محل جایگذاری و بزهکاری نوجوانان: آیا محلات در رفاه کودکان اهمیت دارد؟

Abstract This study aims to advance the knowledge base by investigating where foster youth are placed in terms of neighborhood characteristics and whether specific neighborhood characteristics were associated with delinquency for adolescents in the child welfare system. This study followed the placement experiences of 2360 foster youth in Chicago from birth to 16 years of age. The study used State administrative data, census data, and the community survey of the Project of Human Development in Chicago Neighborhoods. The results indicated that foster care placements cluster in neighborhoods characterized by high concentrated disadvantage, low ethnic heterogeneity, low collective efficacy, prevalent neighborhood disorder and violent culture. The results indicated that neighborhood ethnic heterogeneity is positively associated with delinquent offending. The implications for policy and practice are discussed.

. Introduction Victims of child maltreatment show a higher risk of juvenile delinquency than their non-maltreated peers (English et al., 2002, Ryan and Testa, 2005, Smith and Thornberry, 1995, Widom, 1989 and Zingraff et al., 1993). English et al. (2002) reported that abused and neglected children were 11 times more likely to be arrested for a violent crime as a juvenile, as compared to the matched control group. In addition to first time offending, victims of child maltreatment show a higher risk of recidivism (Halemba et al., 2004, Huang et al., 2012 and Ryan, 2006). The increased delinquency rate is especially true for the child welfare youth placed in substitute care settings (Doyle, 2007 and Ryan and Testa, 2005). Yet, to date, no study has examined the neighborhood effects associated with such placements. The current study addresses this critical gap in the literature. 1.1. Child welfare placement and juvenile delinquency According to national statistics, 61.2% of maltreated youth received post-response child welfare services and 36% received out of home placement services (U.S. Department of Health and Human Services, 2012). Regarding the impacts of placement on juvenile crime, findings are often mixed, which might be due to the diverse measurements of delinquency (e.g. formal arrest records, self-reports). Widom (1991) reported that out-of-home placement was not associated with delinquency rates. The author reported that delinquency arrests occurred among 15.1% of maltreated children never placed, 17.8% of maltreated children whose placement was related only to maltreatment, and 92.7% of children placed due to both delinquent behavior and maltreatment. It appeared that maltreated children never placed and those whose placement was related only to maltreatment had similar risks of arrest, and that these youth had much lower risks as compared with children placed due to both delinquent behavior and maltreatment. With regard to mixed findings, some authors (Jonson-Reid and Barth, 2000 and Lemmon, 2006) reported that child welfare placements might actually help reduce the risk of juvenile justice involvement. Jonson-Reid and Barth (2000a) analyzed administrative data of 159,549 maltreated children in ten counties in California. The authors reported that the provision of child welfare services, including in-home and foster placement, did not change the risk of incarceration for European American children. However, for African American and Latino children, the receipt of child welfare services significantly decreased the risk of incarceration. Similarly, studying a cohort (N = 632) receiving financial supports or other services from the Pennsylvania Department of Public Welfare's Office of Income Maintenance, Lemmon (2006) reported that having a child welfare placement reduced the odds of a delinquency referral. Moreover, the author also found that placement reduced the continuation and severity of delinquency. In contrast, others (Runyan and Gould, 1985 and Ryan and Testa, 2005) reported the deleterious effects of placement experiences. Using administrative data from Cook County, Illinois, Ryan and Testa (2005) investigated the relationship between placement, placement instability, and juvenile delinquency. Their findings indicated that children in placement were at an increased risk of delinquency as compared with children not entering placement. The type of placement might also matter in child welfare, in particular group care and kinship care. Findings from criminal justice studies (Dishion, McCord, & Poulin, 1999) indicate that that congregate care, such as group homes, increased the risk of delinquency (Jonson-Reid and Barth, 2003 and Ryan et al., 2008). Using a sample of 20,309 children with at least one placement episode from Los Angeles County Ryan et al. (2008) compared the delinquency outcome between group home and foster care children. To reduce selection bias, the authors used propensity score matching. The authors included the variables race, gender, maltreatment