Schools and neighborhoods: Moderating the counter-delinquency effect of school belonging with perceived collective efficacy

Abstract

The purpose of this study was to determine whether perceived collective efficacy moderated the prospective relationship between school belonging and delinquency. Analyses were performed on a sample of 4048 youth (2020 boys, 1936 girls) from the Longitudinal Study of Australian Children (LSAC-K). Linear and negative binomial regression analyses performed with maximum likelihood (ML) and maximum likelihood with robust standard errors (MLR) estimators produced consistent results. Bootstrapped and normal theory analyses disclosed a significant interaction between school belonging and collective efficacy after age, sex, indigenous status, physical condition of dwelling, physical condition of surrounding housing, household income, weak parental monitoring, perceived peer delinquency, and prior delinquency were controlled. Further review of the significant interactive effect revealed that the increased levels of school belonging predicted decreased levels of future delinquency, but only when perceived collective efficacy was also elevated. These results support the presence of a small but significant conditional promotive effect.

Keywords

Neighborhood collective efficacy school belonging delinquency moderation Australian schoolchildren

Although there is a great deal of research on risk factors for delinquency and crime (Bonta & Andrews, 2017), we know much less about protective and promotive factors. The situation is further complicated by the fact that researchers often confuse protective and promotive factors. According to Farrington et al. (2016), risk factors predict an increased rate of offending, whereas protective factors interact with risk factors to predict a decreased rate of offending. Protective factors serve to weaken or nullify the positive relationship that exists between a risk factor and offending behavior. Promotive factors, by contrast, predict a decreased rate of offending without interacting with a risk factor. Farrington et al. acknowledge that risk, protective, and promotive effects are not always causal and that promotive effects are frequently, though not always, the opposite end of a risk effect, depending on whether the relationship is linear or nonlinear. Hence, all three factors and their relationships need to be taken into account when predicting behavior. We already know that risk and protective factors interact to protect against delinquency; the purpose of the current investigation was to determine whether two promotive factors—parent-rated neighborhood collective efficacy and child-rated school belonging—combine to promote nondelinquency.

Neighborhood disorganization

Neighborhood disorganization was first identified as a putative risk factor and potential cause of delinquency in research conducted by members of the Chicago school of criminology (Park et al., 1925; Shaw & McKay, 1942). These ideas were then elaborated on by other researchers who introduced such critical theoretical concepts as social capital (Coleman, 1988), informal social control (Ross & Jang, 2000), and collective efficacy (Sampson et al., 1997) into the neighborhood ecological literature. Research has demonstrated that neighborhoods with stronger amounts of social capital, informal social control, and collective efficacy experience lower rates of criminal victimization than neighborhoods with weaker amounts of social capital, informal social control, and collective efficacy (Bellair, 2000; Hipp, 2016; Takagi et al., 2012). Neighborhood criminal victimization is the result of negative influences originating from outside the neighborhood, primarily in the form of infiltrative nonresident criminal acts, and negative influences originating from within the neighborhood, primarily in the form of local youth crime. Although research supports the value of neighborhood unity, control, and observation in preventing negative influences from infiltrating into a neighborhood (Brown & Weil, 2020; Sampson et al., 2002), it is less clear if and how these factors lead to reduced criminality in neighborhood youth.

In extending neighborhood disorganization theory to local youth, it may be helpful to conceptualize neighborhood disorganization as a risk factor and neighborhood solidarity and cohesion as protective or promotive factors. Walters (2022) tested neighborhood social capital and nonverbal intelligence as protective factors in children with a history of early conduct disorder. Results showed that while nonverbal intelligence interacted with early conduct disorder to reduce the level of subsequent delinquency, neighborhood social capital did not. Consequently, the results of this study suggest that while nonverbal intelligence may serve as a protective factor vis-à-vis early conduct disorder, neighborhood social capital displayed no such effect. There is evidence, however, that positive neighborhood influences may serve as promotive factors. Neighborhood cohesion, as reported by a group of early adolescents, was found to correlate with lower levels of delinquency, although it failed to interact with a risk factor (i.e. exposure to violence), in a study by Chen et al. (2016), and a positive neighborhood environment where 28-to-33-year-olds felt safe and supported predicted decreased levels of intimate partner violence in a study by Thulin et al. (2021). What these results do not tell us is whether neighborhood factors interact with other promotive factors.

School belonging

School belonging is one variable that may interact with neighborhood cohesion. That is because both variables can be viewed as forms of social bonding. In developing his social control theory of crime, Hirschi (1969) argued that delinquency was inhibited by attachment to conventional people, involvement in conventional activities, commitment to conventional goals, and belief in the conventional social and moral order. School belonging or bonding demonstrates a sense of attachment to teachers, school officials, and fellow students; involvement in academic and extracurricular activities; commitment to academic success and positive peer relations; and belief in the legitimacy of school as preparation for later life. Neighborhood cohesion, on the other hand, highlights attachment to neighbors both individually and collectively; involvement in neighborhood activities; commitment to neighborhood objectives, such as maintaining a safe environment for everyone in the neighborhood; and belief in the validity of neighborhood values in terms of how one should respond to various situations. Despite the similarities, there are important differences between schools and neighborhoods that should also be noted. Schools, for instance, tend to be more ethnically, racially, and culturally diverse than neighborhoods, whereas neighborhoods provide less supervision and more opportunities for serious victimization than do schools (Cortright, 2018).

Because of the differences that exist between schools and neighborhoods, they may each serve as risk or promotive factors, yet provide somewhat different perspectives and input. This, in turn, may give rise to an interactive relationship in which the magnitude of their combined effect is greater than the sum of their individual effects. In a study that compared school and neighborhood belonging in a group of low-income Latino youth, both variables correlated negatively with depression (Maurizi et al., 2013). Similar results were obtained in a sample of low-income Black students where maternal warmth, parental monitoring, school belonging, and neighborhood connectiveness all predicted positive youth outcomes, although only parental monitoring and school belonging predicted decreased school expulsions and community arrests (Johnson et al., 2020). In a study examining the factors contributing to homophobic name-calling in middle-school students, Valido et al. (2021) determined that neighborhood violence was positively associated and school belonging negatively associated with subsequent homophobic name-calling perpetration and victimization. The results of these various studies indicate that while school belonging and a supportive neighborhood environment may promote positive behaviors and inhibit antisocial activities, their interactive effect remains untested.

