, peer factors remained as consistently strong and significant predictors of crime, even after controlling self control constant. Cross Cultural Studies Comparing Social Learning, Social Bonding and Self Control T heory The relatively greater importance and significance of social learning variables over social boding variables and self control measures were consistently found among studies conducted in cross cultural settings. First, studies conducted in a South Kor ean context strongly supported the greater importance of social learning variables in explaining deviance and crime than social bonding and self control theories (Hwang & Akers, 2003, 2006). Hwang and Akers (2006) found that peer association has a direct positive effect on substance use and that peer effects were greater than that of parental influence. They concluded that the findings were compatible with the previous research in the United States that supported social 003) also compared the empirical validity of social leaning, social bonding, and self control theory by using a sample from South Korean youths use ranged from 58 p ercent to 67 percent explained variance while social bonding variables accounted for less than 20 percent and self control theory less than 12 percent of the variance. T hey concluded that the variables from social learning theory had greater significance a nd power than that of social bonding and self control theory in explaining the variance of South
78 Other studies using samples from Asian countries also reveled social learning theory has greater predictability in exa mining delinquency and crime than social bonding and self control theory. Zhang and Messner (1995) compared social learning and social bonding theories by examin ing t he influence of family harshness and deviance on adolescent delinquency in China where rel atively strong emphasis on family values remains. They family in China. Contrary to their prediction, the peer variables turned out to be the key predictors of delinquen cy in China Cheung and Cheung (2008) tested self control theory against social variables including social bonding variables, differential associations, and strain variables with data from Hong Kong. They found that self control was significantly correlat ed to social bonds and differential associations, and when entered into the model with these variables, self control became insignificant. In sum, what limited literature exists on delinquent youth prediction using data collected from Far Eastern co untries seems to support most of the findings from American studies. The research indicates that differential peer associations were a strong predictor of delinquency, as were family and school attachments, and familial strains. Studies conducted i n various other social cultures, such as Western Europe, South America, and Iran, found that the social learning variables explain adolescents; drug use and delinquency relatively better than social bonding and self control variables. Hartjen and Priyadars ini (2003) compared social bonding and social learning theory with a national sample of 387 French female and male college students on various types of delinquent behaviors. They utilized the four social learning measures (i.e., attitude toward deviance, e xposure to delinquent peers, and an index of exposure to delinquent peers) as well as social bonding measures (involvement with family, and school). The findings revealed that all of
79 social learning measures were significant and reliable predictors for the French college with their delinquent behaviors. Particularly, they found that social learning theory explains Another study conducted in Sweden, Europe by Svensson (2003) who compared behaviors. He conducted a cross sectional study with a sample of 859 adolescent drawn from Sweden to examine whether parental monitoring (social bonding theory) and delinq uent peer those theories address the gender difference in drug use. This study found that when delinquent peer association measure was taken into account in the analysis model, gender became insignificant while parental monitoring failed to reduce the significance of gender. That indicates the social learning variable was a stronger predictor than the social bonding r both boys and girls. The study found that boys tended to have higher levels of contact with delinquent peers than girls, while females tended to receive higher levels of parental monitoring than boys. Also, using an interaction term (parental monitoring delinquent peers association), this study revealed that girls who receive less parental monitoring were at a greater risk of becoming associated with delinquent peers and, in turn, this association with delinquent peers led them to engage in drug use. Ho wever, he found no significant interaction effects for boys. Ghanizadeh (2005) examined relative predictability of social learning variables and social control variables with a sample of 173 college students in one university, Shiraz, Iran. Research
80 cigarettes, cannabis, alcohol and opiate, heroin with variables of social learning theory (i.e., substance using peers) and social bonding theory (i.e., religious attachment). The findings use across all types of substances and drugs. On the other hand, only some of social bonding variables are indirectly related with substance and drug use behaviors. Furthermore, the significant variable of social bonding theory is religious attachment which Hirschi (1969) had originally rejected as having relevance as one of the concept of social bonding. On the other hand, social learning theory consi stently includes the religious variable either as differential location of social structure in SSSL model or one of differential association variables (i.e., association with church group members). Therefore, the significance found among social bonding the ory can be interpreted, in fact, as an evidence for greater predictability of social learning theory than social bonding theory. These findings provide strong evidence supporting social learning theory as having better applicability in explaining substance and drugs use among Iranian college students than social bonding theory and accordingly as having greater generalizability. Meneses (2009) examined cross cultural applicability of the four major criminological theories on deviant and substance use behaviors among college students in South American societies. These four theoretical perspective used in this study include social learning, strain, self control and social bonding theory that have been widely applied for explaining deviant and criminal b ehaviors as well as substance use behaviors in the United States. For analysis, he assessed the findings from a sample of Bolivian college students collected in two universities and compared the findings to the analysis results of a sample of American coll ege students drawn from one large university in southeastern United States. Explanatory variables from the four theories utilized to explain the variances of five dependent variables (alcohol, cigarette, marijuana, other drugs use, hitting, and beating
81 so meone). This study found that measures of social learning theory are strong and significant predictors of substance use (alcohol, cigarette, marijuana, and other drug use) and violent behavior (hitting and beating someone) not only the Bolivian sample but also the American sample. However, they found mixed support for social bonding, strain, and self control theories. Interestingly, this study found some mediation effects of social learning process for the three variables effects on deviant and substance us e behaviors. That is, when social learning variables are added in the same equation with the variables from the three theories, most of the effects of variables from these theories disappear and do not operate as similarly as social learning variables in b oth societies. As such, several studies have compared the relative strength in explaining cross cultural context. However, there are not many studies ex amining the mediation effects of the three theories, social learning, social bonding and self control theory in one combined model. In fact, there are only two of these studies by Krohn et al (1984) and Akers and Lee mediation effects on structural (community) variations or age impact (differential social location) in adolescent substance use compared to mediation effects of the social learning theory. Therefore, again, it is important to assess if there are any media tion or moderation effects among social bonding and self control variables that are comparable to the same effects of social learning variables to expand the propositions of the SSSL theory.
82 Figure 2 1. The Social Structure and Social Learning Model [A dopted from Akers, R.L. 1998. Social Learning and Social Structure (Page 331, Figure 12.2). Transaction Publisher, New Brunswick, New Jersey]
83 CHAPTER 3 METHODS This dissertation analyzes secondary data on adolescence substance use among South Korean yout hs attending schools in Busan (Hwang, 2000) and adds an original data set comprised of structural level variables constructed from census data collected by from the Korean National Statistical Office and Busan City government on the 15 districts of Busan. The school districts and the census districts are coterminous. The analysis uses micro and macro level data to test the SSSL model through hierarchical linear modeling. Individual Level Data The first set of multilevel data is individual level data, which is drawn from a self report survey data from South Korean youths in Busan, South Korea, originally collected by Dr. Sunghuyn Hwang for his dissertation in 2000. The data were collected by using a two staged stratified sampling method. In the first stage o f the research, the key stratification criterion for the selection of schools was geographical location (district) and type of schools. Busan exhibits different soc ial characteristics, geographical dispersion is important for vocation al training curricula, was used as a criterion for selection of schools. On the basis of these criteria, students of 26 schools were initially selected as target population from the total h schools (38.5%), 6 industrial high schools (19.2%) from fifteen districts in Busan. Twenty four of the high schools are in a small town outside the metropolitan Bu san, while the rest of schools are within the metropolitan area.
84 In the second stage, classes were randomly selected and all of students in those selected classes became target sample. In general, the average class size in each high school classroom in Bu san ranged from 40 to 45 students. The questionnaire was administrated to all of the students in the selected classes present at that day. The number of students present was 1,035 and all completed the survey. The typical number of absentees and the number of absentees on that day is not known. The final total number of students included in the study was 1,021 students because 23 students who took part in the survey were excluded due to the incompleteness of their survey questionnaires. Thus, the response r ate from those present was 100% and of these 98% completed usable questionnaires. The data from the survey includes measures of all core concepts of social learning, social bonding and self control theories. The measures of social learning variables use d by Hwang were adapted from the questionnaire from the Boys Town Study conducted by Akers and his associates (Akers et al., 1979; Hwang, 2000). The instrument therefore contains very good measures of social learning variables. Also, measures of social bon ding variables were adapted from the Boys Town instrument and from the instrument used in the study conducted (1986). Therefore there are also very good measures of s ocial bonding concepts in the data. The self control construct was measured by a set of survey questions used by Arnekev et al survey conducted by Grasmick et al., (1993) in spring 1991. The questionnaires used in translation) by the original researcher. For more detailed information in terms of measurement and the survey questionnaire re liability, pleases refer to the dissertation of Hwang (2000, pp. 49 50).
85 Structural Level Data The data collected by Hwang (2000) contains the school name and location imbedded This number allow s identificati on of the schools which the respondents attended. The geographical location of the school s and the residential areas from which it draws students with census data are ideal because the Gu for the school district is exactly the same at the Gu for collection of census and other sociological data. The structural level of data for the Gu is drawn from two different governmental resources in South Korea. First, census data, such as population density, residential mobility, and gender and age composition in each district in Busan, were downloaded from the Korean National Statistical Office (KNSO) to be used for this dissertation research. Second, several important social, economical, and cultural characteristics of districts in Busan (i.e., crime rates, compositi on of low income single family in each district, etc.) were obtained from the Busan Statistical Yearbook (BSY) published by the Busan city government annually. The Korea National Statistical Office (KNSO) which collects national census data across South K orea is a governmental organization which provides a primary service of annually collected national statistics. The KNSO offers services of overall planning and coordination of Korea national statistics, as well as the production and distribution of variou s economic and social statistics operating under the Statistics Acts of 1962 (and subsequent amendments) in South Korea. A total of 34 different government organizations take part in collection of national level census data (i.e., Korean National Police Ag ency, Ministry of Labor, and Korean Intelligence Service). Therefore, the data collected by the KNSO are the most reliable official statistics in South Korea. Fortunately, access to these statistical data is readily available online through the Korean Stat istical Information System (KOSIS) ( www.kosis.kr ). The census data are considered to be open public data and are downloadable
86 from this site directly by theme and by regions (i.e., population density, gender and age com position of each region). Using the KOSIS, several structural data were downloaded by selecting social and cultural characteristics of each district in Busan. The data of population density and residential mobility in each district were downloaded The dow nloaded data were originally collected in 1999 to correspond to the same year in which the self report data were collected by Hwang. The selected structural level data were saved initially in excel file format before being merged with the other structural data set. To me rge this data a corresponding identification number which reflected each district in Busan was assigned (1 through 1 5 ). Second, several important structural level data were also drawn from the Busan Statistical Yearbook (BSY) published b y the Busan city government (http://www.busan.go.kr/main). The BSY is an annual official publication of statistics which has been publicly available since 1993 according to Article 18 of the Statistics Law. The BSY contains various statistics related with Busan city. The BST is available at National Assembly Library (NAL) in South Korea as PDF file format. The BST include statistics such as percent of residents on government welfare, crime rates (i.e., the number of crimes committed by youths, including the crime clearance rates), composition of foreigners and statistics regarding the number of low the BST were also originally collected in year 1999 by the Busan city government. Although several different types of structural variables were available through the BST, due to the lack of level 2 units (15 districts), only the percent of residents on government welfare was used in the analyses. T he BST data of percent of residents on government welfa re in 1999 was downloaded through the NAL online website directly ( http://www.nanet.go.kr/main.jsp ) Those two set of structural level data were merged into one data set for analysis. Since the statistics fr om BSY are in table form, I entered the selected structural data into an excel
87 file to merge it with the census data obtained from the KNSO. To merge the two different data sets into one file, the Busan district names were used as an identifier. To develop an identifier, each district in Busan was assigned a unique number from 1 through 1 5 After the BST data were entered and coded into an excel file according to the district identification number (ID), it was merged with the census data, which has the same district ID. Next, the merged excel file was transferred into a SPSS data format in order to use it with the individual level self report data in analysis. Because the original self report data had been saved as in SPSS format, the structural data were sa ved in the same format. Therefore, I intentionally matched the two different level data sets into one type of statistical software format, SPSS. Before conducting multi level analysis with the structural level data, the self report data must have an identi fier to match individual level data with structural level data. Since self report data were collected by school which are nested in the 1 5 administrative districts (Gus) in Busan, the district identification number, which was used to merge the two structu ral data sets, is the perfect identifier for multilevel analysis. Therefore, the self report data were reentered according to the school identification number which indicated where the school was located in the districts in Busan. Utilizing census data co llected by South Korean government and Busan city government as structural level data will increase the reliability and validity of the structural level data. Using these data from South Korea to measure structural level variables has a great advantage in this dissertation research. As noted above, in South Korea, the census tract attends a school located in the same geographical district where the student has a curr ent address, according to the education policy in South Korea. Therefore, the students who took part in the self report survey in 1999 had their address in the same regions where the schools were located. The schools randomly selected by Dr. Hwang represen t 1 5 different
88 geographical districts which is the same unit of census tracts in Busan, South Korea. Unlike the research using American samples and census data from the United States in multilevel analysis, this proposed dissertation will not have issues a nd problems using census data collected from different geographical units from where respondents reside. Additionally, using the existing data save tremendous amount of times, efforts, and money for the researcher Measures of Variables Dependent variable s: Alcohol, Tobacco, and Depressant Use The original data set by Hwang (2000) includes measures of the frequency of tobacco, depressants (geborin, sardon, penjal, se daphin, nubain), stimulants (timing, night, esnanine, reglin), tranquilizers (atiban, barium, seconal, ruminal, barbital), inhalants (bond, sinna, butane gas), marijuana, and other drugs (LSD, cocaine, and narcotics). The focus of the dissertation research however, is and tobacco, and depressants. Consideration was given to including other types of substance st udy, is at very low frequency levels and the frequency distribution is too skewed to include this study reported having used marijuana and none of the respondents reported use of serious types of illicit drugs (LSD, cocaine, and narcotics) in their life time. Therefore, in this study, the dependent variables are the frequency of use of alcohol, tobacco, and depressants (geborin, sardon, penjal, sedaphin, nubain). It is believed that the inclusion of the use of the three substances as dependent variables is sufficient to address the central goals of the dissertation to conduct tests of the relatively explanatory power and cross cultural generalizability of three major theories of crime and deviance.
89 Each type of substance use (alcohol, depressants, and tobacco) was measured by which comes closest to h ow often have you ever used it in your lifetime? measured on a 7 item Likert response scale: never (0), once or twice (1), several times (2), less than once a month (3), once or twice a month (4), at least once a week (5), but not every day, and every day or nearly e very day (6). Alcohol and depressant use were basically measured as ordinal level data. First, the dependent measure of lifetime alcohol use frequency is created. At the time of data collection, each of students responded to a question asking how often the y used each of substances in their life time. The frequency is measured in a seven Likert item scale (0= and Participants are requested to choo se only one of the seven categories reflecting the frequency of alcohol use for that student. Therefore, the all of students were, then, collapsed into one outcom e measure and became one single measure of frequency of alcohol use. In the same way, a single depressant use frequency measure was created. The distribution of responses to the alcohol and depressant questions were approximately normal. However the dis tribution of responses to the tobacco question were essentially binomial and therefore The tobacco use as a dependent variable in this study was converted to a dichotomous variable (0=never use, 1=used one or more times). The problem with non normality of the frequency of tobacco use distribution is that it does not allow use multiple Hierarchical Linear Models. To conduct multilevel analysis with a dichotomous dependent variable, Hierarchical Generalized Linear Models (HGLM) are developed for tobacco use.
90 Independent Variables: Individual Level Social learning variables The first set of individual level variables is intended to measure the four main concept of social leaning theory. D ifferential association in the family and with peers is measured by differential each substance. Peer Association indicates that differential association with substance using and abstaining friends which was measured by two questio to report how often their close friends used each type of substance (seven point response type of substance (four created by summing the two variables specific to each substance. This summed scale is created because th e two items are highly correlated each other, the two items cannot be used as a separate measure in one analysis model at the same time due to the possibility of multicollinearity. Further, it is one of the common ways to create a summed measure to capture the concept of peer association. The resulting scales ranged from no friends who use the substance (0) to all close friends who used every day or nearly every day (9). The reliability of the scale is Chronbach alpha ( ) = 0.76 on alcohol, =0.77 on tobacc o, and =0.82 on depressants, respectively. This scale of proportion and frequency of substance using friends and all of the individual level independent variables are changed into standardized Z scores to facilitate comparison among the three sets of vari ables of their relative mediating effects. It is noteworthy that there are some methodological issues regarding the measures of concept of differential association with peers in testing social learning theory (Kurbin et al., 2009). One of the issues is tha t asking the individual to describe his or her perceptions of the attitudes or delinquency of his or her friends are not
91 peer approval or disapproval on delinquen cy (Kurbin et al., 2009; p.149). Although it is they perceive t o be the behavior of friends. Akers (1998) argues that while other direct appropriate in testing the concept of peer influence proposed in social learning (as wel l as the to test the SSSL model. Differential association in the fam ily was measured by each specific substance and D efinitions favorable and unfavorable to use of specific substance are measured by a aski ng attitudes approving or disapproving each of the three different types of (2) and dependent variables of definitions indicate greater probability of engaging substance use among adolescents as the SSSL theory predicts. D ifferential reinforcement is measured by the reinforcement balance of perceived rewards and costs that have resulted (if the respondent had ever used the substance) or which would result (if the respondent had never used the substance) from using the specific substance. This is a scale constructed by responses to two items: One asks the participants to
92 on a seven e of point for this variable. Therefore, i f the score is greater than 0, that indicates the participant perceived on balance that more rewards of using were greater than the costs for using the substance. If the value of this variable is less than 0, it means that the participant perceived more co sts than rewards from the using the substance. Therefore, a positive correlation with a dependent variable indicates support for the social learning theory in the expected direction. Differential reinforcement is also measured by one item asking about anticipated and one item asking about anticipated to the respondents )). These scales were redirected and, re measured on a five point response scale ranging from turning into authorities (1) to encouraging use (5). That is, a higher score of these measures mean greater rewarding reactions and a lower score means greater punishing reactions. Therefore, positive correlations with a dependent varia ble are in the direction expected by the SSSL theory. I mitation is an index created by summing four indicators measuring if participants observation of substance uses of a list of significant others or all the admired models have
93 influenced their substanc e behaviors (same age peers; parents or other adults whom the participants admire; admire models on TV and the moves or web sites; and advertisements). This variable is measured on a four (4). Measures of imitation were also redirected and re corded imitation measures with a dependent variable indicate greater exposure for substance use models whi ch proposed to lead to higher level of substance use among adolescents. This scale is also standardized as a Z score ( =0.95). Social bonding variables The second set of individual level variables is intended to measure the four main concepts of social b onding theory. All four of the social bonding measures are standardized as Z scores. A ttachment is operationalized by three items measuring parental attachment, peer a items summed into one scale and standardized as like the social learning variables ( =0.60). B elief is measured by a single item asking if respondents believe that they have moral duty to obey the law (from measured on a four point response scale and redirected as the social bonding theory predicted Commitment to school and future employment are used to meas ure this concept. The participants responded on a four corded two items
94 educational aspirations, and occupational aspirations were summed into one scale. ( =0.78). In volvement is measured by a single item asking the participants how much time they spent each day in studying after school. This item is measured on a six point response scale Self control variable : The concept of self control variable is measured by 12 items reflecting six components of what Gottfredson and Hirsc h considered core elements for the low self control concept. The six components of low self control include impulsivity, preference for simple tasks, risk seeking, preference for physical activities, self centeredness, and volatile temper. Each of the 12 i tems asked the participants to judge themselves in terms of possessing low self control. Each of the items is measured on a four point response scale scale indicati ng the overall low self control. The summed self control variable is also standardized as a Z score ( =0.68). Independent Variables: Structural Level Structural level variables include differential social organization, differential location in the social s tructure, theoretically defined social structural variables, and differential location in primary and secondary group. Although various types of structural variables are available, only limited number of structural level of predictors are used for this stu dy. That is because of a small number of district units (15 Gus) in the data. To conduct multilevel model analysis (HLM and HGLM), it is required to have a substantial number of cases for every variable included in the analysis model. In fact, Raudenbush a nd Bryk (2002) recommend at least 10 cases for every independent variable to develop models. In this dissertation, the student
95 accommodate a number of predictors. H owever, with just 15 Gus of district level units, the district level model can be limited to approximately no more than 2 independent variables, a limited number of st ructural variables drawn from South Korean census data and the BST. Although it might be appropriate to include only two structural variables, in the final model, se the three variables are all important to examine the propositions of the SSSL model appropriately. Further, to assess mediation effects of social learning, social bonding and self control variables, at least one or two structural level variables represe nting each dimensions of social structure described in the SSSL are necessary. Although percent of residents on public welfare is not a significant variable in the overall structural models, it must be included as an important criminogenic indicator of eco nomic disadvantage of districts. Additionally, due to the lack of significant variance across districts of the other structural variables (i.e., percent of foreigners and percent of young males) from census data, the three have been selected as the most ap propriate structural level variables in this study. Differential Social Organization Population density of the district is used as a measure differential social organization. Population density of each Gu, is operationalized by the number of population per area (Km) of each Gu as reported in the census data. Differential Location in the Social Structure. Sex (gender) is drawn from the self report data (female=0, male=1) variables (i.e., race, class, gende r in social structure (Akers, 1988: p.333). Therefore, although sociodemographic variables such as gender are often used in research simply as control variables and are social characteri stics of individuals, they are viewed in the SSSL model as indicators of the
96 hierarchical and horizontal location of individuals and groups in the overall social structure (Lanza Kaduce et al., 2006, p.129). Theoretically Defined Structural Variable Res idential Mobility and Percent of Residents on Public Welfare. These two variables are similar to measures that have been included in the literature on social disorganization, the first as an indicator of population instability and the second as an indicato r of relative economic disadvantage (level of poverty) of the community or neighborhood. These are not direct measures of social disorganization but rather indirect indicators that are commonly found in the literature. That is they are seen as antecedent c onditions of social disorganization as the breakdown in social control and social cohesion (Sampson et al., 1997) They will be considered in this dissertation research as measuring relative social disorganization among the Gu of Busan. Percent on public w elfare is a variable reflecting economical disadvantage in Busan, South Korea drawn from the BSY. It measures the proportion of residents who were on public welfare in each Gu in 1999 Residential mobility is measured by the percent of residents who move i nto and out of each Gu during the previous year. This measure was drawn from the South Korea census data. Differential Location in Primary and Secondary Groups Type of School and Religiosity Two variable s drawn from the self report data will be used to measure the immediate social context of location in different groups in society. Type of school is measured by two qualitatively different characteristics of the schools where the participants attended -either Liberal type or Industrial type 1) high school. Religiosit y is measured by a single item questioning how often the respondents participated in religious service. The response scale is a four
97 Each of these structura l variables is considered as likely to have an effect on deviant behavior based on prior criminological research as reviewed in literature review. Whether they will have an effect on adolescent substance use in this study remains to be seen. The SSSL model (Akers, 1998) proposes that if a structural variable is related to the deviant behavior of individuals then that relationship will be substantially mediated by social learning processes, but if there is no relationship, there is nothing to mediate and the model proposes that the structural variable also will not be related to the social learning variables. Hypotheses with adolescent substance use in Korea as the dependent variable. T he central proposition of SSSL that is tested here is that the effects of social structural variables on the dependent variables will be substantially mediated by the social learning variables. The second major purpose is to extend the test of the SSSL mod el by testing the proposition that the relationships between the structural variables and adolescent substance use will be mediated substantially more by social learning variables than by social psychological variables taken from control theory, both soci al bonding and self control variables. On the basis of these two main purposes and proposition, t here are several sub hypotheses developed as follows: Between District (Gu) Differences in Adolescent Substance Use Structural Level Hypothesis 1: Alcohol, depressant and tobacco use will significantly vary across districts (Gus). SSSL Measures Structural Level Hypothesis 2a: Social structural variables such as population density, residential mobility and percent of residents on public welfare will be asso depressant, and tobacco) use.
98 Hypothesis 2b: Male adolescents, students attending industrial type of school, and students with low level of religiosity (social structural variables) will be more likely to use al cohol, depressants and tobacco. Social Learning, Social Bonding, and Self Control Measures Individual Level Hypothesis 3 : Social learning, social bonding and self control variables will be associated obacco) use in the direction predicted by each of the theories. Mediation Hypothesis Hypothesis 4: The relationships between substance use and all of the social structural variables will be partially and substantially mediated by social leaning variables social bonding variables, and self control variables. Comparison Hypothesis Hypothesis 5: Social structural effects on substance use will be more substantially mediated by social learning variables than by social bonding and self control variables. Ana lysis Plan First, descriptive statistics from the observed measures are estimated ( Table 4 1 and Table 4 2). A set of bivariate correlations among dependent variables (alcohol use, depressant use, and tobacco use) and independent variables (structural vari ables, social learning variables, social bonding variables, and self control variable) are run to examine main effects of each independent variable on each dependent variable used to demonstrate the bivariate relationship between those measures (see Table 4 3). Second, multiple multilevel models are utilized for testing the hypotheses from the SSSL model with data nested in districts (Raudenbush, & Bryk, 200 2). Specifically, Hierarchical Linear Modeling (HLM) is utilized to examine the association between the structural level factors and students use taking into account the
99 impact of individual characteristics. Although the available dependent variables have been measured with 7 items Likert scales ( 0= 3= and these measures are not grouped into separate categories but summed as one dependent variable. Therefore, it still reflects a continuous measure the frequency of alcohol and depressant use of adolescents in their lifetime. In this case, if dependent variables are measured more than 7 or 8 item scales and the dependent variable is summed as one variable, HLM is still applicable for this type of dependent variable. With reg ard to the tobacco use models, Hierarchical Generalized Linear Modeling (HGLM) is utilized due to the distribution of the frequency of tobacco use (outcome variable) in this sample. Although the distribution of alcohol use and depressant use appear to be r of participants in this sample consists of either those who have ne ver used tobacco or those who use almost every day. Such a distribution of responses on the dependent variable violates the assumption of normality in HLM. Therefore, to obtain unbiased estimations for tobacco use, the response of tobacco use, which was me asured with 7 item Likert scales was summed into two categories (0 or 1), a dichotomous variable. After summing the variable as a binary times, HGLM is the most ap propriate form of statistical analysis. HGLM is quite similar to the linear hierarchical models (HLM) but differs in the distribution of the dependent variable and the use of the link function. The link functions transform the values of the dependent varia bles so that they adhere to linear model assumptions and the outcome is still values of the dependent variable (e.g., a HGLM model with a dichotomous outcome can only have a predicted value of 0 or 1).
100 Using this multilevel analytical strategy of HLM (an d HGLM) has benefits for this dissertation because the data collected by Dr. Hwang (2000) are nested within 26 different schools which are also nested in 1 5 different districts (Gus) It has been shown that if we use OLS regression with nested data, it ent ails two major problems (Raudenbush & Bryk, 2002). First, multilevel analysis using regression modeling risks violation of two regression assumptions, namely, the independence of error terms and homoscedasticity. When higher level variables are disaggregat ed to individual characteristics, individuals nested within groups will no longer represent independent observations because they share many of the same characteristics. This results in correlated error terms and biased estimate of standard errors. In addi tion, when the regression equation error terms are compared for individual and group effects the results between groups will be independent, but the results within groups will be perfectly correlated. Yet, HLM (and HGLM) is an enhancement to traditional re gression techniques for analyzing nested or multilevel data by taking structural level data into account and allows for variations within and across group level units. In other words, it corrects for these violations by including random components at all levels of the analysis. In addition, the binary correlation matrix presented here are two tailed tests which examined significance at 0.05 level or lower. However, all multilevel models analyzed in this dissertation were run on the basis of one tailed tes t which uses significance levels of 0.10, or lower. Although it seems to be inconsistent in using significance test level for different set of analyses, using a significance level of 0.10 for multilevel models in this dissertation is justifiable. That is b ecause it has been frequently shown that the majority of variations in dependent variables are from individual level independent variables, while community level influences have little or minor variances in dependent variables in multilevel analysis. Since in multi level analysis, finding significance community level influences is very important it is not unusual to adopt one tail significance test which using a significance level of 0.1
101 instead of 0.05 previous studies using cross level data and examinin g multilevel models (Ennett et al., 1997). Moreover, since the major purpose of this current dissertation is to examine mediation effects of the social learning process on the relationship between social structural influences and individual substance use b ehavior, it is justifiable to use a significance level of 0.1 for multi level models. In fact, it is consistent with previous studies because some of previous studies testing the SSSL model (Holland Davis, 2006; Verill, 2008) have adopted a significance le vel of 0.1 for their multilevel analyses. The current dissertation developed two level models consists of level 2 units (districts or Gus) in which the level 1 units (individuals) are nested. The units of observation in the first level are students that ar e grouped within the second level units, districts (Gus). Generally, a simple two level model equation begins with an equation at level l: Y ij 0j 1j (X ij ij. (1a) In equation (1a), i represents each individual and j represents each structural level unit. The error ij is the level 1 random effect. This equation produces the fixed effects estimates which has similar interpretation to the results of OLS regression analysis In addition to these individual level fixed effects, this two level modeling allows both the intercept and slope to vary across soci al units producing random effects estimates. The example of random effects equations at level 2 is: 0j 00 + 01 W j + u 0j (1b) 1j = 10 11 W j + u 1j (1c) Finally, the equation (1d) represents the combined model of level 1 and level 2 model: Y ij 00 + 01 W j + 10 X ij + 11 W j X ij + u 0j + u 1j X ij ij (1d) The error ij is the level 1 random effect and the error u 0j and u 1j X ij are level 2 random effects. The s parameters in the level 1 model are level 1 coefficients and the s are the level 2 coefficients (Raudenbush & Byrk, 2002:pp.19 23).