type, reason for placement change (i.e. runaway and behavior problems), age at first placement, length of stay in placement, and total changes in placement in the matching. After the matching, the results from Cox regression showed that the odds ratio of delinquency outcome for children with at least one group home episode was 2.40 times greater than for children in foster care. The authors suggested that peer contagion and group home policies pertaining to contacting law enforcement might contribute to the higher risk of delinquency. The kinship care literature is less conclusive (Rubin et al., 2008 and Ryan et al., 2010). Using a national sample of 1309 children placed in out-of-home care from NSCAW, Rubin et al. (2008) found that kinship care reduced the risk of behavioral problems. Depending on the timing of entering kinship care, the authors grouped children into three types of placement, early kinship care (kinship care within 1 month), late kinship care (kinship care beyond 1 month), and general foster care (never in kinship care). The authors reported abnormal behavioral outcomes among 32% of early kinship care, 39% of late kinship care, and 46% of general foster care children after 36 months. However, Barth (2008) suggested that the relationship between kinship care and better behavioral outcomes may be limited by measuring behavioral problems reported by caregivers, since relatives may be less likely to report problematic behaviors than foster parents. More recently, Ryan et al. (2010) studied the relationship between kinship care and juvenile delinquency. Their finding was that among males, kinship care was associated with higher likelihood of delinquency for African Americans, European Americans, and Asians, while it was associated with lower likelihood of delinquency for Hispanics. Among females, kinship care was associated with lower likelihood of delinquency for Hispanics, while it was not associated with delinquency for other race groups. Their findings highlighted that kinship care effects varied with race/ethnicity and gender. 1.2. Neighborhoods and juvenile delinquency Child welfare systems are required to consider the location of foster placements, but not necessarily specific to neighborhood characteristics. As stated in the Adoption Assistance and Child Welfare Act of 1980 (U.S. Public Law 96-272), agencies should find “the least restrictive (most family-like setting) and most appropriate setting available and in close proximity to the parents' home, consistent with the best interest and special needs of the child”. In practice, placing children in their neighborhoods of origin is the placement priority. Yet this practice is not absent in debate. Berrick (2006) acknowledged that placing children in their neighborhoods of origin may minimize academic disruptions, encourage cultural continuity, and encourage parents' visits to their children in care, which increases the likelihood of reunification. However, Berrick also argued that placing children in their neighborhoods of origin extended their exposure to distressed neighborhoods, which compromised their potential for achievement in various aspects of life, such as academic performance, health and mental health, and delinquent behavior. In contrast, Crampton (2007) argued that placing children in their neighborhoods of origin facilitated the ability of child welfare agencies to change neighborhoods. Child welfare agencies could develop partnerships with concerned citizens in those neighborhoods, such as asking them to become foster parents, and provide mentoring or respite for struggling families. Crampton believed that these efforts could strengthen social integration and community support for families, which, in turn, could reduce child maltreatment and the need for foster care. Both Berrick (2006) and Crampton (2007) agree that more research is needed to better understand how neighborhoods may influence children and youth's well-being. One of the important indicators of well-being is behavior outcomes like delinquency. To date, no study has examined how neighborhoods associated with foster homes might influence delinquency. The current study addresses this critical gap in the literature. Juvenile justice scholars have a long tradition of studying neighborhood impacts. Numerous studies have found that neighborhood conditions were associated with delinquency (Abrams and Freisthler, 2010, De Coster et al., 2006, Grunwald et al., 2010, Mennis and Harris, 2011, Mennis et al., 2011, Sampson et al., 2005 and Shaw and McKay, 1942). As early as 1942, Shaw and McKay published their empirical study on several big cities. The authors reported that juvenile delinquency was concentrated in the neighborhoods characterized by social disorganization. Social disorganization theory and social norm theory are used to explain the neighborhood-delinquency relationships. Social disorganization theory emphasizes the inability of a community structure to realize the common values of its residents and maintain