Combining promotive factors

Although a great deal of research has been completed on risk factors and how they combine to increase the risk of future offending, there is very little research addressing the possibility that promotive factors accumulate or interact to reduce risk. In studies examining multiple risks and promotive factors, there is evidence of a cumulative effect that either increases or decreases risk depending on the number, strength, and pattern of risk and promotive factors involved (Stoddard et al., 2012; Stouthamer-Loeber et al., 2002; van der Laan et al., 2010). In one of the more ambitious studies conducted in this area, Whitney et al. (2010) used latent profile analysis to create a low risk/high promotion trajectory marked by higher intelligence and socioeconomic status; fewer early social, behavioral, and attentional problems; a more positive future orientation; and greater parental or peer disapproval of antisocial acts. Youth falling into this particular trajectory engaged in significantly fewer episodes of delinquency than youth from any of the other trajectories identified in this study. Beyond the accumulated effect, however, there is no indication of whether different promotive factors interact to provide an effect that is greater than the sum of the individual promotive factors.

Present study

The research question addressed in this study asked whether a facilitative interactive relationship exists between two promotive factors believed to reflect a strong sense of conventional social bonding—perceived neighborhood cohesion or collective efficacy, as rated by the parents, and school belonging, as rated by the child. Several control variables were included in this study for conceptual and empirical reasons. Conceptually, household income, physical condition of the dwelling the child was living in, and physical condition of the surrounding buildings served as a control for relative deprivation. Empirically, parental monitoring has been found to correlate with and predict outcomes in several studies where neighborhood cohesion and school belonging have been evaluated (Johnson et al., 2020; Valido et al., 2021). Peer delinquency was also controlled based on studies suggesting that it may confound the school belonging–delinquency relationship (Boers et al., 2010; Schreck, 2002). A lagged dependent variable that assessed a change in delinquency from baseline to follow-up was used to provide the study with a dynamic outcome measure. The hypothesis tested in this study held that perceived collective efficacy would moderate the relationship between school belonging and subsequent delinquency after age, sex, indigenous status, physical condition of the dwelling and neighboring buildings, household income, weak parental monitoring, perceived peer delinquency, and prior delinquency were controlled. It was further predicted that the interaction between perceived collective efficacy and school belonging would show delinquency at its lowest when collective efficacy and school belonging were both elevated.

Method

Participants

Data for this study were provided by 4048 mid-adolescents from the Longitudinal Study of Australian Children (LSAC; Australian Institute of Family Studies, 2018). The LSAC is a large representative sample of Australian youth organized into two cohorts, a birth (B) cohort and a kindergarten (K) cohort, obtained using multistep cluster sampling. The K cohort was used in the current investigation because it covered the age range and variables of principal interest in the present study. Children and their parents were initially interviewed for the K cohort when the child entered kindergarten. Participants for the current investigation included all 4048 members (2020 boys, 1936 girls, 92 failed to report their sex) of the LSAC (K cohort) with data on at least one of the twelve variables employed in this study. The sample was comprised of White, Asian, and African participants (97.1%), accompanied by a small indigenous component (2.7% aboriginal and 0.2% Torres Strait Islander).

Weighting

The LSAC assigns cross-sectional and longitudinal sample weights to each participant. The rationale for using sample weights is two-fold. First, the weights account for the child's probability of being selected for and included in the study. Second, the weights adjust for nonresponse. Because the current study was conducted over a two-year period (ages 14/15 to ages 16/17), longitudinal sample weights up through Wave 7 (age 16/17) were utilized for the correlational and regression analyses.

Measures

School Belonging. There were two independent variables included in this study: school belonging and perceived collective efficacy. School belonging at Wave 6 of the LSAC-K (age 14/15) served as the focal predictor of the dependent variable (i.e. delinquency). It was assessed with the 12-item Psychological Sense of School Membership (PSSM: Goodenow, 1993) scale. The PSSM is designed to measure the extent to which the respondent feels personally accepted by, included in, and connected to the school social environment. All 12 items were assessed on a five-point scale (1 = not at all true, 2 = not very true, 3 = neither not at all true nor completely true, 4 = somewhat true, 5 = completely true). Eight items (e.g. “other students in this school take my opinions seriously”; “most teachers at this school are interested in me”) were scored directly from the rated values and four items (e.g. “sometimes I don't feel as if I belong here”) were reverse coded before being totaled. Scores on the PSSM can range from 12 to 60 and the scale itself displayed very good internal consistency in the current sample of participants (α = 0.87; ω = 0.87).

Perceived Collective Efficacy. The second independent variable was an eight-item measure of perceived neighborhood collective efficacy completed by the parent most familiar with the child's behavior. This scale served as a moderator of the school belonging–delinquency relationship and like the school belonging measure, was administered during Wave 6 of the LSAC (age 14/15). Items on the perceived collective efficacy scale assessed either social cohesion and trust (“close-knit neighborhood”; “people don't generally get along” [reverse coded]; “people do not share the same values” [reverse coded]; “trust most people) or social control (e.g., “neighbors would react if…kids were skipping school…kids were spraying graffiti…kids disrespected an adult…there was a fight in the street”) and were scored using a five-point scale (1 = very unlikely, 2 = unlikely, 3 = neither likely nor unlikely, 4 = likely, 5 = very likely). The perceived collective efficacy scale achieved good internal consistency in the LSAC-K (α = .83, ω = .84).

Delinquency. Participants rated their past-year involvement in 19 different delinquent acts at Wave 7 (age 16/17) of the LSAC-K using a six-point rating scale (0 = not at all, 1 = once, 2 = twice, 3 = three times, 4 = four times, 5 = five or more times). When summed, these items produced a total score that could potentially range from 0 to 95. The 19 delinquent acts were as follows: “got into physical fights in public”; “skipped school for a whole day”; “stole something from a shop”; “drew graffiti in public places”; “carried a weapon like a knife, gun or piece of wood”; “took a vehicle for a ride/drive without permission”; “stole money or other things from another person”; “ran away from home and stayed away overnight or longer”; “purposely damaged or destroyed others’ property”; “damaged a parked car”; “went around with a group of three or more kids damaging/fighting”; “been suspended or expelled from school”; “broke into a house, flat or vehicle”; “stole something out of a parked car”; “started a fire in a place where you should not burn”; “used force/threats to get money/things from someone”; “been caught by police for something you had done”; “sold illegal drugs”; and “attacked someone with the idea of harming them”). Internal consistency for the 19-item Delinquency-7 scale was good (α = 0.82, ω = 0.87).