102 Model ing The procedures of HLM and HGLM c onsist of five stages of modeling: 1) specifying unconditional models at the level 2; 2) level 2 conditional models with structural level of predictors only; 3) level 1 conditional models with individual level of predictors included without consideration o f structural level of predictors; 4) the fully conditional level 1 and level 2 model; and finally, 5) the full comparison model including all level 1 and level 2 variables to compare the relative mediation effects among the three different set of social/ps ychological variables. The statistical significance of the effects of predictors on outcome variables as well as the between Gu variances are tested at each phase of analysis before proceeding to the next phase. To use multilevel modeling, it is necessary to consider how variables are centered. Centering subtracts the mean value of a variable from the value of each individual observation (Porter & Umbach, 2001). There are two centering methods, grand mean centering and group mean centering. First, grand me an centering subtracts the mean value of observation. On the other hand, group mean centering estimates the mean of a variable for all observations within the speci fic group or unit and subtracts it from that the value of the variable for each observation. For this dissertation, grand mean centering is used for all variables the all multi level analyses. Centering variables in the analyses facilitates the interpretat ion of the intercept in the model (Raudenbush & Bryk, 2002). Therefore, on the independent variable from the mean of that variable across the mean of all other stude nts in the sample. When grand mean centering is used, the intercepts in HLM equations represent the predicted score of an individual whose value for that independent variable is equal to the grand mean. Also, by using grand mean centering, the intercepts i n HGLM
103 equations represent the likelihood of use of an individual whose value for that independent variable is equal to the grand mean. Using the HLM 6.02 software package, this dissertation develops three separate sets of models for each of the dependen t variables alcohol use, depressant use and tobacco use (Bryk, Raudenbush, & Congdon, 1996). The symbolic representation of models will be presented to help understand the overall steps of analyses and the models that are used in this analysis. However, i t should be pointed out that the analysis can be done with any other multilevel or mixed model analysis software that has a capability of modeling multilevel analysis such as multi level analysis with STATA, GLLAMM program, SEM, MIXOR(Hedecker & Gibbosn, 1996), MLWin (Goldstein et al., 1998), and SAS (Kamata, 2002;p.30) HLM Models Unconditional (Random ANOVA) Models The first model of the analysis will be developed to estimate the amount of variation in average adolescent alcohol and depressant use amon g 1 5 different districts This unconditional one way random ANOVA model provides preliminary information about how much variation in the frequency of alcohol and depressant use within and between districts mple mean is reliable as an estimate of its true population mean. In this unconditional model, reliability is a function of sample size in each of the districts and intraclass correlation is the proportion of the total variance, which is between districts relative to the amount that is within districts (Raudentbush & Bryk, 2002; Hwang, 2006). Significant variation will be assessed by examining 0j the level 2 variance component. If there is no significant variation across the 1 5 districts found, it indica tes that I will not need
104 to add structural level variables to the level 2 model. To assess the magnitude of variation among districts (Gus) in the absence of predictor variables, it is possible to denote this model as follows; Level 1 model: Y ij 0j i j. (2a) Level 1 model: 0j 00 + u 0j (2b) The combined mode l: Y ij 00 + u 0j ij. (2c) At level 1, Y ij is the frequency of alcohol or depressant use of student i in district j, and the intercept 0j is the average of frequency of alco hol or depressant use of the jth Gus. At level 2, 00 Structural Level M odel s This model includes structural level variables only, while it does not contain any individual le vel variables such as social learning, social bonding and self control variables. In this model, only three district level variables (i.e., population density, percent on public welfare, and percent involved in residential mobility) are included because th e small number of districts limit the number of district level variables may be included. Also, according to the arguments of the SSSL model, three variables, which were drawn from questionnaire data, sex and religiosity, and from the school level informat ion on type of school which is coded as a school number for each individual respondent in each school, are used as structural level variables as explained above. Therefore, three structural predictors drawn from Busan census data (population density, perce nt of residents on public welfare, and percent population that have moved into or out of the district), and survey (sex, religiosity, and type of school) and can be modeled as follows: Level 1 model: Y ij 0j 9 j (Sex) + 10 j (Type of School) + 11 j (Religiosity) + ij (3a) Level 2 model: 0j = 00 01 (Population Density) 0 2 (Percent on Public
105 Welfare)+ 0 3 (Residential Mobility) 0j (3b) 9 j 9 0 9 j (3c) 10 j 10 0 10 j (3d) 11 j 11 0 11 j (3e) The combined model: Y ij = 00 01 (Population Density) 0 2 (Percent on Public Welfare)+ 0 3 Resi dential Mobility) + 9 0 (Sex) 10 0 (Type of School) 11 0 (Religiosity) + 0j + 9 j + 10 j 11 j + ij (3f) All predictors are grand mean centered. In terms of the symbols in this model, 0j indicates the mean of substanc e use for each Gu after addressing the effects of sex, type of school, and religiosity; 9 j is the change in substance use for being a male; 10 j is the change in alcohol and tobacco use for attending industrial schools; 11 j is the change in substance tha t is predicted by each unit of change in religiosity. Finally, 0j 9j 10 j and 11 j indicate that the intercept and the slopes are treated as percent who are resi 00 is the mean substance use for Gus at the grand mean for population density, percent on public welfare, and percent residential mobility within one year. 9 0 10 0 and 11 0 are the mea n of the Gu specific slopes between substance use and sex, type of school, and religiosity, respectively. Random Coefficients Models: Individual Level Predictor Models This is a random coefficient regression model in which level 1 predictors are entered into the model and allowed to vary across Gus. A random coefficient model examines the multivariate association between individual level variables and substance use and shows whether any of the individual level slopes vary significantly across Gus. This ra ndom coefficient regression model helps determine whether a slope is to be fixed within districts or
106 should be specified as random across districts depending on the significance of variance across districts (Hwang, 2006). If a district level slope varies a cross districts, the slope can be estimated using district level predictors (Rountree, Land, & Miethe, 1994). Social learning variables ( eight variables), social bonding variables (four variables) and self control variable (one variable which is a scale of several items), were included in this third stage of analysis respectively. However, in this model, no structural level predictor is included. Social learning model The first model is social learning only model. Th is model is written as ; Level 1 model: Y ij 0j 1 j (Peer Association) + 2 j 3 j + 4 j (Definitions) + 5 j (Reinforcement Balance) + 6 j 7 j Reactions) + 8 j (Imitation) + ij (4a) Level 2 model: 0j 00 0j (4b) 1 j 1 0 1 j (4c) 2 j 2 0 2 j (4d) 3 j 3 0 3 j (4e) 4 j 4 0 4 j (4f) 5 j 5 0 5 j (4g) 6 j 6 0 6 j (4h) 7 j 7 0 7 j (4i) 8 j 8 0 8 j (4j) The combined model: Y ij 00 1 0 (Peer Association) + 2 0 3 0 4 0 (Definitions) + 5 0 (Reinforcement Balance) + 6 0 7 0 eactions) + 8 0 (Imitation) 0j 1 j 2 j 3 j 4 j 5 j 6 j 7 j 8 j + ij (4k)
107 All district and individual level variables are grand mean centered. 0j represent the mean of subst ance use when the eight individual level social leaning variables equal to their respective grand means. 1 j 2 j 3 j 4 j 5 j 6 j 7 j and 8 j indicate the change in substance use predicted by each social learning variable, differential association wit 00 is the grand mean substance use, and 1 0 2 0 3 0 4 0 5 0 6 0 7 0 and 8 0 are the mean of the Gu specific slopes between alcohol or depressant use and differential association with friends, father use, mother use, definition, differential reinforcement of substance use, reactions from parents, reactions from friends, an d imitation, respectively. The same procedure is applicable for both the social bonding only model and the self control only model. Social bonding model The second model is social bonding only model. This model is basically in the same structure with soci al learning model, but the variables used in this model are social bonding variables only. Th is model is written as ; Level 1 model: Y ij 0j 12 j (Attachment) + 13 j (Belief)+ 14 j (Commitment) + 15 j (Involvement) + ij (5a) Level 2 model: 0 j 00 0j (5b) 12 j 1 0 12 j (5c) 13 j 13 0 13 j (5d) 14 j 14 0 14 j (5e) 15 j 15 0 15 j (5f) The combined model:
108 Y ij 00 12 0 (Attachment) + 13 0 (Belief) + 14 0 (Commitment) + 15 0 (Involvement) 0j + 12 j 13 j 14 j 15 j + ij (5g) Self control model The third model is self control only model. This model is basically in the same structure with social learning and social bonding model, but the variables used in this model is self control variable only. Th is model is written as ; Level 1 mo del: Y ij 0j 16 j (Self Control) + ij (6a) Level 2 model: 0j 00 0j (6b) 16 j 16 0 16 j (6c) Th e combined model: Y ij 00 16 0 (Self Control) 0j 16 j + ij (6d) F ull M odel s The SSSL model The full SSSL model include s both the individua l level (social learning variables) and the structural level variables can be denoted as follows; Level 1 model: Y ij 0j 1 j (Peer Association) + 2 j 3 j 4 j (Definitions) + 5 j (Reinforcement Balance) + 6 j eactions) + 7 j Reactions) + 8 j (Imitation) + 9 j (Sex) + 10 j (Type of School) + 11 j (Religiosity) + ij (7a) 1 j 1 0 1 j (7b) 2 j 2 0 2 j (7c) 3 j 3 0 3 j (7d) 4 j 4 0 4 j (7e) 5 j 5 0 5 j (7f) 6 j 6 0 6 j (7g)
109 7 j 7 0 7 j (7h) 8 j 8 0 8 j (7i) Level 2 model: 0j = 00 01 (Population Density) 0 2 (Pe rcent on Public Welfare) + 0 3 (Residential Mobility) 0j (7j) 9 j 9 0 9 j (7k) 10 j 10 0 10 j (7l) 11 j 11 0 11 j (7m) The combined model: Y ij = 00 01 (Population Density) 0 2 (Percent on Public Welfare) + 0 3 (Residential Mobility) + 1 0 (Peer Association) + 2 0 3 0 4 0 (Definitions) + 5 0 (Reinf orcement Balance) + 6 0 7 0 8 0 (Imitation) 9 0 (Sex) 10 0 (Type of School) 11 0 (Religiosity) + 0j 1 j 2 j 3 j 4 j + 5 j 6 j 7 j 8 j + 9 j + 10 j 11 j + ij (7n) Social structure and social bonding model The social structure and social bonding model include s both the individual level (social bonding variables) and the str uctural level variables can be denoted as follows; Level 1 model: Y ij = 0j + 9 j (Sex) + 10 j (Type of School) + 11 j (Religiosity) + 12 j (Attachment) + 13 j (Belief) + 14 j (Commitment) + 15 j (Involvement) + ij (8a) 12 j 12 0 12 j (8b) 13 j 13 0 13 j (8c) 14 j 14 0 14 j (8d) 1 5 j 16 0 16 j (8e) Level 2 model: 0j= 00 01 (Population Density) 0 2 (Percent on Public Welfare) + 0 3 (Residential Mobility) 0j (8f)
110 9 j 9 0 9 j (8g) 10 j 10 0 10 j (8h) 11 j 11 0 11 j (8i) The combined model: Y ij = 00 01 (Population Density) 0 2 (Percent on Public Welfare) + 0 3 (Residential Mobility) + 9 0 (Sex) 10 0 (Type of School) 11 0 (Religiosity) )+ 0j + 9 j + 10 j 11 j + 12 j + 13 j + 14 j + 15 j + ij (8j) Social structure and self control model The social structure and se lf control model contains both the individual level (self control variable) and the structural level variables can be denoted as follows; Level 1 model: Y ij 0j 9 j (Sex) + 10 j (Type of School) + 11 j (Religiosity) + 16 j (Self Control) + ij (9a) 161 j 1 6 0 16 j (9b) Level 2 model: 0j = 00 01 (Population Density) 0 2 (Percent on Publi c Welfare) + 0 3 (Residential Mobility) 0j (9c) 9 j 9 0 9 j (9d) 10 j 10 0 10 j (9e) 11 j 11 0 11 j (9f) The combined model: Y ij = 00 01 (Population Density) 0 2 (Percent on Public Welfare) + 0 3 (Residential Mobility) 9 0 (Sex) 10 0 (Type of School) 11 0 (Religiosity) + 0j + 9 j + 10 j 11 j + 16 j + ij (9g) Comparison Models This full comparison model for alcohol and depressant use includes all of the social structural variables as well as social learning, social bonding and self control variables. This model intends to examine relati ve mediation impact of these individual level variables from
111 the three different theories of the social structural variables on alcohol and depressant use. It is expected that variables with relatively stronger mediation effect would remain significant wit h greater regression coefficients compared to variables having weaker mediation effects. The model can be written as follows: Level 1 model: Y ij = 0j 1 j (Peer Association)+ 2 j 3 j 4 j (Definitions)+ 5 j (Reinforcement Balan ce) 6 j 7 j + 8 j (Imitation) + 9 j (Sex) + 10 j (Type of School) + 11 j (Religiosity) + 12 j (Attachment) + 13 j (Belief) + 14 j (Commitment) + 15 j (Involvement)+ 16 j (Self Control) + ij (10a) 1 j 1 0 1 j (10b) 2 j 2 0 2 j (10c) 3 j 3 0 3 j (10d) 4 j 4 0 4 j (10e) 5 j 5 0 5 j (10f) 6 j 6 0 6 j (10g) 7 j 7 0 7 j (10h) 8 j 8 0 8 j (10i) 12 j 12 0 12 j (10j) 13 j 13 0 13 j (10k) 14 j 14 0 + 14 j (10l) 15 j 15 0 15 j (10m) 161 j 16 0 16 j (10n) Level 2 model: 0j = 00 01 (Population Density) 0 2 (Percent on Public Welfare)+ 0 3 (Residential Mobility) 0j (10o) 9 j 9 0 9 j (10p)
112 10 j 10 0 10 j (10q) 11 j 11 0 11 j (10r) The combined model: Y ij = 00 01 (Population Density) 0 2 (Percent on Public Welfare) + 0 3 (Residential Mobility) + 1 0 (Peer Associatio n) + 2 0 3 0 4 0 (Definitions) + 5 0 (Reinforcement Balance) + 6 0 7 0 8 0 (Imitation) 9 0 (Sex) 10 0 (Type of School) 11 0 (Religiosity) + 12 0 (Attachment)+ 13 0 (Belief )+ 14 0 (Commitment)+ 15 0 (Involvement)+ 16 0 (Self Control)+ 0j 1 j 2 j 3 j + 4 j 5 j 6 j 7 j 8 j + 9 j + 10 j 11 j + 12 j + 13 j + 14 j + 15 j + 16 j + ij (10s) HGLM Models While a standard HLM model uses a normally distributed dependent variable and an identity link function, the binary outcome model uses a binomial sampling model and a logit link (Raudenbush & Bryk, 2002; p.295). In this dissertati on, the sampling model of HGLM is Bernoulli. It is a binary model used if the outcome is only measured once and the predicted value is the probability of success. For example, it estimates if a student use tobacco or not in their life time and there is onl y one outcome per individual. As explained earlier in this chapter, the distribution of the frequency of tobacco use is such that it requires making it into a dichotomously measured variable. The participants who responded never used tobacco remained as a Also, the link function of the HGLM model is a logit link function. That is, ij = log ij ) (1 ij ) (11a) ij is the log of the odds of success. If the probability of success is ij the odds of success is ij /(1 ij ) = 0.5/0.5 =1.0and the log
113 probability of success is less than 0.5, the odds are less than 1.0 and the logit is negative. When the probability is greater than 0.5, the odds are greater than 1.0 and the logit is positiv e. The structural model of HGLM predicting which is a predicted log odds. The log odds can ij ). Also, the predicted log odds can be converted to a predicted probability (Raudenbush & Bryk, 2002; p.295). The equation converting predicted logs odds to pre dicted probability is ij = 1/[1 +exp( ij )]. The procedure of HGLM modeling is similar to the HLM models explained above. Also, all independent variable are grand mean centered. Therefore, the five stages of modeling for the tobacco use models are discus sed briefly. Unconditional (Random ANOVA) Model To justify the use of HGLM model, tobacco use must vary across districts (Gus). In this estimate if the average lik elihood of tobacco use vary across Gus, the first HGLM model is the random unconditional which include no predictors at either level. Given the Beroulli sampling model and a logit link function, the level 1 model is as following, Level 1 model: Prob ( ij = 1 | i j ) = ij Log [ ij /(1 ij ) ] = 0j (12a) Level 1 model: 0j 00 + u 0j (12b) ij 00 + u 0j ij. (12c) At level ij is the log odds of tobacco use of student i in district j, and the intercept 0j is the expected average log odds of tobacco use of the jth Gus. At level 2, 00 represents the average log 2 variance, the interclass correlation for linear models does not applicable for nonlinear models due to the heteroscedastic nature of level 1 variance.
114 Structural Level M odel As like the HLM structural level model, this model includes structural level variables only (population dentistry, poverty, and mobility), while it does not contain any individual level variables. However, sex, type of school, and religiosity are treated as structural level predictors as suggested in the SSSL theory. The structural HGLM model is as follows: Level ij = 1 | i j ) = ij Log [ ij /(1 ij ) ] = 0j 9 j (Sex) + 10 j (Type of School) + 11 j (Religiosity) + ij (13a) Level 2 model: 0j = 00 01 (Population Density) 0 2 (Percent on Public Welfare)+ 0 3 (Residentia l Mobility) 0j (13b) 9 j 9 0 9 j (13c) 10 j 10 0 10 j (13d) 11 j 11 0 11 j (13e) The combined model: ij = 00 01 (Populati on Density) 0 2 (Percent on Public Welfare)+ 0 3 (Residential Mobility) 9 0 (Sex) 10 0 (Type of School) 11 0 (Religiosity) + 0j + 9 j + 10 j 11 j + ij (1 3f) All of predictors were grand mean centered. The paramete r in these equations describes the expected log odds of tobacco use at Gu j for the various structural independent variables. 0j effects of sex, type of school, and rel igiosity. In this model, not only intercept but also all independent variables are allowed be vary across Gus. Each s refer to the log odds of each s will be converted to odds ratio in order to improve interpretability. In this dissertation, odds ratios designate the change in unit change in a certain independent variable when holding constant other var iable in the model (Hedeker & Gibbons, 1996). Odds ratios greater than one indicate an increase in
115 using tobacco (Hedeker & Gibbons, 1996). Finally, 0j 4 j 5 j and 6 j indicate that the intercept and the slopes are treated as random coefficients. Random Coefficients Models: Individual Level Predictor Models It includes level 1 predictors only and allows these variables to be varied across Gus. As l ike the three HLM conditional models which contain separate groups of i ndividual level predictors only, there are three HGLM conditional models at level 1. Th is example model includes the social learning variables only can be written as ; Level 1 model: Pro b ( ij = 1 | i j ) = ij Log [ ij /(1 ij ) ] 0j 1 j (Peer Association)+ 2 j 3 j 4 j (Definitions)+ 5 j (Reinforcement Balance) 6 j 7 j + 8 j (Imitation) + ij (14a) Level 2 model: 0j 00 0j (14b) 1 j 1 0 1 j (14c) 2 j 2 0 2 j (14d) 3 j 3 0 3 j (14e) 4 j 4 0 4 j (14f) 5 j 5 0 5 j (14g) 6 j 6 0 6 j (14h) 7 j 7 0 7 j (14i) 8 j 8 0 8 j (14j) The combined model: ij 00 1 0 (Peer Association) + 2 0 3 0 4 0 (Definitions) + 5 0 (Reinforceme nt Balance) + 6 0 7 0
116 8 0 (Imitation) 0j 1 j 2 j 3 j 4 j 5 j 6 j 7 j 8 j + ij (14k) All individual level variables are grand mean centere d and 0j represent the average likelihood of tobacco use for each Gu, after controlling each set of individual level social leaning variables. 00 is the average likelihood of using tobacco, while 1 0 2 0 3 0 4 0 5 0 6 0 7 0 and 8 0 represent the mea n of the Gu specific slopes between tobacco use and each independent variables at level 1. The same procedure is applicable for social bonding model and self control model. The structure of the social bonding and self control models are consistent with the social learning model described above, but the variables used in the each of the models. For the social bonding model, social bonding variables are only used and for the self control model, the only variable used in the model is self control variable. The refore, F ull M odel s The full model include s both the individual level and the structural level variables The following model is an example of the SSSL model with the social learning variable s. The model can be described as follows; Level ij = 1 | i j ) = ij Log [ ij /(1 ij ) ] 0j 1 j (Peer Association)+ 2 j 3 j 4 j (Definitions)+ 5 j (Reinforcement Balance) 6 j 7 j + 8 j (Imitation) + 9 j (Sex) + 10 j (Type of School) + 11 j (Religiosity) + ij (15a) 1 j 1 0 1 j (15b) 2 j 2 0 2 j (15c) 3 j 3 0 3 j (15d) 4 j 4 0 4 j (15e) 5 j 5 0 5 j (15f)
117 6 j 6 0 6 j (15g) 7 j 7 0 7 j (15h) 8 j 8 0 8 j (15i) Level 2 model: 0j 00 01 (Population Density) 0 2 (Percent on Public Welfare)+ 0 3 (Residential Mobil ity) 0j (15i) 9 j 9 0 9 j (15k) 10 j 10 0 10 j (15l) 11 j 11 0 + 11 j (15m) The combined model: ij = 00 01 (Population Density) 0 2 (Percent on Publi c Welfare)+ 0 3 (Residential Mobility) + 1 0 (Peer Association) + 2 0 3 0 4 0 (Definitions) + 5 0 (Reinforcement Balance) + 6 0 7 0 8 0 (Imitation) 9 0 (Sex) 10 0 (Type of Sc hool) 11 0 (Religiosity) + 0j 1 j 2 j 3 j + 4 j 5 j 6 j 7 j 8 j + 9 j + 10 j 11 j + ij (15n) Comparison Model As like the HLM full comparison model, the final HGLM compar ison model also includes all of structural level predictors and individual level predictors. The model intended to compare the relative mediation effects of the three different set of individual level variables. The model can be denotes as follows: Level 1 ij = 1 | i j ) = ij Log [ ij /(1 ij ) ] 0j 1 j (Peer Association)+ 2 j 3 j 4 j (Definitions)+ 5 j (Reinforcement Balance) 6 j 7 j Reactions)+ 8 j (Imitation)+ 9 j (Sex)+ 10 j (Type o f School)+ 11 j (Religiosity)+ 12 j (Attachment) + 13 j (Belief)+ 14 j (Commitment)+ 15 j (Involvement)+ 16 j (Self Control)+ ij (16a)
118 1 j 1 0 1 j (16b) 2 j 2 0 2 j (16c) 3 j 3 0 3 j (16 d) 4 j 4 0 4 j (16e) 5 j 5 0 5 j (16f) 6 j 6 0 6 j (16g) 7 j 7 0 7 j (16h) 8 j 8 0 8 j (16i) 12 j 12 0 12 j (16j) 13 j 13 0 13 j (16k) 9 j 14 0 14 j (16l) 10 j 15 0 15 j (16m) 11 j 16 0 16 j (16n) Level 2 model: 0j 00 01 ( Population Density) 0 2 (Percent on Public Welfare)+ 0 3 (Residential Mobility) 0j (16o) 9 j 9 0 9 j (16p) 10 j 10 0 10 j (16q) 11 j 11 0 11 j (16r) The combined model: ij = 00 01 (Population Density) 0 2 (Percent on Public Welfare)+ 0 3 (Residential Mobility) + 1 0 (Peer Association)+ 2 0 3 0 4 0 (Definitions) + 5 0 (Reinforcement Balance ) + 6 0 7 0 8 0 (Imitation) 9 0 (Sex) 10 0 (Type of School) 11 0 (Religiosity) + 12 0 (Attachment)+ 13 0 (Belief)+ 14 0 (Commitment)+ 15 0 (Involvement)+ 16 0 (Self Control) 0j 1 j 2 j 3 j
119 4 j 5 j 6 j 7 j 8 j + 9 j + 10 j 11 j + 12 j + 13 j + 14 j + 15 j + 16 j + ij (16s) Data L imitations The limitation of the study concern one issue: suitability of these data for te sting multilevel hypotheses Although the data used in this dissertation are ideally suited for testing social learning theory and are also well suited to test social bonding theory of substance use they do not contain many of structural level of measures that will be used in estimating structural variation of substance and drug use among students. The individual level data contain a limited number of social structural measures, mainly the socio demographic variables such as gender, family structure, SES, and age. The current study utilize s measures from census data in B usan, South Korea, and as noted these correspond exactly to the residential areas in which the students are located. However, the district level data set in B usan is limited in size (1 5 Gu) compared to the individual data set. Accordingly, the limited district sample size reduces the number of predictors of between school variation and random slope coefficients the level 2 model can support. Although the issue of suitability of the data set i s important, it is n either an unusual limitation of data in utilizing level 1 and level 2 data sets in combined models nor does it completely prohibit conducting multilevel of analysis. First, f or example, Holland Davis (2006) had the same issue in her di ssertation. According to Holland Davis (2006), her dissertation corrected this issue by including only one or two predictor for each structural dimension in the structural and full models (p.67). Therefore, this study use d the same analytic strategy to ove rcome the data limitation. Second, having 1 5 districts as level 2 sample sites is a feasible sample size to conduct multilevel of analysis as long as there is an appropriate number in the sample at the level 1 per site. There are several situational studi es examining how to choose the optimal sample size at the macro and at the micro level to ensure a desired level of power given a relevant (hypothesized) effect size and a chosen significance level ( Mass & Hox, 2004 ). With regard
12 0 to multilevel analysis, the primary concern is sample size at the level 2 because the sample size at the group or district level is smaller than the sample size at the level 1. In general, therefore, it is well known that if the goal of a study is to maximize power in testing the average effect of treatment, then the larger the variation in the treatment impact across sites the more sites (level 2 sample) are needed to attain adequate power (Mass & Hox, 2004). 000) report that with as few as six to twelve groups, Restricted ML (RLM) estimation provides reasonable variance estimates and, with 48 groups, both RML and Full Information ML (FML) p. 128). Furthermore, Raudenbush & Liu (2000) suggest that the temptation simply to maximize sample size at level 2 in designing a study must be tempered by the relative cost ratio, the cost of sampling sites relative to the cost of sampling participan ts within sites. Accordingly, Raudenbush and Liu (2000) introduce an equation used to choose optimal sample sizes at the level 1 and level 2 along with estimated power for varying values of the cost ratio and the variance of the treatment effect in designi ng a randomized study 1 In their simulation analysis, they found that if a study has 17 groups or primary site unit at level 2, its optional sample size per site is 20 at level 1, which would have an effect of site covariate ranging from 0.2 to 0.6 and pow er for site covariate effect ranged .109 to .561 (see Raudenbush & Liu, 2000, p. 209). Considering the moderate effect size of a study to detect average impact of a treatment across varied settings and moderating effect of a site 1 known formula for the optimal cluster si ze in a two stage cluster sample (Cochran, 1977) and the optimal sample size per cluster in a cluster examining optimal sample size (cf. Allison et al., 1997; Overall & Dala, 1965; Waters & Chester, 1987; Raudenbush, 1997).
121 characteristic 2 this stud y has data from 1 5 Gus, with data from at least 40 individuals per district. 2 Cohen viewed standard effect sizes of .2 and .5 and .8 as small, medium, and large, respect ively. Raudenbush & Liu also suggest rules of thumbs for variances of the treatment effects of .05, .10, and .15 as small, medium and large variances (Raudenbush & Liu, 2000, pp.203 204).
