تاثیرات فردی و زمینه بر بزهکاری: نقش خانواده های تک والد
|کد مقاله||سال انتشار||مقاله انگلیسی||ترجمه فارسی||تعداد کلمات|
|38547||2002||13 صفحه PDF||سفارش دهید||7520 کلمه|
Publisher : Elsevier - Science Direct (الزویر - ساینس دایرکت)
Journal : Journal of Criminal Justice, Volume 30, Issue 6, November–December 2002, Pages 575–587
Abstract Research indicates that children are at risk for delinquency if they live in a single-parent family and if they live in areas with high levels of family disruption. Although there is a substantial amount of research on both the individual and aggregate relationships, examining delinquency at either of these two levels alone is not appropriate. Specifically, families do not exist in isolation as individual-level research inherently assumes, and aggregate research is concerned with explaining rates of delinquency as opposed to explaining influences on individual behavior. The current research used data from thirty-five schools, an important adolescent context, to determine the individual- and school-level effects of single-parent families on delinquency. The results from an overdispersed Poisson HLM regression model suggest both individual and aggregate effects, with a potential buffering effect of intact families regardless of any adolescents' specific family structure.
Introduction Both individual- and aggregate-level theories suggest that family structure is an important factor related to delinquency. Although both levels have been examined in isolation, there is no research explicitly modeling both effects simultaneously. Furthermore, each level of explanation implies different things, increasing the importance of properly separating the two levels of effects. This study advanced the criminological literature concerning the link between single-parent families and delinquency by exploring the theoretically important relationships in a methodologically appropriate way, using a multilevel design in order to separate the two levels of effects. Furthermore, this research argues that schools are an appropriate unit for assessing contextual effects on delinquency. This argument is based on (1) research suggesting that delinquency is often a group event and (2) the nature of school in terms of both pulling together same-aged children and creating a context favorable towards friendship formation and maintenance. Individual-level research There are many possible reasons why the absence of a parent in the home is associated with an adolescent's risk for delinquency, such as lower income (McLanahan, 1985) or higher residential mobility (Astone & McLanahan, 1994). The prevailing criminological notion follows research indicating that two parents are better able to care for, supervise, and socialize children than one parent Amato & Keith, 1991, Hirschi, 1969 and McLanahan & Sandefur, 1994. In general, both parents are important and the absence of one weakens family functioning. These general concepts are reflected in the notion of social control, particularly informal social control. The family is an important socializing and supervision agent, and the notion is that a child is exposed to lower levels of both types of social control when one of the two parents is missing Gottfredson & Hirschi, 1990, Mulligan, 1960 and Nye, 1957. Lower levels of supervision and effective socialization are associated with delinquency whether the child has one parent or two. Children with one parent are at higher risk of delinquency, then, because there is one less person capable of supervision. Thus, parental absence, in a very broad sense, is likely to reduce the level of social control to which the child is exposed. Recent criminological research concerning the effect of single-parent families on delinquency is rather sparse. Primarily due to backlash over the Moynihan Report in 1965, criminologists stopped looking directly at the relationship between broken homes and delinquency (Wilkinson, 1974). Instead, researchers focused on which family processes increased an adolescent's risk for delinquency when weakened. This shift in focus created a base of research with mixed results because studies did not examine the direct effect of single-parent families. This variation is partly attributable to the difference between older research that examined the total effect of living in a single-parent family and current research that usually reports the family structure effect after controlling statistically for other variables. Two meta-analyses illustrate this variability. First, Loeber and Stouthamer-Loeber (1986) found thirty-three of forty analyses with a strong, statistically significant association between broken homes or parental absence and a child's delinquency or aggression for cross-sectional studies. In contrast, Lipsey and Derzon (1998) reported broken homes as one of the weakest predictors of violent or serious delinquency for ages six to fourteen. Neither of these reviews discussed whether the effect sizes for family structure on delinquency are direct, indirect, or total effects. Both reviews suggested that living in a single-parent family increased an adolescent's risk for delinquency as compared with living with two parents. In sum, most individual-level studies of delinquency include a measure of family structure, but only as a control variable. Nevertheless, there is evidence that living in a single-parent family does increase an adolescent's risk for delinquency as compared with living with two parents. Unfortunately, current research has not adequately explored this relationship. Aggregate-level research There is also an aggregate-level relationship posited between family structure and the delinquency rates of an area. The aggregate-level relationship is different from the individual-level relationship discussed above. Specifically, whereas individual-level research explores the effects of living with one parent on the delinquent behavior of a particular adolescent, aggregate-level research explores how the proportion of single-parent families within some social unit potentially place a social unit, and by extension all children within that unit, at risk for