Control Variables. Three demographic control variables were included in this study: age (in years), sex (1 = male, 2 = female), and indigenous status (1 = nonindigenous, 2 = indigenous). The condition of the dwelling (1 = badly deteriorated, 2 = poor condition with peeling paint and need of repair, 3 = fair condition, 4 = well-kept with good repair and exterior surface) and the condition of the surrounding homes (1 = deterioration/poor condition in more than 50% of the structures, 2 = a fair bit of deterioration or poor condition—more than 20% of the structures, 3 = one or two such structures, 4 = none at all). Ratings were made by trained raters who conducted the interviews. Monthly household income (in Australian dollars) was also included as a control variable. A weak parental monitoring scale was completed by both parents with three items (“know where child is”; “know who child is with”; “child goes out without telling you” [reverse coded]) on a five-point scale (1 = always, 2 = almost always, 3 = about half the time, 4 = almost never, and 5 = never). The summed score achieved modest internal consistency (α = 0.65–0.66, ω = 0.67–0.68) and was averaged across mothers and fathers. The eighth and final control variable, perceived peer delinquency, was assessed with five items (“Kids you know…get into fights…smoke cigarettes…drink alcohol…have broken the law…try drugs”) rated on a five-point scale (1 = none of them, 2 = one or two of them, 3 = some of them, 4 = most of them, 5 = all of them) that when summed generated a score that could range from 5 to 25 (α = 0.86, ω = 0.88).

The use of longitudinal data with no overlap between waves afforded the current investigation proper temporal order between the independent and dependent variables. Proper temporal direction (Cole & Maxwell, 2003) was attained by using a lagged dependent variable. The use of a lagged dependent variable also provided for a dynamic outcome measure (i.e. instead of predicting a static future outcome, the objective was to predict a change in behavior over time), eliminated a major source of omitted variable bias and misspecification error, and removed autocorrelation from the residuals. Lagging was achieved by including Wave 6 (age 14/15) delinquency as a covariate in the regression equation predicting Wave 7 (age 16/17) delinquency. The same rating scale (0–5) and time frame (past year) were used for the Wave 6 and Wave 7 delinquency measures, and item content was identical except for two items that were not part of the Wave 6 measure but were subsequently added to the Wave 7 measure (“sold illegal drugs” and “attacked someone with the idea of harming them”). Scores on the two delinquency scales (Wave 6 and Wave 7) correlated 0.58, indicating good test–retest reliability (r) for the delinquency measure over a period of two years.

Analytic strategy

Descriptive statistics and correlations were computed with SPSS: Version 26 (IBM, 2019), whereas the regression analyses were performed with MPlus 8.3 (Muthén & Muthén, 1998–2017). For the regression analyses, the dependent variable, Wave 7 delinquency, was regressed onto two independent variables, school belonging and perceived collective efficacy, and their interaction, belonging × efficacy, along with nine control/precursor measures (age, sex, indigenous status, condition of dwelling, condition of surrounding buildings, household income, weak parental monitoring, perceived peer delinquency, and Wave 6 delinquency). The belonging × efficacy interaction was created by multiplying school belonging and collective efficacy after the values for both measures had been centered. The regression analyses were performed with both maximum likelihood (ML) and maximum likelihood with robust standard errors (MLR) estimators. Bias-corrected bootstrapped confidence intervals (5000 replications) were also calculated where the ML estimator was used given the dependent variable's highly non-normal distribution (Kennet-Cohen et al., 2018). A negative binomial regression model was then estimated even though it was not a true negative binomial distribution (i.e., each item on the Wave 7 delinquency scale was truncated at a count of 5).

Missing data

There was a moderate amount of missing data in this study. Slightly more than half the participants (50.9%) had complete data on all 12 study variables and two-thirds (66.0%) had complete data for the three main variables (school belonging, perceived collective efficacy, Delinquency-7). Individual variables with more than 5% missing data included: household income (11.5%), nearby buildings (11.6%), parental monitoring (14.6%), perceived collective efficacy (17.0%), perceived peer delinquency (17.3%), Delinquency-6 (17.4%), school belonging (18.2%), and Delinquency-7 (27.5%). Missing data were handled with full information maximum likelihood (FIML), a procedure that estimates population parameters and standard errors for the entire sample using existing relationships between nonmissing data. Research indicates that FIML is superior to traditional missing data procedures like simple imputation and listwise deletion when it comes to maximizing precision and minimizing bias (Allison, 2002).

Results

Descriptive statistics and correlations pertaining to the 12 study variables are listed in Table 1. Given an extensive amount of non-normality (skew = 6.67, kurtosis = 62.29) in the dependent variable (Delinquency-7), bootstrapping was the principal means by which significance was determined in the main regression analysis. The main analysis was then followed by three supplemental analyses: one in which an MLR estimator was used to calculate the Wald Z values, one in which a negative binomial regression was performed with an MLR estimator and Monte Carlo integration, and one restricted to participants with complete data on the three main variables (i.e. school belonging, perceived collective efficacy, and Delinquency-7). The zero-order correlations between school belonging and perceived collective efficacy, on the one hand, and Delinquency-7, on the other hand, were in the predicted direction (i.e. inverse), but only the school belonging–delinquency correlation was significant.

Table 1.
Descriptive statistics and correlations for the 12 variables included in this study.