122 CHATPER 4 RESULTS OF ANALYSES To begin, descriptive statistics for all explanatory and dependent variables are presented. Also, three bivariate correlation analyses between th e explanatory variables and each type of substance use are done. Finally, multilevel analyses proceed with the estimation of a series of hierarchical linear models for alcohol and depressant use and hierarchical generalized linear models for tobacco use. Descriptive Statistics Table 4 across Gu (district). The percent for each Gu indicates the proportion of students who have ever used each type of substanc. In Table 4 1, Sa Sang G u shows the highest percentage of students who used alcohol. Among 39 respondents, 38 students reported that they ever used use (50%). With regard to depressant use, Sooyoung Gu has the highest proportion of depressant use (84%), while Seu Gu has the lowest proportion of students who had used depressant (25%). However, Seu Gu is the district where the highest proportion of students reported tobacco use (77%). Consiste nt with the alcohol use distributions, Saha Gu has the low proportions of students using alcohol and tobacco, but it has the third highest proportion depressant use (74%). In contrast, Sa Sang Gu has the highest proportion of ed to the sample characteristics in each Gu. In Saha Gu, the sample consists of female students attending liberal school only, while there are only male respondents attending an industrial school in Sa Sang Gu. Therefore, it indicates gender and type of sc hool may influence
123 substance use differently. Overall, 84% of respondents reported alcohol use, 55% reported depressant use, and 46% reported tobacco use. Table 4 2 presents descriptive statistics for the students and district characteristics (see also, H wang 2006, p.137). In Table 4 2, all of the explanatory variables, individual level and structural level variables, as well as three dependent variables (i.e., the frequency of alcohol and depressant use and tobacco use (yes=1)) are included to demonstrate the variable characteristics. The mean level of alcohol use for the students is 2.17 on a seven point scale of use which indicates the frequency of alcohol use by students in this sample is skewed toward the lower end, an average of using several times in their life time. The mean frequency of depressant use is even lower at 1.22, indicating that students in this sample report that their lifetime use of depressants is once or twice time on average. However, there is enough dispersion across the frequency s cale to allow using these variables as ordinal level data. As noted, this was not true of tobacco use, and that measure was dichotomized. The mean of the binary tobacco use variable indicates that about 46% of students report that they have ever used tobac co in their life time. In this sample, 52% of the participants are male students while 48% of students are female; 53 % of the participants were attending liberal type of high school at the time of the survey. The participants report that they attend relig ious service, on average, about once or twice a month. With regard to the structural level characteristics, the mean population density across the 15 Gus in Busan is 11472.67 per kilometer which ranged from 367 to 21318. The average proportion of residents on public welfare is 6.45%. That indicates on average 6.45% of residents of each 15 Gus in Busan were under the poverty level and therefore were receiving public welfare. The range of the proportion of welfare recipients in districts are from 1.42% to 12. 32 %. Finally, residential mobility measured by the average proportion of
124 residents in 15 districts in Busan moved in and out of each district during the previous year of 1999, the year when these data were collected. Bivariate Correlations Table 4 3 reports the zero order correlations of each measure of substance use with social structure, social learning, social bonding, and self control variables. For alcohol use, seve nteen out of nineteen variables the bivariate correlations are statistically significant. All of the individual level variables are statistically significant predictors for alcohol use. Four of the social structure and community level predictors, sex, type of school, religiosity, and population density have significant bivarite relationships with alcohol use. The percent of the population on public welfare and percent who have moved in the past year are not significant predictors of the level of use of alco hol. With regard to depressant use, for eleven out of nineteen variables bivariate correlations are statistically significant. Unlike the bivariate correlations between the explanatory variables and alcohol use, none of the social bonding variables are si gnificantly correlated with depressant use. Also, only two of the social structure and community level predictors, sex and residential mobility have significant bivariate relationships with depressant use. Finally, sixteen of nineteen variables are statis tically significant bivariate correlates of tobacco use. At the individual level, all of the explanatory variables, except for the measure of commitment (social bonding theory) are significantly correlated with tobacco use at the bivariate level. At the so cial structural level, four variables, sex, type of school, population density and percent on public welfare are significant bivariate correlates of tobacco use. Combined the social learning variables are strongly correlated with each of the dependent vari ables.
125 Each of the social learning variables is at least moderately correlated, with fairly strong correlations of peer association, imitation, and differential reinforcement variables, with the three substances (alcohol, depressant, and tobacco). Social bonding and self control variables tend to be more weakly correlated with use of the three substances, but some are quite substantially correlated with the measures of substance use. Overall, social learning, social bonding, and self control variables are all statistically significantly correlated with supports Hypothesis 3. In addition, gender, type of school, and population density show noticeable effects with religiosi ty, welfare, and mobility having low to negligible effects, on the three substances. All correlations are in the hypothesized direction. The correlation matrix for all independent and dependent variables are reported in Table A 1, A 2, and A 3 (see Appendi x ). Hierarchical Linear Model Analyses: Alcohol and Depressant Use The core proposition of the Social Structure Social Learning model is that social learning variables substantially mediate the effects of any structural influences on delinquent or devian t behaviors. To test the arguments of the SSSL model, several hierarchical linear regression models were analyzed for both alcohol use and depressant use behaviors among South Korean youths. The results are presented in this chapter. Due to the complexity of the models, the interpretations of a set of HLM models are presented separately by type of substance, alcohol and depressant, respectively. Here the two dependent variables, frequency of alcohol and depressant use are treated as continuous measures. Al cohol Use Unconditional (Random ANOVA) Model This first model intended to examine whether the frequency of alcohol use of students are significantly different across districts (Gus). This finding that there is significant
126 variations in alcohol use among th The results of the unconditional one way random ANOVA model are presented in Table 4 4. In this fully unconditional model, the grand mean alcohol use is 2.14. The coefficient representing the amount of variation in the mean frequency of alcohol use across the 15 districts (Gus) is 00 =0.218 which is statistically significant, In addition, the variation at the individual level is 2 = 0. 91. The interaclass correlation coefficients estimated by within ( 2 ) and between ( 00 ) variations ( = 00 / ( 00 + 2 ) = 0.21897/ (0.21897+ 2.20539) =0.09), shows a portion of between group variance in the total variance (Duncan & Raudenbush, 1999). Approximately 9percent of the variation in alcohol use comes from di fferences between Gus. Although this between group variation is small, it is still reasonably acceptable compared to other studies that have found small variance between macro level units ranging from around 5 to10 percent (Reisig & Parks, 2000; Sampson & Bartusch, 1998; Sampson, Morenoff, & Earls, 1999). The district level reliability (0.844) indicates that the sample mean was a reliable measure of the true district mean for alcohol use. If reliability is close to 1, the group means, 0j vary substantially across level 2 units holding constant the sample size per group (Raudenbush & Bryk, 2002, p.257). This reliability indicates that district level differences can be modeled with a reasonable degree of precision, another encouraging r esult for multilevel analyses along with the interclass correlation. Structural Level Model Table 4 5 shows the findings from the structural model analysis. In this model, structural level predictors, population density, percent of residents on public wel fare, percent of residential mobility during the past year, sex, type of school, and religiosity are introduced. Although three of the structural level factors, sex, type of school, and religiosity are drawn from individual level of data, Akers (1998, Lee et al, 2004) defines the three factors as
127 indicators of social structural dimensions in the SSSL model. Sex is an indicator for differential locations in the social structure, and type of school and religiosity are indicators for differential location in p rimary and secondary groups. Sex, type of school, and religiosity may vary across districts as well as population density and unemployment do. However, they are measured by responses to the questionnaire not by census data. This model reveals that all of t hese predictors but percent of residents on public welfare are significantly associated with the frequency of alcohol use. Population density ( 01 = 0.00004) and sex ( 90 = 0.47) have positive and significant relationships with alcohol use, while residenti al mobility ( 03 = 0.08), type of school ( 100 = 0.76) and religiosity ( 110 = 0.10) have negative and significant relationship with alcohol use. The findings indicate that, on average, the more densely populated the district in which respondents reside the less frequently resident moving in and out of the district during the past year. This findings support Hypothesis 2a. Further, being male and attending industrial school are associated with increased frequency of alcohol use within Gus (districts). A nd the less respondents participate in religious services the greater the frequency of alcohol use in that district. The direction of each of the coefficients is consistent with the Hypothesis 2b. With these variables in the model, the between Gus v ariance in average alcohol use is 0.0 6 and it is not significant. The variance of the district specific slopes for sex is the only significant variance component. That is, in Gus with low average alcohol use the relationship between sex and alcohol use is strong. In fact, when a structural level model only contains the three community level predictors, population density, percent of residents on public welfare and percent of mobility, the R 2 variance explained based on the level 2 predictor is [ = ( 00 (unconditional model) 00(the intercept only structural model) )/ 00(unconditional model) = [(0.21897 0.08036)/ 0.21897=0.633009] and remains significant. Thus, including structural variables explains approximately 62% of the between Gu varianc e in alcohol use. However, after taking into
128 account the three predictors, sex, type of school, and religiosity, the between variance at district level became insignificant. This findings suggest that meso level social structural factors such as gender, (d ifferential location in the social structure), type of school, and religiosity (differential location in primary and secondary groups) explain greater proportion p opulation density (differential social organization) and percent of residents on public welfare, and percent of residential mobility (theoretically defined structural variables). Random Coefficient Model Social learning model Prior to modeling the effec ts of all the variables on alcohol use, to test the separate effects of each set of social psychological variables measuring concepts taken from social learning, bonding, and self control theories respectively are run. These are followed by models examinin g the net effects of structural variables and each set of the social psychological variables. I begin with a random coefficient regression model containing social learning variables only shown in Table 4 6. All social learning variables are grand mean cent ered, centered around the mean of districts (Gus). The expected frequency of alcohol use among students when each of social learning variables are entered is 2.17. The findings in the hen students: 1) have greater proportion of peers who use alcohol ( 10 = 0.97 *** ); 2) have fathers who use alcohol more frequently( 20 = 0.08 ); 3) have definitions favorable toward use of alcohol( 40 = 0.08 ); 4) perceive that on balance overall effects of u sing alcohol are good( 50 = 0.11 ** ); 5) perceive that parental reactions to their alcohol use are not discouraging ( 70 = 0.08 ); and 6) have a greater chance of imitating use of alcohol by behavioral models ( 80 = 0.2 0 *** ). The directions of the effects of a ll of the variables are consistent with the social learning hypotheses (Hypohtesis3).
129 The between district variance ( 00 ) in average alcohol use is 0.002 but it is not significant with eight social learning variables in the model. In terms of the Gu spec ific slopes variance components of the social learning variables, only the two slopes of differential association with friends and imitation are significant. The other six social learning variable slopes variance components are not significant, indicating that the relationships between these variables and individual alcohol use do not significantly differ between Gus. However, the between Gu variance is not significant; the significance of each slope may not contain importance and need not be interpreted in this model. Within Gus, 2 based on the level 1 predictors [(2.20539 0.84746)/ 2.20539 =0.615732] is 0.62. This tells us that social learning variables account for about 62% of the variance in student level frequency o f alcohol use. The within learning variables explain around 60 % of the variation in alcohol use. Social bonding model This model includes only a set of social b onding variables without containing any other variables. The findings are presented in Table 4 7 and only partially support Hypothesis 3 for social bonding variables. All of the social bonding variables are grand mean centered. Averaged across Gus, there i s a significant relationship between belief and involvement and 130 = 0.09 150 = 0.29 *** respectively). These findings suggest that having higher level of approval for general law and conforming belief and spending more time for study are negatively associated with alcohol use. The between district (Gus) varia nce in mean alcohol use is 0.11 and is significant with the four social bonding variables in the model. However, only one of the Gu level slopes variance components of the social bonding variables, attachment, is significant ( 120 = 0.02 *) That indicates the relationship between attachment and alcohol use is stronger in Gus with
130 low mean alcohol use, while the relationships between the three other social bonding variables and individual alcohol use do not significantly differ between Gus. Within Gus, the R 2 of this model [ = ( 2 (unconditional model) 2 (the social bonding model) )/ 2 (unconditional model) = (2.20539 2.04967)/ 2.20539 =0.070608] is 0.07. That indicates that 7% of the variance in student level frequency of alcohol use is explained by social bonding variabl es. Self control model This model includes only the self control variable. Table 4 8 shows the findings of the multilevel test of self control theory. The self control variable is grand mean centered. Averaged across Gus, the self control variable, as exp ected by the theory, is negatively associated with alcohol use, indicating having lower level of self control is associated with increased alcohol use ( 160 = 0.43 *** ).This finding support Hypothesis 3 for self control. The between Gu variance ( 00 ) in mean alcohol use is 0.14 and is significant. Within Gus, 2 of this model [ = ( 2 (unconditional model) 2 (the self control mod el) )/ 2 (unconditional model) = (2.20539 2.01097)/ 2.20539 =0.088156] is about 0.09. That means the measures of the self control construct explain approximately 9% of the variance of student level alcohol use. The between district variance in the Gu speci fic slopes of self control is not significant, indicating that there is no statistically significant difference among Gus in the relationship between the self control variable and alcohol use. Models with Structural Variables and Variables from the Three Theories The social structure social learning (SSSL) model The full SSSL alcohol model includes all predictors at the structural level and all of the social learning variables to assess the central hypothesis of the SSSL that the social learning variables will mediate the effects of social structure on alcohol use (Hypothesis 4). The findings of the model presented in Table 4 9 are consistent with the SSSL model predictions.
131 At the structural level, four of the structural level factors, which were significa nt in the structural model (Table 4 5), population density, percent of residential mobility, type of school and religiosity, became insignificant once the social learning variables are taken accounted for the model. Also, the regression coefficients of the structural level factors are reduced compared to the structural model ( 01 = 0.000004, 03 = 0.0 1, 100 = 0.1 1, 110 = 0.05, respectively). The social learning variables, however, remain significant with similar magnitudes of effects, consistent with t hose in the social learning only model (Table 4 6). That indicates social structural measures are substantially mediated by the inclusion of the social learning variables as the SSSL model predicted. This finding supports Hypothesis 4 for the mediation eff ects of social learning variables. The net effect of sex on alcohol use in the SSSL model is significant, but is also substantially reduced by the inclusion of the social learning variables (by more than half) from its effect shown in Table 4 5 (from 90 = 0.4 7 ** to 90 = 0.18 ) Here it is noteworthy to compare the findings from Holland In her dissertation, she found that percent on welfare which was measured in the same way with the percent of residents on public welfare as in this dissertation, was not a significant predictor of school specific mean alcohol use in the structural level only model. Yet, after the inclusion of the social learning variables it became a significant predictor (p.80). Thus, sh e concluded that the change of significance in this variable may imply there may be moderation effects in addition to mediation effects For the social learning variables, however, the findings here are that Gu specific slopes on alcohol use in this SSSL mo del, their significance and strength of regression coefficients are relatively unchanged from the estimates of the slopes in the social learning only model
132 The between Gu variance in average alcohol use has been reduced from the unconditional model (from 00 = 0.21 to 00 = 0.05), and it is insignificant in this full model. That indicates this model explains the between Gu variance substantially. Social structure with social bonding model This model includes all of the social structure and social bonding v ariables to test the extent to which the bonding variables mediate the effects of the structural level variables on alcohol use (see Table 4 10). The findings in this table show that the social bonding variables have some but not very substantial mediating effects. Only one structural variable, religiosity appears to be mediated by the bonding variables. Population density and percent of residential mobility are not much changed from the structural model in Table 4 5 ( 01 = 0.00004 03 = 0.08 from 0.06 ), sex effects actually increase ( 90 = 0.58 ** from .47** ).Type of school effect is reduced ( 100 = 0. 42 from 0.76** ) but remains significant. Oddly, percent on public welfare, which was not sign ificant in the structural level model, became significant ( 20 = 0.01 from = 0.02) and the coefficient of sex has been slightly increased in this model with the social bonding variables are ( 90 = 0.58 ** from 0.4 7 ** ) Furthermore, the relationship betwe en belief and alcohol use was insignificant in the social boning only model, but the coefficient in this model is stronger and statistically significant ( 130 = 0.1 from 0.09). The relationship between involvement and alcohol use remains relatively unch anged but its coefficient has been slightly reduced from the social bonding only model ( 150 = 0.23 ** from 0.29 *** ). Therefore, it appears that social bonding variables did not mediate the effects of social structural variables on alcohol use. Therefore, this finding does not support Hypothesis 4 which states that the social bonding variables will have mediation effects. The between district variance in mean alcohol use is 0.0 4 and is no longer significant. However, the between Gus variance is reduced from the unconditional model by including the social bonding variables ( 00 = 0.21 versus. 00 = 0. 04 )
133 Social structure with self control model The model in Table 4 11 includes all the social structure and self control variables to examine if self control mediates the influence of social structural variables on alcohol use. The effects of population density, percent of mobility, sex, and type of school in this model are essentially the same as in the structural level model (Table 4 5), but the effect of religiosity became insignificant. The regression coefficients of the four significant structural level predictors have been slightly reduced in the model (from 01 = 0.0000 4 *** to 01 = 0.00002 ** 03 = 0.0 8 to 03 = 0.07 10 = 0.4 7 ** 10 = 0.35 ** 20 = 0.7 6 *** 20 = 0.66 *** ). In terms of the self control variable, the relationship between low self control and alcohol use remains signi ficant but the coefficient has been slightly reduced from the self 160 = 0.43 *** 160 = 0.36 *** ). These findings suggest that while self control is significantly related to alcohol use, it does not substantially mediate the struct (Hypothesis 3). The between Gu variance in mean levels of alcohol use is reduced from the structural level model (Table 4 5) ( 0. 21 versus. 0. 03) and becomes insignificant. Full Comparison Model fo r Alcohol Use with All Social Structural, Social Learning, Social Bonding, and Self Control variables This final full comparison model for alcohol use includes all social structural and social psychological variables. This model allows examining the relati ve mediation impact of the variables from the three different theories on the effects of social structural variables on alcohol use. It is expected that the variables with relatively stronger mediation effect in the separate models would remain significant with higher regression coefficients compared to variables having weaker mediation effects in the separate models. The findings are presented in Table 4 12. In this full comparison model, all of the structural predictors but sex are significantly reduced t oward zero and became insignificant in this model. Differences
134 between male and female respondents are significant but the magnitude of the difference is substantially reduced (by about 75%) from the structure only model (from 90 = 0.4 7 ** 90 = ). The relationships of alcohol use to differential association with alcohol using alcohol use, and imitation remain significant and relatively unchanged from their relationships found in the social learning only model (Table 4 6). The effect of the measures of self control remains significant but the regression coefficients in this model has been substantially reduced from the self con trol only model shown in Table 4 8 (from 160 = 0.43 *** 160 = 0.0 7 ) and from the model with social structure and self control variables ( from 160 = 0. 36 *** 160 = 0.0 7 ) as shown in Table 4 11. Furthermore, none of the social bonding variables have significant effects in this full comp arison model. Among social psychological predictors in this full comparison model, differential association with alcohol using friends has the strongest net effect. As has been found in previous research this differential peer association measure is a robu st predictor, which in these data retains essentially the same impact found in the series of HLM analyses for alcohol use with social learning variables only model and the SSSL model as shown in Table 4 6 and Table 4 9 ( 10 = 0. 97 *** 10 = 0. 93 *** 10 = 0. 93 *** respectively) These findings support Hypothesis 5 indicating that the relative mediation effects of the social learning variables are substantially greater than the mediation effects of the self control and soc ial bonding variables. The between Gu variance in mean alcohol use is 0.004 which is reduced substantially from the variance of the unconditional model in Table 4 4 ( Depressant Use Unconditional (Random ANOVA) Model This model examines the extent to which the frequency of use of depressants varies across Gus (see Table 4 13). The grand mean for depressant use is 1.26, indicating that each
135 students has used depressant at least once or twice in their lifetime on averag e. The between Gu variance in depressant use is 0.14 and is significant, suggesting that mean use of depressant significantly differs between Gus in this sample. This finding also provides support for Hypothesis 1. In terms of within district variance, it was 1.68 which is greater than the between Gu variance. Therefore, the between Gu variance ( interaclass correlation coefficients) accounts for approximately 7.7 % of the total variance in depressant use, where the vast majority of variance is explained by the variance between students (92.3%). The coefficient of reliability for the sample mean of depressant use is .82. It suggests that the sample means are reliable estimates of the true district (Gu) means. Structural Level Model for Depressant Use This st ructural model includes all the structural level variables outlined in the SSSL model (see Table 4 14). Among the structural level predictors, population density and percent on residential mobility are significant predictors of Gu specific mean depressant use ( 01 = 0.0000 2 03 = 0.06 increases. In terms of testing Hypothesis 2b, unlike the structural model for alcohol use, sex 90 = 0. 59 *** ) among structural variables drawn self report data. Furthermore, the coefficient for the effects of sex is in the opposite direction as predicted. That is, female s are more likely than males to use depressant drugs in this sample. These findings provide only partial support for Hypothesis 2a and 2b, in that percent on public welfare, type of school and religiosity do not have significant effects on depressant use a nd while the gender effects is significant as expected it is not in the direction expected. However, the finding that the adolescent girls in the sample are somewhat more likely than the boys is not really surprising considering that it is in the use of st imulants and depressants that the gender ratio is often small and sometimes is tilted toward females.
136 Depressant drugs, affect mood and emotions. It may be that adolescent girls in Korea are more prone to self medication for depression and emotional issues and therefore are more likely to find the use of depressants positively or negatively reinforcing. The between district variance in mean depressant use in this model is 0.001 which has been reduced substantially from the unconditional model and has becom e insignificant. The inclusion of sex, type of school and religiosity, into the structural model has reduced the between Gu variation in individual depressant use. In this model, the R 2 based on the level 2 predictor is [ = ( 00 (unconditional model) 00 (the intercept only structural model) )/ 00(unconditional model) = [(0.13930 0.11741)/ 0.13930=0.15714]. That is, including structural variables explains approximately 15% of the between Gu variance in depressant use. However, again similar to the structu ral model for alcohol use, after taking into account the three predictors of sex, type of school, and religiosity, the between variance at district level became insignificant. Random Coefficient Models Social learning model The social learning model for d epressant use contains the eight social learning variables only (Table 4 15). The significant predictors in this model include differential association with depressant using friends ( 10 = 0.58*** 30 = 0. 18 *** ), definition favorable to depressant use ( 40 = 0.20** ), reinforcement balance for depressant use ( 50 = 0.11 ), parental reaction to the use of depressant ( 70 = 0. 11 ) and imitation ( 80= 0. 07 ). That is, having 1) greater proportion of friends using depressants, 2) mothers who use depressant more frequently, 3) more favorable attitudes toward the use of depressant use, 4) greater perceived rewards of using depressants, 5) less punitive parental r eaction to use of depressants, and 6) greater exposure to depressant using behavioral models are associated
137 use. This finding is compared to the findings from the social learning model for alcohol use, which found significant influence of alcohol use. That indicates that parental influence is important but for those substances used more apparent for those substances used more by girls. Overall, most of social learning variables are theory. The between district variance in mean depressant use is 0.00 8 and not significant. But the between Gu variance has been reduced substantially from the unconditional mod el ( 00 =0.14). Also, by including the social learning variables, 62% of the variation in student depressant use is explained within districts. Social bonding model Table 4 16 shows the findings from the multilevel analysis of social bonding theory. As can b e seen from the table none of the social bonding variables significantly predicts depressant use. This finding does not support Hypothesis 3 which states that the social bonding variables will have significant effects on substance use. The between distric t variance in mean depressant use is 0.15 which has not been reduced from the unconditional model ( 00 =0.14), and is significant. These findings suggest that social bonding variables explain very little the variance of depressant use among adolescents in this sample. Within districts, only 7% of the variance in depressant use is explained by including the social bonding variables. Also, none of the Gu specific slopes of the social bonding variables are significantly vary among districts.