higher rates of delinquency. Currently, most aggregate-level research stems from social disorganization theory and uses the neighborhood as the unit of analysis. The essence of social disorganization theory is that, at the aggregate level, informal social control mechanisms are not working effectively; the community is not able to manage the behavior of its residents Bursik, 1988, Bursik and Grasmick, 1993, Sampson & Groves, 1989 and Sampson & Lauritsen, 1994. The ability of a community to overcome common problems is hindered when formal and informal ties are not developed. Thus, the community is not able to rally against criminogenic forces that result from the weak levels of control because there is a deficit in the collective monitoring of individuals. For example, one important dimension of Shaw and McKay's (1969) framework of social disorganization was the ability to control peer groups (see Sampson & Groves, 1989 and Sampson & Lauritsen, 1994). Peer groups are of considerable importance because delinquency often is a group event Shaw & McKay, 1942 and Thrasher, 1927. The ability of a community to control these peer groups is thought to be directly and positively correlated with the community's delinquency rates. Shaw and McKay (1942, p. 20) suggested that residential mobility, poverty, and ethnic/racial heterogeneity undermined formal and informal community ties by decreasing communication and increasing anonymity among residents, thereby decreasing the chances someone will intervene to control the behavior of children (Bursik & Grasmick, 1993, p. 7). In 1987, Sampson argued that family disruption affects crime in a similar fashion. One reason was that a high level of single-parent families in a community weakens informal social controls Sampson, 1987 and Sampson & Lauritsen, 1994. Informal controls are those controls most likely to affect unsupervised peer groups, one of the leading predictors of higher neighborhood delinquency rates (Sampson & Groves, 1989, p. 778). Essentially, due to the presence of many households with absent adults, there are fewer adults available for the day-to-day monitoring of their own children and other children in the area Bursik and Grasmick, 1993 and Sampson, 1987. Informal social controls are weakened when there are high numbers of missing parents. This reasoning suggests an emergent property of single-parent families. That is, the adolescent in the setting with many single-parent families is at higher risk for delinquency than the adolescent in the setting with few single-parent families, regardless of any specific adolescent's family structure. Interestingly, some aggregate-level research points to family disruption as a leading predictor of juvenile delinquency, above and beyond the “usual” social disorganization variables, such as unsupervised peer groups and other formal and informal social controls. For example, Sampson and Groves (1989, p. 789) tested social disorganization theory using neighborhood characteristics and found that family disruption was significantly related to rates of various types of crime after controlling for mediating measures of social disorganization (local friendship networks, unsupervised peer groups, and organizational participation). A re-examination by Veysey and Messner (1999) found that after including other structural and mediating variables generally thought to influence levels of social organization, the structural effect of living in a single-parent family is still large. Those neighborhoods with high proportions of single-parent families also experience higher rates of delinquency. Moreover, this effect also holds for rural areas. Osgood and Chambers (2000) tested social disorganization theory in rural areas and found that a high proportion of female-headed households are associated with higher levels of rural violent delinquency, excluding homicide. Thus, there is aggregate-level support for a significant relationship between the proportion of single-parent families in a setting and the delinquency rates of that setting. In sum, social disorganization theory indicates that supervision of children by families is an important buffer against high rates of delinquency. At the individual level, a child in a single-parent home may be at higher risk for delinquency because fewer controls are placed over the child due to the absence of an adult in the home. Within the social disorganization framework, this notion is extended to a collective level: if there is a high proportion of single-parent homes in a social setting (e.g., neighborhood), the youths within the setting are at higher risk for delinquency regardless of their particular family arrangement. In such settings, delinquency rates are higher due in part to the ineffectual level of control over all the residents including, and especially, adolescents. Alternatively, there is some benefit for all children when intact families surround them. Multilevel research As discussed above, there is evidence that children in single-parent families are at higher risk for delinquency and that areas with high proportions of single-parent families have higher delinquency rates. Research conducted at either the individual or aggregate level, however, is susceptible to cross-level misspecification. Cross-level misspecification refers to assuming the unit of analysis determines the unit of causation (Sampson & Lauritsen, 1994, p. 80). In statistical terms, the coefficients that are obtained with research conducted at either level are pooled estimates and may contain any proportion of influence from either level. The impossibility of differentiating the two levels is also the source of the ecological fallacy, which is improperly inferring individual effects from aggregate data (Robinson, 1950). Aggregate studies of delinquency explain variation in delinquency rates by the collective characteristics of individuals in the social unit, but say nothing about which people are involved or affected (see Sampson & Lauritsen, 1994, p. 3). The current research used individual and aggregate or contextual level data in order to separate the two different effects and to explore