effective social controls. Accordingly, neighborhoods characterized by high poverty, residential instability, and ethnic heterogeneity have limited social control over the behaviors of the residents, and therefore, experience high crime rates. Collective efficacy is an important concept in the theory and is defined as social cohesion among neighbors, combined with their willingness to intervene on behalf of the common good (Sampson, Raudenbush, & Earls, 1997). Shaw and McKay (1942) first proposed social disorganization theory. Since the 1980s, researchers have started to directly measure social disorganization and test its mediation effect (Elliott et al., 1996, Sampson, 1997, Sampson and Groves, 1989 and Sampson et al., 1997). There have been two types of studies, neighborhood-level and multi-level studies. The first group of empirical studies used neighborhood level data. Researchers demonstrated that social disorganization mediated the effect of neighborhood conditions on neighborhood delinquency rates at the neighborhood level (Sampson & Groves, 1989). The second group of empirical studies used both neighborhood-level and individual-level data. These studies benefited from the development of the hierarchical linear model (HLM), which takes into account the dependence between individuals nested in the same neighborhoods (Raudenbush & Bryk, 1992). HLM separates the effect of neighborhoods from the effect of individuals and families, and therefore, reduces selection bias. Using HLM, Elliott et al. (1996) reported that the organizational and cultural characteristics of neighborhoods mediated the effect of neighborhood disadvantages on problem behavior, which included delinquent behavior, drug use, and arrest. The authors analyzed data from Chicago and Denver and reported that informal control accounted for 60% of the variance in problem behavior between neighborhoods in Chicago, and informal networks accounted for 26% of the variance in problem behavior between neighborhoods in Denver. Sampson et al. (1997) also reported that collective efficacy mediated the effects of concentrated disadvantages and residential stability on violence. Social norm theory emphasizes the effect of subculture on delinquency. Social norm theory was proposed in Anderson's (1999) ethnographic study of neighborhoods in Philadelphia. Anderson found that neighborhood subculture mediated the association between neighborhood conditions and violent delinquency. The high rates of male joblessness, poverty, substance abuse, and the lack of institutional resources among poor inner-city black neighborhoods fostered the violence-prevalent “code of street”, i.e. a set of informal rules governing interpersonal public behavior. As the poor inner-city black neighborhoods became alienated from mainstream society and ignored by institutions like the police. The residents relied on violence to defend themselves and earn respect (Anderson, 1999). It is important to note that neighborhood effects require time to operate. A review of neighborhood effects studies (Dietz, 2002) suggested that the magnitude of neighborhood effects depends on the duration of neighborhood membership. The longer an individual stays in the same neighborhood, the greater he/she is influenced by the neighborhood (Ellen & Turner, 1997). Using data from the Moving To Opportunity (MTO) Demonstration, Kling, Liebman, and Katz (2007) weighted neighborhood poverty by duration in neighborhood. Their findings indicate that duration-weighted neighborhood poverty was negatively associated with an overall index of adult outcomes. Clampet-Lundquist and Massey (2008) included the measure of the cumulative amount of time spent in different neighborhood environments, and reported that neighborhood is associated with financial outcomes such as employment, earnings, TANF receipt, and use of food stamps. In a qualitative study at the MTO Baltimore site, Turney et al. (2006) addressed the question why the experimental group did not show significantly better adult economic self-sufficiency. One reason is that the experimental group had yet to remain in the low poverty neighborhoods for a sufficient period of time. 1.3. Neighborhood of child welfare population Few child welfare studies examined the effect of neighborhoods on delinquency. The only two studies (Schuck and Widom, 2005 and Yonas et al., 2010) in this area reported that neighborhood conditions moderated the relationship between maltreatment and delinquency. Each study used different neighborhood measures. Schuck and Widom (2005) reported that neighborhood disadvantages and residential stability moderated the relationship between early child maltreatment and offending. The authors showed that, in the more disadvantaged and more stable neighborhoods, early child maltreatment had a greater impact on later juvenile and adult criminal behavior. Yonas et al. (2010) also reported that neighborhood conditions moderated the relationship between neglect and externalizing behavior. The authors