138 Self control model Table 4 17 presents the findings for the self control model for depressant use. As with alcohol use, self control is a significant predictor for the use of depressants ( 160= 0. 11 ** ). control, the higher likelihood of using depressant as the theory predicts. This finding supports Hypothesis 3 for self control. The between Gu variance in mean depressant use is 0.14 and is sig nificant. Similar to the social bonding model, the between Gu variance has not been reduced from the unconditional model ( 00 =0.14). Within Gus, only 9% of the variation in depressant use is explained due to the self control variable. The between Gu varia nce in the Gu specific slope of self control and depressant use is not significant, indicating that there is no difference among Gus in the relationship between depressant use and the self control variable. Models with Structural Variables and Variables from the Three Theories The social structure social learning (SSSL) model The SSSL model includes all of the social structural variables and the social learning variables to assess the mediating impact of the social learning variables on the social structu re predictors of depressant use (see Table 4 18). Population density, percent residentially mobile and sex, which are significant in the structural model, are no longer significant after the social learning variables are included. Furthermore, the coeffici ents of population density, percent of residential mobility, and sex have been reduced substantially from the structural model (from 01 = 0.00002* to 01 = 0.000001, from 03 = 0.06 to 03 = 0.04, and from 90 = 0. 59 *** to 90= 0.10 respectively). The relationships between the social learning variables that were significant in the social learning model remain relatively unchanged, inclu ding the magnitude of their coefficients. These findings support the social learning mediation Hypothesis 4; that is, the social learning variables substantially mediate the social structural predictors of
139 dicted. The between district variance in mean depressant use is 0.00 4 and is not significant. Social structure and social bonding variables model The findings of this social bonding and social structure model are presented in Table 4 19. This model was de signed to examine if there is mediation effects of social bonding variables on the structural level effects on depressant use. The three structural level predictors, population density, percent on residential mobility and sex remained unchanged from the st ructural model. The three predictors are significant and their coefficients are the similar in magnitude (from 01= 0.0000 2 to 01= 0.0000 1 from 03= 0.06 to 03= 0.07 and from 90= 0. 59 *** to 90= 0.6 2 *** respectively). As with the social bonding only model, none of the relationship between social bonding variables and depressant use is significant. That me ans that the social bonding variables do not mediate the structural level effects on depressant use is 0.00 6 and not significant. Social structure and self control variables model control would mediate the influence of structural level predictors on individual depressant use (Table 4 20). Although population density is no longer significant, resi dential mobility and sex remains relatively unchanged from the structural model and is not reduced toward zero. Oddly, the coefficient for gender effects is slightly increased compared to the social structure only model ( 90= 0. 59 *** to 90= 0.6 3 *** ). As with the self control model, the relationship between low self control variable and depressant use is significant. Also, the coefficient of self control is also slightly increased ( 160= 0. 11 ** to 160= 0. 14 ** ) That indicates living in Gus where people move in and out more frequently and being a female increase the use of depressant. Similarly, having low self control increases individual use of depressant. In sum, this finding
140 suggests little support for the Hypothesis 4 proposing mediating effects of the self control variable. The between Gu variance in mean depressant use is not significant ( 0.007 ). Full Comparison Model with Social Structural, Social Learning, Social Bonding, and Self Control Variables As was true for the full comparison model for alcohol use, this full model for depressant use also contains all social structural variab les, social learning, social bonding and self control variables. This model also was developed to assess the Hypothesis 5 examining relative mediation impact of these social psychological variables on the relationships between social structural level varia bles and depressant use. Table 4 21 presents the findings of the full comparison model. Consistent with the full comparison model for alcohol use (Table 4 12), sex is the only significant predictor among social structural variables in the full comparison m odel. However, the coefficient and the level of significance of the gender difference has been substantially reduced in this model compared to the coefficient in the structural model (from 90= 0. 59 *** to 90= 0.13 e ffects of the gender variable on depressant use ). The level of reduction of the coefficients of sex as well as other structural variables are similar to the social structure social learning model (the SSSL model). That indicates the significant reduction i n coefficients and the level of significance among the structural variables are explained by the mediation effects of the social learning variables more so than any other set of variables. It should also be noted that the effects of the sex variable, which were mediated to insignificance in the SSSL model, but significant in this full comparison model, may indicate that the social bonding and self control variables somehow counteract rather than add to the mediating effects of the social learning variables. As shown in the social bonding and self control models (Table 4 16 and Table 4 17), it can be suggested that there are some moderation effects on social bonding and self control variables by social structural variables. Furthermore, the relationships betw een
141 the social learning variables and depressant use remained relatively unchanged from the social learning model and the SSSL model. On the other hand, involvement, a social bonding measure, became significant and the coefficient is slightly increased in this full comparison model, but the direction of the relationship between involvement and depressant use is opposite to what social bonding theory would predict (from 140= 0. 08 to 140= 0.10 ). That is, the more time in homework study, the greater probability of depressant use. Finally, the self control measure is significant but the coefficient and level of significance of this variable has been substantially reduced in this comparison model (from 160= 0. 11 ** to 160= 0.06 ). Overall, the findings from this comparison model support Hypothesis 5, indicating strongest mediation effects of social learning variables than both social bonding and self control variables. T he between Gu variance in this model is 0.00 4 and not significant. Hierarchical Generalized Linear Model Analyses: Tobacco Use In the entire HGLM model for tobacco use, the results from the analyses are reported as odds ratios in order to improve interpre tability. The odds ratios indicate the change in the unit change in a certain independent variable when holding constant other variables in the model (Hedeker & Gibbo ns, 2006). Odds ratios greater than one suggest an increase in students likelihood of using tobacco, where values less than one indicate a reduction in their likelihood of tobacco use (Hedecker & Gibbons, 2006). Unconditional (Random ANOVA) Model Table 4 22 presents the findings of a HGLM Bernoulli unconditional model for differences in tobacco use across Gus. The estimates result from the unit specific model with robust standard errors. The estimated average mean (or logit or log odds) of tobacco use acro ss Gus is 0.2 odds of tobacco use to be approximately
142 normally distributed with a mean of 0.21 and variance of 0.65, one would expect about 95% The variance between Gus in Gu average log odds of tobacco use is 0.65 ( 2 = 93.383*** ) and is significant. This finding supports Hypothesis 1. That indicates there are significant differences in tobacco use between Gus. The coefficient of reliability for the intercept of tobacco use is 0.87, indicating that the mean probabili ty of tobacco use among Gus are quite reliable estimates of the true probability of tobacco use across Gus. Structural Level Model Table 4 23 presents the results for the model analysis with social structural variables only. Population density, percent of residential mobility, sex and type of school are significantly associated with the odds of tobacco use. Residential mobility and type of school are associated with a lower expected probability (log odds) of tobacco use ( 03 = and 100 = 1. 12 *** respectively), holding constant the other predictors in the model and the random effects. The interpretation is that, the expected odds of tobacco use for students who live in Gus with higher level of mobility is reduced by exp ( 0.1) = 0.90 (odds ratio) times the odds of tobacco use for an otherwise similar student who lives in Gus with lower level of mobility. Also, the expected odds of tobacco use of students who attend liberal type of school reduced by exp ( 1.12) = 0. 32( odds ratio) times the odds of tobacco use than students who attend industrial type of school. That is, living in Gus where people move in and out frequently during the last year and attending liberal school significantly reduced the odds of tobacco use com pared to their counterparts living in Gus with lower level of residents move in and out and students attending to industrial school. The direction of the effects of residential mobility is opposite of what would be expected to the extent that such mobility is seen as an indicator of less integrated or more disorganized districts. On the other hand, population density and sex are positively associated with the odds of tobacco use
143 ( 01 = 0.0000 3 and 90 = 1. 20 ** respectively). That indicates male students have significantly greater odds (odds ratio = 3.34 ) of tobacco use than female students. In terms of population density, it has significant and positive direction, but its odds of t obacco use is negligible (odds ratio s =1.0). That indicates that, in fact, there is not a substantial odds difference between students who live in Gus where population is highly dense and their counterparts for the expected odds of tobacco use. These find ings give mixed support Hypothesis 2a and 2b. The between Gu variance in the odds of tobacco use is 0.0 3 ( 2 = 23.74 ) and is significant. The variance of the Gu specific slope for sex and the log odds of tobacco use is significant. That is, in Gus with high log odds of tobacco use, there is a significant and strong relationship between sex and tobacco use. There fore, it appears that in Gus with more tobacco use, the group that is at higher risk of using tobacco is male students Random Coefficient Models The social learning model Table 4 24 presents the social learning model for tobacco use. For this model, the results suggest that differential association with tobacco using friends, definitions, perceived effects of tobacco use, parental reaction to the use of tobacco and imitation are significantly associated with the log odds of tobacco use ( 10 = 1.78*** 40 = 0.40** 50 = 0.51 *** 70 = 0.29 and 80 = 0.47 ** respectively). Having more friends who use tobacco greatly increases the odds of tobacco use ( odds ratio = 5.97 ) compared to those with fewer tobacco using friends. Students who hav e definitions approving the use of tobacco are more likely to use tobacco than students holding disapproving attitudes toward tobacco use ( 1.5 odds ratio) Also perceiving more rewards than costs from smoking is associated with a 1.65 (odds ratio) times in crease in the log odds of tobacco use. Students who expect more encouraging parental reactions to their use of tobacco have increased log odds of tobacco use by 1.33 (odds ratio) times of the odds that of their counterparts. Finally, having greater exposur e to
144 role models using tobacco increases the odds of tobacco use by 1.60 (odds ratio) times the odds of the students who have less exposure to role models using tobacco. This finding also supports Hypothesis 3 for social learning variables. The between Gu variance in the odds of tobacco use is 0.02 ( 2 = 11.15) and is not significant. Social boding model Table 4 25 presents the findings of the social bonding model for tobacco use. Consistent with social bonding theory, attachment and involvement are significantly and negatively associated with the o dds of tobacco use ( 120 = 0.27 *** 150 = 0.44 *** ). Greater attachment of students to friends and parents lowers odds of tobacco use by 0.75 time than the odds of students less attached to friends and parents ( odds ratio = 0.75 ). Spending more time in study significantly reduces the odds of tobacco use by 0.64 ( odds ratio ) times, compared to students who spend less time in study. The between Gu variance of the mean odds of tobacco use is 0.43 ( 2 = 62.04 *** ) and is significant. The variance of the Gu sp ecific slopes for involvement and the odds of tobacco use is significant. That indicates, Gus with high mean log odds of tobacco use tend to have strong relationship between time spent in studying and the odds of individual tobacco use. Self control model The findings of this model are presented in Table 4 26. In this model, self control is a significant predictor for the log odds of tobacco use ( 160 = 0.61 *** ). Students with high self control have reduced log odds of tobacco use by 0.53 ( odds ratio ) times the odds of tobacco use by students with low self control. The results are consistent with self control theory and supports Hypothesis 3. The between Gu variance in the odds of tobacco use is 0.47 ( 2 = 71.06 *** ) and is significant. However, the variance of the Gu specific slope for the relationship between self control and tobacco use is not significant.
145 Models with Structural Variables and Vari ables from the Three Theories The social structure social learning (SSSL) model The SSSL model includes all of the social structure and social learning variables to examine if the social learning variables mediate the influence of structural predictors on 27. As the SSSL model predicted, all of the structural variables which were significant in the structural model became no longer significant and their coefficients ar e substantially reduced by including the social learning variables (population density, residential mobility, sex, and type of school). Particularly, the coefficients of, sex and type of schools are very substantially reduced (from 90 = 1. 20 ** to 90 = 0.4 2, from 100 = 1. 12 *** to 100 = 0.3 7 ). In terms of social learning variables, differential association with friends using tobacco, definitions favorable for using tobacco, differential reinforcement for tobacco use, encouraging reactions to tobacco use by parents and greater exposure for smoking models are all positively and 10 = 1. 78 *** to from 10 = 1.6 7 *** from 40 = 0.40 ** to 40 = 0.4 0 ** from 50 = 0.51 *** to 5 0 = 0.5 0 *** from 70 = 0.29 to 70 = 0.2 5 from 80 = 0.47 ** to 80 = 0.4 5 ** respectively). These variables are significant predictor of tobacco use and the magnitudes of their coefficients also remain relatively unchanged. The findings support the argument s of the SSSL model suggesting the social learning variables substantially mediate the impact of structural predictors on deviant behaviors (Hypothesis 4). Interpretations of this model is as follows: Having higher proportion of friends who use tobacco inc reases the odds of the tobacco use exp ( 1.6 7) = 5.36 (odds ratio) times the odds of students who do not have friends use of tobacco by 1.53 times the odds of tobacco use. Students who reported a balance of reinforcement with greater perceived rewards than costs from smoking have greater odds of
146 tobacco use by 1.68 times than the odds of their counterparts. Students who expected encouraging parental reactions to their smoking increased the log odds of tobacco use by 1.28 (odd ratio) than the odds of students who expected discouraging parental reaction to their smoking. Lastly, students who reported that they received greater influence by observing role e have greater odds of tobacco use by 1.61 (odds ratio) times compared to their counterparts who reported less influence from observing others smoking. The between Gu variance in mean log odds of tobacco use is 0.0 3 and not significant. Social structure a nd social bonding model The findings on the model with social structure and social bonding variables are presented in Table 4 28. In this model, all social structural predictors and social bonding variables are included to examine if the social bonding var iables would mediate the impact of mobility and tobacco use became insignificant after including the four social bonding variables, while the relationships for popu lation density, sex and type of school remain significant ( 01 = 0.00003 90 = 1.48 ***, 100 = 0.68 **, respectively ). Oddly, although the coefficient for type of school reduced substantially (from 20 = 1.12 ** to 20 = 0. 68 ** ), the coefficient for sex increased (from 90 = 1.20 *** to 90 = 1.48 *** ). That indicates t he social addition, the findings suggest that there are may be some moderation effects between social structural variables and social bonding variables. That is, in addition to the differences in effect of sex on tobacco use, the effects of belief and commitment became significant when social structural variables are entered in this full model. The level of significance and level of magnitude of coefficients of att achment and involvement variables remained relatively unchanged from the social banding only model (from 120 = 0.27 ** to 120 = 0.2 7 from 150 = 0.44 ** to 150 = 0.4 3 ** respectively). The findings of this model can be interpreted as
147 follows: Attending l iberal schools reduces the odds of smoking by 0.50 times the odds of students attending industrial school. Also, male students have 4.42 times greater odds of smoking compared to the odds of smoking for female students. Furthermore, the coefficients for be lief and commitment have been increased by including the structural predictors ( 130 = 0.04 *** to 130 = 0.1 0 140 = 1.0 to 140 = 0.13 respectively). That is, having greater belief in general law reduced the odds of smoking among students by 0.9 times. The direction of commitment is positive relationship with smoking, which is opposi te to the expectation of the social bonding theory. That is, students who are more committed to their school work and future employment are more likely to use tobacco by 1.14 times than their counterparts. The findings of these two variables suggest some m oderation effects of social bonding variables in that the effects of some social bonding variables on the tobacco use increased once the structural variable are included. The between Gu variance in mean log odds tobacco use is 0.04 and is not significant. Social structure and self control model The social structure with self control model includes all social structure predictors and the self control variable (see Table 4 29). By including the self control variable, population density and percent on mobilit y are no longer significant. However, sex and type of school are significant and maintain the similar magnitude their coefficients to the structural model (from 90 = 1.20 *** to 90 = 1.13 *** from 100 = 1.12 *** to 100 = 1.03 *** respectively). As in the self control only model, the self control variable is significant but has slightly reduced coefficient from the self control only model (from 160 = 0.61*** to 16 0 = 0.52** ). Therefore, the self control variable substantially mediates the impact of community level structural predictors, but it does not mediate the social structural variables, sex and type of school. The findings indicate that for being male student s, attending industrial school increase
148 the odds of tobacco use among students regardless of their level self control. The between district variance in the mean log odds of tobacco use is 0.02 but it is not significant. Full Comparison Model with Structu ral, Social Learning, Social Bonding, and Self Control Variables As with the full comparison model for alcohol and depressant use, this full model for tobacco use also contains all social structural variables, social learning variables, social bonding vari ables and self control variables. This model compares the relative mediation impact of these variables from the three theories on the social structural level variables on tobacco use. Table 4 30 presents the findings from the full comparison model for toba cco use. Unlike with the full comparison model for alcohol use and depressant use (Table 4 12 and 4 21), sex became not significant in the full comparison model even. Also, its coefficient and the level of significant is substantially reduced (by more than half) compared to the structural model ( 90 = 0.53 from 1.20 ** ). The rest of social structural predictors are also not significant. The relationships between the social learning variables and tobacco use remained relatively unchanged from the social learning model and the SSSL model. Differential association with tobacco using friends, definition, reinforcement balance of rewards and costs of tobacco use, and imitation remain significant predictors, while parental reaction toward tobacco use became insignificant. In addition, the magnitudes of the ir coefficients have been slightly increased from the social structure social learning model. With regard to the social boning variables, attachment, belief, and involvement are no longer significant in this model. Commitment is the only significant predic tor from social bonding theory. These changes in significance are not consistent with the social bonding only model, in which attachment and involvement are significant while belief and commitment are not significant. In the social structure and social bon ding model, however, belief and commitment became significant and the magnitudes of coefficients are also increased. In the final comparison model, only
149 commitment, which is not the strongest predictor in the previous social bonding only and social structu re social bonding models, became the only significant predictor among the four social bonding variables. Furthermore, the coefficient for commitment increased substantially from the full social bonding model after including social learning variables and se lf control variable (from 140 = to 140 = 0. 34 ). Finally, self control in this model is not a significant variable and the coefficient is substantially reduced from the self control and social structure model (from 160 = 0.52** to 160 = 0.07 ). These findings indicate that th e mediation effects of social learning variables are greater than the effects of social bonding and self control variable. The between Gu variance in mean tobacco use is 0.0 5 which is reduced greatly from the variance of the unconditional model ( 0.6 5 ) and is insignificant.
150 Table 4 1. Percentages of students who ever used different substances in their lifetime District ID (Gu name) Alcohol Depressants Tobacco Total Sample Freq. =Yes (%) Freq.=Yes (%) Freq.=Yes (%) 01 (Jung Gu) 34 (85%) 23(57%) 26(65%) 40 02 (Dongrae Gu) 83 (90%) 56(61%) 43(46%) 92 03 (Sa Sang Gu) 38 (97%) 15(38%) 24(62%) 39 04 (Nam Gu) 129 (89%) 71(49%) 76(52%) 145 0 5 (Seu Gu) 35 (90%) 10(25%) 30(77%) 39 0 6 (Jin Gu) 115 (87%) 71(54%) 70(53%) 132 0 7 (Yeunjae Gu) 30 (83%) 2 5(70%) 22(61%) 36 0 8 (Sooyoung Gu) 36 (88%) 34(83%) 11(27%) 41 0 9 (Haeyundae Gu) 90 (83%) 49(45%) 53(49%) 108 10 (Buck Gu) 27 (79%) 26(76%) 20(59%) 34 1 1 (Keumjung Gu) 62 (76%) 54(66%) 23(28%) 82 1 2 (Dong Gu) 61 (85%) 41(57%) 27(37%) 72 1 3 (Youngdo G u) 61 (86%) 32(45%) 35(49%) 71 1 4 (Kangse Gu) 35 (76%) 26(56%) 6(13%) 46 15 (Saha Gu) 21 (50%) 26(74%) 3(8%) 35 Total 859/1,012 559/1,012 469/1,012 1,012
151 Table 4 2. Descriptive statistics on students and districts Variables Mean S.D. Min. Max. Le vel 1 measures a Frequency of alcohol use 2.17 1.54 0 6 Frequency of depressant use 1.22 1.35 0 5 Tobacco use (0 1) 0.46 0.49 0 1 Sex (male=1) 0.52 0.49 0 1 Type of school (liberal =1) 0.53 0.49 0 1 Religiosity 1.81 1.10 1 4 Social learning vari ables Differential association Peer association Alcohol 6.04 2.56 0 9 Depressant 3.36 1.91 0 8 Tobacco 5.02 3.65 0 9 Alcohol 4.40 1.97 0 6 Depressant 1.53 1.11 0 6 Tobacco 4.61 2.80 0 6 Alcohol 2.53 1.64 0 6 Depressant 1.80 1.30 0 5 Tobacco 1.15 0.90 0 6 Definitions Alcohol 1.75 0.55 1 3 Depressant 2.19 0.64 1 3 Tobacco 2.45 0.69 1 3 Differential reinforcement Reinforcement balance Alcohol 1.83 0.64 1 3 Depressant 2.42 0.58 1 3 Tobacco 2.52 0.62 1 3 Alcohol 2.81 0.93 1 7 Depressant 3.56 0.95 1 7 Tobacco 3.41 0.97 1 7 Alcohol 2.60 0.67 1 5 Depressant 2.75 0.93 1 5 Tobacco 3.24 0.68 1 5 Imitation 2.87 0.80 1 4 Social bonding variables Att achment 6.18 1.62 3 12 Belief 6.81 1.63 3 12 Commitment 3.02 1.20 2 8 Involvement 2.35 1.38 1 6 Self control variable Self control 2.64 0.35 1 5 Level 2 Measures b Population density 11472.67 6422.30 367 21318 Percent on public welfare 6.47 % 3.10% 1.42 % 12.32 % Residential mobility 18.23% 1.34% 15.8 % 20.3 % a Source: Hwang (2000) dissertation; b 1999 South Korea Census data and BST data.
152 Table 4 3. Bivariate correlations of the independent variables with alcohol, depressant and t obacco use Alcohol Depressant Tobacco Variable Coefficients Coefficients Coefficients Social l earning v ariables Differential a ssociation 1 Peer association 0.77* 0.64* 0.71* 0.14* 0.31* 0.11* 3 0.15* 0.45* 0.0 9* 4 Definitions 0.31* 0.45* 0.44* Differential reinforcement 5. Reinforcement balance 0.37* 0.42* 0.52* 0.34* 0.30* 0.43* 0.21* 0.37* 0.28* 8. Imitation 0.49* 0.12* 0.50* Social bondi ng variables 9. Attachment 0.13* 0.002 0.16* 10.Belief 0.12* 0.01 0.06* 1 1.Commitment 0.09* 0.004 0.03 1 2.Involvement 0.24* 0.02 0.24* Self c ontrol v ariable 1 3.Self Control 0.31* 0.06* 0.28* Social s tructural v ariables 1 4.Sex (male = 1) 0 .19* 0.22* 0.29* 15 .Type of school (liberal = 1) 0.25* 0.005 0.26* 16.Religiosity 0.06* 0.03 0.0 1 17. Population density 0.19* 0.01 0.16* 18. Percent on public welfare 0.00 2 0.03 0.06* 19. Residential mobility 0.02 0.10* 0.006 *P < 0.05 (o ne tailed t test)
153 Table 4 4. Unconditional r andom ANOVA m odel for variation in a lcohol u se across Gus Parameter Coefficient se Fixed e ffects Grand m ean Frequency of u se 2 .14*** 0.127 Variance D.F. Random e ffects Between G u s ( 00 ) 0.21 a 92.75*** 14 Within Gu s 2.20 b value<.1; *p value<.05; **p value<.01; ***p value<.001 a = 00, b = 2 Table 4 5. Structural level m odel for alcohol use Parameter Coefficient Se Fixed e ffects Between d istricts (Gus) Inte rcept 2.15*** 0.07 3 Population d ensity 0.0000 4 *** 0.000008 Percent on p ublic welfare 0.0 2 0.0 18 Residential mobility 0.08 0.034 Within d istricts (Gus) Sex 0.4 7 ** 0.1 39 Type of s chool 0.7 6 *** 0.1 00 Religiosity 0.10 0.04 7 V ariance D.F. Random e ffects Between d istricts (Gus) ( 00 ) 0.0 8 a 1 3.40 11 Sex 0. 11 ** 7. 56 14 Type of s chool 0. 07 2.15 14 Religiosity 0.0 1 1. 82 14 Within districts (Gus) 2.03 b Note: the three variables, sex, type of school, & religiosity, are grand mean centered. p value<.1; *p value<.05; **p value<.01; ***p value<.001 a = 00, b = 2
154 Table 4 6. Social learning m odel for alcohol use Parameter Coefficient Se Fixed e ffects Within d istricts (Gus) Social learning variables I ntercept 2 .17*** 0.029 Differential association Peer a ssociation 0.97*** 0.047 u se 0.08* 0.026 u se 0.01 0.033 Definitions 0.07* 0.031 Differential reinforcement Reinforcement balance 0.11** 0.027 r eactions 0. 03 0.023 r eactions 0.08* 0.029 Imitation 0.2 0 *** 0.042 Variance DF Random e ffects Between d istricts (Gus) ( 00 ) 0.002 a 16.91 14 Differential association Peer a ssociation 0.01 7 22.30 14 u se 0.001 15.77 14 u se 0.007 15.18 14 Definitions 0.00 3 14.86 14 Differential reinforcement Re inforcement balance 0.00 3 7.41 14 r eactions 0.0007 6.83 14 r eactions 0.00 4 14.65 14 Imitation 0.01 0 22.52 14 Within districts (Gus) 0.8 5 b % Reduction in within school conditional error variance 62% Note: Level 1 predictors were grand mean centered. p value<.1; *p value<.05; **p value<.01; ***p value<.001 a = 00, b = 2
155 Table 4 7. Social b onding m odel for alcohol use Parameter Coefficient Se Fixed e ffects Within d istricts (Gus) Intercept 2 .13*** 0.10 0 Attachment 0.07 0.062 Belief 0. 09 0.04 1 Commitment 0.06 0.048 Involvement 0.29*** 0.0 53 Variance D.F. Random e ffects Between d istricts (Gus) ( 00 ) Intercept 0.117*** a 52.84 14 Attachment 0.02* 24.42 14 Belief 0.009 13.12 14 Commitment 0.01 13.28 14 Involvement 0.019 19.50 14 Within districts (Gus) 2.0 5 b % Reductio n in within school conditional error variance 7% Note: Level 1 predictors were grand mean centered. p value<.1; *p value<.05; **p value<.01; ***p value<.001 a = 00, b = 2 Table 4 8. Self c ontrol m odel for alcohol use Parameter Coefficient Se Fixe d e ffects Within d istricts (Gus) Intercept 1 2 .13*** 0.109 Self control 0.43*** 0.055 Variance D.F. Random e ffects Between d istricts (Gus) ( 00 ) Intercept 0.1 4 *** a 76.31 14 Self control 0.013 20.93 14 Within districts (Gus) 2.0 1 b % Reduction in within school conditional error variance 9% Note: Level 1 variable is grand mean centered. p value<.1; *p value<.05; **p value<.01; ***p value<.001 a = 00, b = 2
156 Table 4 9. T he SSSL m odel with social structure and social learni ng variables for alcohol use Parameter Coefficient Se Fixed e ffects Between d istricts (Gus) Intercept 2.18 *** 0.03 8 Population density 0.000004 0.000007 Percent on public welfare 0.0 1 0.013 Residential mobility 0.0 1 0.032 Within Dis tricts (Gus) Sex 0.18* 0.080 Type of s chool 0.1 1 0.084 Religiosity 0.05 0.030 Differential association Peer a ssociation 0.93*** 0.046 u se 0.0 7 0.035 u se 0.00 3 0.042 Definitions 0.09* 0.034 Differential reinf orcement Reinforcement balance 0.11* 0.038 r eactions 0.0 3 0.034 r eactions 0.0 9 0.035 Imitation 0.2 0 *** 0.044 Variance D.F. Random effects Between d istricts (Gus) ( 00 ) 0.0 5 0.296 11 Sex 0.02 1.109 14 Type o f s chool 0.0 3 3.772 14 Religiosity 0.003 2.12 14 Differential association Peer a ssociation 0.01 1.10 1 14 u se 0.004 0.527 14 u se 0.01* 4.627 14 Definitions 0.002 0.472 14 Differential reinforcement Reinforcement balance 0.003 2.659 14 r eactions 0.001 1.128 14 r eactions 0.00 4 3.662 14 Imitation 0.01 1.667 14 Note: All predictors are grand mean centered. value<.1; *p value<.05; **p value<.01; ***p value<.001
157 Table 4 10. Model with social str ucture and s ocial b onding variables for alcohol use Parameter Coefficient se Fixed e ffects Between d istricts (Gus) Intercept, 2.19 *** 0.067 Population d ensity 0.00004*** 0.000005 Percent on p ublic w elfare 0.009 Residential mobility 0.029 Within d istricts (Gus) Sex 0.58** 0.136 Type of s chool 0. 42 0.149 Religiosity 0.06 0.050 Attachment 0.0 6 0.053 Belief 0.10* 0.055 Commitment 0.0 5 0.044 Involvement 0.23** 0.052 Variance D.F. Random effects Between d istricts (Gus) ( 00 ) 0.0 4 11.80 1 1 Sex 0.10** 6.97 14 Type of s chool 0.12* 5.39 14 Religiosity 0.01 2.29 14 Attachment 0.02 0.13 14 Belief 0.02 4.00 14 Commitment 0.008* 3.93 14 Involvement 0.01 1.59 14 Withi n districts (Gus) 1.90 Note: All predictors are grand mean centered. value<.1; *p value<.05; **p value<.01; ***p value<.001
158 Table 4 11. M odel with social structure and s elf control variables for alcohol use Parameter Coefficient se Fixed e ffects Between d istricts (Gus) Intercept 2 .15*** 0.068 Population d ensity 0.00002** 0.000006 Percent on p ublic w elfare 0.01 0.011 Residential mobility 0.07 0.02 9 Within d istricts (Gus) Sex 0.35** 0.109 Type of s chool 0.66*** 0.099 R eligiosity 0.08 0.054 Self control 0.36*** 0.051 Variance D.F. Random effects Between d istricts (Gus) ( 00 ) 0.03 11.19 1 1 Sex 0.07** 7.50 14 Type of s chool 0.08 1.88 14 Religiosity 0.02 2.51 14 Self control 3.69 14 Note: All pr edictors are grand mean centered. value<.1; *p value<.05; **p value<.01; ***p value<.001
159 Table 4 12. Full m odel comparing the relative mediating effects of s ocial l earning, s ocial b onding and s elf c ontrol variables for alcohol use Parameter Coefficie nt se Fixed Effects Between d istricts (Gus) Intercept 2 .17*** 0.037 Population Density 0.00000 1 0.000007 Percent on Public Welfare 0.0 2 0.013 Residential Mobility 0.0 2 0.032 Within d istricts (Gus) Sex 0.15 0.078 Type of s chool 0.0 5 0.089 Religiosity 0.0 5 0.03 0 Differential association Peer a ssociation 0.93*** 0.050 u se 0.08 0.03 4 u se 0.006 0.041 Definitions 0.0 7 0.03 5 Differential reinforcement Reinforcement balance 0. 11 0.036 r eactions 0.02 0.034 r eactions 0.0 5 0.036 Imitation 0.1 6 ** 0.04 7 Attachment 0.00 9 0.036 Belief 0.04 0.033 Commitment 0.00 8 0.038 Involvement 0.0 6 0.03 6 Self control 0.0 7 0.038 Variance D.F. Random effects Between districts (Gus) ( 00 ) 0.004 0.284 1 1 Sex 0.0 2 1.78 14 Type of s chool 0.02* 5.80 14 Religiosity 0.002 1.06 14 Differential association Peer a ssociation 0.01 1.65 14 u se 0.003 0.50 14 u se 0.01* 6.12 14 Definitions Differential r einforcement 0.003 0.59 14 Reinforcement b alance 0.005 3.05 14 r eactions 0.005 0.90 14 r eactions 0.006* 4.85 14 Imitation 3.33 14 Attachment 5.14 14 Belief 0.0001* 0.004 14 Commitment 0.0 1 5.53 14 Involvement 0.00 2 2.17 14 Self control 0.006** 6.24 14 Within districts (Gus) 0.78 Note: All predictors are grand mean centered. value<.1; *p value<.05; **p value<.01; ***p value<.001