the relationship between single-parent families and delinquency. Multilevel research has not examined the total relationship between both levels of family structure and delinquency. One reason is that many researchers take a factor analytic approach, which combines a number of aggregate-level variables into one “factor” that represents the variables' dominant theme Elliott et al., 1996, Sampson et al., 1999, Sampson et al., 1997 and Taylor & Covington, 1988. The factor analytic approach is useful particularly in ecological research because many aspects of neighborhoods are related (e.g., minority population and poverty rates). Specifically, collapsing several correlated variables into a single factor adjusts for the multicollinearity, yielding stable regression coefficients and greater statistical power. The result of this approach, however, is that the independent effects of particular components of a factor cannot be determined. Such is the case for family disruption; multilevel studies include some aggregate measure of family disruption, but it is combined with other variables in a larger factor rendering interpretation impossible. Thus, there is a lack of multilevel research concerning the relationship between single-parent families and delinquency. The school as a meaningful context As mentioned earlier, most contextual research is geographic in nature. The assumption is that individuals in close proximity to each other are likely to be similarly affected by the structural conditions of the spatial unit. Large spatial areas may be too geographically dispersed, however, to provide meaningful insight into contextual effects. Alternatively, macrolevel (or emergent) effects are evident in many contexts that are not necessarily rooted in residence. Based on the group nature of delinquency and the structure of schools for creating and maintaining friendship networks, the current research argues that schools are an appropriate context for examining delinquency. The family is a primary context for child development but it does not exist in isolation; rather, contexts of development are nested within other contexts (Bronfenbrenner, 1986). Literature on adolescent development suggests that three of the most important contexts in which adolescents are embedded are the family, peers, and school Steinberg & Darling, 1994 and Vazsonyi & Flannery, 1997. Due to the natural interaction between the family and school, these two contexts are often used to predict adolescent developmental outcomes such as school achievement Astone & McLanahan, 1991 and Thompson et al., 1988. For example, Pong, 1997 and Pong, 1998 examined the individual and school contextual effect of single-parent families on the reading and math achievement of each student. She found that schools with high proportions of single-parent families created a significant negative contextual effect on the math and reading achievement of eighth and tenth graders, over and above the effect of an adolescent's membership in a single-parent family. Criminologists seldom examine schools in multilevel research although they may be important for understanding delinquency. Schools bring together large groups of same-aged children to an environment with a highly skewed adult–child ratio, leaving children more room to operate away from watchful adults Corsaro & Eder, 1990 and Corsaro & Eder, 1995. Importantly, schools offer an opportunity for adolescents to form peer groups or to maintain peer groups already formed before they enter into school. Indeed, Ennett and Bauman (1993) reported that 95 percent of friendship's ties were between students attending the same school. There is extensive criminological literature concerning the group nature of delinquency Sampson & Groves, 1989, Shaw & McKay, 1942, Warr, 1996 and Zimring, 1981. Moreover, studies of social disorganization suggest that the presence of unsupervised peer groups is one of the strongest predictors of juvenile delinquency Sampson & Groves, 1989 and Veysey & Messner, 1999. Given that unsupervised peer groups contribute to high neighborhood delinquency rates, that delinquency is often a group event, and school is the dominant setting for friendship formation and/or maintenance, it seems reasonable to expect the school to play a part in an adolescent's risk for delinquency. The purpose in the current study was not to determine the impact of the school environment; rather, schools were used to identify aggregates of adolescents who were available to one another as potential companions (see also Osgood & Anderson, 2002). To date, two multilevel studies in criminological research used the school as a context for delinquency. The first examined the effects of academic values and the subculture of violence on interpersonal violence, theft/vandalism, and rule breaking (Felson, Liska, & South, 1994). A measure of family stability was included at both the individual- and school-level as control variables and both relationships were small and insignificant across all three dependent variables. The second employed an HLM analysis in order to determine individual, school, and community effects on school disorder (Welsh, Greene, & Jenkins, 1999). A factor analytic approach was taken with the concentration of single-parent families included in the community poverty factor, rendering interpretation impossible. It was evident that multilevel studies had not addressed the question of family structure effects on delinquency. The current research fills an empirical gap by describing the total effects of single-parent families on delinquency from both an individual and structural perspective. This research addresses two questions: First, are children from single-parent families more at risk for delinquency compared with children who live with two parents? Second, are adolescents who attend a school with a high proportion of single-parent families at higher risk for delinquency than those who attend school with mostly two-parent families? This question implies that all children are at risk for delinquency in schools with a high proportion of single-parent families, regardless of family structure.