used a sample of 861 caregivers and 823 youth from a longitudinal study of child abuse and neglect (LONGSCAN). The neighborhood measure was the mean of caregivers' answers to 12 items accessing collective efficacy (i.e. active participation by neighbors to provide a close, responsible, and accountable neighborhood). The authors demonstrated that in neighborhoods with higher levels of collective efficacy, neglect youth had lower externalizing behavior scores in CBCL. Both studies presented significant limitations with respect to their neighborhood measurements. They only measured the neighborhood conditions of children's original neighborhoods. They did not take into consideration that some children were moved out of their homes after maltreatment. As shown in the annual national report (U.S. Department of Health and Human Services, 2012), 36% of victims of maltreatment were placed into substitute care. Some of the children remain out of their homes for a long time and experienced multiple placement changes. It is possible that their placements are located in neighborhoods outside their original neighborhoods. They in turn may socialize with residents in the new neighborhoods, and adapt to the culture of their new neighborhoods. There is a need of research to examine the impact of new neighborhoods. Developing knowledge in this aspect can inform the debate on location based placement criteria (Berrick, 2006 and Crampton, 2007). Our review of empirical work to date indicates that neighborhoods do in fact affect delinquency. Social disorganization theory and social norm theory provide the framework for understanding this relationship. The current study addresses the following research questions: 1) How were foster care placements distributed in Chicago? 2) Do the neighborhoods associated with foster care placement affect the risk of delinquent offending? 3) Do social disorganization and social norms mediate the relationship between neighborhood demographics and delinquency? 4) Does length of time in care moderate the relationship between neighborhood characteristics and delinquency among foster youth?

. Results 3.1. Exploratory spatial presentation Fig. 2 shows the spatial distribution of the sample of 2360 foster youth on the Chicago map color shaded by neighborhood poverty rate. Each point indicates the weighted mean center of one foster youth. Fig. 2 shows that foster youth cluster in the neighborhoods of high poverty rates. To provide context for understanding the results, we will compare neighborhood variables between all neighborhoods and our study sample in Chicago in Table 1. Weighted mean centers on the map color shaded by neighborhood poverty rates. Fig. 2. Weighted mean centers on the map color shaded by neighborhood poverty rates. Figure options Table 1. Compared means between all neighborhoods in Chicago and the neighborhoods associated with our study sample. Chicago (N = 343) Study sample (N = 2360) p-Value for equal mean test p-Value for equal variance test Mean (std) Mean (std) Census variables Below poverty line 0.23 (0.17) 0.32 (0.17) ≤ 0.001 0.79 On public assistance 0.17 (0.15) 0.28 (0.15) ≤ 0.001 0.40 Female-headed families 0.23 (0.15) 0.34 (0.13) ≤ 0.001 ≤ 0.001 Unemployed 0.14 (0.10) 0.21 (0.10) ≤ 0.001 0.75 Less than age 18 0.27 (0.09) 0.32 (0.07) ≤ 0.001 ≤ 0.001 African American 0.42 (0.44) 0.80 (0.33) ≤ 0.001 ≤ 0.001 Latino 0.20 (0.26) 0.10 (0.20) ≤ 0.001 ≤ 0.001 Foreign born 0.17 (0.16) 0.06 (0.11) ≤ 0.001 ≤ 0.001 Same house as in 1985 0.56 (0.13) 0.60 (0.11) ≤ 0.001 ≤ 0.001 Owner-occupied house 0.42 (0.24) 0.39 (0.22) 0.01 0.05 Neighborhood social process Collective efficacy 7.24 (0.56) 6.97 (0.45) ≤ 0.001 ≤ 0.001 Disorder 1.83 (0.34) 2.01 (0.27) ≤ 0.001 ≤ 0.001 Violent culture 2.49 (0.13) 2.56 (0.12) ≤ 0.001 ≤ 0.001 Table options In the table, we showed both the p-values of the t-test for equality of means and Levene's test for equality of variances. The t-test for equality of means showed that the means of all variables were significantly different between all Chicago neighborhoods and the neighborhoods associated with our sample. Comparatively, foster youth in our sample lived in neighborhoods of higher rates of poverty, residents on public assistance, female-headed families, unemployment, residents less than 18 years of age, African American residents, residents living in the same house as in 1985, and lower rates of Latino residents, foreign born residents, and owner-occupied houses. For the neighborhood social processes, the neighborhoods associated with our sample scored lower on collective efficacy, higher on neighborhood disorder, and higher on violent culture. In summary, the quality of neighborhoods associated with our sample is significantly below the average of Chicago neighborhoods. The Levene's test for equality of variances (Table 1) showed that the variances of most variables were significantly different between all Chicago neighborhoods and the neighborhoods associated with our sample. As compared with all