160 Table 4 13. Unconditio nal r andom ANOVA m odel for variation in depressant use across Gus Parameter Coefficient se Fixed e ffects Grand m ean Frequency of u se 1.26*** 0.106 Variance D.F. Random Effects Between districts (Gus) ( 00 ) 0. 14 a 78.94*** 14 Within d istricts (Gus) 1.6 8 b value<.1; *p value<.05; **p value<.01; ***p value<.001 a = 00, b = 2 Table 4 14. Structural level m odel for depressant use Parameter Coefficient Se Fixed e ffects Between d istricts (Gus) Intercept 1.20 *** 0.047 Population d ensity 0.0000 2 0.00000 8 Percent on p ublic w elfare 0.002 0.012 Residential mobility 0.06 0.032 Within d istricts (Gus) Sex 0.59 *** 0.068 Type of s chool 0.10 0. 144 Religiosity 0.0 3 0. 0 44 Variance D.F Random e ffects Between d istricts (Gus) ( 00 ) 0.00 1 a 0. 30 11 Sex 0.00 9 0. 37 14 Type of s chool 0. 20 ** 7 .0 3 14 Religiosity 0.01 1.7 5 14 Within districts (Gus) 1.62 b Note: All variables are grand mean centered. value<.1; *p value<.05; **p val ue<.01; ***p value<.001 a = 00, b = 2
161 Table 4 15. Social l earning m odel for d epressant use Parameter Coefficient se Fixed e ffects Within d istricts (Gus) Social learning variables Intercept 1.23*** 0.033 Differential association Pe er Association 0.58*** 0.0 34 0.04 0.0 41 0. 18 *** 0.03 5 Definitions 0.2 0 ** 0.046 Differential reinforcement Reinforcement balance 0. 11 0.041 0.0 01 0.02 8 0. 11 0.0 38 Imitation 0. 07 0.04 0 Variance DF Random e ffects Between d istricts (Gus) ( 00 ) 0.00 8 a 18.74 14 Differential association Peer a ssociation 0.001 9.38 14 u se 0.00 8 14 u se 0.003 14.13 14 Definitions 0.01 19.03 14 Differential reinforcement Reinforcement balance 0.01 18.84 14 r eactions 0.00 1 10.27 14 r eactions 0.00 9 12.88 14 Imitation 0.01 20.62* 14 Within districts (Gus) 0.8 7 b % Reduction in within school conditional error variance 62% Note: Level 1 pre dictors were grand mean centered. value<.1; *p value<.05; **p value<.01; ***p value<.001 a = 00, b = 2
162 Table 4 16. Social b onding m odel for d epressant u se Parameter Coefficient se Fixed e ffects Within d istricts (Gus) Intercept 1.26*** 0.1 10 Attachment 0.005 0.046 Belief 0.03 0.020 Commitment 0.01 0.03 3 Involvement 0.06 0. 047 Variance D.F. Random e ffects Between d istricts (Gus )( 00 ) 0.15 74.00*** 14 Attachment 0.008 13.87 14 Belief 0. 0002 3.91 14 Commitment 0.009 9.70 14 Involvement 0.007 15.25 14 Within districts (Gus) 1.65 b % Reduction in within school conditional error variance 7% Note: Level 1 predictors were grand mean centered. value<.1; *p value<.05; **p value<.01; ***p value<.001 a = 00, b = 2 Table 4 17. Self c ontrol m odel for d epressant u se Parameter Coefficient se Fixed e ffects Within d istricts (Gus) Intercept 1 1.26*** 0.10 7 Self control 0. 11 ** 0.0 28 Variance D.F. Random e ffects Between d istricts (Gus) ( 00 ) 0.14 a 77.11*** 14 Self control 0.001 6.22 14 Within sistricts (Gus) 1.66 b % Reduction in within sc hool conditional error variance 9% Note: Level 1 variable is grand mean centered. value<.1; *p value<.05; **p value<.01; ***p value<.001 a = 00, b = 2
163 Table 4 18. The SSSL m odel, with social structure and social learning variables for depressant use Parameter Coefficient se Fixed e ffects Between d istricts (Gus) Intercept 1.25*** 0.0 35 Population d ensity 0.00000 1 0.00000 7 Percent on p ublic w elfare 0.006 0.01 Residential mobility 0.04 0.03 Within d istricts (Gus) Sex 0.10 0 .0 81 Type of s chool 0.0 9 0.0 79 Religiosity 0.01 0.0 31 Differential association Peer a ssociation 0.5*** 0.0 45 u se 0.04 0.0 42 u se 0.1 9 *** 0.037 Definitions 0.2 0 ** 0.0 50 Differential reinforcement Reinforcement balanc e 0. 11 0.0 42 r eactions 0.00 5 0.027 r eactions 0.10* 0.03 6 Imitation 0.0 9 0.04 0 Variance D.F. Random effects Between districts (Gus) ( 00 ) 0.00 3 a 2.05 11 Sex 0.0 1 0.408 14 Type of School 0.0 2 4.00 14 Religiosity 0.0 03 4. 84 14 Differential association Peer a ssociation 0.007 4.46* 14 u se 0.00 8 10.73** 14 u se 0.001 0.07 14 Definitions 0.01 1.10 14 Differential reinforcement Reinforcement balance 0.004 0.35 14 r eactions 0.00 09 0. 37 14 r eactions 0.00 7 2.5 4 14 Imitation 0.0 2 4.12 14 Within districts (Gus) 0.84 b Note: All predictors are grand mean centered. value<.1; *p value<.05; **p value<.01; ***p value<.001 a = 00, b = 2
164 Table 4 19. Model with social structure and s ocial b onding variables for depressant use Parameter Coefficient se Fixed e ffects Between d istricts (Gus) Intercept 1.2 1 *** 0.048 Population d ensity 0.0000 1 0.000008 Percent o n p ublic w elfare 0.00 2 0.01 2 Residential mobility 0.07 0.032 Within d istricts (Gus) Sex 0.6 2 *** 0.0 61 Type of s chool 0.06 0.131 Religiosity 0.03 0.048 Attachment 0.02 0.047 Belief 0.0 01 0.041 Commitment 0.0 01 0.028 Inv olvement 0.044 Variance D.F. Random effects Between districts (Gus) ( 00 ) 0.00 2 0.37 1 1 Sex 0.006 0.57 14 Type of s chool 0.1 2 3.17 14 Religiosity 0.01 0.93 14 Attachment 0.006** 7.29 14 Belief 0.0002 0.84 14 Commitment 0.0007 0.10 14 Involvement 3.17 14 Within districts (Gus) 1.60 Note: All predictors are grand mean centered. value<.1; *p value<.05; **p value<.01; ***p value<.001 a = 00, b = 2
165 Table 4 20. Model with social structure and s elf control variables for depressant use Parame ter Coefficient se Fixed e ffects Between d istricts (Gus) Intercept 1.2 0 *** 0.0 49 Population d ensity 0.0000 1 0.00000 8 Percent on p ublic w elfare 0.004 0.0 12 Residential mobility 0.06 0.029 Within d istricts (Gus) Sex 0.63 *** 0.067 Type of s chool 0.12 0.144 Religiosity 0.0 3 0.043 Self control 0.14** 0.0 32 Variance D.F. Random effects Between districts (Gus) ( 00 ) 0.007 a 0. 62 1 1 Sex 0.01 0. 37 14 Type of s chool 0.13** 7.4 0 14 Religiosity 0.0 1 1.7 1 14 Self control 0.0 03 1.24 15 Within districts (Gus) 1.60 b Note: All predictors are grand mean centered. value<.1; *p value<.05; **p value<.01; ***p value<.001 a = 00, b = 2
166 Table 4 21. Full m odel comparing the relative mediating effects of s ocial l earning, s ocial b onding and s elf c ontrol variables for depressant use Parameter Coefficient se Fixed e ffects Between d istricts (Gus) Intercept 1.23*** 0.03 5 Population d ensity 0.00000 8 0.000007 Percent on p ublic w elfare 0.011 0.0 12 Residential mobility 0.02 0.034 Within d istricts (Gus) Sex 0.13 0.073 Typ e of s chool 0.0 2 0.07 8 Religiosity 0.0 2 0.037 Differential association Peer a ssociation 0.56*** 0.043 u se 0.03 0.042 u se 0.20*** 0.038 Definitions 0. 20 ** 0.04 8 Differential reinforcement Reinforceme nt balance 0.11* 0.04 3 r eactions 0.01 0.04 1 r eactions 0. 10 0.0 49 Imitation 0.04 1 Attachment 0.0 6 0.0 40 Belief 0.03 0.043 Involvement 0.057 Commitment 0.01 0.038 Self control 0.03 8 Variance D.F. Random effects Between districts (Gus) ( 00 ) 0.00 4 a 3.50 1 1 Sex 0.004 0.1 4 14 Type of s chool 0.00 6 2.0 7 14 Religiosity 0.007 2.26 14 Differential a ssociation Peer a ssociation 0.00 5 4.56 14 u se 0.01** 1 1.21 14 u se 0.002 0.02 14 Definitions 0.01 1.79 14 Differential r einforcement Reinforcement b alance 0.00 6 0.81 14 r eactions 0. 005 2.71 14 r eactions 0.0 1 0.84 14 Imitation 0.009 2.52 14 Attachmen t 0.0 1 5.14 14 Belief 0.001 1. 67 14 Involvement 0.02 2.5 2 14 Commitment 0.3 5 14 Self control 0.003 0. 68 14 Within districts (Gus) 0.7 9 b Note: All predictors are grand mean centered. value<.1; *p value<.05; **p value<.01 ; ***p value<.001
167 Table 4 22. Unconditional r andom ANOVA m odel for variation in tobacco u se across Gus Unit specific model p arameter Coefficient se Odds ratio Fixed e ffects Intercept 0.2 1 0.214 0.81 Variance 2 D.F. Random e ffects Betwe en districts (Gus) ( 00 ) 0.65 93.383*** 14 Note: All predictors are grand mean centered. value<.1; *p value<.05; **p value<.01; ***p value<.001 Table 4 23. Structural level m odel for tobacco use Unit specific model p arameter Coefficient se Odds ratio Fixed e ffects Between d istricts (Gus) Intercept 0.2 1 0.08 4 0. 80 Population d ensity 0.0000 3 0.00001 1.00 Percent on public welfare 0.001 0.0 18 1.0 0 Residential mobility 0.051 0.90 Within Districts (Gus) Sex 1. 20 ** 0.25 6 3.34 T ype of s chool 1. 12 *** 0.159 0. 32 Religiosity 0.02 0.048 0.97 Variance 2 D.F. Random e ffects Between d istricts (Gus) ( 00 ) 0.0 3 23.74 1 1 Sex 0. 5 18. 6 1*** 14 Type of s chool 0.0 006 1.81 14 Religiosity 0.000 1 0. 50 14 Note: All predictors are grand mean centered. value<.1; *p value<.05; **p value<.01; ***p valu e<.001
168 Table 4 24. Social l earning m odel for tobacco use Unit specific model p arameter Coefficient se Odds ratio Fixed e ffects Within d istricts (Gus) Social learning variables Intercept 0.17 0.09 5 1.18 Differential association Peer As sociation 1.78*** 0.114 5.97 0.11 0.092 1.12 0.08 0.094 0.92 Definitions 0.40** 0.123 1.50 Differential reinforcement Reinforcement balance 0.51*** 0.113 1.65 0.22 0.164 1.24 0. 29* 0.114 1.33 Imitation 0.47** 0.109 1.60 Variance DF Random e ffects Between d istricts (Gus) ( 00 ) 0.02 11.15 14 Differential a ssociation Peer a ssociation 0.06 4.82 14 u se 0.02 8.68 14 u se 0.06 13.38 14 Definit ions 0.09 11.39 14 Differential r einforcement Reinforcement b alance 0.05 8.21 14 r eactions 0.14 14 r eactions 0.06 12.35 14 Imitation 0.04 7.10 14 Note: Level 1 predictors were grand mean centered. value<.1; *p valu e<.05; **p value<.01; ***p value<.001
169 Table 4 25. Social b onding m odel for tobacco use Unit specific model p arameter Coefficient se Odds ratio Fixed e ffects Within d istricts (Gus) Intercept 0.25 0.186 0.78 Attachment 0.27** 0.086 0.75 Belief 0.04 0.06 1 0.96 Commitment 0.10 0.06 0 1.11 Involvement 0.44** 0.124 0.64 Variance 2 D.F. Random e ffects Between d istricts (Gus) ( 00 ) 0.43 60.02*** 14 Attachment 0.04 21.27 14 Belief 0.01 15.18 14 Commitment 0.0003 13.08 14 Involvement 0. 13 34.75** 14 Note: Level 1 predictors were grand mean centered. value<.1; *p value<.05; **p value<.01; ***p value<.001 Table 4 26. Self c ontrol m odel for tobacco use Unit specific model p arameter Coefficient se Odds ratio Fixed e ffects Within d i stricts (Gus) Intercept 0.21 0.195 0.80 Self control 0.61*** 0.058 0.53 Variance 2 D.F. Random e ffects Between d istricts (Gus) ( 00 ) 0.47 71.06*** 14 Self control 0.00006 10.29 14 Note: All predictors are grand mean centered. value<.1; *p value<.05; **p value<.01; ***p value<.001
170 Table 4 27. The SSSL Model, with social st ructure and s ocial l earning variables for tobacco use Unit specific model p arameter Coefficient se Odds ratio Fixed e ffects Between d istricts (Gus) Intercept 0.1 3 0.133 1.13 Population d ensity 0.00000 3 0.0000 2 1.0 0 Percent on p ublic w elfare 0.0 08 0.050 1.0 06 Residential mobility 0.1 0.118 0.83 Within d istricts (Gus) Sex 0.4 2 0.2 91 1.58 Type of s chool 0.3 7 0. 271 0.68 Religiosity 0.04 0.105 0.95 Differential association Peer a ssociation 1.6 7 *** 0.173 5.36 u se 0.1 2 0 .114 1.14 u se 0.10 0.141 0.90 Definitions 0.4 0 ** 0.158 1.53 Differential reinforcement Reinforcement balance 0.5 0 *** 0.146 1.68 r eactions 0.12 0.163 1.18 r eactions 0.2 5 0.147 1.28 Imitation 0.4 5 ** 0.140 1.61 V ariance D.F. Random effects Between districts (Gus) ( 00 ) 0.0 3 0.33 11 Sex 0.36 1.8 5 11 Type of s chool 0.2 1.82 11 Religiosity 0.0 2 0.32 14 Differential a ssociation Peer a ssociation 0.04 1.56 14 u se 0.0 2 1.80 14 u se 0 .07 0.46 14 Definitions 0.1 0.02 14 Differential r einforcement Reinforcement b alance 0.0 5 1.36 14 r eactions 0.1 14 r eactions 0.1 1.97 14 Imitation 0.0 4 2.13 14 Note: All predictors are grand mean centered. value <.1; *p value<.05; **p value<.01; ***p value<.001
171 Table 4 28. Model with social structure and s ocial b onding variables for tobacco use Unit specific model p arameter Coefficient se Odds ratio Fixed e ffects Between d istricts (Gus) Intercept 0.2 5 0.083 0.77 Population d ensity 0.0000 3 0.00001 1.00 Percent on p ublic w elfare 0.02 0.014 1.02 Residential mobility 0.0 3 0.050 0.9 7 Within d istricts (Gus) Sex 1.48 *** 0.229 4. 42 Type of s chool 0. 68 ** 0.169 0. 50 Religiosity 0.0 1 0.065 1.0 1 A ttachment 0.2 7 0.106 0.75 Belief 0.1 0 0.076 0.90 Commitmen t 0.068 1.14 Involvement 0.4 3 ** 0.123 0.65 Variance D.F. Random effects Between districts (Gus) ( 00 ) 0.04 7.12 1 1 Sex 0.3 13.95*** 14 Type of s chool 0.07 1.31 14 Religiosity 0.01 1.63 114 Attachment 0.09 0.44 114 Belief 0.02 2.73 114 Involvement 0.1 0.31 114 Commitment 0.01 0.82 114 No te: All predictors are grand mean centered. value<.1; *p value<.05; **p value<.01; ***p value<.001
172 Table 4 29. M odel with social structure and s elf control variables for tobacco use Unit specific model p arameter Coefficient se Odds ratio Fixed e ffect s Between d istricts (Gus) Intercept 0.22* 0.078 0.79 Population d ensity 0.0000 2 0.0000 1 1.00 Percent on p ublic w elfare 0.006 0.0 15 1.00 Residential mobility 0.09 0.0 52 0.91 Within d istricts (Gus) Sex 1.13 *** 0.244 3.12 Type of s chool 1.03 *** 0.158 0.35 Religiosity 0.01 0.049 0.98 Self control 0.52** 0.0 52 0.58 Variance D.F. Random effects Between districts (Gus) ( 00 ) 0.02 7.22 1 1 Sex 0.4 15.99*** 14 Type of s chool 0.0008 2.24 14 Religiosity 0.00005 0.42 14 Self co ntrol 0.0001 0.31 14 Note: All predictors are grand mean centered. value<.1; *p value<.05; **p value<.01; ***p value<.001
173 Table 4 30. Full m odel comparing the relative mediating effects of s ocial l earning, s ocial b onding and s elf c ontrol variables f or tobacco use Unit specific model p arameter Coefficient se Odds ratio Fixed e ffects Between d istricts (Gus) Intercept 0.14 0.148 1.15 Population d ensity 0.00001 0.00002 0.99 Percent on p ublic w elfare 0.02 0.039 1.0 2 Residential mobility 0.0 8 0.103 0.92 Within d istricts (Gus) Sex 0.53 0.22 1.85 Type of s chool 0.20 0.401 0.77 Religiosity 0.05 0.128 1.02 Differential association Peer a ssociation 1.85*** 0.197 6.37 u se 0.05 0.127 1.05 u se 0.01 0.177 1.02 De finitions 0.42* 0.156 1.53 Differential reinforcement Reinforcement balance 0.63** 0.187 1.89 r eactions 0.15 0.199 1.24 r eactions 0.18 0.167 1.24 Imitation 0.54** 0.151 1.72 Attachment 0.08 0.185 0.91 Belief 0. 01 0.1 67 1. 01 Commitment 0. 34 0.159 1.41 Involvement 0.2 8 0. 211 0.7 4 Self control 0.04 0.1 60 0.9 6 Variance D.F. Random effects Between districts (Gus) ( 00 ) 0.05 0.18 11 Sex 0.4 0.72 14 Type of s chool 0.4 0.45 14 Religiosity 0.0 3 0.13 14 Differential a ssociation Peer a ssociation 0.1 2.13 14 u se 0.03 0.62 14 u se 0.1 0.11 14 Definitions 0.09 0.39 14 Differential r einforcement Reinforcement b alance 3.17 14 r eactions 0.2 2.46 14 Parent r eactions 3.47 14 Imitation 0.04 2.29 14 Attachment 0.2 0.49 14 Belief 0.1 0. 29 14 Commitment 0.1 0.89 14 Involvement 0. 3 2. 09 14 Self control 0. 1 0.31 14 Note: All predictors are grand mean centered. value<.1; *p value<.05; **p v alue<.01; ***p value<.001
174 CHPATER 5 SUMMARY AND CONCLUSI ONS This chapter summarizes findings from the models run for use of alcohol, depressants, and tobacco. The findings from several HLM models for alcohol and depressants use and HGLM models for tobacco use will be interpreted and conclusions stated in relation to the hypotheses of this study. Moreover, the limitations of the current study and the implications for future research are discussed in this chapter. Variations in Alcohol, Depressant and Tobacco Use across Gus depressant and tobacco use found supportive evidence for Hypothesis 1which states that alcohol, tobacco and depressant use will significantly vary acr oss districts (Gus). The uses of these substances varied significantly vary across fifteen Gus (districts) in Busan, South Korea (Tables 4 4, 4 13, and 4 22). These findings are consistent with previous studies examining substance and drug use variations b etween districts, communities or schools in American studies (Holland Davis, 2006; Ennett et al., 1997; Krohn, Lanza Kaduce & Akers, 1984) and the relatively small size of the differences across macro level group or geographical units is consistent with wh at is typically found in empirical studies (Reisig & Parks, 2000; Sampson & Bartusch, 1998; Sampson, Morenoff, & Earls, 1999). Therefore, although the variances are small, they are sufficient to allow testing the SSSL model and the comparable models using social bonding and self control variables. The majority of the variations in substance use among adolescents in Busan is found in between individual differences. The differences in variation between groups and between individual sources are consistent with findings in previous studies (Holland Davis, 2006; Ennett et al., 1997).
175 Social Structural Dimensions of Use of Alcohol, Depressants and Tobacco Among Adolescents in Busan Following three unconditional ANOVA models that estimated between Gus differenc es for each of three substances, three social structural models were run that included social structural variables to assess their impact on differences in alcohol, depressant, and tobacco use. These models examined Hypotheses 2a, which states that social structural variables such as population density, residential mobility and percent of residents on public welfare will be Hypothesis 2b, which states that male adolescents, s tudents attending industrial type of school, and students with low level of religiosity (social structural variables) will be more likely to use alcohol, depressants and tobacco. These models were analyzed a as a prior step to examining the central proposi measured at the individual level will mediate the effects of the structural level variables 5, 4 14, and 4 23). The findings support th e conclusion that some social structural dimensions were social structural variables representing differential location in the social structure and differential loca tion in primary and secondary groups, such as sex, type of school and 2b). With regard to the community level measures of differential social organization and theo retically defined dimensions of social structure, two of the variables, population density and residential mobility were significantly associated with adolescents drinking behavior but percent on public welfare was not (Hypothesis 2a). Also, two community level variables (population density, and residential mobility) and one measure of differential location in the
176 support for Hypotheses 2a and 2b. The model examin partial support for the Hypothesis 2a and 2b. The findings on this model revealed that population density, residential mobility, sex and type of school, were significant predictors of The conclusion is that population density was a consistent and positive predictor for alcohol, depressant and tobacco use among adolescents. These findings seem to be consistent with some previous studies ( Sundq u ist & Frank, 2004); however, there are als o still some differences from previous studies that found positive relationships between population density and adolescent use of marijuana and other illicit drugs, but not with alcohol and tobacco use (Holland this finding from South Korean youths support the arguments by Osgood and Chambers (2000) that communities with greater population density pose greater opportunities for youth to engage in delinquent behaviors such as substance use. Residential mobility was also a statistically significant predictor of all three types of substance use, but the direction of the effects was not consistent in the three models. Residential mobility was positively associated with depressant use, but it was negatively associa ted with alcohol and tobacco use. That is, adolescents living in Gus where higher levels of residential mobility are more likely to use depressants but less likely to use alcohol and tobacco. These findings suggest that residential mobility does not have c onsistent negative influence on substance use for adolescents, as predicted by social disorganization theory ( Sampson & Groves, 1989) These findings are consistent with previous studies that found some mixed effects of residential mobility or residential instability ( Clark & Loheac, 200; DeWit, 1998; Ennett et al., 1997). Overall, community level social structural predictors, which are at the most macro level and more distal causal factors, such as population density, percent on public welfare and
177 r esidential mobility have less predictive power than differential location in the social structure and differential location in the primary and secondary groups variables (which are more meso level elements of social structure. Again these findings are not surprising and are consistent with previous research findings that structural predictors are often only weakly associated with individual behaviors (Holland Davis, 2006, p.103; Verill, 2008). Sex of the respondent was a consistently significant and strong predictor for all three types of substance use. Such a strong gender effect is consistent with previous studies (Landza Kaduce et al., 2006; Lee, et al., 2004; Holland Davis, 2006; Verill, 2008). More interestingly, gender effects were different by type o f substances. Male students were significant more likely to use alcohol and tobacco than female students, but female students were significantly more likely to report higher level of depressant use compared to male students. It is not clear how much these findings for depressant use are in line with previous studies (Jensen, 2003; Mears et al., 1998; Piequero et al., 2005; Sevensson, 2003). Type of school attended by the respondents (industrial or liberal high school) was also a significant predictor for a lcohol and tobacco use, but not depressant use, among South Korean adolescents. Such findings appear to be relevant to the gender composition of liberal and industrial schools. In this sample, 59.6 percent of students were attending liberal type of school. Among those students attending liberal high schools, 20.6 percent were female students. That might attenuate the differences of type of schools in using depressants. Generally, this type of school effect is consistent with findings from South Korean resea rch 2007). For example, the National Youth Commission (2005) found that students attending industrial schools more frequently use alcohol and are at greater risk of binge drinking than students attending liberal type of schools. South Korean studies also suggest that students attending industrial schools are more likely to be under greater peer pressure for using
178 substances, more exposure to role models using substanc es compared to counterparts and less supervision by parents (Kim et al., 1990; National Youth Commission, 2005; Yun et al, 1999). Therefore, the type of school where students attend may provide certain circumstances that certain social learning process may operates to lead increased or decreased adolescents engagement in substance use. Furthermore, this finding may indicate that type of school may be an indicator of social status for adolescence years. That is, generally students attending industrial scho ol have lower grades and their parents tend to be in lower levels of socioeconomical status compared to the students attending liberal school. Therefore, students attending industrial school may perceive inability to achieve academic or occupational aspira tion than students attending liberal school. That is because it is expected that students attending industrial schools are headed for blue collar jobs after graduation, and will earn less income compared to the students graduating from liberal schools who have more opportunity to achieve college level education or more and expect to earn more income and respects. According to Braithwaite (1981), self report studies may find more significant findings in studies of adolescents if the social class measure actu social class. Therefore, type of school can be considered as an indicator measuring expectations from t ype of school where students attending. In this sense, this measure also would be considered as one of theoretically defined variables indicated in the SSSL theory in South Korean context. Unlike some American studies found which found evidence for the ne gative religiosity insulate them from using substances (Holland Davis, 2006, Johnson et al., 2000). In the bivaraite analysis, religiosity was negatively associated with alcohol use but it was not
179 significantly associated with depressant use and tobacco use. This may reflect in part a weak measure of religiosity. The measure of religiosity here imply how often students go to church or temple. The mean frequency of chu rch or temple attendance was less than once a month, and this question does not directly ask if the respondent attend for worship services or other purposes. On the other hands, other studies that have found significant effects of religiosity used differen t question in operationalizing the variable. For example, Holland Davis (2006) measured religiosity by using question requiring respondents to assess how religious they are (p.67). Furthermore, some studies suggest that the concept of religiosity is compos ed of multiple components such as emotion, knowledge, and behaviors (Cornwall, Albrecht, Cunningham & Pitcher, 1986). Finally, the between Gus variance of the two structural level models for alcohol and depressant use (but not tobacco use) was not signifi cant models were run including all of the social structural variables (population density, residential mobility, percent on public welfare, sex, type of school, and religiosity). In fact, when the two structural level models analyzed the intercept only mo dels containing the three community level predictors only, such as population density, percent of residents on public welfare and residential mobility, the between Gus variations in alcohol and depressant uses models were all statistically significantly. M ore specifically, for example, the models containing population density, percent on public welfare, and residential mobility explained about 62 % and 15 % of the between Gu variance in alcohol and depressant use respectively. However, once sex, type of sch ool, and religiosity are entered into the structural models, the significance of the between variance at district level were disappeared. Social Learning, Social Bonding and Self Control Variables in Multi Level Analyses The next step in the research was to Hypothesis 3 drawn from social learning, social bonding and self control theories. The models testing Hypothesis 3 examine which of the
180 major variables taken from each the three theories have the most apparent mediating effects on the structural relatio nships. The findings fairly substantially support the conclusion that the social learning variables have the most consistent and strong mediating effects. First, social learning theory received strong support with regard to all three substances use (see Ta bles 4 6, 4 15, and 4 24). Among the social learning variables, differential peer association measures indicating association with friends using each of the substances (alcohol, depressant, and tobacco) were found to be the strongest predictors for all thr ee substance models This finding is consistent with a large body of previous study findings (Akers & Sellers, 2009; Pratt et al., 2000, 2006; Warr, 2002). Interestingly, this study found that imitation was one of the consistently strong and significant e xplanatory variable for adolescents substance use in the three models. This is a research, that the influence of imitation is relatively weaker than that of other social learning variables. This may be due to the modifications in the measure of imitation used here from ons on TV, or advertisement) has behavior of observing others using substances than do previous measures which asked about own substance use. This study found that parental use of substances (which is intended as a measure of differential association but is used in some studies as a measure of imitation of parents) affected the use b y their adolescent children but the influence differed by type of addi
181 findings are consistent with previous South Korean research that reported higher level of cohol use problems. Moreover, studies contains narrative interview information reported that fathers sometimes support drinking or smoking by their male children while they do not encourage using alcohol and tobacco for their female children ( Kang & Kim, 2 005, National Youth Commission, 2005). In previous structural level models, it was found that girls are more likely to use depressant while boys are more likely to use alcohol and tobacco. That indicates t relevant influence on the same sex children. Additionally, these findings would suggest that there are some other theoretical interpretations, such as biological theories and general strain theories besides social learning theory, possibly applicable for interpreting these findings. Therefore, these findings suggest that more scrutiny is necessary for future studies. For example, future studies examining separately b y gender of their children and type of substance use and apply different criminological theories. substance use and these findings support Hypothesis 3 for the socia l learning theory. These findings are consistent not only with previous American studies but also other cross cultural studies testing social learning theory (Akers, et al., 1979; Hwang & Akers, 2003). The between district variance in mean alcohol, depres sant and tobacco use were not significant with eight social learning variables in the three models That means none of the social learning models examining random variance components of the social learning variables are significant and, therefore, the rela tionships between these variables and individual alcohol, depressant and tobacco use do not significantly differ across Gus. In other words, while social learning variables mediate structural effects, Gus do not moderate the
182 effects of the social learning variables on individual behavior. Social learning variables explain most of between district differences in substance use. That is because social learning models considerable reduced the between Gus variances in substance uses from the random ANOVA models. These conclusions differ somewhat from those of Holland Holland Davis (2006) found significant between school variance from the multi level social learning models in examining drug and substance use and reported that the some relationship b etween some social learning variables and for marijuana and overall illicit drug use significantly differ by schools and neighborhoods. She concludes that the finding of interaction effects of social learning variables with community characteristics shows moderating effects on the impact of social learning processes for some drug use behaviors. On the other hands, this current study found that the social learning process works in the same manner across social contexts in the three substances in Busan, South Korea. The According to the calculation of the reduction in between Gus variance accounted for by the individual level social learning variables, social learning variables account for about 62% of the variance in student level frequency of alcohol and depressant use. These within district learning variables explain around 60 % of the variat ion in alcohol and depressant use. Next, the findings from the hierarchal models testing the mediating effects of social bonding and self control variables lead to a conclusion of mixed support for Hypotheses 3 (see Table 4 7, Table 4 16, and Table 4 25) Belief and involvement were significant predictors for tobacco use, but none of the social bonding variables were significantly associated with depressant use. The models r evealed that greater attachment to friends and parents, spending more time for study, and higher level of approval for general law and
183 directions of these significan t relationships are in the direction expected by social boding theory they were generally weak. Compared to the social learning variables, the social bonding variables explained very little (about 7%) variance in alcohol and depressant use. With regard to between Gus variances, influences of some social bonding variables differ by community context. The relationship between attachment and alcohol use were different across Gus. That may suggest that Gu had moderating effects on the relationship between socia l bonding and substance use but testing interactions or moderation effects for these and other variables are left to future research. Also, the relationship between involvement and tobacco use varied across Gus. This finding also suggests that Gus with hig h mean of log odds of tobacco use tend to have strong relationship between time spent studying and the odds of individual tobacco use. These findings imply that social bonding variables interact with community context and, therefore, operate differently de pending on average levels of use for alcohol and tobacco use. That is, the impacts of attachment and involvement are different across districts, again suggesting that the influence of these variables may be moderated to some extent by the structure of the districts. More specifically, attachment is more important in predicting alcohol use in Gus that have high levels of alcohol use and less important in Gus with low levels of alcohol use. Also involvement is more important in Gus where students mean use of tobacco is high, while it is less important in Gus with low level of tobacco use. In addition, although the between Gu variance in mean alcohol, depressant and tobacco use were significant for the social bonding models, the variance have not been reduced f rom the unconditional model, indicating social bonding variables mediate very little of the between district variances in adolescents substance use. S elf control, on the other hand, was significantly and negatively associated with pressant and tobacco use as predicted by the theory supporting