نتیجه گیری انگلیسی
Results The statistical technique used for this analysis was a Poisson (overdispersed) HLM regression analysis (see Bryk et al., 1996, p. 150). The coefficients correspond to mean levels of offending and are in a log linear metric, which should be exponentiated to convert them to a multiplicative model for the original metric (Liao, 1994). The interpretation is “a one-unit increase in xij multiplies the expected incidents by a factor of exp(βj) and a one-unit decrease divides the expected incidents by the same amount” (Gardner et al., 1995, p. 396). Put simply, any negative coefficient is a fraction once exponentiated, and in a multiplicative model a fraction will decrease the expected delinquent incidents while a positive coefficient will increase the expected count (i.e., negative numbers reduce the odds while positive numbers increase the odds). Null model The top of Table 2 presents results from an analysis without any explanatory variables. The null model provides baseline information on the average number of offenses per adolescent for each of the three types of delinquency. For example, the coefficient 1.39 for status offenses means that the average number of status offenses for an adolescent is 1.39, in a log linear metric. The exponentiated values for the log linear coefficients are provided in the second column to ease interpretation (i.e., odds ratio). The coefficient of 1.39 converts to an average of 4.0 status offenses for an adolescent (i.e., e1.394, represented by the 4.03 in the second column) on a scale of zero to twenty-four. The average number of property offenses for an adolescent is about 3.4 out of forty-eight and about 3.5 out of forty-eight for person offenses. Table 2. Null model: regression coefficients and variance components Fixed effects γ Odds ratio Status offense, γ00 1.39*** (0.06) 4.03 Property offense, γ00 1.21*** (0.06) 3.35 Person offense, γ00 1.26*** (0.06) 3.51 Random effects (error variance components) Variance componentsa Proportion of total variance Status offense Between schools Mean status offenses, uoj 0.093*** 0.012 Within schools, rij 7.76 0.988 Property offense Between schools Mean property crime, uoj 0.076*** 0.005 Within schools, rij 14.53 0.995 Person offense Between schools Mean person crime, uoj 0.078*** 0.007 Within schools, rij 10.61 0.993 Figures based on an overdispersed Poisson HLM regression. Standard errors are in parentheses. *** P<.001 (two-tailed tests). a The variance components and the proportion of the total is in the log linear metric of the Poisson distribution. Table options The variance components or random effects for the null model are reported in the bottom part of Table 2. One feature of HLM is that it divides total variance into between-school and within-school variation. As is always the case in contextual analysis, there is more unexplained variance within schools than between schools (represented by the proportions in the second column). The between-school variance is about 1 percent of the total unexplained variance (in a nonlinear metric); however, there is still significant variation between schools across all three dependent variables. The overdispersed Poisson regression alters the Level 1 distribution rendering the Level 1 variance components unusable. For this reason, Fig. 1 presents the substantive meaning of the Level 2 variance components. The figures were obtained by taking the standard deviation for each dependent variable (i.e., the square root of the variance component) and adding/subtracting it from the mean to obtain the range, which is then exponentiated. The first row for each dependent variable shows the range for that variable in a log metric, while the second row shows the exponentiated, or standard metric, values. The point is to demonstrate the meaningful differences between schools in the average level of delinquency. Indeed, students in some schools commit two to three times as much delinquency as students in other schools. The magnitude of school differences across three types of delinquency implied by ... Fig. 1. The magnitude of school differences across three types of delinquency implied by an overdispersed Poisson regression. The figures are computed by taking the standard deviation for each dependent variable (the square root of the variance component) and adding/subtracting it from the mean to obtain the range, which is then exponentiated. Figure options Full model Table 3 presents a full model that includes the Levels 1 and 2 family structure variables, the individual-level control variables of sex, race/ethnicity, mother's education, and a dummy variable representing missing values for the mother's education variable (not reported). The coefficient for the school-level single-parent family variable is the school contextual effect, over and above the individual-level effect because the independent variables are all grand-mean centered (see Bryk & Raudenbush, 1992, p. 26). The reader should bear in mind that significant contextual effects are hard to obtain in a small sample such as this (thirty-five schools) and P-values less than .10 are reported