Variable n M SD Range 2 3 4 5 6 7 8 9 10 11 12

1. Age 4034 14.41 0.49 14–15 −0.00 −0.01 0.01 0.02 0.01 0.07* 0.05 0.02 0.04 0.01 0.01

2. Sex 3956 1.49 Male = 51.1% 0.02 −0.00 −0.00 0.00 −0.08* 0.04 −0.11* −0.08* −0.02 −0.13*

3. Indigenous Status 3956 1.03 Nonindigenous = 97.1% 0.00 −0.13* −0.09* 0.08* 0.06 0.04 −0.04 −0.06 0.06

4. Condition of Dwelling 3847 3.67 0.58 1–4 0.03 0.02 0.03 0.05 0.04 −0.02 −0.00 −0.02

5. Nearby Buildings 3579 3.72 0.50 1–4 0.25* −0.04 −0.12* −0.14* 0.12* 0.15* −0.09*

6. Household Income 3583 16.83 10.83 0–89.1 0.00 −0.08* −0.09* 0.15* 0.15* −0.06

7. Parental Monitoring-6 3457 1.50 0.42 1–4.3 0.25* 0.29* −0.15* −0.03 0.20*

8. Peer Delinquency-6 3346 2.92 6.46 5–25 0.47* −0.36* −0.05 0.37*

9. Delinquency-6 3345 1.32 3.98 0–74 −0.30* −0.09* 0.58*

10. School Belonging-6 3312 49.41 7.56 2–60 0.10* −0.21*

11. Collective Efficacy-6 3361 3.65 0.61 1–5 −0.03

12. Delinquency-7 2936 2.24 6.35 0–92

Note: All correlations are Pearson Product Moment Correlations except for the point-biserial correlations between continuous variables and either sex or indigenous status and the phi coefficients between sex and indigenous status; Age = chronological age in years; Sex = 1 (male) and 2 (female); Indigenous Status = 1 (nonindigenous) and 2 (indigenous); Condition of Dwelling = physical condition of participant's home (higher scores indicate better condition); Nearby Buildings = physical condition of buildings near the participant's home (higher scores indicate better condition); Household Income = monthly income for household measured in hundreds of Australian dollars; Parental Monitoring-6 = parent-rated weak parental monitoring at Wave 6 (age 14/15); Peer Delinquency-6 = perceived peer delinquency at Wave 6 (age 14/15); Delinquency-6 = self-reported delinquency at Wave 6 (age 14/15); School Belonging = child-rated school belonging at Wave 6 (age 14/15); Collective Efficacy-6 = parent-rated perceived collective efficacy at Wave 6 (age 14/15); Delinquency-7 = self-reported delinquency at Wave 7 (age 16/17); n = number of nonmissing cases; M = mean; SD = standard deviation; Range = range of scores in current sample.

p < 0.00076 (Bonferroni-corrected alpha: 0.05/66 correlations).

Prior to conducting the regression analyses, multicollinearity was tested and found not to be an issue (tolerance = 0.722–0.999, variance inflation factor = 1.001–1.385). As summarized in Table 2, the regression results revealed the presence of significant normal theory (Wald Z) coefficients for sex, perceived peer delinquency, Wave 6 delinquency, school belonging, and the belonging × efficacy interaction, and significant bias-corrected bootstrapped confidence intervals for sex, condition of dwelling, perceived peer delinquency, Wave 6 delinquency, school belonging, and the belonging × efficacy interaction. The ƒ² effect size was 0.009, which is classified as small (Aiken & West, 1991).

Table 2.
Regression analysis of Wave 7 delinquency.

Predictor b[95% BCBCI] β Z p

Delinquency-7 on

Age −0.010 [−0.047, 0.031] −0.009 −0.52 0.606

Sex −0.091 [−0.129, −0.056] −0.080 −4.87 <0.001

Indigenous Status 0.011 [−0.042, 0.079] 0.011 0.36 0.720

Dwelling Condition −0.042 [−0.095, −0.001] −0.037 −1.79 0.074

Nearby Buildings 0.004 [−0.047, 0.047] 0.004 0.19 0.852

Household Income 0.006 [−0.029, 0.043] 0.005 0.35 0.729

Parental Monitoring-6 0.010 [−0.033, 0.056] 0.010 0.43 0.667

Peer Delinquency-6 0.136 [0.062, 0.220] 0.133 3.42 <0.001

Delinquency-6 0.552 [0.427, 0.713] 0.569 7.59 <0.001

School Belonging-6 −0.050 [−0.108, −0.003] −0.048 −1.93 0.053

Collective Efficacy-6 0.006 [−0.032, 0.050] 0.006 0.29 0.773

Belonging x Efficacy −0.070 [−0.148, −0.011] −0.064 −2.06 0.040

R² 0.448 7.65 <0.001

Note:* Delinquency-7 = self-reported delinquency at Wave 7 (age 16/17); Age = chronological age in years; Sex = 1 (male) and 2 (female); Indigenous Status = 1 (nonindigenous) and 2 (indigenous); Condition of Dwelling = physical condition of participant's home (higher scores indicate better condition); Nearby Buildings = physical condition of buildings near the participant's home (higher score indicate better condition); Household Income = monthly income for household measured in hundreds of Australian dollars; Parental Monitoring-6 = parent-rated weak parental monitoring at Wave 6 (age 14/15); Peer Delinquency-6 = perceived peer delinquency at Wave 6 (age 14/15); Delinquency-6 = self-reported delinquency at Wave 6 (age 14/15); School Belonging = child-rated school belonging at Wave 6 (age 14/15); Collective Efficacy-6 = parent-rated perceived collective efficacy at Wave 6 (age 14/15); Belonging x Efficacy = interaction between Wave 6 school belonging and Wave 6 collective efficacy; R² = R-square for the entire regression equation; b[95% BCBCI] = unstandardized coefficient and 95% bias-corrected bootstrapped confidence interval [in brackets]; β = standardized coefficient; Z = Wald Z statistic; p = significance level of Wald Z statistic; N = 4048.

Replacing the ML estimator with an MLR estimator and conducting a second regression analysis indicated that sex, perceived peer delinquency, Wave 6 delinquency, school belonging, and the belonging × efficacy interaction all predicted Wave 7 delinquency (p < 0.05). Still using an MLR estimator, significant Z score effects (p < 0.05) were obtained for sex, perceived peer delinquency, parental monitoring, Wave 6 delinquency, school belonging, and the belonging × efficacy interaction in a negative binomial analysis. When the main regression analysis was restricted to cases with complete data on school belonging, perceived collective efficacy, and Delinquency-7 (n = 2672), significant bias-corrected bootstrapped confidence intervals surfaced for sex, perceived peer delinquency, Wave 6 delinquency, school belonging, and the belonging × efficacy interaction (the Z score for the belonging × efficacy interaction, on the other hand, approached significance, p = 0.05).