184 Hypothesis 3 (see Tables 4 8, 4 17, and 4 26). Individuals with lower self control are more likely to be engaged in increased alcohol, depressant and tobacco use. However, the amount of varianc e explained by self control for each substance was not great. The self control models explained only about 9 % the variance of student level alcohol and depressant use. The three self control models found that the between district variance in each substanc e use are significant, indicating the influences of self control variables on substance use vary significantly across Gus. However, the variance of the Gu specific slopes for the relationship between self control and each of the three substances were not s ignificant. That is, the influence of self control variable on substance use is the same across Gus regardless of the differences of level of substance use of each district. Also, as with the case of social bonding variables, the between Gu variance in mea n alcohol, depressant and tobacco use have not been reduced from the unconditional model, indicating those self control variables explain very little of the between Gu differences in adolescents substance use. Again, it is reasonable to conclude that self control has limited mediation of the effects on substance use by differences across Gus. The SSSL Model and Models with Social Structure and Social Psychological Variables All the social structural variables used in the previous structural level models w ere included in each set of social learning, social bonding, and self control models to test Hypothesis 4. Hypothesis 4 states that the relationships between substance use and all of the social structural variables will be partially and substantially media ted by social leaning variables, social bonding variables, and self control variables. More specifically, these models not only examined the SSSL theory but also expanded the SSSL to include other social psychological variables. Therefore, comparing social structure and social learning models with social structure and social bonding models and social structure and self control
185 models allows conclusions about whether other social psychological variables better able to mediate social structural effects than s ocial learning variables. With regard to the SSSL models that used social structural variables and social learning variables, all three SSSL models for alcohol, depressant and tobacco use provide strong support for Hypothesis 4 (see Tables 4 9, 4 18, and 4 27). As the SSSL model predicted, alcohol, depressant and tobacco use. In all the three multilevel analyses models, social learning variables substantially re duced the magnitude of the coefficients of all structural predictors, which was significant in the structural level models, towards zero and made them insignificant. For the alcohol use model, the coefficients for population density, percent of residential mobility, type of school and religiosity were reduced and became insignificant with social learning variables in the model. Gender effects (sex) remained significant but the magnitude of the coefficient is reduced substantially. Therefore, gender effects are not fully mediated but they are consistent with SSSL theory substantially mediated by social learning variables. Moreover for the depressant and tobacco use models, the coefficients for all of the social structural variables including sex not only were substantially reduced but are rendered statistically insignificant. The three models generally revealed that social learning variables, as compared with social bonding and self control variables, mediated the four social structural dimensions of different ial social organization, theoretically defined variables, differential social location and differential location in primary and secondary groups successfully. Such conclusions are generally consistent with those reached by previous researchers testing the SSSL model with American samples (Holland Davis, 2006; Lanza Kaduce & Capeace, 2003; Lee et al., 2004). This supports the conclusion that the mediation of structural effects on adolescent deviancy by social learning variables is not confined to the United States where the previous research has been done. In fact, the mediation effects of the
186 social learning process found in this study among Korean adolescents were stronger than found in previous studies directly testing the SSSL theory in America (Holland D avis, 2006; Lanza Kaduce & Capeace, 2003; Lanza Kaduce et al, 2006; Lee et al., 2004; Gibson et al., 2010; Verill, 2008). This flies in the face of the assumption that to the extent a theory is culture bound it would apply more appropriately in the society in which it was developed. Some have concluded that gender effect are not mediated by social learning variables but rather interacts with or moderates the effects of social learning variables (Mears, Ploeger & Warr, 1998; Piquero, Gover, MacDonald & Pique ro, 2005). That is, the learning process is with alcohol use friends and differential reinforcement of alcohol use than girls or vice versa. Since the structural m odel for alcohol use revealed that boys are more likely to use alcohol than girls, the significant gender effects on drinking in the SSSL model may support the argument. Also, the depressant use model found that girls are more likely to use depressant than boys, the insignificant gender effects in the SSSL model for depressant may be linked to the gender effects that operate differently in the social learning process. But this argument is not supported by the finding of strong mediation effects of social le arning variables on gender and little support for moderation of social learning by gender for the three substances among South Korean youths compared to American studies. Furthermore, this study mostly found significant mediation effects rather than moder ation effects compared to some previous studies (Holland Davis, 2006; Verill, 2008). For example, Holland Davis (2006) found some moderation effects of the social learning verty on alcohol use increased and became significant which was not significant in the structural which is unmasked when the social learning variable are contro
187 found that in the SSSL model for alcohol use, age and gender remained significant, while the coefficient of socioeconomic status increased and became significant (it was not significant in the structural level only model) (se e, p.106). Another direct test of the SSSL model by Verill (2008) also reported most of moderation effects rather than mediation effects of social Overall, this current study provides as strong or stronger evidence for the mediation proposition of the SSSL theory than any other previous studies. These findings generally factors are in some way impac ting the normative and cultural climate in which the learning Davise, 2006, p.106) On the other hand, conclusions about the social structure and social bonding and self control models wi th regard to Hypothesis 4(see Tables 4 11, 4 20, and 4 29 for social bonding variables, see Tables 4 10, 4 19, and 4 28 for self control variables) are more limited. Indeed, one could conclude that these models are not supported. Rather some findings may s uggest that there may be some moderation effects in addition to or rather than mediation effects For the all three substance use models, most of structural variables remained significant and the magnitude of coefficients were either unchanged or increase d by including social bonding variables, except for religiosity and residential mobility which became insignificant for alcohol use model and for tobacco use model, respectively. The coefficient for gender effects increased by including social bonding vari ables for alcohol and tobacco use models. Particularly, social bonding variables appear to neither mediate nor moderate any structural factors in the depressant use model. Notably, percent on public welfare which was not significant in the structural level model became significant when social bonding variables are controlled in this model. Moreover, some of variables which were not significant in the social bonding models became significant when social structural
188 variables were taken accounted for in this m odel. For the alcohol use model, the magnitude of coefficient for belief slightly increased and it became significant. Also, belief and commitment became significant in the social structure and social bonding model for tobacco use. These findings support t he conclusion that social bonding variables do not mediate social structural variables. Hypothesis 4 regarding social structure and self control also has little support. Most of the social structural variables remained significant after self control is ta ken into accounted for the models. Only one or two social structural variables are mediated by self control variable: religiosity for alcohol use model, population density for depressant use model, and population density and residential mobility for tobacc o use model. Additionally, however, unlike the social structure and social bonding models, there is little suggestion in these findings that there are moderation effects in the social structure and self control models. Additionally, the regression coeffici ents of these significant structural level predictors have been slightly reduced in the models while self control variables remained significant. Therefore, the social structure and self control models have weak mediation effects compared to the social lea rning variables in the SSSL models. Overall, these social psychological variables taken from social bonding and self control theory do not mediate, or do so weakly, social learning variables do in this study. Moreover, these findings suggest that social bonding variables may be affected significantly by differences of district context. The between Gu variance in mean substance use of all three social structure and s ocial bonding models and all three social structure and self control models became insignificant. Also, these models substantially reduced the variance of between Gu from the unconditional random ANOVA models. That means these models explained a lot of bet ween Gu differences in each of substance use. However, these reductions in between district difference variations
189 may be due to the influence of the social structural variables that explained considerable portion of between Gus differences in substance use in structural level models. Comparison Models of Social Learning, Social Bonding and Self Control Variables This final set of models are full comparison models containing all social structural variables as well as all social learning, social bonding and self control variables. This model aims to examine Hypothesis 5 that predicts relatively stronger mediation impact of social learning variables on the social structural variables on each substance use compared to other social psychological variables. It is expected that social learning variables which have relatively stronger mediation effects would exert significant effects on each type of substance use with greater magnitude of regression coefficients than social bonding and self control variables. The f indings from the three full comparison models for alcohol, depressant and tobacco use support Hypothesis 5: Social learning variables have greater mediation effects than the other two sets of social psychological variables. In the three comparison models, most of the structural predictors were substantially reduced toward zero and became insignificant. Sex was significant for the comparison models for alcohol and depressant use. In fact, the significance of sex variable in the depressant use comparison mode l is not consistent with the SSSL model for depressant use because the sex variable became insignificant after social learning variables mediated the gender effect. However, the significance of sex may be attributed to the moderation effects of social bond ing variables included in the comparison models. Furthermore, most social learning variables maintained significant effect coefficients found in the social learning only models and the SSSL models for the three substance use. The relationships between dif ferential association with substance using friends, definitions favorable to substance use, differential reinforcement of each of substance use and having
190 more role models using substances were consistently significant predictors for all the three substanc e uses in the comparison model with all other variables entered. These social learning variables in the three comparison models remain significant and the magnitudes of coefficients were relatively unchanged from the social learning only models and the SSS L depressant use also remained significant in the full comparison m odel predicting depressant depressant comparison models but not for tobacco use model. Also, consistent with previous models testing the SSSL models and the social l earning only models, differential association with substance using friends was the strongest predictor among the all individual level predictors in all three substance use comparison models. Overall, the effects of social learning variables in the compari son models mirrored the results of the social learning only models and the SSSL models for alcohol, depressant, and tobacco use. Such strong influences of social learning variables in the comparison models strongly support Hypothesis 5. On the other hand, social psychological variables do not appear to mediate social structural variables as they have shown in the models with social structure and social psychological models. Moreover, the results for social bonding variables in the comparison models are mor e confusing. First, most of social bonding variables became insignificant in the models with social learning and self control variables. For alcohol use comparison model, none of social bonding variables remained significant. For depressant and tobacco use comparison models, only one variable for each model, involvement for depressant use and commitment for tobacco use, were significant. Yet, the coefficients of these significant variables are substantially reduced from the social bonding only models and mo dels with
191 social structural variables. These findings indicate that social bonding variables do not have mediation effects. Also, some of the measures became significant in the final model were not a significant predictor in the social bonding only model ( commitment in the tobacco use model, and involvement in the depressant use model were not significant). As suggested above, these findings may suggest some moderation effects of social structural variables on social bonding variables in the comparison mode ls. However, it is not clear because this study does not explicitly examine moderation effects of the social psychological theories on the relationship between social structural factors and individual substance use behaviors. Therefore, it is left to futur e research to consider more directly examining not only mediation effects but also moderation effects of these social psychological variables. In addition, the coefficients of commitment in tobacco use in the comparison model are substantially increased co mpared to the social bonding only model and the model with social structure variables for tobacco use. Moreover the direction of the effects of commitment in the comparison model for tobacco use was in the opposite direction from the theory prediction. As such, social bonding variables revealed somewhat inconsistent patterns in the series of multilevel models. Overall, the conclusions are that there are relatively and substantially stronger mediation effects of the social learning variables on the relation ships between social structural variables and substance use than self control and social bonding variables. Implications, Contributions, and Suggestions for Future Research This study tested the main proposition of the SSSL theory by testing five Hypothes es. Overall, the study found strong support for the SSSL theory contentions. As hypothesized, districts. Most social structural variables representing the four soci al structural dimensions in
192 tobacco use. Importantly, these significant social structural influences were substantially mediated by social learning variables reflectin g the major four concepts, differential association, definition, differential reinforcement and imitation. The mediation effects held for all the three substance uses and in comparison with other social psychological variables, such as social bonding and self control variables. Most interestingly, the mediation effects of social learning process found in this Korean study was stronger with regard to certain structural variables than has been found in previous American studies. That is, because most of th ese American studies reported not only mediation effects but also some moderation effects of social learning variables on several structural variables, or the other way around. That is, social structural variables conditions the way of social learning vari ables impacts individual behaviors. Particularly, studies noted considerable moderation effects of gender that interact with social learning variables (Landza Kaduce et al., 2006; Lee et al., 2004; Holland Dvais, 2006; Verill, 2008). However, in fact, the direct gender effect on alcohol use was found in the SSSL model, yet not for depressant and tobacco use models. The gender effects, although substantially reduced, al learning variables were included. This is consistent with previous studies that found strong p.30). Lanza Kaduce et al (2006) found that gender effect were not as strongly mediated by Kaduce et ediated like the effects of other social structural variables. However, their study found significant gender interaction effects with other social structural variable which does not fully support for the SSSL model. Therefore, they suggest that the SSSL hy
193 Similarly, Holland Davis (2006) also found some evidence of moderation effects on soci al learning variables. In her multilevel analyses, she found that variations in social learning influence on individual drug and substance use were different depending on characteristics of the social structure across schools. She also found that some soci al structural variables, such as social status and gender were not mediated as much by social SSSL theory (pp.110 111). treatment on the outcome depends on the values of a mo an individual difference variable, then it would mean that the mediating process that intervenes between the treatment and the outcome is different for people who differ on that individual difference. If the moderator i s a contextual variable, then it would mean that the Dvais, 2006, p.111). That is, according to Holland Davis (2006), social learning mediates social structure influences and, at the same time, s ocial structure moderates the social learning influence on behaviors (p.110). These are reasonable suggestions for future research to consider, but in the current study there was not much support for structural moderation of social learning but there was evidence for such moderation of social bonding and self control variables that were not included in the earlier research. Also, there is a difference in findings in that for two of the three substances, depressant and tobacco, in this Korean study, gender effects were reduced to non significance, and even for the alcohol use for which it remained statistically
194 significant, its initial effect was quite substantially reduced. According to South Korean haracteristics of social context, particularly family context, are significantly associated with the initiation and the frequency studies found that boys frequently report that they are sometimes encouraged to drink with father or other adults. Also boys have more opportunity to drink alcohol when there is a family traditional serv a family to remember and honor their forefathers, such as great grandfather or grand fathers who are already passed away. This service is based on the belief that dead ancestors may protect and bless their offspring so that some families still practice Jesa at least several times a year. Jesa service is managed by men in the family only and, men are s upposed to drink a glass of alcohol as one of the procedures of this Jesa, while women do not allowed to be a part of this Jesa service. Research also reported that some male adolescents are reported to drink alcohol after encouragement by fathers and othe r adults in their families. Needless to say, most of the cases, girls are not encouraged to drink alcohol. In addition to this Jesa case, male adolescents reported various occasions that parents or other adults allow them to drink small amount of alcohol b ut prohibit drinking by girls in their families. Therefore, the cultural normative social context encourages drinking by boys or, at least, do not consider it as problematic behaviors, but discourage drinking by girls is an example where cultural practices and traditions produce gender differences in drinking. This example fits well with the SSSL model because it is illustrates the way in which sociocultural contexts operate through social learning processes to affect individual behavior reflected in gender differences in rates of behavior. From a social learning perspective differences in gender rates of drinking
195 in this cultural context is produced by the operation of the social learning process in which differential reinforcement (encouragement from fathe rs and other adults), modeling (being in the company of admired adults who are drinking), and promotion of definitions favorable to drinking by boys but not for girls based on tradition of Jesa. The circumstances found in the South Korean context, therefor Taken together, the gender effects on the social learning process of substance use behaviors as well as the relationship between other social structural variables and social learning variables on various outcome behaviors may need further investigations in the future studies. The direct gender effects are substantially mediated by social learning variables in the SSSL models. Further, all social structural variables, including gender, were mediated should be noted that other research has produced different findings. This may result from differences in the dependent variables, measures of independent variab les, and other methodological differences. It may be that the model works better for less serious deviance, such as adolescent substance use and not as well with more serious criminal behavior, such as violence in which there are stronger gender and other structural effects. Moreover, this current study does not aim to test moderation effects of social structural variables on social learning variables as previous studies do (for example, Lanza Kaduce et al., 2006 and Verill, 2008). No interaction terms were included in the models and moderation effects are only suggested not found by analyses in this dissertation. More research is needed to develop a larger body of findings to see well the SSSL model or variations on it supported by research on a range of de viant behavior and in different societal, social, and cultural contexts.
196 Comparison of Mediation Effects of Social Learning and Other Social Psychological Variables The research in this dissertation is done in line with the suggestion by Lee et al (2004) t hat future studies need to examine not just the mediating effects of social learning variables but the mediating effects of other theoretical variables. This research follows the suggestion ocial learning variables, good measures of other potential mediating processual or micro level variables. The most obvious of these would be social bonding, or other social psychological or personality his colleagues conducted before the formulation of the SSSL model produced findings that in retrospect are relevant to testing the relative mediating effects of social learning compared to other potential mediating variables. They found that social learnin g variables had stronger mediating effects on the relationship between community context and adolescent substance use than did social bonding variables. Consistent with these findings, the present study also found that social learning variables demonstrate greater mediation effects on the relationship between social structural variables with regard to the mediating effects of self control variables. Social bonding and self control variables did not mediate social structural influences or did so only weakly. Consequently, the conclusion is that the SSSL model is supported both on its own and in comparison with how much social structural effects on the same se t of dependent variables are mediated by variables from other theoretical models perspectives. Furthermore, the analysis findings in this study suggest that the mediating effects of social bonding variables may be moderated to some extent by social struct ural variables. Therefore, Holland
197 be reasonable (p.112). Such findings have implications for future research which should address the issue of which social structural relationships are substantially mediated, which are not mediated, which variables are moderators and so forth in relation with not only to social learning variables but also other s ocial psychological theoretical variables as well (p.139). In addition, future research should examine these issues in cross cultural contexts in different societies. The methodology and findings from this dissertation study, therefore contributes to and goes beyond the existing literature on the SSSL model and should provide some background and guidelines for this future research on and the development of the SSSL. The series of multi level analysis models for three different dependent variables in this study addresses methodological issues that occur when uses cross level data. The current study significantly extended the testing and developing of the SSSL theory by comparing the mediation effects of social learning variables with social bonding and sel f control variables on the same set of depended variables. In this way, the mediation effects of social learning variables were appropriately interpreted by comparing the mediation effects of other social psychological theory variables. Most of all, this current study provides stronger evidence for the mediation proposition of the SSSL theory than previous studies. These findings do not eliminate the issues regarding development of the SSSL model to incorporate structural moderation of social learning pro cesses as well as social learning mediation of structural effects. Yet, this current study found mediation effects of the social learning variables; even in the case of gender effects for which the issue of moderating effects have most often been raised, t he social learning variables substantially mediated the effects of gender on all of the dependent variables.
198 Finally, this present study utilized a sample of South Korean adolescents to examine the genalizability of the SSSL model in cross cultural contex t. As explained earlier, none of the previous studies has examined the SSSL model with a sample drawn from non Western culture. The evidence supporting the SSSL theory in a sample of South Korean youths, therefore, is also supportive of the contention that the SSSL model is generalizable beyond American society. Consequently, t his current study contributes significantly to the still small Limitations of the Current Research and Implications f or Future Research As shown, the methodology, findings, and conclusions of the present research provides implications for what future research should address. However, there are limitations of the present study that could be overcome in future research. On e of the limitation s is the small num ber of districts (15 Gus) at the macro level Because the secondary data utilized in the study were collected from only one city, it was not possible to include a larger number of macro level units. In hierarchal models the optimal number of such units is 30 or more. The relevant issue of using less than 30 sites is that it may reduce the statistical power in multi level analysis. In general, it is well known that the larger the variation in the treatment impact across s ites, the more sample sites (level 2 sample) are needed to maximize power in examining the average effect of treatment (Mass & Hox, 2004). Also, the limited number of districts limited the number of available structural level variables in HLM and HGLM anal ysis models. In multi level analysis, having one predictor per ten sample sites or individual samples is used as rule of thumb. That is, even the data used in this study include various types of social structural variables representing four social structur al dimensions proposed in the SSSL theory, in the analysis models, only one structural variable per one or two of social structural dimensions was able to be included in the analysis models (i.e., population density for differential social organization, re sidents on public welfare and
199 residential mobility for theoretically defined variables). Therefore, it is important in future research to utilize a larger number of level 2 units to enable a larger number of measures of structural variables in testing the SSSL model Better measurements for the level 2 units than were available in the census data for this study should be sought. Although this study had measures of all four social structural components in the SSSL theory, the measures utilized in this study were not an ideal. For example, location in social groups is not well indicated by the measure of religiosity in this study and better indicators of religion affiliation and membership should be devised in future studies. Also, this study does not include a measure of collective efficacy which is one of structural measures representing theoretically defined social structural dimension in the SSSL theory. As Akers (1998) suggested the concept of collective efficacy is considered theoretically and empirically related with the SSSL model. In some studies, it has been found deviance and criminal behaviors in American studies (Gibson et al, 2010; Holland Davis, 2006; Verill, 2 009). Therefore, it would be beneficial for future research on SSSL to measure collective efficacy and examine. In addition, other measures of population characteristics beyond population density need to be found. In this study, this structural factor did not show great differences between districts in Busan. The structural variables drawn from the South Korean census data do not explain a lot the variance in behaviors across districts. In addition to stronger measures of the structural independent variabl es, different dependent variables should be included in research. The dependent variables used this study represent rather minor type of deviant acts, alcohol, tobacco use and depressant use among adolescents. Future research should test structural and soc ial psychological effects on various types of criminal and delinquent acts, including more serious types of offending. Future studies should consider collecting data on structural
200 variables in addition to census data. In particular theoretically defined st ructural conditions such as social disorganization might be better measured by interviewing or surveying residents in each neighborhood, district, communities, or perhaps nations (Sampson et al., 1997). Additionally, it might be considerable that using sm aller level of level 2 units such as smaller unit of neighborhood or community level and classroom for the future study. Partially, since the data used in this dissertation has been collected from students attending high schools, it might be able to examin e if there are any classroom differences influencing Using this kind of smaller level 2 units would benefit the current research in a different way if it was po ssible. However, since this current data utilized secondary data and the original researcher does not have classroom identifier for the participants, it was not possible to use the smaller level of unit, such as classroom as level 2 unit in multilevel anal yses. This study compared the findings from South Korean youths to the previous studies used American samples, such as Holland Davis (2006), Lanza Kaduce et al (2006), and Lee et al. (2004) to examine the cross cultural applicability of the SSSL theory. This attempt is very valuable, but the next step would be to conduct more direct comparisons by collecting data from multiple country sites with the same measures and instruments in order to test the validity and applicability of the SSSL theory. Those ins truments should allow for testing effects of other social psychological variables beyond the social learning, social bonding, and self control variables and theoretically defined structural variables beyond the social disorganization in this study. Ideally future research should also include measures of strain, labeling, and personality factors as well as indicators of conflict, anomie, and patriarchy. Finally, fut ure research needs to pay more attend to additional ways of examining direct and indirect m ediation effects of the social learning (and other) variables on the social
201 structural influences. For example, more direct investigation of mediation estimation is also applicable for multilevel modeling. Raudenbush and Sampson (1999) conducted HLM analys is to estimate indirect effects of structural level of factors (i.e., poverty) on perceived violence through perceived informal social control with data drawn from Chicago. In an attempt to address measurement error in hierarchical regression models, they adopted the approach that presented measurement error as a level within the hierarchical model. It makes it possible to examine if the social learning variables have more mediation effects or moderation effects on the relationship between social structural influences on individual behaviors more closely Although there is only one study that has attempted this analysis (Raudenbush & Sampson, 1999), future research benefit greatly by utilizing this approach. Because such a rigorous analytic approach focuses on distinguishing mediation effects from moderation effects it may serve better, where it can be used, than the approach used here to confirm or suggest modification in the SSSL model.
202 APPENDIX CORRELATION MATRIX F OR ALCOHOL USE SCALE RELATED EXOGE NOUS VARIABLES Table A 1 through A 3 presents correlation matrixes for each set of exogenous variables run on the three difference substances, alcohol, depressant and tobacco. The three correlation matrixes reveal that there are no correlation coefficients exceeding more than 0.70 between exogenous variables and each dependent variable. These findings indicate that there are no mulitcolinearity issues in the data used for this study.