Chicago neighborhoods, the neighborhoods associated with our sample had smaller variances on rates of female-headed families, rates of residents less than 18 years of age, rates of African American residents, rates of Latino residents, rates of foreign born residents, rates of residents living in the same house as in 1985, rates of owner-occupied houses, collective efficacy, neighborhood disorder, and violent culture. The smaller variances for our sample indicate that the neighborhoods associated with foster care placements were more homogeneous. There are limited variances on the neighborhood variables. When independent variables cluster at the highest values, dependent variables are restricted by ceiling effects (Shadish, Cook, & Campbell, 2002, p. 50), which can threaten the power of study. In summary, the comparison indicates that foster care placements cluster in neighborhoods of high concentrated disadvantage, low ethnic heterogeneity, low collective efficacy, prevalent neighborhood disorder and violent culture. 3.2. Descriptive analysis The results of descriptive analysis are displayed in Table 2. We conducted descriptive analysis of the whole sample, the subsample of 5 + years in care, and the subsample of no more than 5 years in care, respectively. The columns on the left show the frequencies and the ones on the right show the percentages. Table 2. Descriptive analysis. Frequency Percentage All (N = 2360) 5 + years (N = 767) No more than 5 years (N = 1593) All (N = 2360) 5 + years (N = 767) No more than 5 years (N = 1593) Age at initial maltreatment (numeric) 6.45 4.68 7.30 Age at initial maltreatment Early childhood 1136 525 611 48.14 68.45 38.36 Late childhood 1074 229 845 45.51 29.86 53.04 Adolescence 150 13 137 6.36 1.69 8.60 Male 1131 369 762 47.92 48.11 47.83 Race/ethnicity African American 2114 729 1385 89.58 95.05 86.94 Latino 170 23 147 7.20 3.00 9.23 White 76 15 61 3.22 1.96 3.83 Maltreatment type Sexual abuse 275 88 187 11.65 11.47 11.74 Physical abuse 330 107 223 13.98 13.95 14.00 SEI 34 13 21 1.44 1.69 1.32 Neglect 1882 665 1217 79.75 86.70 76.40 Risk of harm 996 235 761 42.20 30.64 47.77 Congregate care 286 64 222 12.12 8.34 13.94 Placement instability One placement 843 207 636 35.72 26.99 39.92 Two placements 616 215 401 26.10 28.03 25.17 Three placements 384 141 243 16.27 18.38 15.25 Four + placements 517 204 313 21.91 26.60 19.65 Delinquency petition 264 88 176 11.19 11.47 11.05 Table options Regarding the whole sample, 48.14% of the youth had their first indicated maltreatment report in their early childhood (0–5 years of age); 45.51% of them had their first indicated maltreatment report in late childhood (6–11 years of age); and 6.36% had their first indicated maltreatment report in adolescence (at least 12 years of age). Only a small percent of youth had first report in adolescence, since we excluded the foster youth who entered placement at age 14 or older. Slightly less than half (47.92%) of the youth were male. The vast majority (89.58%) of the sample were African American, 7.2% were Latino, and 3.22% were European American. Maltreatment type indicates the types of all indicated maltreatment allegations associated with each person. Each person can have more than one maltreatment type. The most prevalent types were neglect (79.75%) and risk of harm (42.20%). Few cases (1.44%) were substance exposed infants. Over one tenth (12.12%) of the sample had at least one episode of stay in congregate care placements (i.e. group home). Placement instability indicates the number of placements that lasted 14 days or longer. Slightly over one third of the sample stayed in one out-of-home placement, 26.10% stayed in two placements, 16.2% stayed in three placements, and 21.91% stayed in four or more placements. For the dependent variable, 11.19% of the sample had at least one delinquency petition at age 14 or older, which is higher than the rate among the general population, 3.48% (Puzzanchera et al., 2000). The subsamples of different lengths of stay have some statistics substantially different from the whole sample. First, nearly 70% of the subsample of 5 + years in care experienced early childhood maltreatment, which is 20% higher than the whole sample, and 30% higher than the subsample of no more than 5 years in care. It is not surprising, since those of 5 + years in care entered care at least 5 years before 14 years of age, which means that they were most likely to have experienced maltreatment at an early age. Second, 95.05% the subsample of 5 + years in care was African American, which is 5% higher than the whole sample, and 8% higher than the subsample of no more than 5 years in care. Previous literature also reported that African Americans tend to stay in care longer than other races (Courtney & Wong, 1996). Third, the subsample of 5 + years in care had more neglect cases but fewer risk of harm cases than the whole sample, while the subsample of no more than 5 years in care had fewer neglect cases but more risk of harm cases than the whole sample. Fourth, the subsample of 5 + years in care had fewer cases of one placement only than the whole sample, while the subsample of no more than 5 years in care had more cases of one placement only than the whole sample. This is not surprising, since foster youth in care for longer times are likely to experience placement change. 