The interaction between school belonging and collective efficacy is depicted in Figure 1. A review of this figure indicates that the only time Wave 7 delinquency was below average was when perceived collective efficacy and school belonging were both high. In each of the other three combinations, Wave 7 delinquency was above average. Table 3 provides another perspective on the moderating effect of perceived collective efficacy on the prospective school belonging–delinquency association. When the moderator, perceived collective efficacy, was medium, medium-high, or high, above-average levels of school belonging served a promotive function, but when perceived collective efficacy was low or medium-low, school belonging demonstrated a nonsignificant augmenting effect on subsequent delinquency in a pattern referred to as disordinal interaction (Rogers, 2002). Hence, while perceived collective efficacy alone was ineffective in shielding a child from subsequent delinquency, it did have a counter-delinquency effect when it was of at least medium magnitude and paired with high school belonging.

Figure 1.
Interactive relationship between school belonging and perceived collective efficacy as predictors of Wave 7 delinquency after controlling for age, sex, indigenous status, condition of dwelling, weak parental monitoring, perceived peer delinquency, and Wave 6 delinquency: Lagged dependent variable. Note: Low school belonging is defined as one standard deviation below the mean and high school belonging as one standard deviation above the mean; the same for low and high collective efficacy.

Table 3.
Conditional effects of the focal predictor (school belonging) at different levels of the moderator variable (perceived collective efficacy).

Level of moderator b SE Z p 95% BCBCI

Low Collective Efficacy 0.088 0.065 1.35 0.176 [−0.035, 0.222]

Medium−Low Collective Efficacy 0.018 0.036 0.49 0.621 [−0.050, 0.091]

Medium Collective Efficacy −0.052 0.026 −1.98 0.047 [−0.108, −0.003]

Medium−High Collective Efficacy −0.123 0.049 −2.49 0.013 [−0.237, −0.040]

High Collective Efficacy −0.193 0.080 −2.40 0.017 [−0.380, −0.060]

Note: Moderator (perceived collective efficacy) levels were calculated as 2 standard deviations (SD) below the mean or −2 SD for low, −1 SD for low-medium, 0 SD for medium, + 1 SD for high-medium, and + 2 SD for high; b = unstandardized coefficient; SE = standard error of the unstandardized coefficient; Z = Wald Z statistic; p = significance level of the Wald Z statistic; 95% BCBCI = 95% bias-corrected bootstrapped confidence interval; N = 4048.

An auxiliary analysis was performed in an effort to determine the effect of school belonging and perceived collective efficacy on Wave 7 delinquency when the belonging × efficacy interaction term was removed from the regression equation. When the regression was computed without the belonging × efficacy interaction, the main effect for school belonging (Z = −1.63, p = 0.10; 95% BCBCI = −0.092, 0.009) became nonsignificant and the main effect for perceived collective efficacy (Z = 0.44, p = 0.66; 95% BCBCI = −0.034, 0.056) remained nonsignificant. This indicates that after the interaction term was removed from the regression equation, school belonging and perceived collective efficacy failed to achieve a promotive effect, once age, sex, indigenous status, condition of dwelling and neighboring buildings, household income, parental monitoring, peer delinquency, and prior delinquency were controlled.

Discussion

The research hypothesis tested in this study received support when a lagged dependent variable (delinquency) served as the outcome measure in a series of regression analyses. As predicted, a significant interaction was recorded between school belonging and perceived collective efficacy regardless of which estimator (standard ML or MLR) was used, whether linear regression or negative binomial regression was employed, and when both the entire sample and just those participants with complete data on the three main variables (school belonging, perceived collective efficacy, and Wave 7 delinquency) were studied. It should also be noted that in none of the analyses, including when the zero-order correlation was calculated, did perceived collective efficacy predict delinquency, whereas school belonging only predicted delinquency in about half the analyses. Still, the two variables produced a significant moderating effect when their interaction was regressed onto future delinquency. A graph of this interaction revealed, consistent with predictions, that delinquency was lowest when collective efficacy and school belonging were highest. Thus, while school belonging seemed to have a stronger solitary promotive effect, at least with respect to its zero-order correlation with Wave 7 delinquency, the former was still conditional on the latter, such that school belonging only impeded delinquency when perceived collective efficacy was of medium to high magnitude.

Theoretical implications

The theory tested in this study proposed that influences relevant to delinquent development can be organized into risk, promotive, and protective factors (Farrington et al., 2016), and that while some protective factors achieve their effect by interacting with known risk factors, it is also possible that some promotive effects achieve their effect by interacting with one another. The results of several studies have shown that promotive factors can be combined to achieve an effect greater than any one factor is capable of producing on its own (Stoddard et al., 2012; Stouthamer-Loeber et al., 2002; van der Laan et al., 2010; Whitney et al. (2010). The current study sought to take this a step further by illustrating how promotive factors are capable of generating effects that are greater than the sum of their individual parts. This was demonstrated by testing the interaction between two putative promotive factors, school belonging, and perceived collective efficacy. The interaction, while small, was significant and may explain how promotive effects achieve a portion of their influence: namely, by interacting with and potentiating the effect of other promotive factors. In the current instance, living in a neighborhood perceived by a parent as cohesive led to a significant reduction in delinquency when the child also felt accepted and appreciated at school.

In several prior studies (Chen et al., 2016; Walters, 2022), neighborhood cohesion fell short of interacting with a risk factor and in so doing, failed to qualify as a protective factor. Other studies, however, showed that neighborhood cohesion may act as a promotive factor alongside school belonging and other documented promotive factors (Johnson et al., 2020; Maurizi et al., 2013; Valido et al., 2021). When paired with school belonging in the current investigation, perceived collective efficacy failed to achieve a promotive effect, yet it promoted low delinquency by interacting with school belonging, which raises the question of how best to explain this effect. In the current study, delinquency dropped in students with a strong sense of school belonging, but only when the child was living in a neighborhood their parent perceived as exhibiting moderate to high collective efficacy. Given that collective efficacy was based on parental ratings rather than child ratings, shared method variance (Shadish et al., 2002), and perceptual (Young et al., 2013) explanations appear implausible. It is more likely that there is something about living in a neighborhood with moderate to high perceived collective efficacy that stimulates or empowers promotive factors like school belonging. What is required next is a study that investigates whether perceived collective efficacy moderates the counter-delinquency effects of other well-known promotive factors such as parental support and positive peer relations.