below. Table 3. Full model overdispersed Poisson HLM regression coefficients by type of delinquency Fixed effects Status Property Person γ Odds ratio γ Odds ratio γ Odds ratio School mean (constant) Base (constant), γ00 1.34 (0.05) 3.82 1.11 (0.06) 3.04 1.17 (0.05) 3.23 Single-parent family, γ01 0.48† (0.27) 1.61 −0.65† (0.36) 0.52 0.65* (0.28) 1.92 Sex, γ10 0.25*** (0.042) 1.29 0.64*** (0.06) 1.89 0.59*** (0.05) 1.80 Minority, γ20 0.20** (0.05) 1.23 0.10 (0.07) 1.10 0.21** (0.06) 1.23 Single-parent family, γ30 0.24*** (0.05) 1.27 0.36*** (0.06) 1.43 0.26*** (0.05) 1.29 Mother's education, γ40 −0.05* (0.03) 0.95 −0.07† (0.04) 0.93 1.17 (0.05) 3.23 Error variance components Random effects Status Property Person Variance components R2 a Variance components R2 a Variance components R2 a Between schools 38.7% −11.8%b 39.7% Mean status offenses, uoj 0.057*** 0.085*** 0.047*** Mother's education, u4j 0.010** 0.032** 0.011* Within schools, rij 7.407 12.741 9.416 Standard errors are in parentheses and the coefficients are grand-mean centered. Figures based on an overdispersed Poisson HLM regression that also included a dummy variable representing missing values for mother's education (not reported). † P<.10. * P<.05 (two-tailed test). *** P<.001 (two-tailed test). ** P<.01 (two-tailed test). a R2's are computed only for the between-schools variance (i.e., Level 2) using the formula (uojNULL−uojFULL)/uojNULL. The Level 1 variance components do not lend themselves to this method because this model is overdispersed at Level 1, which alters the Poisson distribution. b Negative R2's are possible in multilevel modeling (see Kreft & De Leeuw, 1998, p. 118) because the inclusion/withdrawal of particular variables may increase/decrease the residual variance of school-level delinquency means. Table options The Level 1 single-parent family variable (γ30) maintains significance across all three models (P<.001). This finding suggests that adolescents living in a single-parent family were at significantly higher risk for status, property, and person delinquency than adolescents who lived with two parents. This finding was significant controlling for sex, minority group status, and mother's education, a proxy measure of income. The school-level single-parent family variable (γ01) was significant for person offenses (P<.05), and was marginally significant for status and property offenses (P<.10). This suggests that students attending school where a high proportion of students live with only one parent are at significantly higher risk for person delinquency than those attending school where a low proportion of students live with only one parent, and they were somewhat more at risk for committing status offenses. The results also indicate that adolescents attending schools with a high proportion of single-parent families were less likely to commit property offenses. For the control variables, being male was significantly related to all three types of delinquency. Minority group members were significantly more at risk for status and person offenses but was not significantly related to property offending. Mother's education has a borderline significant relationship to status and property offending, with adolescents whose mothers had more education being at lower risk. The random effects or variance components for the full model are reported in the bottom of Table 3. The residual variance of school-level averages decreased by approximately 40 percent for status offenses and person crime. The residual variance at Level 2 for the full model actually increased from the null model for property offenses (see Kreft & De Leeuw, 1998), meaning there is more unexplained variance than there was before including any variables (i.e., 11.8 percent more). This is a function of shifting some of the Level 1 variance to Level 2 once contextual variables were introduced. The adjusted means now have more variance around them than did the original means from the null model. Fig. 2 presents a graphical substantive interpretation of the results by demonstrating the mean predicted delinquency counts. These counts are based on the range of single-parent families between schools for the current sample and holding the control variables constant at their mean. Status and person offenses follow the same general pattern, with adolescents living with two parents and attending school where only 9 percent of students who live with one parent are least at risk. Adolescents who live with two parents but attend school where 74 percent of the students live with only one parent commit slightly more status and person offenses than adolescents who live with only one parent and attend school where 91 percent of the students live with two parents. This suggests that attending school where a higher proportion of the students have two-parent families can reduce some of the risk involved with an adolescent living in a single-parent family for these two offenses. Additionally, the average number of property offenses is lowest for those living with two parents but attending a school where 74 percent of the adolescents live with only one