3.3. Path analysis We first modeled delinquency petition using collective efficacy as the mediator, and regressed collective efficacy on three neighborhood sociodemographic factors. Model fit indices show a good fit (RMSEA = 0.02, CFI = 1.00, TLI = 0.99, WRMR = 0.77). The variance explained in overall delinquency is 14% and collective efficacy is 57%. According to Cohen (1988), the benchmarks for small, medium, and large effect size for R2 are 0.02, 0.13, and 0.26, respectively. Therefore, the R2 for overall delinquency is medium, and for collective efficacy, the R2 is large. The standardized path coefficients are displayed in Table 3. Since the models examine both the direct and indirect effects of neighborhood sociodemographics, we show them separately in two different columns. Table 3. Path analysis modeling delinquency with collective efficacy as mediator. Variables Whole sample (N = 2360) 5 + years sample (N = 767) No more than 5 years sample (N = 1593) Direct Indirect Direct Indirect Direct Indirect Probit regression on delinquency African American 0.36 0.00 − 0.13 0.01 0.53 0.00 Neighborhood sociodemographics Concentrated disadvantage 0.00 0.02 − 0.05 0.05 0.02 0.01 Ethnic heterogeneity 0.08 0.00 − 0.02 0.01 0.13⁎ 0.00 Residential stability 0.00 − 0.01 0.05 − 0.02 − 0.02 − 0.01 Neighborhood mediator Collective efficacy − 0.03 − 0.07 − 0.02 Individual characteristics Age at first maltreatment (Reference: Early childhood) Late childhood 0.01 0.02 0.01 Adolescence − 0.22 − 0.31 − 0.23 Male 0.63⁎⁎⁎ 0.58⁎⁎⁎ 0.66⁎⁎⁎ Race/ethnicity (Reference: White) Latino − 0.22 0.10 − 0.28 Maltreatment type Physical abuse 0.02 − 0.13 0.11 Neglect − 0.01 − 0.02 0.00 Maltreatment report count − 0.02 0.04 − 0.05 Congregate care 0.17 0.40 0.07 Placement instability (Reference: One placement) Two placements 0.11 0.19 0.09 Three placements 0.20 0.25 0.19 Four + placements 0.28⁎⁎ 0.17 0.37⁎⁎ Linear regression on mediator Concentrated disadvantage − 0.63⁎⁎⁎ − 0.66⁎⁎⁎ − 0.63⁎⁎⁎ Ethnic heterogeneity − 0.12⁎⁎⁎ − 0.12⁎⁎⁎ − 0.11⁎⁎⁎ Residential stability 0.26⁎⁎⁎ 0.22⁎⁎⁎ 0.27⁎⁎⁎ African American − 0.12⁎⁎⁎ − 0.10⁎⁎ − 0.14⁎⁎⁎ R square Delinquency 0.14 0.12 0.18 Collective efficacy 0.57 0.58 0.57 Model fit RMSEA 0.02 0.00 0.02 CFI 1.00 1.00 0.99 TLI 0.99 1.03 0.98 WRMR 0.77 0.39 0.77 ⁎ p ≤ 0.05. ⁎⁎ p ≤ 0.01. ⁎⁎⁎ p ≤ 0.001. Table options For the whole sample, none of the neighborhood sociodemographic factors showed significant effects on delinquency. Collective efficacy, the mediator, was also not significant in the model. Two individual variables were significantly associated with delinquency. Male foster youth are more likely to offend than female foster youth (β = 0.63, p ≤ 0.001). The interpretation of coefficients in the probit model is not as straightforward as linear regression. Unlike the logit model, the coefficient in the probit model cannot be directly reported as changes on odds ratio. Therefore, we manually calculated the average effects of each significant variable with setting the other significant variables at their mean, which is a common practice (Long, 1997). We also had to use unstandardized coefficients that are not reported in the table for the calculations. For example, the effect of being a male is: pr(y = 1|male = 0)=Φ(0.681 × 0 + 0.307 × 0.22 1) = Φ(0.07) = 0.52790 pr(y = 1|male = 1)=Φ(0.681 × 1 + 0.307 × 0.22) = Φ(0.75) = 0.77337 0.77337 − 0.52790 = 0.24547_ So the probability of delinquency is 0.25 higher for male foster youth than for female foster youth. Using the same approach, the effect of four + placements can be interpreted as follows: having four + placements is associated with 0.112 higher probability of delinquency as compared with foster youth with one placement experience only. In the regression on collective efficacy, the mediator, three neighborhood sociodemographic factors, and being African American were all significant. Each standard deviation increase on concentrated disadvantage is associated with a 0.63 standard deviation decrease on collective efficacy. Each standard deviation increase on ethnic heterogeneity is associated with a 0.12 standard deviation decrease on collective efficacy. Each standard deviation increase on residential stability is associated with a 0.26 standard deviation increase on collective efficacy. African American foster youth are associated with a 0.12 standard deviation lower on collective efficacy than foster youth of other races. As mentioned earlier, we ran the same model for two subsamples separately to examine the moderation effect of length of stay. We hypothesized that neighborhood effects of the longer stay subsample are more likely to be statistically significant and