بزهکاری، دخالت والدین، جرم و جنایت بزرگسالان اولیه و جنسیت: مدارک و شواهد از میانجیگری تعدیل شده

Abstract One purpose of this study was to determine whether parental involvement, measured in late adolescence, mediates the relationship between delinquency in mid-adolescence and crime in early adulthood. This study's second purpose was to ascertain whether this relationship is moderated by sex, such that late adolescent parental involvement mediates the delinquency–crime relationship in females but not in males. A secondary analysis of data provided by 579 (272 males, 307 females) members of the National Longitudinal Survey of Youth-Child (NLSYC) was conducted in an effort to evaluate the possibility of moderated mediation in the relationship between delinquency at age 16, parental involvement at age 18, and criminality at age 24. Moderated mediation analysis, path analysis, and causal mediation analysis revealed the presence of a conditional indirect relationship between delinquency, parental involvement, and adult crime moderated by sex. These results are consistent with views on cumulative disadvantage and gendered pathways to crime.

Results Moderated mediation and path analysis using MPlus Descriptive statistics and sample correlations achieved by the control variables (early antisocial behavior, early parental involvement), independent variable (delinquency at age 16), mediator variable (parental involvement at age 18), and dependent variable (crime at age 24) in the total sample, male subsample, and female subsample are listed in Table 1. It should be noted that the parental involvement score was significantly higher for girls than for boys, t (577) = 3.69, p < 0.001. Moderated mediation analyses were then conducted using maximum likelihood (ML) as the estimator. Results obtained from testing Preacher et al.'s (2007) Models 2 (W moderating X → M and X → Y through interactions with X) and 3 (W moderating M → Y through interactions with M) are summarized in Table 2. The results of these moderated mediation analyses disclosed a significant interaction effect (moderated mediation) for Model 2 (sex moderated the delinquency-involvement relationship by interacting with delinquency). Path analysis of the delinquency → parental involvement → crime relationship further clarified the significant moderated mediation effect by exposing the presence of a significant mediation effect in females but not in males (see Fig. 1). Table 1. Descriptive statistics for and correlations between the control variables, independent variable, mediator variable, and outcome variable in the total sample, male subsample, and female subsample. Variable M SD Range TV Delinquency Involvement Crime Total sample (N = 579) Early Antisocial 1.52 1.64 0–11 0.01 0.10* 0.13** 0.00 Discuss TV 0.78 0.42 0–1 0.01 0.11** −0.02 Delinquency 0.58 0.93 0–4 0.14*** 0.11** Involvement 10.76 3.26 0–15 0.10* Crime 0.12 0.48 0–3 Male subsample (n = 272) Early Antisocial 1.74 1.76 0–11 −0.07 0.08 0.09 −0.04 Discuss TV 0.77 0.42 0–1 −0.03 0.06 −0.06 Delinquency 0.69 1.04 0–4 0.06 0.09 Involvement 10.24 3.26 0–15 0.02 Crime 0.22 0.38 0–3 Female subsample (n = 307) Early Antisocial 1.32 1.50 0–9 0.08 0.10 0.14* 0.01 Discuss TV 0.78 0.41 0–1 0.05 0.15** 0.07 Delinquency 0.48 0.81 0–3 0.22*** 0.10 Involvement 11.23 3.20 0–15 0.20*** Crime 0.04 0.28 0–3 Note. Early Antisocial = BPI Antisocial score at age 6; Discuss TV = parents discussed television programs with participant (1) or did not discuss television programs with participant (0) at age 6–8 (for the correlations reported in this table scores on this scale were inverted so that higher scores indicated less parental involvement); Delinquency = self-report of involvement in four delinquent activities at age 16 years; Involvement = score on the five-item Parental Involvement scale taken at age 18 years (for the correlations reported in this table and the for the remainder of the analyses scores on this scale were inverted so that high scores indicated less parental involvement and correlated positively with the independent and dependent variables); Crime = convicted of a crime, placed on probation, or served time in jail or prison in the last two years; M = mean; SD = standard deviation; Range = high and low scores on this particular measure in the current sample. *p < 0.05; **p < 0.01; ***p < 0.001. Table options Table 2. Results of moderated mediation analyses of the moderator variable effects of gender on the delinquency → parental involvement → adult crime mediated relationship: Models 2 and 3. Model/Effect Estimate (95% CI) S.E. T Model 2 Involvement on Delinquency −0.508 (−1.465 to 0.381) 0.469 −1.08 Involvement on Sex −1.282 (−1.910 to −0.683) 0.311 −4.12*** Involvement on Delinquency × Sex 0.679 (0.066–1.339) 0.323 2.10* Crime on Involvement 0.010 (−0.001 to 0.024) 0.006 1.59 Crime on Delinquency 0.082 (−0.100 to 0.291) 0.099 0.83 Crime on Sex −0.146 (−0.246 to −0.061) 0.047 −3.12** Crime on Delinquency × Sex −0.029 (−0.137 to 0.092) 0.058 −0.49 Model 3 Involvement on Delinquency 0.501 (0.203–0.810) 0.155 3.23** Crime on Involvement −0.008 (−0.049 to 0.040) 0.023 −0.35 Crime on Delinquency 0.041 (−0.014 to 0.107) 0.031 1.31 Crime on Sex −0.037 (−0.345 to 0.283) 0.159 −0.23 Crime on Involvement × Sex 0.012 (−0.015 to 0.038) 0.013 0.88 Note. Delinquency × Sex = interaction between delinquency and gender (moderator variable); Involvement × Sex = interaction between parental involvement and gender (moderator variable); Estimate = unstandardized beta coefficient; 95% CI = 95 percent confidence interval; S.E. = standard error of estimate; T = asymptotic t-test calculated by dividing the estimate by the standard error of estimate; the significant moderated mediation effect is in bold. *p < 0.05, **p < 0.01, ***p < 0.001. Table options Path analysis of the mediating effects of parental involvement on the ... Fig. 1. Path analysis of the mediating effects of parental involvement on the delinquency-crime relationship in the total sample, male subsample, and female subsample Note. Standardized beta coefficients are reported; the first figure is for the total sample, the second figure is for the male subsample, and the third figure is for the female subsample *p < 0.05; **p < 0.01; ***p < 0.001. Figure options When bootstrapping is used to test conditional indirect effects, only two assumptions are required: the relationships should be linear and the observations should be independent (Preacher et al., 2007). The independent observation assumption was met both conceptually (proper temporal order with no overlap between variables) and statistically (Tolerance = 0.96–0.97, Variance Inflation Factor = 1.03–1.04). A linear relationship between the independent and dependent variables is more difficult to establish given the fact that each variable was only composed of 4 or 5 ordered categories and the dependent variable may have been zero-inflated (92.9% in the total sample, 87.1% in males, 98.0% in females). Additional moderated mediation analyses and path analyses were consequently conducted in which the dependent variable (crime) was classified as categorical and then analyzed with a WLSMV (weighted least square parameter estimates with a diagonal weight matrix) estimator. Whether crime was treated as a four-level categorical or two-level dichotomous variable (0 = absent, 1 = present) the Model 2 results for the delinquency-involvement interaction, t = 2.11, p < 0.05, and the pathways between delinquency and involvement (standardized beta coefficient = 0.22) and between involvement and crime (standardized beta coefficients = 0.49–0.51) remained significant. Causal mediation analysis using R A bootstrapped causal mediation analysis was conducted using algorithms contained in an R language package authored by Imai et al. (2010). According to the results of this analysis, parental involvement failed to mediate the relationship between delinquency and crime in the full sample and in the male subsample, though it did mediate the relationship between delinquency and crime in the female subsample, accounting for 37% of the total effect (see Table 3). A sensitivity analysis of the mediation and outcome models in the female subsample, with early antisocial behavior and early parental involvement serving as control variables, revealed a rho (ρ) at which mediation = 0 of 0.20. By using the coefficients of determination (R2) for the mediator and outcome models as axes in a sensitivity graph it was possible to estimate the proportion of variance an unobserved confounding covariate or group of confounding covariates would need to explain before the