203 Table A 1. Correlation matrix for alcohol use related exogenous and dep endent variables (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14) (15) (16) (17) (18) (19) (20) (1) 1 (2) .77* 1 (3) .14* .12* 1 (4) .15* .13* .25* 1 (5) .31* .30* .08* .12* 1 (6) .43* .43* .03 .12* .37* 1 (7) .34* .37* .03 .10* .23* .37* 1 (8) .21* .17* .00 .16* .19* .26* .22* 1 (9) .49* .50* .05 .08* .18* .33* .28* .06* 1 (10) .13* .12* .00 .02 05 .07* .07* .00 .18* 1 (11) .10* .07* .04 .04 .06* .09* .02 .08* .11* .18* 1 (12) .09 .90* .03 .00 .03 .06* .02 .03 .11* .28* .20* 1 (13) .24* .24* .01 .02 .08* .12* .10* .05 .18* .20* .11* .16* 1 (14) .31* .28* .03 .03 .10* .20* .25* .09* .03* .15* .20* /07* .18* 1 (15) .25* .27* .00 .00 .05 .08* .08* .05 .19* .11* .08* .14* .47* .16* 1 (16) .19* .15* .02 .06* .03 .05 .15* .00 .18* .01 .10* .02 .07* .14* .03 1 (17) .06* .04 .13* .08* .03 .03 .00 .01 .01 .04 .06* .03 .12* .01 .06 .04 1 (18) .19* .20* .01 .02 .13 .13* .10* .09* .14* .08* .01 .04 .17* .12* .34* .21* .02 1 (19) .00 .01 .02 .02 .04 .04 .03 .01 .04 .05 .01 .05 .00 .0 0 .06 .09* .02 .18* 1 (20) .02 .05 .02 .02 .04 .04 .02 .01 .00 .02 .02 .02 .07* .05 .17* .08* .01 .41* .45* 1 s reaction;(8) control; (15) Sex; (16) Type of school; (17) Religiosity; (18) Population density; (19) Percent of residents on public welfare; (20) Residential mobility Sources: Hwang (2000), 1999 South Korean Census, 1999 Busan Yearly Statistics Note: all of exogenous variables are standardized as Z scores p< 0.5
204 Table A 2. Correlation matrix for depressant use related exogenous and dependen t variables (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14) (15) (16) (17) (18) (19) (20) (1) 1 (2) .64* 1 (3) .31* .35* 1 (4) .45* .49* .43* 1 (5) .45* .41* .22* .31* 1 (6) .41* .39* .19* .30* .44* 1 (7) .30* .28* .15* .20* .39* .44* 1 (8) .37* .33* .17* .26* .44* .48* .59* 1 (9) .12* .09* .07* .08* .12* .14* .00 .01 1 (10) .00 .01 .03 .04 .10* .02 .03 .02 .18* 1 (11) .03 .00 .03 .01 .00 .07* .05 .08* .11* .18* 1 (12) .00 .00 .04 .01 .02 .02 .00 .01 .11* .28* .20* 1 (13) .02 .01 .07* .07* .01 .05 .05 .07* .18* .20* .11* .16* 1 (14) .06* .00 .05 .02 .09* .04 .00 .00 .39* .15* .10* .07* .18* 1 (15) .00 .00 .05 .04 .05 .03 .04 .02 .19* .11* .08* .14* .47* .16* 1 (16) .22* .29* .04 .13* .08* .15* .15* .18* .18* .01 .10* .02 .07* .14* .03 1 (17) .03 .05 .01 .04 .01 .04 .02 .02 .01 .04 .06* .03 .12* .01 .06 .04 1 (18) .01 .00 .00 .01 .01 .04 .01 .01 .14* .08* .01 .04 .17* .12* .34* .21* .02 1 (19) .03 .03 .01 .01 .04 .06* .00 .00 .04 .05 .01 .05 .00 .00 .06 .09* .02 .18* 1 ( 20) .10* .12* .05 .05 .04 .04 .02 .04 .00 .02 .02 .02 .07* .05 .17* .08* .01 .41* .45* 1 action; (9) Imitation; (10) Attachment; (11) Belief; (12) Commitment; (13) Involvement; (14) Self control; (15) Sex; (16) Type of school; (17) Religiosity; (18) Population density; (19) Percent of residents on public welfare; (20) Residential mobility Sou rces: Hwang (2000), 1999 South Korean Census, 1999 Busan Yearly Statistics Note: all of exogenous variables are standardized as Z scores p< 0.5
205 Table A 3. Correlation matrix for tobacco use related exogenous and dependent variables (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14) (15) (16) (17) (18) (19) (20) (1) 1 (2) .71* 1 (3) .11* .11* 1 (4) .09* .13* .01 1 (5) .44* .46* .11* .08* 1 (6) .49* .52* .06* .07* .45* 1 (7) .28* .32* .00 .11* .16* .21* 1 (8) .37* .33* .17* .26* .44* .48* .59* 1 (9) .50* .56* .04 .13* .40* .45* .00 .17* 1 (10) .16* .15* .07* .08* .15* .16* .03 .08 .18* 1 (11) .06* .03 .00 .01 .04 .12* .05 .05 .11* .18* 1 (12) .03 .09* .00 .01 .07* .09* .00 .04 .11* .28* .20* 1 (13) .24* .28* .13* .10* .11* .21* .05 .16* .18* .20* .11* .16* 1 (14) .28* .31* .06* 11* .21* .30* .00 .13* .39* .15* .10* .07* .18* 1 (15) .26* .34* .09* .11* .08* .23* .04 .19* .19* .11* .08* .14* .47* .16* 1 (16) .29* .31* .02 .01 .13* .18* .15* .21* .18* .01 .10* .02 .07* .14* .03 1 (17) .00 .01 .11* .04 .02 .06 .02 .00 .01 .04 .06* .03 .12* .01 .06 .04 1 (18) .16* .21* .02 .03 .09* .17* .01 .11* .14* .08* .01 .04 .17* .12* .34* .21* .02 1 (19) .06* .03 .00 .02 .03 .00 .00 .06* .04 .05 .01 .05 .00 .00 .06 .09* .02 .18* 1 (20) .00 .04 .05 .01 .00 .01 .02 .02 .00 .02 .02 .02 .07* .05 .17* .08* .01 .41* .45* 1 ion; (9) Imitation; (10) Attachment; (11) Belief; (12) Commitment; (13) Involvement; (14) Self control; (15) Sex; (16) Type of school; (17) Religiosity; (18) Population density; (19) Percent of residents on public welfare; (20) Residential mobility Source s: Hwang (2000), 1999 South Korean Census, 1999 Busan Yearly Statistics p< 0.5
2 06 LIST OF REFERENCES Adlaf, E.M., Korf, D.J., Harrison, L., & Erickson, P. ( 2006 ). Cross national differences in drugs and violence among adolescents: Preliminary findings of th e DAVI study. The Journal of Drug Issues 67, 597 618. Agnew, R. (1985). A revised strain theory of delinquency, Social Forces 64:151 167. ( 1991b ) A longitudinal test of social control theory and delinquency. Journal of Research in Crime and Delinque ncy 28, 126 156. ( 1993 ) Why do they do? An examination of the intervening mechanisms between social control variables and delinquency. Journal of Research in Crime and Delinquency, 30, 245 266. (1994). The techniques of neutralization and violence. Cr iminology 32, 555 580. Akers, R. L. ( 1973 ) Deviant Behavior: A Social Learning Approach Belmont, CA: Wadsworth Publishing. ( 1985 ) Deviant behavior: A social learning approach Belmont, CA: Wardsworth. (1992). Drugs, alcohol, and society: S ocial st ructure, process, and policy. Belmont, California: Wadsworth. ( 1998 ) Social learning and social structure: a general theory of crime and deviance. Boston: Northeastern University Press. ( 2005 ) Sociological theory and practice: the case of criminology Journal of Applied Sociology/Sociological Practice: A Journal of Applied and Clinical Sociology. 22, 24 41. Akers, R.L., & Jenson, G.F. (2006). The empirical status of social learning theory of crime and deviance: The past, present, and future. Pp:37 7 6. In F. Cullen, Wright, J. and Blevins, K (eds.) Taking stock: The status of criminological theory: Advances in criminological theory : New Brunswick, NJ: Transaction Publishers. Akers, R. L., & Cochran, J. ( 1985 ) Adolescent marijuana use: A test of th ree theories of deviant behavior. Deviant Behavior 6 323 346. Akers, 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. Ak ers, R. L., La Greca, A. J., Cochran, J., & Sellers, C. (1989). Social learning theory and alcohol behavior among the elderly. Sociological Quarterly, 30 625 638.
207 Akers, R. L., & La Greca, A.J. (1991). Alcohol use among the elderly: Social learning, comm unity context, and life events. Pp.242 262 in D.J. Pittman and H.R. White (eds)., Society, culture, and drinking patters re examined. New Brunswick, NJ: Rutgers Center of Alcohol Studies. Akers, R. L., & Lee, G. ( 1996 ) A longitudinal test of social lear ning theory: Adolescent smoking. Journal of Drug Issues 26, 317 346. Akers, R. L., & Lee, G. (1999). Age, social learning, and social bonding in adolescent substance use. Deviant Behavior, 19 1 25. Akers, R. L. & Sellers, C. S. ( 2009 ) Criminological theories: Introduction, evaluation, and application 5 th ed. New York: Oxford University Press. Akers, R.L., & Silverman, A. (2004). Toward a social learning model of violence and terrorism. Pp. 19 35. In M.A. Zahn, H.H. Brownstein, and S.L. Jackson (eds. ), Violence: From theory to research. Cincinnati: LexisNexis Anderson Publishing. Akers, R.L., Skinner, W.F., Krohn, M.D., & Lauer, R.M. (1987). Recent trends in teenage tobacco use: Findings from a five year longitudinal study. Sociology and Social Rese arch 71, 110 114. Alarid, L.F., Burton, V.S., & Cullen, F.T. ( 2000 ) Journal of Research in Crime and Delinquency 37, 171 199. Allahverdipour H., Hidarnia, A., Kazamnegad, A., Shafii, F., Fallah, P.A., & Emani, A. (2006) The status of self control and its relation to drug abuse related behaviors among Iranian male high school students. Social Behavior and Personality 34, 413 42 4. Allison, K. W., I. Crawford, P. E. Leone, E. Trickett, A. Perez Febles, L. M. Burton, and R. and neighborhood context. American Journal of Community Psychology, 27, 11 4 1. Andrews, D. A., & Bonta, J. (1994). The psychology of criminal conduct Cincinnati, OH: Anderson Publishing Co. Andrews, D. A., & Bonta, J. (2006). The psychology of criminal conduct (4th ed.).Cincinnati, OH: Anderson. Ameklev, B. J., Grasmick, H. G., Tittle, C. R., & Bursik, R. J., Jr. (1993). Low self control and imprudent behavior. Journal of Quantitative Criminology 9 225 247. Bahr, S. J., Hoffmann, J. P., & Yang, X. ( 2005 ) Parental and peer influences on the risk of adolescent drug use. Journa l of Primary Prevention 26, 526 551.
208 Bahr, S.J., Marcos, A.C., and Maughan, S.L. (1995). Family, educational, and peer influences on the alcohol use of female and male adolescents. Journal of Studies on Alcohol 56, 457 469 Barnes, G. E., Barnes, M. D., & Patton, D. ( 2005 ) marijuana use in a Canadian youth sample. Substance Use and Misuse 40 1849 1863. Barnes, G. M., Welte, J. W., & Hoffman, J. H. ( 2002 ) Relationship of alcohol use to delinquency and illicit dru g use in adolescents: Gender, age, and racial/ethnic differences. Journal of Drug Issues 32, 513 178. Baron, S. (2003). Self control, social consequences and criminal behavior: Street youth and the general theory of crime. Journal of Research in Crime an d Delinquency, 40, 403 425. Battin, S.R., Hill, K.G., Abbott, R.D., Catalano, R.F., & Hawkins, J.D. (1998). The contribution of gang membership to delinquency: Beyond delinquent friends Criminology 36, 93 115. Batton C., & Ogle, R.S. ( 2003 t gonna be you or me?: The potential of social learning for intergared homicide suicide theory. pp.85 108. In R.L. Akers and G.F. Jenson (eds.) Social learning theory and the explanation of crime: A guide for a new century. Advances in criminological theor y, Vol, 11. New Brunswick, NJ: Transaction Publishers. Belenko, S., & Sprott, J.B. ( 2002 ) Comparative recidivism rates of drug and nondrug juvenile offenders: Results from three jurisdictions. Paper presented at the Academy of Criminal Justice Sciences Annual Conference. Anaheim, CA. Bellair, P. E., Roscigno, V.J.,& Velez, M.V. ( 2003 ) Occupational structure, social learning and 225 in Social Learning Theory and the Explanation for Crime: A Guide for the New Century, edite d by Ronald L. Akers and Gary F. Jensen. New Brunswick, NJ: Transaction Publishers. Benda, B. B., & DiBlasio, F. A. (1994). An integration of theory: Adolescent sexual contacts. Journal of Youth and Adolescence, 23 403 420. Beaver, K.M., Wright, J.P., & Delisi, M. (2007). Self contro l as an executive function: Justice and Behavior 34, 1345 1361. taking behavior in early adolescence: cross sectional study (Croatia). Public Health 50, 157 164.
209 Boeringer, S. B., Shehan, C.L., & Akers, R.L. ( 1991 ) Social contexts and social learning in sexual coercion and aggression: Assessing the contribution o f fraternity membership. Family Relations, 40, 58 64. Brenzina, T, & Riquero, A.R. (2003). Exploring the relationship between social and non social reinforcement in the context of social learning theory. Pp. 265 288 in R. L. Akers and G.F. Jensen (eds.), Social learning theory and the explanation of crime: A guide for the new century. Advances in criminological theory. New Brunswick, NJ: Transaction Publishers Brook, J.S., Whiteman, M., Gordon, A.S., Nomura, C., & Brook, D.W. (1986). Onset of adolescent d rinking: A longitudinal study of intrapersonal and interpersonal antecedents. In Alcohol and Substance Abuse in Women and Children. Advances in Alcohol & Substance Abuse New York: Haworth Press. pp. 91 110. Brook, J.S., Gordon, A.S., Brook, A., Brook, D .W. ( 1989a ) The consequences of marijuana use on intrapersonal and interpersonal functioning in Black and White adolescents. Genet. Soc. Gen. Psychol. Monogr 115, 349 369. Brooks, J., Pahl, T. Balka, E., & Fei, K. 2004. Smoking among New Yorican adolesc ents: Time 1 predictors of time 2 tobacco use. Journal of Genetic Psychology 165, 324 340 Brook, J.S., Whiteman, M., Gordon, A.S., & Brook, D.W. (1984). Parental determinants of female adolescent marijuana use. Developmental Psychology 20:1032 1043. Brook, J.S., Whiteman, M., Finch, S.J., & Cohen, P. ( 1996 ) Young adult drug use and delinquency: Childhood antecedents and adolescent mediators. Journal of the American Academy of Child Adolescent Psychiatry 35, 1584 1592. Brownfield, D., & Sorenson, A. M. (1993). Self control and juvenile delinquency: Theoreitical issues and an empirical assessment of selected elements of a general theory of crime. Deviant Behavior, 14,243 264. Browning, S.E. ( 2008 ) Neighborhood, school, and individual effects on sub stance use, violence and the drug/violence nexus : A study of Toronto secondary students. A dissertation. University of Toronto. Browne, W.J. &, & Draper, D. ( 2000 ) Implementation and performance issues in the Bayesian and likelihood fitting of multileve l models. Computational Statistics 15, 391 420 Brown, B. B., Mounts, N., Lamborn, S.D., and Setinberg, L. ( 1993 ) Parenting Practices and Peer Group Affliction in Adolescence. Child Development 64, 469 482 Bruinsma, G. J. N. (1992). Differential assoc iation theory reconsidered: An extension and its empirical test. Journal of Quantitative Criminology, 8 29 49
210 Bryk, A. S., Raudenbush, S. W., & Congdon, R. ( 1996). HLM: Hierarchical linear and nonlinear modeling with the HLML/L and HLM/3L programs [Comp uter Program]. Chicago: Scientific Software International Burgess, R. L., & Akers, R. L. ( 1966 ) Social Problems, 14, 128 147. Burkett, S.R., & Jensen., E.L. (1975). Conventional tie s, peer influence, and the fear of apprehension: A study of adolescent marijuana use. Sociological Quarterly 16, 522 533. Burkett, S.R., & Warren. B.O. (1987). Religiosity, peer associations, and adolescent marijuana use: A panel study of underlying cau sal structures. Criminology. 25, 109 131. Bursik, R. J., & Grasmick, H.G., ( 1993 ) Neighborhoods and crime: The dimensions of effective community control. New York: Lexington Books. Burt, C.H., Simons, R.L., & Simons, L.G. ( 2006 ). A longitudinal test of the effects of parenting and the stability of self control: negative evidence for the general theory of crime. Criminology, 44, 353 396. Burton, V.S., Cullen, F.T., Evans, D. & Dunaway, R.G. (1994). Reconsidering strain theory: Operationalization, riv al theories, and adult criminality. Journal of Quantitative Criminology, 10: 213 239. Burton, Jr., V.S., Cullen, F.T., Evans, T.D., Dunaway, R.G., Kethineni, S.R., & Payne, G.L. (1995). The impact of parental controls on delinquency, Journal of Criminal Justice 23, 111 126. Cao, L. (2004) Major criminological theories: Concepts and measurement. Belmont, CA: Wadsworth. Cattell, V. (2001). Poor people, poor places, and poor health: the mediating role of social networks and social capital. Social Scien ce & Medicine, 52 1501 1516. Cauffman, E., Steinberg, L., & Piquero, A.R. ( 2005 ). Psychological, neuropsychological, and physiological correlates of serious antisocial behavior. Criminology 43, 133 176. Chassin, L., Clar, C., Presson, M., Bensenberg, E., Corty, R.W., Olshavsky, & Sherman, S.J. Journal of Health and Social Behavior 22, 4455 455. Cheung N. W. T., & Cheung, Y. W. ( 2008 ). Self control, social factors, and delinquency: A t est of the general theory of crime among adolescents in Hong Kong. Journal of Youth Adolescence 47, 412 430.
211 Cheung, C., Ngai, N., & Ngai, S. S. (2007). Family strain and adolescent delinquency in two Chinese cities, Guangzhou and Hong K ong. Journal of Child and Family Studies, 16 626 641. Choi, Y.S. (2003). The trends of juvenile drug abuse in Korea by adolescents. Journal of Health Economics, 26 763 7 84. Cochran, W.G. (1977). Sampling Techniques (3rd ed.). New York: Wiley. Cochran, J. Wood, P., Sellers, C., Wilerson, W., & Chamlin, M. (1998). Academic dishonesty and low self control: An empirical test if a general theory of crime. Deviant Behavior 19, 227 255. Conger, R. (1976). Social control and social learning models of delinquency: A synthesis, Criminology 14, 17 40. Conger, R. & Simons, R.L. (1995). Life course contingencies in the development of adolescent antisocial behavior: A matching l aw approach. In T.P. Thornberry (ed.), Developmental theories of crime and delinquency. New Brunswick, NJ: Transaction Books. Coombs, R. H., & Landsverk, J. (1988). Parenting styles and substance use during childhood and adolescence. Journal of Marriage & the Family 50, 473 482. Cooper, K, May, D., Soderstrom, I, & Jarjoura, G.R. (2009). Examining theoretical predictors of substance use among a sample of incarcerated youth. Journal of Offender Rehabilitation 48, 669 695 Costello, B. J. ( 2000 ) Techni ques of neutralization and self esteem: A critical test of social control and neutralization theory. Deviant Behavior 21, 307 330. Cornwall, M., Albrecht, S.L., Cunningham, P.H., & Pitcher, B.L (1986). The dimensions of religiosity: A conceptual model w ith an empirical test. Review of Religious Research 27, 226 244. Crane, J. 1991. The epidemic theory of ghettos and neighborhood effects on dropping out and teenage childbearing. American Journal of Sociology 96, 1226 1259. Crum, R. M., Lillie Blanton M., Anthony, J. C. 1996. Neighborhood environment and opportunity to use cocaine and other drugs in late childhood and early adolescence. Drug & Alcohol Dependence 43, 155 161. Cullen, F., & Agnew, R. 2003. Criminological theory: Past to present (2 nd ed.).Los Angeles: Roxbury.
212 Cullen, F. T., & Gendreau, P. (2000). Assessing correctional rehabilitation: Policy, practice, and prospects. In J. Horney (Vol. Ed.), Criminal justice 2000: Vol. 3. Policies, processes, and decisions of the criminal justice sys tem, pp. 109 175. Washington, DC: U.S. Department of Justice, National Institute of Justice. Cullen, F. T., Wright, J. P., Gendreau, P., & Andrews, D. A. (2003). What correctional treatment can tell us about criminological theory: Implications for social learning theory. In R. L. Akers & G. F. Jensen (Vol. Eds.), Social learning theory and the explanation of crime: Advances in criminological theory (Vol. 11) pp. 339 362. New Brunswick, NJ: Transaction. Currey, G.D., Decker, S.H., & Egley A. ( 2002 ). Gang invovlment and delinquency in a middle school population. Justice Quarterly 19, 275 292. Dabney, D, (1995). Neutralizaiton and deviance in the wo rkplace: Theft of supplies and medicines by hospital nurses. Deviant Behavior 16, 313 331. Dembo, R., Blou nt, W.R., Schmeidler, J., & Burgos, W. ( 1986 ) Perceived environmental d rug use risk and the correlates of early drug use or nonuse among inner city youth: The motivated actor, International Journal of the Addictions 21, 977 1000. Deng, S., & Roosa, M. W. (2007). Family influences on adolescent delinquent behaviors: Applying the social development model to A Chinese sample. American Journal of Community Psychology, 40 333 344. DeWit, D.J. (1998). Frequent childhood geographic relocation: Its impact on d rug use initiation and the development of alcohol and other drug related problems among adolescents and young adults. Addictive Behaviors. 23, 623 634. DiBlasio, F. A., & Benda, B. B. (1990). Adolescent sexual behavior: Multivariate analysis of a social l earning model. Journal of Adolescent Research, 5 449 466. Dillon, F. R., Pantin, H., Robbins, M. S., & Szapocznik, J. (2008). E xploring the role of parental monitoring of peers on the relationship be tween family functioning and delinquency in the lives o f African American and Hispanic adolescents. Crime & Delinquency, 54 65 94. Dishion, T. J., & Loeber, R. (1985). Adolescent marijuana and alcohol use: The role of parents and peers revisited. American Journal of Drug and Alcohol Abuse 11, 11 25. Dishi on, T. J., Patterson, G. R., Stoolmiller, M., & Skinner, M. L. ( 1991 ) Family, school, and behavioral antecedents to early adolescent involvement with antisocial peers. Developmental Psychology 27 172 180.
213 Donovan, J.E., ( 1996 ) Problem behavior theo ry and the explanation of adolescent marijuana use. Journal of Drug Issues 26, 379 404. Ellickson, P., Hays, R., & Bell, R. M. ( 1992 ). Stepping through the drug use Sequence: Longitudinal scalogram analysis of initiation and regular use. Journal of Abnor mal Psychology, 10 441 451. Elliott, D.S., Huizinga, D., Suzanne, S., & Ageton, S. (1985). Explaining delinquency and drug use Beverly Hills, CA: Sage. Elliott, D.S., Wilson, W.J., Huizinga, D., Sampson, R. J., Elliott, A., & Rankin, B. ( 1996 ) The ef fects of neighborhood disadvantage on adolescent development. Journal of Research in Crime and Delinquency 33, 389 426. Ennett, S.T., Flewelling, R.L., Lindrooth, R.C., Norton, E.C. ( 1997 ) School and neighborhood characteristics associated with school rates of alcohol, cigarette, and marijuana use. J. Health Soc. Behav. 38, 55 71. Esbensen F.A., & Deschenes, E.P. ( 1998 ). A multisite examination of youth gang membership: Does gender matter? Criminology 36, 799 827. Evans, T.D., Cullen, F.T., Burton, J.R., Dunaway, R.G. (1997). The social consequences of self control: Testing the general theory of crime. Criminology 35, 475 504. Fergusson, D. M., & Horwood, L. J. ( 199 9) The role of adolescent peer affiliations in the continuity between childhood b ehavioral adjustment and juvenile offending. Journal of Abnormal Child Psychology 24 205 221. Fergusson, D, M., Swain Campbell, N. R., & Horwood, L. J. (2002). Deviant peer affiliations, crime and substance use: A fixed effects regression analysis. Jou rnal of Abnormal Child Psychology 30 419 430. Fishbein, M., & Ajzen, I. (1975). Belief, attitude, intention and behavior: An introduction to theory and research. Don Mills, Canada: Addison Wesley. Erickson, K. G., Crosnoe, R., & Dornbusch, S. M. ( 2000 ) A social process model of adolescent deviance: Combining social control and differential association perspectives. Journal of Youth and Adolescence 29 395 425. Ford, J.M., and Beveridge, A.A. ( 2006 ) Varieties of substance use and visible drug probl ems: Individual and neighborhood factors. Journal of Drug Issues 68, 377 392. Gibbs, J. J., & Giever, D.(1995). Self control and its manifestations among university students: An empirical test of Gottfredson and Hirschi's general theory. Justice Quarterl y 12, 231 256.
214 Gendreau, P., Goggin, C. E., & Law, M. A. (1997). Predicting prison misconducts. Criminal Justice and Behavior, 24 414 431. Gendreau, P., Little, T., & Goggin, C. E. (1996). A meta analysis of the predictors of adult offender recidivism: What works! Criminology, 34 575 607. Ghanizadeh A. (2005) Journal of Substance Use, 10, 263 271. Gib so n, C.L., Poles, T.B. & Akers, R.L. ( 2010 ). A partial test of social structure social learning : Neighborhood disadvantage, differential association with delinquent peers, and delinquency. Pp.133 148 in (eds), Matt Delisi and Kevin M. Beaver, Criminological theory: A life course approach London: Jones and Bartlett Publishers Ginsberg, I., & Greenl y, J.R. (1978). Competing theories of marijuana use: A longitudinal study. Journal of health and Social Behavior, 19, 22 34. Goldstein, H., Rashbash, J., Plewis, I., Draper, D., Browne, W., Yang, M., Woodhouse, G., & Healy, H. (1998). LwiN London: University of London, Institute of London Golub, A. & Johnson, B. D. ( 2001 ). Variation in youthful risks of progression from alcohol and tobacco to marijuana and to hard drugs across generations. American Journal of Public Health, 97, 225 23 2. Gottfredson, M. R. and Hirschi, T. ( 1990 ) A general theory of crime Stanford, CA: Stanford University Press. Grasmick, H.G., Tittle, C.R., & Bursik, R.J. (1993). Testing the core empirical implications of f crime. Journal of Research in Crime and Delinquency 30, 5 29. Greenwood, P.W. ( 1992 ) Substance abuse problems among high risk youth and potential intervention. Crime & Delinquency 38, 444 458. Green, K.M., Ensminger, M.E., ( 2006 ) Adult social beha vioral effects of heavy adolescent marijuana use among African Americans. Development of Psychology, 42 1168 1178. Guo, G., Elder, G. H., Cai, T., & Hamilton, N. (2008). Gene alcohol use moderates genetic contribution to adolescent drinking behavior. Social Science Research Hagan, J. (1991). Destiny and drift: Subcultural preferences, status attainments, and the risks and rewards of youth. American Sociological Review 56, 567 582.
215 Harwood, H. ( 2000 ) Updating estimate s of the economic costs of alcohol abuse in the United States: Estimates, update methods and data, National Institute on Alcohol Abuse and Alcoholism. Ha w kins, J.D., Catalano, R.F., & Miller, (1992). Risk and protectiv e factors for alcohol and other drug problems in adolescence and early adultho od: Implications for substance abuse prevention. Psychological Bulletin, 112, 64 105. Hawkins, J. D., Herrenkohl, T. I., Farrington, D. P., Brewer, D., Catalano, R. F., Harachi, T. W., & Cothern, L. (2000). Predi ctors of youth violence. Juvenile Justice Bulletin 4, 1 11. H awkins, J.D., & Weis, J.G. ( 1985 ). The social development model: An integrated approach to delinquency prevention, Journal of Primary Prevention, 6, 73 97. Hartjen, C.A., & Priyadarsini, S., 2003, Gender, peers, and delinquency: A study of boys and girls in rural France. Youth and Society, 34, 387 414. Hay, C, & Forrest, W. (2006). The development of self control: Examining self stability thesis, Criminology, 44,739 774. H aynie, D. L., Silver, E., & Teasdale, B. ( 2006 ) Neighborhood characteristics, peer networks, and adolescent violence. Journal of Quantitative Criminology, 22, 247 169 Hedeker, D., & Gibbons, R. D. (1996). MIXOR: A computer pro gram for mixed effects ordi nal regression analysis. Computer Methods and Programs in Biomedicine, 49, 157 176. Hirschi, T. ( 1969 ) Causes of delinquency Berkeley, CA: University of California Press. 2004. Self control and crime, (pp.537 552) in R. F. Baumeister and K. D. Vohs (eds .), Handbook of self regulation: Research, theory, and applications New York: Guilford. Hoffman, J. P. ( 1994 ) Exploring the direct and indirect family effects on adolescent drug use. Journal of Drug Issues 23, 535 557. Holland Davis, L. ( 2006 ) Put ting Behavior in Context: A Test of the Social Structure Social Learning Model. Ph. D. Dissertation. Gainesville, FL: University of Florida. Houghton, S., & Carroll, A. (2002). Longitudinal rates of self reported delinquency of at risk and not at risk hig h school students. Australian and New Zealand Journal of Criminology, 42, 112 134. Huizinga, D., Loeber, R., & Thornberry, T. P. ( 1995 ) Urban delinquency and substance abuse Washington, D.C: Office of Juvenile Justice and Delinquency Prevention, Office of Justice Programs.
216 Hundleby, J.D., & Mercer, G.W. (1987). Family and friends as social environments and their Journal of Marriage and the Family 49, 151 164. Hwang, S. H. ( 2000 ) Substance use in a sample of South Korean adolescents: A test of alterative theorie s. Ph.D Dissertation. University of Florida Hwang, S., & Akers, R. L. ( 2003 ) Substance use by Korean adolescents: A cross cultural test of social learning, social bondi ng, and self control theories. In R. L. Akers & G. F. Jensen (Eds.), Social learning and the explanation of crime: A guide for the new century, advances in criminological theory pp. 39 63. New Brunswick, NJ: Transaction. Hwang, S., & Akers, R. L. ( 2006 ) Parental and peer influences on adolescent drug use in Korea. Asian Journal of Criminology 1, 51 69. Inciardi, J.A., Horwitz, R. & Potttiger, A. E. (1993). Street Kids, street drugs, street crime: An examination of drug use and serious delinquency in Miami Belmont, CA: Wadsworth. Institute of Medicine. (1996). To err is human building a safer health system Washington DC: National Academy Press. Jang, U. S. ( 2003 ) Investigation of factors relevant to smoking among juvenile high school student i n South Korea. Journal of Korean Academic Family Medicine. 23 894 903. Jang, S.U., & Johnson, B.R. ( 2001 ) Neighborhood disorder, individual religiosity, and Criminology 39, 109 144. J e, K.J., Hu, Y. H., Kim, S. H., & Lee, S. M. (2004). Characteristics of substance use adolescents and its implication. Youth Protection Committee. Seoul, South Korea. Jensen, Gary F. (1972). Parents, peers, and delinquent action: A test of the differenti al association perspective. American Journal of Sociology 78, 63 72. ( 2003 ) Gender variation in delinquency: Self images, beliefs, and peers as mediating mechanisms, pp. 151 177 in R.L. Akers and G. F. Jensen, (eds.). Social Learning Theory and the Exp lanation of Crime: A Guide for the New Century. Advances in Criminological Theory Volume 11. New Brunswick, NJ: Transaction Publishers
217 Jensen, Gary F. & Akers, R. L. (2003). Taking social learning global: Micro macro transitions in criminological theory pp.9 38. in R.L. Akers and G. F. Jensen, (eds.). Social Learning Theory and the Explanation of Crime: A Guide for the New Century.Advances in Criminological Theory, Volume 11 New Brunswick, NJ: Transaction Publishers Jeynes, W. (2002). The relationship between the consumption of various drugs by adolescents and their academic achievement. American Journal of Drug & Alcohol Abuse, 28, 15 35. Jo, S.K. (2001). Alcohol use among elementary school students and adolescents. Korean Institutes of Substance Us e Research. Seoul, Korea. Johnson, R.E. Marcos, A.C., & Bahr, S.J. (1987). The role of peers in the complex etiology of adolescent drug use. Criminology 25: 323 340. l, illicit drug use by American teens continues gradual decline in 2007. National Press Release 57. Retrieved from http://www.monitoringthefuture.org/pressreleases/07drugpr_complete.pdf E. (2008a). Monitoring the Future national results on adolescent drug use: Overview of key findings, 2007, NIH Publication No. 08 6418 news on teen smoking: Rates at or near record lows. National press release 23 pp.: Retrieved from http://www.monitoringthefuture.org/pressreleases/08cigpr_complete.pdf Junger M., & Marshall I.H. (1997) The interethnic generalizability of social control theory. Anemphircal test. Journal of Research in Crime and Delinquency 34:79 112. Junger Tas, J. (1992). An empirical test of social control theory. Journal of Quantitative Criminology 38,25 46. Kandal, D.B. ( 1984 ) Marijuana users in young adulthood. General Psychiatry, 41, 200 209 ( 2002 ) Examining the gateway hypothesis: stages and pathways of drug involvement In: Kandel, D.B. (Ed.), Stages and Pathways of Drug Involvement: Examining the Gateway Hypothesis. Cambridge University Press, New York, NY, pp. 3 18. Kandel, D.B. & Adler, I. (1982). Socialization into marijuana use among French adolescents: A cross cultural comparison with the United States. Journal of Health and Social Behavior, 23, 295 309.