have greater magnitude. To facilitate the comparison, we presented the results of both subsamples on the right columns of the same table as the whole sample. Surprisingly, the subsample of 5 + years in care did not show any significant neighborhood effects on delinquency, while the subsample of no more than 5 years in care showed a significant neighborhood effect of ethnic heterogeneity. Ethnic heterogeneity is positively associated with delinquency. Using the same approach as mentioned above, the effect of ethnic heterogeneity can be interpreted as follows: each standard deviation increase on ethnic heterogeneity is associated with a 0.053 higher probability of delinquency. To compare results across samples, we will report on the unstandardized coefficients in the text, since unstandardized coefficients rather than standardized coefficients should be used for cross sample comparison (Stage et al., 2004). The coefficient for males varies among samples. As compared with the whole sample, the coefficient for males is smaller in the subsample of 5 + years in care (0.29 vs. 0.32), and slightly greater in the subsample of no more than 5 years in care (0.33 vs. 0.32). That is, gender difference on overall delinquency is greater for foster youth who stay in care for a shorter time. Gender difference decreases as foster youth stay in care for over 5 years. Another difference across samples is the findings on placement instability. The subsample of 5 + years in care did not show a significant effect of four + placements. In other words, placement instability is not significantly associated with overall delinquency for foster youth who stayed in care for more than five years. The results of the regression on collective efficacy showed consistent results in three samples. The three neighborhood sociodemographic factors and being African American consistently showed significant effects on collective efficacy. That is, concentrated disadvantage, ethnic heterogeneity, and being African American were negatively associated with collective efficacy, while residential stability was positively associated with collective efficacy. The results of testing neighborhood disorder as the mediator are displayed in Table 4. The model fit indices show good model fit (RMSEA = 0.02, CFI = 1.00, TLI = 0.99, WRMR = 0.76). The variance explained in overall delinquency is 14% and neighborhood disorder is 68%, which indicates that the R2 for overall delinquency is medium, and for neighborhood disorder, the R2 is large. Table 4. Path analysis modeling delinquency with neighborhood disorder as mediator. Variables Whole sample (N = 2360) 5 + years sample (N = 767) No more than 5 years sample (N = 1593) Direct Indirect Direct Indirect Direct Indirect Probit regression on delinquency African American 0.37 0.00 − 0.11 0.00 0.54 0.00 Neighborhood sociodemographics Concentrated disadvantage 0.01 0.01 − 0.02 0.02 0.03 0.00 Ethnic heterogeneity 0.08 0.00 − 0.01 0.00 0.13⁎ 0.00 Residential stability 0.00 0.00 0.04 − 0.01 − 0.02 0.00 Neighborhood mediator Neighborhood disorder 0.01 0.03 0.00 Individual characteristics Age at first maltreatment (Reference: Early childhood) Late childhood 0.01 0.02 0.01 Adolescence − 0.22 − 0.31 − 0.23 Male 0.63⁎⁎⁎ 0.58⁎⁎⁎ 0.66⁎⁎⁎ Race/ethnicity (Reference: White) Latino − 0.22 0.11 − 0.28 Maltreatment type Physical abuse 0.02 − 0.13 0.11 Neglect − 0.01 − 0.02 0.00 Maltreatment report count − 0.02 0.04 − 0.05 Congregate care 0.16 0.40 0.07 Placement instability (Reference: One placement) Two placements 0.11 0.19 0.09 Three placements 0.20 0.25 0.19 Four + placements 0.29⁎⁎ 0.17 0.37⁎⁎ Linear regression on mediator Concentrated disadvantage 0.76⁎⁎⁎ 0.74⁎⁎⁎ 0.77⁎⁎⁎ Ethnic heterogeneity 0.13⁎⁎⁎ 0.10⁎⁎⁎ 0.14⁎⁎⁎ Residential stability − 0.19⁎⁎⁎ − 0.19⁎⁎⁎ − 0.19⁎⁎⁎ African American 0.09⁎⁎⁎ 0.05 0.11⁎⁎⁎ R square Delinquency 0.14 0.11 0.18 Neighborhood disorder 0.68 0.68 0.68 Model fit RMSEA 0.02 0.02 0.02 CFI 1.00 1.00 1.00 TLI 0.99 0.99 0.99 WRMR 0.76 0.65 0.81 ⁎ p ≤ 0.05. ⁎⁎ p ≤ 0.01. ⁎⁎⁎ p ≤ 0.001. Table options Most results were similar to the models of testing collective efficacy as the mediator. For the whole sample, none of the neighborhood sociodemographic factors showed significant effects on delinquency. Neighborhood disorder, the mediator, was also not significant in the model. Two individual characteristic variables were significantly associated with delinquency. Male foster youth are more likely to offend than female foster youth (β = 0.63, p ≤ 0.001). Foster youth with four + placements are more likely to offend than foster youth with one placement experience only (β = 0.29, p ≤ 0.01). In the regression on neighborhood disorder, the mediator, three neighborhood sociodemographic factors and being African American were all significant. Each standard deviation increase on concentrated disadvantage is associated with a 0.76 standard deviation increase on neighborhood disorder. Each standard deviation increase on ethnic heterogeneity is associated with a 0.13 standard