Practical implications

A major conclusion that can be drawn from these results is that scholars and clinicians need to look beyond single risk, protective, and promotive factors and begin the difficult but vital task of examining the relationships between factors. Researchers involved in the study of protective factors as defined by Rutter (1985) and Farrington et al. (2016) have a head start in this regard given that Rutter and Farrington et al. define protective factors as variables that interact with known risk factors to reduce the negative impact of these risk factors on behavior. Although promotive factors have traditionally been studied in isolation, the current results suggest that the relationships that form between individual promotive factors should also be investigated. In the present study, perceived collective efficacy did not qualify as a promotive factor, even as a zero-order correlation or when the interaction term was removed, yet it had an impact on delinquency through its interaction with school belonging. This would seem to indicate that relationships between variables may be just as important as the variables themselves in gauging risk and promotive effects. As such, more time and effort need to be spent on factor relationships, not only as a way of advancing general theory but also as a way of reducing risk in individual children as Turner et al. (2005) discovered when they examined the cross-facilitating relationship between school socialization factors and child self-control.

What does it mean that the interaction between two promotive factors, school belonging and perceived collective efficacy, in this case, predicted a drop in delinquent behavior in a group of early to mid-adolescents? From a conceptual standpoint, it means that school belonging and perceived collective efficacy may both need to be present for there to be a significant promotive effect. In an earlier study (Walters, 2022), it was determined that perceived neighborhood social capital, a construct related to perceived collective efficacy, failed to exert a protective effect on future delinquency when crossed with early antisocial behavior. In the current study, perceived collective efficacy likewise failed to achieve a solitary promotive effect but did display an effect when paired with school belonging. Hence, perceived collective efficacy may need to be in interaction with a variable like school belonging to be maximally effective. One way this interaction can be enriched for the purpose of preventing future delinquency is to arrange for parents and teachers to work together for the betterment of the child. Keyes (2000) described such an approach in a symposium on early child education. In her presentation, Keyes noted that parent–teacher partnerships led to improved communication and a more productive working relationship between parents and teachers and that high self-efficacy on the part of both the parent and teacher was a key factor in the formation of a strong parent–teacher bond.

Limitations

A principal weakness of this study is that the effects were small. This is common in moderation research, although an effect size (ƒ²) of 0.009 is small even by moderation research standards. A persistent challenge in conducting research on moderation is obtaining sufficient power to observe an effect, even in samples as large as the present one. Interactive effects are notoriously small and fragile when based on data from a nonexperimental study (McClelland & Judd, 1993). This is due, in large part, to increased measurement error and the possibility of nonlinear trends in the data (Aguinis et al., 2005). Measurement error in an interaction term is the product of the error found in each of the contributing main effects. The internal consistency reliability of the interaction for school bonding (α = 0.87) and perceived collective efficacy (α = 0.83) is calculated as 0.72 (0.87 × 0.83), and while this indicates an acceptable level of measurement error, there was significantly more measurement error in the interaction than there was in either of the two main effects. It should also be noted that moderation, like mediation, is a theory-based confirmatory approach and that a conceptual rationale was given for why perceived collective efficacy and school belonging alone or in combination would likely predict a change in delinquency. The predicted effect, while small, surfaced and, at the very least, indicates that this and similar relationships require further investigation.

The two-year gap between waves can be considered a second potential weakness of this study. Researchers normally prefer a smaller gap, somewhere in the neighborhood of 6–12 months, so as to limit the number of alternative factors that might intervene between the independent and dependent variables and thereby explain the results. On the other hand, the use of a lagged dependent variable could be viewed as both a strength and weakness of this study. It has been argued that lagged dependent variables can bias path coefficients downward toward null (Achen, 2000), although they have the advantage of minimizing residual serial correlation, ensuring proper temporal direction between variables, and capturing the dynamic nature of the dependent variable (Keele & Kelly, 2006; Wilkins, 2017). A fourth limitation of this study is that collective efficacy was measured at the individual (parental) level rather than at the group (neighborhood) level. This was why the collective efficacy variable was labeled perceived collective efficacy rather than neighborhood collective efficacy. Further research is required to determine whether these results generalize to situations where collective efficacy is measured for the entire neighborhood and the data analyzed with a statistical procedure like multilevel modeling.

Summary and conclusion

The risk–need–responsivity (RNR) model (Bonta & Andrews, 2017) is one of the more popular and research-based models used in clinical prediction. The risk–protective–promotive model proposed by Farrington et al. (2016) and examined in the present study also appears to have value, in part because it covers areas not currently encompassed by the RNR model. In light of the seeming compatibility of the two models, it would make sense for researchers to investigate how these two models might be effectively integrated or merged. Not only could risks and needs be better understood by incorporating information on protective and promotive factors but the relationships between the different variables may open up a number of new areas of inquiry. For instance, certain types of factors or factor relationships may be more important at one age than another, a clear sign of a developmental process. In our efforts to understand, predict, and reduce future crime, delinquency, and recidivism it is imperative then that all relevant sources of information be explored and that this entail not only risk, need, and responsivity factors, but protective and promotive factors as well as the relationships that form between them.

Variable	n	M	SD	Range	2	3	4	5	6	7	8	9	10	11	12
1. Age	4034	14.41	0.49	14–15	−0.00	−0.01	0.01	0.02	0.01	0.07*	0.05	0.02	0.04	0.01	0.01
2. Sex	3956	1.49	Male = 51.1%			0.02	−0.00	−0.00	0.00	−0.08*	0.04	−0.11*	−0.08*	−0.02	−0.13*
3. Indigenous Status	3956	1.03	Nonindigenous = 97.1%				0.00	−0.13*	−0.09*	0.08*	0.06	0.04	−0.04	−0.06	0.06
4. Condition of Dwelling	3847	3.67	0.58	1–4				0.03	0.02	0.03	0.05	0.04	−0.02	−0.00	−0.02
5. Nearby Buildings	3579	3.72	0.50	1–4					0.25*	−0.04	−0.12*	−0.14*	0.12*	0.15*	−0.09*
6. Household Income	3583	16.83	10.83	0–89.1						0.00	−0.08*	−0.09*	0.15*	0.15*	−0.06
7. Parental Monitoring-6	3457	1.50	0.42	1–4.3							0.25*	0.29*	−0.15*	−0.03	0.20*
8. Peer Delinquency-6	3346	2.92	6.46	5–25								0.47*	−0.36*	−0.05	0.37*
9. Delinquency-6	3345	1.32	3.98	0–74									−0.30*	−0.09*	0.58*
10. School Belonging-6	3312	49.41	7.56	2–60										0.10*	−0.21*
11. Collective Efficacy-6	3361	3.65	0.61	1–5											−0.03
12. Delinquency-7	2936	2.24	6.35	0–92