218 Kandel, D.B. & Dvaies, M. (1991). Friendship networks, intimacy, and il licit drug use in young adulthood: A comparison of two competing theories. Criminology 29, 441 469. Kandel, D.B., Kessler, R.C., & Margulies, R.Z. (1978). Antecedents of adolescent initiation into stages of drug use: A developmental analysis. pp.73 99. In Kandel, D.B. (ed.). Longitudinal research on drug use. New York: Wiley. Kang, Y.J., & Kim, H, S. (2005). Examining factors associated with mal e high school smoking. 4, 121 142. Kamata, A. (2002). Procedure t o perform item respon se analysis by hierarchical generalize linear model Paper presented at the annual meeting of the American Educational Research Association, New Orleans, LA. Kaplan, H.B. (1996). Empirical validation of the applicability of an integra tive theory of deviant behavior to a study of drug use. Journal of Drug Issues 26, 345 377. Kaplan, H.B., Jonson, R.J., & Bailey, C.A. (1987). Deviant peers and deviant behavior: Further elaboration of a model. Social Psychological Quarterly 50, 277 28 4. Kaplan, H.B., Martin, S.S., & Robbins, C. (1982). Application o f a general theory of deviant behavior: Self derogation and adolescent drug use. Journal of Health and Social Behavior, 23, 274 294. Keane, C., Maxim, P.S., & Teevan, J.J. (1993). Drinki ng and driving, self control, and gender: Testing a general theory of crime. Journal of Research in Crime and Delinquency, 30, 30 46. Kim, S.H., Choi, Y.S., & Chin, S.M. (1990). Korea Institute of Criminology Seoul, Korea. Kim, E., Kwak, D.H., & Yun, M. (2010). Investigating the effects of peer association and parental influence on adolescent substance use: A study of adolescents in South Korea. Journal of Criminal Justice, 38, 17 24. Kim, J.Y., & Park, S. W. (2009). Predictors of current smok ing among male students in a technical high school: A prospective study. Journal of Prevention Medicine 42, 59 66. Krohn, M. ( 1999 ) Social learning theory: The continuing development of a perspective. Theoretical Cr iminology, 3, 462 476. Krohn, M.D., Akers, R.L., Radosevich, M.J., & Lanza Kaduce, L. (1982). Norm qualities and adolescent drinking and drug behavior. Journal of Drug Issues, 12,343 359.
219 Krohn, M.D., Lanza Kaduce, L., & Akers, R. L. ( 1984 ) Community context and theories o f Sociological Quarterly, 25, 353 371. Krohn, M.D., & Massey, J.L. (1980). Social control and delinquent behavior: An examination of the elements of the social bond. Sociological Quarterly, 21, 529 543. Krohn, M.D., Massey, J.L., Skinner, W.F., & Lauer, R.M. (1983). Social bonding theory and adolescent cigarette smoking: A longitudinal analysis. Journal of Health and Social Behavior 24, 337 349. K rohn, M.D., Skinner, W.F., Massey, J.L., & Akers, R.L. ( 1985 ) Social learning theory and adolescent cigarette smoking: A longitudinal study. Social Problems, 32 455 473. Krosnick, J.A., & Judd, C.M. (182). Transitions in social influence at adolescence: Who induces cigarette smoking. Developmental Psychology, 18, 359 368. Kubrin, C. E., Stucky, T. D., & Krohn, M. D. (2009). Researching theories of crime and deviance. New York: Oxford University Press. LaGrange, R.L. & White, H.R. (1985). Age differenc es in delinquency: A test of theory. Criminology. 23, 19 46. Lanza Kaduce, L., Akers, R. L., Krohn, M. D., & Radosevich, M. (1984). Cessation of alcohol and drug use among adolescents: A social learning model. Deviant Behavior, 5 79 96. Lanza Kaduce, L ., & Capece, M. ( 2003 ) A specific test of an integrated general theory. pp. 179 196 in R. L. Akers and G. F. Jensen, (eds.). Social Learning Theory and the Explanation of Crime: A Guide for the New Century. Advances in Criminological Theory Volume 11. Ne w Brunswick, NJ: Transaction Publishers. Lanza Kaduce, L., Capece, M. & Alden, H. (2006). Liquor is quicker: Gender and social learning among college students. Criminal Justice Policy Review 17, 127 143. Lanza Kaduce, L., & Klug, M. (1986). Learning t o cheat: the interaction of moral development and social learning theories. Deviant Behavior 7, 243 259. Lauer, R. M., Akers, R.L., Massey, J., & Clarke, W. (1982). The evaluation of cigarette smoking among adolescents: The Muscatine study. Preventive Me dicine, 11,417 428. Lazarou, J., Pomeranz B.H., Corey P.N. (1998). Incidence of adverse drug reactions in hospitalized patients: A meta analysis of prospective studies. Journal of the American Medical Association. 279,1200 1205.
220 Lee, G. (1998). Social structure and social learning in adolescent delinquency and substance use: A test of the mediating process of social learning theory Ph.D. Dissertation, University of Florida. Lee, G., Akers, R. L., & Borg, M. J. ( 2004 ) Social learning and structural f actors in adolescent substance use. Western Criminology Review 5:17 34. Leventhal, T., & Brooks Gunn, J. ( 2000 ) The neighborhoods they live in: The effects of neighborhood residence on child and adolescent development. Psychological Bulletin, 126, 309 3 37. Lindstrm, M. (2008). Social capital, political trust and experience of cannabis smoking: A population based study in Southern Sweden. Preventive Medicine 46 599 604. Lipsey, M. W., Chapman, G. L., & Landenberger, N. A. (2001). Cognitive behaviora l programs for offenders. Annals of the American Academy of Political and Social Science, 578,144 157. Lipsey, M.W., & Derzon, J.H. (1998). Predictors of violent or serious delinquency adolescence and early adulthood: A synthesis of longitudinal research In R. Loeber and D.P. Farrington (eds.), Serious and violent juvenile offenders: Risk factors and successful interventions (pp.86 105). Thousand Oaks, CA: Sage. Liska, A.E., Reed, M.D. (1985). Ties to conventional institutions and delinquency: Estimati ng reciprocal effects. American Sociological Review, 50,547 560. Lyerly, R.R., & Skipper,Jr, J.K. (1981). Differential rates of rural urban delinquency: A social control approach. Criminology 19, 385 399. Lynskey, M.T., Coffey, C., Degenhardt, L., Carl in, J.E., Patton, G. ( 2003 ) A longitudinal study of the effects of adolescent cannabis use on high school completion. Addiction 98, 685 692. Lynskey, M.T., Vink, J.M., Boomsma, D.I. ( 2006 ) Early onset cannabis use and progression to other drug use in a sample of Dutch twins. Behavior Genetics 36, 195 200. MaGee, Z.T. (1992). Social class differences in parental and peer influence on adolescent drug use. Deviant Behavior, 13, 349 372. MacKenzie, D. L. (2006). What works in corrections: Reducing the cr iminal activities of offenders and delinquents. New York: Cambridge University Press. Marcos, A.C., Bahr, S.J., Johson, R.E. (1986). Test of bonding/association theory of adolescent drug use. Social Forces, 65, 135 161.
221 Mass, C.J. M., & Hox, J.J. ( 2004 ) Robustness issues in multilevel regression analysis. Statistica Neerlandica, 58,127 137 Matsueda, R.L. & Heimer, K.(1987). Race, family structure and delinquency: A test of differential association and social control theories. American Sociological Revi ew 52, 826 840. Meier, R. Burkett, S., & Hickman, C. (1984). Sanctions, peers, and deviance: Preliminary models of a social control process. Sociological Quarterly 25, 67 82. Meneses, R.A. (2009). A cross cultural test of social learning, self contro l, social bonding and general strain theories of crime and deviance. Phd. Dissertation, university of Florida. Miller, H.V., Jennings, W.G., Alvarez Rivera, L.L., & Miller, J.M. (2008). Explaining substance use among PuertoRican adolescents: A partial te st of social learning theory. Journal of Drug Issues 38, 261 283. Morash, M. 91999). A consideration of gender in relation to social learning and social structure: A general theory of crime and deviance. Theoretical Criminology, 3, 452 461. Nagin, D., & Paternoster, R. (1993). Enduring individual differences and ration choice theories of crime. Low and Society Review, 27, 467 496. Nakhaie M.R., Silverman, R.A., & LaGrange, T.C. (2000 ). Self control and social control: An examination of gender, ethni city, class and delinquency. Canadian journal of Sociology 25, 35 59. National Health Control Institution (2006). Annual national health report in 2006. Seoul, South Korea. National Youth Commission. (2002 ) A study on environments negatively affectin g adolescents. Seoul: National Youth Commission. ( 2004 ) A study on environments negatively affectin g adolescents. Seoul: National Youth Commission. (2005). A study on environments negatively affectin g adole scents. Seoul: National Youth Commission (2007). A study on environments negatively affectin g adolescents. Seoul: National Youth Commission Neff, L. J., & Waite, D. E. ( 2007 ) Male versus female substance abuse patterns among inc arcerated juvenile offenders: Comparing strain and social learning variables. Justice Quarterly 24 106 132
222 Newcomb, M.D., Bentler, P.M. ( 1988 ) Consequences of Adolescent Drug Use: Impact on the Lives of Young Adults. Sage, Newbury Park, CA. Office of N ational Drug Control Policy, 2004. National Drug Control Strategy: High Intensity Drug Trafficking Area Program Annual Report. ONDCP, Washington, D.C. Ngai, N. P., & Cheung, C. K. (2005). Predictors of the likelihood of delinquency: A study of marginal yo uth in Hong Kong, China. Youth & Society 36, 445 470. Differences among high schools 1986 1987. Monitoring the Future Occasional Paper. 24. Ann Arbor, MI: The Universit y of Michigan, Institute for Social Research. Unpublished manuscript. Oetting, E., Deffenbacher, J., Taylor, M., Luther, N., Beauvais, F., & Edwards, R. ( 2000 ). Methamphetamine use by high school students: recent trends, gender, and ethinicty differences, and use of other drugs Journal of Child and Adolescent Substance Abuse 10, 33 50. Office of Applied Studies (2009a). Cigar use among young adults aged 18 to 25. NSDUH Report Retrieved March, 20, 2010 from http://www.oas.samhsa.gov/2k9/cigars/cigars.p df Office of National Drug Control Policy (2009). National drug control strategy FY 2009 budget summary Retrieved February 18, 2009. from http://www.whitehousedrugpolicy.gov/publications/policy/09budget/index.html Orcutt, J.D. (1987). Differential ass ociation and marijuana use: A closer look at Sutherland (with a little help from Becker). Criminology 25, 341 358. Overall, J. E., & Dalai, S. N. (1965). Design of experiments to maximize power relative to cost. Psychological Bulletin, 64, 339 350. Pag e, E.R. (1998). Family structure and juvenile delinquency: The mediating role of social learning variables Ph.D. Dissertation. University of Florida. Park, S., Kim, H.S., Kim, H., & Sung, K. T. ( 2007 ) Exploration of the prevalence and correlates of sub stance use among sheltered adolescents in South Korea. Adolescence, 42 603 616. Patternson, G. R., & Dishion, T.J. (1985). Contributions of families and peers to delinquency. Criminology. 23, 63 79. Patternoster, R., & Brame, R. (1998). The structural similarity of processes of a generation of criminal and analogous behaviors. Criminology, 36, 633 669.
223 Paternoster, R., & Iovanni, L. (1986). The deterrence effect of perceived severity: A re examination. Social Forces 64, 751 777. Perrone, D., Sulliv an, C., Pratt, S.T., & Margaryan, S. (2004). Parental efficacy, self control, and delinquency: A test of a general theory of crime on a nationally representative sample of youth. International Journal of Offender Therapy and Comparative Criminology 48, 29 8 312. Perl, P. (2003, Updated March 6, 2003). Drug control: International policy and approaches. Issue brief for congress. Retrieved February 5, 2007, from http://usinfo.state.gov/usa/infousa/society/crime/crimegun2.pdf Piquero A. R., & Bouffard J.A (2007 ). Something old, something new: A preliminary control, Justice Quarterly 24, 1 27. Piquero, A.R., MacDonald, J., Dobrin, A., Daigle, L.E., & Cullen, F.T. ( 2005 ). Self control, violent infractions, and ho micide victimization: Assessing the general theory of crime. Journal of Quantitative Criminology, 21, 55 71. Piquero, N. L., & Piquero, A. R. ( 2010 ) Contemporary retrospective on self control theory. Criminological Theory (ed). Copes, Heith, & Topalli, Vokan. P p.299 311. Porter, S. R. & Umbach, P. D. (2001). What works best? Collecting alumni data with multiple technologies. Paper presented at the 41st Annual Association for Institutional Research Forum, Long Beach, CA. Pratt, T.C., & Cullen, F.T. (2 theory of crime: A meta analysis. Criminology, 38, 931 964. Pratt, T.C., Cullen, F.T., Sellers, C.S., Winfree, T., Madenson, T.D., Diago, L.E., Fearm, N.E., & Gau, J.M. (2009). The empiric al status of social learning theory: A meta analysis. Justice Quarterly 23, 1 38. Raudenbush, S.W. ( 1997 ) Statistical analysis and optimal design for cluster randomized trial. Psychological Methods 2, 173 185 Raudenbush, S. W. & Bryk, A. S. ( 2002 ) H ierarchical linear models: Applications and data analysis methods Thousand Oaks, CA: Sage Publications. Raudenbush, S.W., & Liu, X. ( 2000 ) Statistical power and optimal design for multisite randomized trials. Psychological Methods 5,199 213 Reed, M. D., & Rountree, P. W. ( 1997 ) Peer pressure and adolescent substance use. Journal of Quantitative Criminology 13 143 180.
224 Rebellon, C.J. (2002). Reconsidering the broken homes/delinquency relationship and exploring its mediating mechanism(s). Criminol ogy 40,103 136. Ritt Olson, A., Unger, J., Valente, T., Nezami, E., Chou, C., Trinidad, D., et al. (2005). Exploring peers as a mediator of the association between depression and smoking in young adolescents. Substance Use & Misuse, 40 77 98. Luengo, R.E. M., & Sobral, J. (2001). Personality and antisocial behavior: Study of temperamental dimensions. Personality and Individual Differences, 31, 329 348. Rosental, R., & DiMatteo, M.R. (2001). Metal analysis: Recent developments in quantities methods f or literature reviews. Annual Review of Psychology 52, 59 82. Rountree, P. W., K. C. Land, and T. D. Miethe, 1994, Macro micro integration in the study of victimization: A hierarchical logistic model analysis across Seattle neighborhoods, Criminology 3 2, 387 404. Rumpold, G., Klingseis, M., Dornauer, K., Kopp, M., Doering, S., Hofer, S., Mumelter, B., Schussler, G. ( 2006 ) Psychotropic substance abuse among adolescents: A structural equation model on risk and protective factors. Substance Use & Misus e 41 1155 1169. Sampson, R.J. ( 1999 ) Theoretical Criminology, 3, 438 451. ( 2002 ) Transcending tradition: New directions in community research, Chicago study. The American Society of Criminology 2001 Sutherland Address. Criminology 40, 213 230. Sampson, R., & Groves, B. W. ( 1989 ) Community Structure and Crime: Testing Social Disorganization Theory. American Journal of Sociology, 94, 774 802. Sampson, R. J., Raudenbush, S. W. & Earls, F. J. ( 1997 ) Neighborho ods and violent crime:A multilevel study of collective efficacy. Science 277, 918 24 Sakai, J., Hall, S., Mikulich Gilbertson, S., & Crowley, T. (2004). Inhalant use, abuse, and dependence among adolescent patients: Commonly comorbid problems. Journal of the American Academy of Child & Adolescent Psychiatry, 43 1080 1088. Sampson, R. J., & Grove, W.B. (1989). Community structur e and crime: Testing social disorganization theory. American Journal of Sociology 94,774 802. Santrock, J.W. (2007). Adole scence (12th ed.). Boston: McGraw Hill. Saxe, L., Kadushin, C., Beveridge, A., Livert, D., Tighe, E., Rindskopf, D., Ford, J., & Brodsky, A. ( 2001 ) The visibility of illicit drugs: Implication for community based drug control strategies. American Journal of Public Health, 91, 1987 1994.
225 Sealock, M.D., Gottfredson, D.C., & Gallagher, C.A. ( 1997 ) Drug treatment for juvenile offenders: Some good and bad news. Journal of Research in Crime and Delinquency, 34, 210 236. Sellers, C. S., Cochran, J. K., & Branc h, K.A. ( 2005 ) Social learning theory and partner violence: A research note. Deviant Behavior 26, 379 395. Sellers, C. S., Cochran, J. K., & Winfree, L.T. (2003). Social learning theory and courtship violence: An empirical test. Pp. 109 128. In R.L. Ake rs and G.F. Jensen (eds.), Social learning theory and the explanation of crime: A guide for the new century. Advances in criminological theory Vol.11, New Brunswick, NJ: Transaction Publishers. Shaw, C. R. & McKay, H. D. ( 1942 ) Juvenile Delinquency in Urban Areas Chicago: University of Chicago Press Skinner W.F., & Fream, A.M. ( 1997 ). A social learning theory analysis of computer crime among college students Journal of Research in Crime and Delinquency, 34, 495 518. Spear, S., & Akers, R. L. (1988 ). Social learning variables and the risk of habitual smoking among adolescents: the Muscatine study. American Journal of Preventive Medicine 4, 336 348. Stephen, K., Flavio Francisco, M., Diane, S., & Tanya, N. ( 2007 ) Neighborhood effects on youth sub stance use in a southwestern city. Sociological Perspectives, 50, 273 301. sexual behaviors. Journal of Korean Alcohol Science 3, 132 144. South Korea Nation al Persecution Office. (2002). South Kore an supreme court drug report. Retrieved April, 22, 2010 from http://www.sppo.go.kr/drug/ Sunder, P.K., Grady, J.J., & Wu, Z.H. ( 2007 ). Neighborhood and individual factors in marijuana and other illicit drug use in a sample of low income women. American Journal of Community Psychology, 40, 167 180. Sundquist, K & Frank, G. (2004). Urbanization and Hosptital admission rates for alcohol and drug abuse: a follow up study of 4 .5 million women and men in Sweden. Society for the Study of Addiction, 99, 1298 1305. Sutherland, E. H. ( 1939 ) Principles of criminology 3rd edition. Philadelphia: J.B. Lippincott. Sutherland, E. H., & Cressey, D.R. ( 1966 ) Principles of criminology 7 th edition. Philadelphia: J.B. Lippincott.
226 Svensson, Robert. 2003. Gender differences in adolescent drug use: The impact of parental monitoring and peer deviance. Youth and Society, 34, 300 329. Swadi, H. ( 1999 ) Individual risk factors for adolescent substance use. Drug and Alcoho l Dependence 55, 209 224. Thonberry, T.P., Lizotte, A.J. Krohn, M.D., Franworth, M. & Jang, S.J. (1994). Delinquent peers, beliefs, and delinquent behavior: A longitudinal test of interactional theory. Criminology 32, 47 84 Tienda, M. ( 1991 ) Poor people and poor places: Deciphering neighborhood effects on poverty outcomes Macro micro linkages in sociology, Edited by Joan Huber, 244 262. Newbury Park, CA: Sage. Tittle, C., & Ward, D. (1993). The interaction of age with the correlates and causes of crime. Journal of Quantitative Criminology, 9, 3 53. Tittle, C., Ward, D., & Grasmick, H. ( 200 3 ) Gender, age, and crime/deviance: A challenge to self control theory. Journal of Research in Crime and Delinquency 40, 426 453. Triplett, R., & Payne, B. ( 2004 ) Problem solving as reinforcement in adolescent drug use: Implication for theory and policy. Journal of Criminal Justice 32 617 630. Unnever, J., Cullen, F., & Pratt, T. (2003). Parental management, ADHD, and delinque ncy Justice Quarterly, 20, 471 500. Valente, T., Gallaher, P., & Mouttapa, M. (2004). Using social networks to understand and prevent substance use: A transdisciplinary perspective. Subst ance Use & Misuse, 39 1685 1712. Valente, T. W., Hoffman, B. R., Ritt Olson, A., Lichtman, K., & Johnson, C. A. (2003). Effects of a social network method for group assignment strategies on peer led tobacco prevention programs in schools. American Journ al of Public Health, 93 1837 1843. Vazonyi, A. T., Pickering, L. E., Junger, M., and Hessing, D. ( 2001 ) An empirical test of a general theory of crime: A four nation comparative study of self control and the prediction of deviance. Journal of Research in Crime and Delinquency, 38 91 131. Vazonyi, A., Witteking, J., Belliston, L, & Van Loh, T. (2004). Extending the general theory of crime to the east: Low self control in Japanese late adolescents. Journal of Quantities Criminology, 20, 189 216. Verri ll, S.W. ( 2008 ) Social structure social learning and delinquency: Mediation or moderation? New York: LFB Scholarly Publishing
227 Vitaro, F., Brendgen, M.,& Tremblay, R. (2000). Influence of deviant friends on delinquency: Searching for moderator variables. Journal of Abnormal Child Psychology 28, 313 325. Wagner, F. A., & Anthony, J.C., ( 2002 ) From first drug use to drug dependence; Developmental periods of risk for dependence upon marijuana, cocaine, and alcohol. Neuropsychopharmacology, 26, 479 488. Wang, G.T., Qiao, H., Hong, S., & Zhang, J. (2002). Adolescent social bond, self control, and deviant behavior in China, International Journal of Contemporary Sociology 39, 52 68. Warner, L. A. and White, H. R. (2003) Longitudinal effects of age at onset and first drinking situations on problem drinking. Substance Use and Misuse 38 1983 2016. Warr, M. ( 1993a ) Age, peers, and delinquency. Criminology 31 17 40. ( 1993b ) Parents, peers, and delinquency. Social Forces 72 247 264. ( 1996 ) Organizatio n and instigation in delinquent groups. Criminology 34 11 38. (2002). Companions in crime: The social aspects of criminal conduct. Cambridge, UK: Cambridge University Press. Warr, M., & Stafford, M. ( 1991 ) The influence of delinquent peers: What they think or what they do? Criminology 29, 851 866. Waters, J. R., & Chester, A. J. (1987). Optimal allocation in multi variate, two stage sampling design. American Statistician, 41 46 50. White, H.R., Johnson, V. & Horowitz, A. (1986). An application of three deviance theories for adolescent substance use. International Journal of the Additions. 21, 347 366. White, H.R., Pandina, R.J., & LaGrange, R.L. (1987). Longitudinal predictors of serious substance use and delinquency. Criminology 25, 715 740. Wiatrowski, M.D., & Anderson, K. (1987). The dimensionality of the social bond. Journal of Quantitative Criminology 3, 65 81. Wilcox, H.C., Wagner, F.A., Anthony, J.C. ( 2002 ) Exposure opportunity: linking youthful cannabis use to hallucinogen use. Drug Alcohol Depend. 66, 127 135.
228 Wilson, N., Syme, S. L., Boyce, W. T., Battistich, V. A., & Slevin, S. ( 2005 ) Adolescent alcohol, tobacco, and marijuana Use: The influence of neighborhood disorder and hope. American Journal of Health Promotion 20, 11 19. Winfree, L. T. & Griffiths, C.T. (1983). Social learning and marijuana use: A trend study of deviant behavior in a rural middle school. Rural sociology 48, 219 239. Winfree L.T., Mays, G.L., & Vigil Backstrom, T. (1994b). Youth gangs and incarcerated delinquents: Exploring the ties between gang membership, delinquency, and social learning theory. Justice Quarterly 11, 229 256. Winfree, L. T., Sellers, C., & Clason, D.L. (1993). Social learning and adolescent deviance abstention: Toward understandin g reasons for initiating, quitting, and avoiding drugs Journal of Quantitative Criminology 9, 101 125. Winfree L.T., Vigil Backstrom, T., Mays, G.L. (1994a). Social learning theory, self reported delinquency, and youth gangs: A new twist on a general theory of crime and delinquency. Youth and Society 26, 147 177. Winstanley, E.L., Steinwachs, D.M., Ensminger, M.E., Latkin, C.A., Stitzer, M.L., & Olsen, Y. ( 2008 ). The association of self reported neighborhood disorganization and social capital with adolescent alcohol and drug use, dependence, and access to treatment. Drug and Alcohol Dependence 92, 173 182 Wood, P., Cohran, J. K., & Pfefferbaum, B. (1995). Sensation seeking and delinquent substance use: An extension of learning theory. Journal of Drug Issues 25, 173 193. Wood, P.B., Pfefferbaum, B., & Arneklev, B.J. (1993). Risk taking and self control: social psychological correlates of delinquency. Journal of Crime and Justice, 16, 111 130. Wright, D.A., Bobashev, G., & Folsom, R. ( 2007 ) Un derstanding the relative influence of neighborhood, family, and youth on adolescent drug use. Substance Use and Misuse 42, 2159 2171. Wright, B., Entner, R., Caspi, A., Moffitt, T., & Silva, P. (1999). Low self control, social bonds, and crime: Social c ausation, social selection, or both? Criminology 37, 479 514. Wu, Z. H., Eschbach, K., & Gardy, J.J. ( 2008 ) Contextual influences on ploydrug use among young, low income women: Effects of neighborhood and personal networks. The American Journal on Addi tions 17, 135 144. Yun, H. M., Kim, Y. S., & Jang, S. Y. (1999). The relationship between expectation of alcohol effects and alcohol use among South Korean high school students. Journal of Korea Social Welfare 38, 153 179.
229 Zhang, L., & Messner, S. F (199 5 ). Family deviance and delinquency in China. Criminology, 33, 359 388.
230 BIOGRAPHICAL SKETCH Eunyoung Kim is originally from Seoul, South Korea, where she was born and raised until the 8 th grade of school. Later, she and her family moved to K young Ju city, and she graduated from Kyoung University in Deagu, South Korea in 1996. After graduation, Eunyoung worked for Lufthansa German Airline Company as a Korean interprete r. Unfortunately, she had a serious injury in an airplane accident which fractured backbone in1998. Because of that accident, she would be disabled and was not expected to be able ever to walk again. However, thanks to the amazing grace of God, she complet ely recovered after a couple of years of hospitalization just like a miracle. Eunyoung, then, began to have a vision of helping and supporting marginalized people in society and spent a couple of years in service for several faith based non governmental or ganizations in South Korea. In 2003, Eunyoung entered graduate school at the University of Maryland at College Park to obtain the M.A. degree in Criminology and Criminal justice. After graduation from the University of Maryland with the M.A, she worked for the Police Foundation, a renowned research institution in Washington D.C. from 2006 to 2007. She also served a non government organization assisting North Korean refugees in the United States as well. Eunyoung continued her graduate education at the Unive rsity of Florida in Sociology and Criminology and Law to earn her Ph.D. in Criminology. With invaluable support from Dr. Akers, my mentor and my adviser, Eunyoung was able to work effectively. Upon completion of her Ph.D. program in 2010, Eunyoung will have a job with the Korean Institute of Criminology in Seoul, South Korea, where her great husband, Minwoo, and family currently live.