mediation effect fell to zero (see Fig. 2). The results indicated that an unobserved confounding covariate or group of covariates would need to account for 16% of the variance in the mediator and 16% of the variance in the outcome to eliminate the mediation effect. Table 3. Results of causal mediation analysis of parental involvement at age 18 as a mediator of the delinquency–crime relationship in the total sample and in subsamples of males and females. Effect type Point estimate 95% CI p Total sample (N = 579) Mediation Effect (Delinq → Involvement → Crime) 0.0064 0.0006–0.0148 0.11 Direct Effect (Delinquency → Crime) 0.0510 0.0082–0.0940 0.02 Total Effect 0.0575 0.0154–0.1000 0.01 Proportion of Total Effect via Mediation 0.1081 0.0054–0.4750 0.28 MALE SUBSAMPLE (n = 272) Mediation Effect (Delinq → Involvement → Crime) 0.0008 −0.0047 to 0.0083 0.59 Direct Effect (Delinquency → Crime) 0.0539 −0.0180 to 0.1282 0.13 Total Effect 0.0547 −0.0168 to 0.1288 0.12 Proportion of Total Effect via Mediation 0.0044 −1.4534 to 2.7793 0.89 Female subsample (n = 307) Mediation Effect (Delinq → Involvement → Crime) 0.0130 0.0034–0.0251 0.04 Direct Effect (Delinquency → Crime) 0.0189 −0.0203 to 0.0569 0.35 Total Effect 0.0318 −0.0070 to 0.0717 0.11 Proportion of Total Effect via Mediation 0.3732 −1.4524 to 2.7793 0.36 Note. Delinq = Delinquency within the last year, taken at age 16; Involvement = Parental Involvement at age 18; Crime = conviction, probation, incarceration within the last two years, taken at age 24; Point Estimate = estimate of the size of the effect; 95% CI = 95% confidence interval of the point estimate; p = probability of the point estimate differing from zero. Table options Sensitivity analysis of the continuous dependent variable, crime at age 24, and ... Fig. 2. Sensitivity analysis of the continuous dependent variable, crime at age 24, and continuous mediator, parental involvement at age 18, with delinquency at age 16 as the independent variable. Contour lines represent the estimated average meditational effect at different levels of an unobserved confounder. The “0” line indicates how strong the unobserved confounder must be to completely eliminate the mediation effect. Figure options To illustrate the robustness of the causal mediation analysis results in the female subsample the mediator and outcome variables were regressed onto the two observed covariate confounders (i.e., early antisocial behavior and early parental involvement) and one additional covariate (race: white vs. non-white). This is only a demonstration because the effects of early antisocial behavior and early parental involvement were already taken into account in the original casual mediation and sensitivity analyses. The observed covariate confounders recorded an R2 of 0.029 in predicting the mediator (parental involvement) and an R2 of 0.002 in predicting the outcome (adult crime) in the female subsample. Hence, before sex and early parental involvement could reduce the mediating effect of parental support on the delinquency–crime relationship to nonsignificance, having already accounted for 2.9% of the variance in the mediator, they would need to explain over 50% of the variance in adult crime. In actuality, these confounding covariates accounted for only 0.2% of the variance in adult crime. Because it could be argued that the dependent variable (crime) violated the continuous variable assumption of multiple regression analysis, a second set of analyses were conducted using a dichotomized dependent variable (present = 1, absent = 0) and replacing linear regression analysis with probit regression analysis. The results were similar to those obtained when crime was treated as a continuous variable. In the total sample the mediating effect failed to achieve overall significance (p = 0.10) but produced a confidence interval that did not include zero (0.0005–0.0074). The mediating effect in the male subsample both failed to achieve overall significance (p = 0.56) and produced a confidence interval that included zero (−0.0026 to 0.0042). As with the linear regression results, the probit results for the female subsample displayed the strongest evidence of mediation. The overall effect approached statistical significance (p = 0.07) and the confidence interval did not include zero (0.0018–0.0125).