Predictor	b[95% BCBCI]	β	Z	p
Delinquency-7 on
Age	−0.010 [−0.047, 0.031]	−0.009	−0.52	0.606
Sex	−0.091 [−0.129, −0.056]	−0.080	−4.87	<0.001
Indigenous Status	0.011 [−0.042, 0.079]	0.011	0.36	0.720
Dwelling Condition	−0.042 [−0.095, −0.001]	−0.037	−1.79	0.074
Nearby Buildings	0.004 [−0.047, 0.047]	0.004	0.19	0.852
Household Income	0.006 [−0.029, 0.043]	0.005	0.35	0.729
Parental Monitoring-6	0.010 [−0.033, 0.056]	0.010	0.43	0.667
Peer Delinquency-6	0.136 [0.062, 0.220]	0.133	3.42	<0.001
Delinquency-6	0.552 [0.427, 0.713]	0.569	7.59	<0.001
School Belonging-6	−0.050 [−0.108, −0.003]	−0.048	−1.93	0.053
Collective Efficacy-6	0.006 [−0.032, 0.050]	0.006	0.29	0.773
Belonging x Efficacy	−0.070 [−0.148, −0.011]	−0.064	−2.06	0.040
R²	0.448		7.65	<0.001

Level of moderator	b	SE	Z	p	95% BCBCI
Low Collective Efficacy	0.088	0.065	1.35	0.176	[−0.035, 0.222]
Medium−Low Collective Efficacy	0.018	0.036	0.49	0.621	[−0.050, 0.091]
Medium Collective Efficacy	−0.052	0.026	−1.98	0.047	[−0.108, −0.003]
Medium−High Collective Efficacy	−0.123	0.049	−2.49	0.013	[−0.237, −0.040]
High Collective Efficacy	−0.193	0.080	−2.40	0.017	[−0.380, −0.060]

Footnotes

Author note

The author would like to express his gratitude to the Australian Government Department of Social Services (DSS), the Australian Institute of Family Studies (AIFS), and the Australian Bureau of Statistics (ABS) for providing access to the Growing Up in Australia: Longitudinal Study of Australian Children database. Address all correspondence concerning this paper to Glenn D. Walters, Department of Criminal Justice, Kutztown University, Kutztown, Pennsylvania, USA.

Declaration of conflicting interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) received no financial support for the research, authorship, and/or publication of this article.

ORCID iD

Glenn D Walters

References

Achen

C. H.

(2000). Why lagged dependent variables can suppress the explanatory power of other independent variables. Presented at the Annual Meeting of Political Methodology, Los Angeles.

Aguinis

Beaty

J. C.

Boik

R. J.

Pierce

C. A.

(2005). Effect size and power in assessing moderating effects of categorical variables using multiple regression: A 30-year review. Journal of Applied Psychology, 90(1), 94–107. https://doi.org/10.1037/0021-9010.90.1.94

Aiken

L. S.

West

S. G.

(1991). Multiple regression: testing and interpreting interactions. Sage.

Allison

P. D.

(2002). Missing data. Sage.

Australian Institute of Family Studies (2018). Longitudinal study of Australian children data user guide – December 2018. Australian Institute of Family Studies.

Bellair

P. E.

(2000). Informal surveillance and street crime: A complex relationship. Criminology; An Interdisciplinary Journal, 38(1), 137–167. https://doi.org/10.1111/j.1745-9125.2000.tb00886x

Boers

Reinecke

Seddig

Mariotti

(2010). Explaining the development of adolescent violent delinquency. European Journal of Criminology, 7(6), 499–520. https://doi.org/10.1177/1477370810376572

Bonta

Andrews

(2017). The psychology of criminal conduct (6th ed.). Routledge.

Brown

K. J.

Weil

F. D.

(2020). Strangers in the neighborhood: Violence and neighborhood boundaries. Journal of Contemporary Ethnography, 49(1), 86–117. https://doi.org/10.1177/0891241619857150

10.

Chen

Voisin

D. R.

Jacobson

K. C.

(2016). Community violence exposure and adolescent delinquency: Examining a spectrum of promotive factors. Youth and Society, 48(1), 33–57. https://doi.org/10.1177/0044118X13475827

11.

Cole

D. A.

Maxwell

S. E.

(2003). Testing mediational models with longitudinal data: Questions and tips in the use of structural equation modeling. Journal of Abnormal Psychology, 112(4), 558–577. https://doi.org/10.1037/0021-843X.112.4.558

12.

Coleman

(1988). Social capital in the creation of human capital. American Journal of Sociology, 94(Suppl.), 95–120. https://doi.org/10.1086/228943

13.

Cortright

(2018, July). America’s most diverse mixed-income neighborhoods. City Observatory.

14.

Farrington

D. P.

Ttofi

M. M.

Piquero

A. R.

(2016). Risk, promotive, and protective factors in youth offending: Results from the Cambridge study in delinquent development. Journal of Criminal Justice, 45, 63–70. https://doi.org/10.1016/j.jcrimjus.2016.02.014

15.

Goodenow

(1993). The psychological sense of school membership among adolescents–scale development and educational correlates. Psychology in the Schools, 30(1), 79–90. https://doi.org/10.1002/1520-6807(199301)30:1<79::AID-PITS2310300113>3.0.CO;2-X

16.

Hipp

J. R.

(2016). Collective efficacy: How is it conceptualized, how is it measured, and does it really matter for understanding perceived neighborhood crime and disorder? Journal of Criminal Justice, 46, 32–44. https://doi.org/10.1016/j.jcrimjus.2016.02.016

17.

Hirschi

(1969). Causes of delinquency. University of California Press.

18.

IBM (2019). IBM SPSS statistics for windows (version 26.0). IBM Corp.

19.

Johnson

E. I.

Kilpatrick

Bolland

(2020). Positive youth development in the context of household member contact with the criminal justice system. Children and Youth Services Review, 114, 105033. https://doi.org/10.1016/j.childyouth.2020.105033

20.

Keele

Kelly

N. J.

(2006). Dynamic models for dynamic theories: The ins and outs of lagged dependent variables. Political Analysis, 14(2), 186–205. https://doi.org/10.1093/pan/mpj006

21.