parent, followed by the adolescents who live with only one parent but attend a school where 74 percent of the adolescents live with only one parent. The highest risk offenders were those that attended school where 91 percent of the kids lived with two parents, with the highest risk offender living with only one parent. Predictive values for single parent families and delinquency based on full ... Fig. 2. Predictive values for single parent families and delinquency based on full model. The figures are derived using uncentered coefficients from an overdispersed Poisson HLM regression and the sample range while holding all other variables constant at their mean. Figure options Structural disadvantage The previous analyses did not include any contextual variables other than the proportion of single-parent families within a school. As is evidenced from ecological and contextual research, the percentage of single-parent families is correlated with other structural characteristics, especially measures of disadvantage. To test for the possibility that the school-level structural effect of single-parent families might be spurious due to structural disadvantage, another model was estimated including both the proportion of minority students and mean mother's education as school-level variables. The school-level coefficients are the only coefficients presented in Table 4 because only they are affected by the addition of the school-level disadvantage measures. Table 4. Disadvantage-overdispersed Poisson HLM regression coefficients Fixed effects Status Property Person School-level variables γ Odds ratio γ Odds ratio γ Odds ratio Base (constant), γ00 1.34 (0.05) 3.81 1.12 (0.06) 3.04 1.16 (0.05) 3.23 Single-parent family, γ01 0.45 (0.41) 1.57 −0.75 (0.55) 0.47 0.91* (0.43) 2.48 (Single-parent family, γ01) a 0.48 †(0.27) 1.61 −0.65 †(0.36) 0.52 0.65 *(0.28) 1.92 % Minority students, γ02 0.05 (0.28) 1.05 0.09 (0.38) 1.89 −0.19 (0.29) 1.80 Mother's education, γ03 0.07 (0.13) 1.07 −0.03 (0.18) 1.10 0.12 (0.15) 1.23 Figures based on an overdispersed Poisson HLM regression. Standard errors are in parentheses and the coefficients are grand-mean centered. * P<.05 (two-tailed test). a The italicized “single-parent family” variables are the figures presented in Table 3. † P<.10 (two-tailed test). Table options The results indicate that an adolescent who attended a school with a high proportion of single-parent families was at significantly higher risk for person delinquency than an adolescent in a school with a low proportion, even after controlling for school-level disadvantage. For status offenses, the magnitude of the single-parent family effect is slightly smaller than in the full model discussed earlier. The coefficient for property offenses remained negative and increased; meaning the effect became smaller and further reducing the risk of committing a property offense for those attending school with a high proportion of single-parent families. Generally, there is a loss in statistical power in this model because the sample size at Level 2 is relatively small and the contextual measures are somewhat collinear. The overall pattern of the results, however, suggests that the school-level single-parent family effect is still important. In particular, the disadvantage measures are small, insignificant, and oscillate between positive and negative in direction across all three types of delinquency. The change in the magnitude of the coefficients in addition to the loss of significance calls for a more thorough investigation of contextual single-parent family effects with a larger sample of schools. Limitations There are several threats to validity that warrant mentioning. First, due to the cross-sectional nature of the data, there is no way of determining whether single-parent families “cause” delinquency, or whether delinquency “causes” single-parent families, creating ambiguity about the direction of causal influence (Cook & Campbell, 1979). Although it is possible that delinquency affects family structure, chances are low that the effect is strong enough to render this analysis futile. Second, there is no way to discern whether the self-reported delinquent acts were committed while attending the present school or a different school. The overall movement of all students among schools, however, may work to cancel each other out, not greatly affecting the results. At worst, the current analysis suffers from a reduction in accuracy. There may be some selection issues concerning the sample and students. Specifically, students filled out the interview in school, and those students who were absent the day it was administered were not included. These absent students might also be more delinquent than the children who were attending that day. Finally, these schools were chosen as part of an evaluation effort. Thus, the sample students may not be representative of eighth grade students and the generalizability of the results is suspect. Nevertheless, the schools vary on many community level dimensions, such as size of school and minority population, potentially minimizing bias (see Esbensen & Osgood, 1999, p. 201).