deviation increase on neighborhood disorder. Each standard deviation increase on residential stability is associated with a 0.19 standard deviation decrease on neighborhood disorder. African American foster youth are associated with a 0.09 standard deviation higher on neighborhood disorder than foster youth of other races. The cross sample comparisons also showed similar results as the earlier models of testing collective efficacy as the mediator. Still, the subsample of 5 + years in care did not show any significant neighborhood effects on delinquency, while the subsample of no more than 5 years in care showed a significant neighborhood effect of ethnic heterogeneity. In the regression on neighborhood disorder, the subsample of no more than 5 years in care showed a different result on the relationship between being African American and neighborhood disorder. Unlike the whole sample and the subsample of 5 + years in care, being African American is not significantly associated with neighborhood disorder for foster youth of no more than 5 years in care. The three neighborhood sociodemographic factors were consistently associated with neighborhood disorder. The results of testing violent culture as the mediator are displayed in Table 5. For the whole sample, the model fit indices show good model fit (RMSEA = 0.01, CFI = 1.00, TLI = 1.00, WRMR = 0.59). The variance explained in overall delinquency is 14% and violent culture is 55%, which indicate that the R2 for overall delinquency is medium, and for violent culture is large. Table 5. Path analysis modeling delinquency with violent culture as mediator. Variables Whole sample (N = 2360) 5 + years sample (N = 767) No more than 5 years sample (N = 1593) Direct Indirect Direct Indirect Direct Indirect Probit regression on delinquency African American 0.37 0.00 − 0.10 0.00 0.53 0.00 Neighborhood sociodemographics Concentrated disadvantage 0.02 0.00 0.01 − 0.01 0.01 0.02 Ethnic heterogeneity 0.08 0.00 − 0.01 0.00 0.13⁎ 0.00 Residential stability − 0.01 0.00 0.03 0.00 − 0.02 0.00 Neighborhood mediator Violent culture 0.00 − 0.02 0.02 Individual characteristics Age at first maltreatment (Reference: Early childhood) Late childhood 0.01 0.02 0.01 Adolescence − 0.22 − 0.31 − 0.23 Male 0.63⁎⁎⁎ 0.58⁎⁎⁎ 0.66⁎⁎⁎ Race/ethnicity (Reference: White) Latino − 0.22 0.11 − 0.28 Maltreatment type Physical abuse 0.02 − 0.13 0.11 Neglect − 0.01 − 0.02 0.00 Maltreatment report count − 0.02 0.04 − 0.05 Congregate care 0.17 0.40 0.07 Placement instability (Reference: One placement) Two placements 0.11 0.19 0.09 Three placements 0.20 0.25 0.19 Four + placements 0.28⁎⁎ 0.17 0.37⁎⁎ Linear regression on mediator Concentrated disadvantage 0.74⁎⁎⁎ 0.68⁎⁎⁎ 0.77⁎⁎⁎ Ethnic heterogeneity 0.11⁎⁎⁎ 0.08⁎ 0.13⁎⁎⁎ Residential stability − 0.07⁎⁎⁎ − 0.12⁎⁎⁎ − 0.06⁎⁎ African American 0.04 0.04 0.04 R square Delinquency 0.14 0.11 0.18 Violent culture 0.55 0.53 0.56 Model fit RMSEA 0.01 0.00 0.02 CFI 1.00 1.00 1.00 TLI 1.00 1.04 0.99 WRMR 0.59 0.41 0.74 ⁎ p ≤ 0.05. ⁎⁎ p ≤ 0.01. ⁎⁎⁎ p ≤ 0.001. Table options Most results were similar to the models of testing collective efficacy and neighborhood disorder as the mediators. For the whole sample, none of the neighborhood sociodemographic factors showed significant effects on delinquency. Violent culture, the mediator, was also not significant in the model. Two individual characteristic variables were significantly associated with delinquency. Male foster youth are more likely to offend than female foster youth (β = 0.63, p ≤ 0.001). Foster youth with four + placements are more likely to offend than foster youth with one placement experience only (β = 0.28, p ≤ 0.01). Unlike the regression on the two mediators above, being African American is not significantly associated with the level of neighborhood violent culture. But three neighborhood sociodemographic factors were still significant. Each standard deviation increase on concentrated disadvantage is associated with a 0.74 standard deviation increase on violent culture. Each standard deviation increase on ethnic heterogeneity is associated with a 0.11 standard deviation increase on violent culture. Each standard deviation increase on residential stability is associated with a 0.07 standard deviation decrease on violent culture. The cross sample comparisons also showed similar results as the earlier models of testing collective efficacy and neighborhood disorder as the mediators. Ethnic heterogeneity is positively associated with delinquency only for the subsample of no more than 5 years in care (β = 0.13, p ≤ 0.05). As compared with the whole sample, gender difference is smaller in the subsample of 5 + years in care (0.29 vs. 0.32), and slightly greater in the subsample of no more than 5 years in care (0.33 vs. 0.32). Another difference across samples is the findings on placement instability. The subsample of 5 + years in care did not show significant effect of four + placements. In the regression on violent culture, the results were consistent across samples.