Kennet-Cohen

Kleper

Turvall

(2018). Standard errors and confidence intervals for correlations corrected for indirect range restriction: A simulation study comparing analytic and bootstrap methods. British Journal of Mathematical and Statistical Psychology, 71(1), 39–59. https://doi.org/10.1111/bmsp.12105

22.

Keyes

C. R.

(2000, November 5–7). Parent-teacher partnerships: A theoretical approach for teachers [Conference session], Issues in Early Childhood Education: Curriculum, Teacher Education, & Dissemination of Information. Proceedings of the Lilian Katz Symposium, Champaign, IL, USA.

23.

Maurizi

L. K.

Ceballo

Espstein-Ngo

Cortina

K. S.

(2013). Does neighborhood belonging matter? Examining school and neighborhood belonging as protective factors for Latino adolescents. American Journal of Orthopsychiatry, 83(2/3), 323–334. https://doi.org/10.1111/ajop.12017

24.

McClelland

G. H.

Judd

C. M.

(1993). Statistical difficulties in detecting interactions and moderator effects. Psychological Bulletin, 114(2), 376–390. https://doi.org/10.1037/0033-2909.114.2.376

25.

Muthén

(1998-2017). Mplus user’s guide (8th ed.). Muthén and Muthén.

26.

Park

R. E.

Burgess

McKenzie

(1925). The city. University of Chicago Press.

27.

Rogers

W. M.

(2002). Theoretical and mathematical constraints of interaction regression models. Organizational Research Methods, 5(3), 212–230. https://doi.org/10.1177/1094428102005003002

28.

Ross

C. E.

Jang

S. J.

(2000). Neighborhood disorder, fear, and mistrust: The buffering role of social ties with neighbors. American Journal of Community Psychology, 28(4), 401–420. https://doi.org/10.1023/A:1005137713332

29.

Rutter

(1985). Resilience in the face of adversity: Protective factors and resistance to psychiatric disorder. British Journal of Psychiatry, 147(6), 598–611. https://doi.org/10.1192/bjp.147.6.598

30.

Sampson

R. J.

Morenoff

J. F.

Gannon-Rowley

(2002). Assessing “neighborhood effects”: Social processes and new directions in research. Annual Review of Sociology, 28, 443–478. https://doi.org/10.1146/annurev.soc.28.110601.141114

31.

Sampson

R. J.

Raudenbush

S. W.

Earls

(1997). Neighborhoods and violent crime: A multilevel study of collective efficacy. Science (New York, N.Y.), 277(5328), 918–924. https://doi.org/10.1126/science.277.5328.918

32.

Schreck

C. J.

(2002). How do social bonds restrain crime? A study of the mechanisms. Journal of Crime and Justice, 25(2), 1–21. https://doi.org/10.1080/0735648X.2002.9721155

33.

Shadish

W. R.

Cook

T. D.

Campbell

D. T.

(2002). Experimental and quasi-experimental designs for generalized causal inference. Mifflin.

34.

Shaw

C. R.

McKay

H. D.

(1942). Juvenile delinquency and urban areas. University of Chicago Press.

35.

Stoddard

S. A.

Zimmerman

M. A.

Bauermeister

J. A.

(2012). A longitudinal analysis of cumulative risks, cumulative promotive factors, and adolescent violent behavior. Journal of Research on Adolescence, 22(3), 542–555. https://doi.org/10.1111/j.1532-7795.2012.00786.x

36.

Stouthamer-Loeber

Wei

Wikström

P.-O. H.

(2002). Risk and promotive effects in the explanation of persistent serious delinquency in boys. Journal of Consulting and Clinical Psychology, 70(1), 111–123. https://doi.org/10.1037/0022-006X.70.1.111

37.

Takagi

Ikeda

Kawachi

(2012). Neighborhood social capital and crime victimization: Comparison of spatial regression analysis and hierarchical regression analysis. Social Science and Medicine, 75(10), 1895–1902. https://doi.org/10.1016/j.socscimed.2012.07.039

38.

Thulin

E. J.

Heinze

J. E.

Kusunoki

Hsieh

H.-F.

Zimmerman

M. A.

(2021). Perceived neighborhood characteristics and experiences of intimate partner violence: A multilevel analysis. Journal of Interpersonal Violence, 36(23–24), NP13162–NP13184. https://doi.org/10.1177/0886260520906183

39.

Turner

M. G.

Piquero

A. R.

Pratt

T. C.

(2005). The school context as a source of self-control. Journal of Criminal Justice, 33(4), 327–339. https://doi.org/10.1016/j.jcrimjus.2005.04.003

40.

Valido

Merrin

G. J.

Espelage

D. L.

Robinson

L. E.

Nickodem

Ingram

K. M.

El Sheikh

A. J.

Torgal

Fairclough

(2021). Social-ecological predictors of homophobic name-calling perpetration and victimization among early adolescents. Journal of Early Adolescence. Online first, https://doi.org/10.1177/02724316211002271

41.

van der Laan

A. M.

Veenstra

Bogaerts

Verhulst

F. C.

Ormel

(2010). Serious, minor, and non-delinquents in early adolescence: The impact of cumulative risk and promotive factors. The TRAILS study. Journal of Abnormal Child Psychology, 38(3), 339–351. https://doi.org/10.1007/s10802-009-9368-3

42.

Walters

G. D.

(2022). Protecting against delinquency in early adolescence with nonverbal intelligence and neighborhood social capital. Journal of Early Adolescence, 42(2), 204–226. https://doi.org/10.1177/02724316211020362

43.

Whitney

S. D.

Renner

L. M.

Herrenkohl

T. I.

(2010). Gender differences in risk and promotive classifications associated with adolescent delinquency. Journal of Genetic Psychology, 17(2), 116–138. https://doi.org/10.1080/00221320903548092

44.

Wilkins

A. S.

(2017). To lag or not to lag? Re-evaluating the use of lagged dependent variables in regression analysis. Political Science Research and Methods, 6(2), 393–411. https://doi.org/10.1017/psrm.2017.4

45.

Young

J. T. N.

Rebellon

C. J.

Barnes

J. C.

Weerman

F. M.

(2013). What do alternative measures of peer behavior tell us? Examining the discriminant validity of multiple methods of measuring peer deviance and the implications for etiological models. Justice Quarterly, 32(4), 626–762. https://doi.org/10.1080/07418825.2013.788730