هنگامی که عشق لطمه می زند: بررسی تقاطع باوری قومیت، اجتماعی، اقتصادی، وضعیت، ارتباط پدر و مادر، کودک آزاری و نگرش جنسیتی در بزهکاری خشونت آمیز نوجوانان

Abstract Researchers have not yet reached agreement about the validity of several competing explanations that seek to explain ethnic differences in juvenile violent offending. Ethnicity cannot solely explain why boys with an ethnic minority background commit more (violent) crimes. By assessing the intersectionality of structural, cultural and individual considerations, both the independent effects as well as the interplay between different factors can be examined. This study shows that aforementioned factors cumulatively play a role in severe violent offending, with parental connectedness and child abuse having the strongest associations. However, since most variables interact and ethnicity is associated with those specific factors, a conclusion to be drawn is that ethnicity may be relevant as an additional variable predicting severe violent offending although indirectly.

Introduction Ethnic differences in juvenile violent crime have been repeatedly observed in different countries across the world. For instance, in the USA, official crime statistics (e.g., Engen et al., 2002, McCarter, 2009, Rossiter and Rossiter, 2009 and Stahl et al., 2007) as well as surveys on juvenile violent delinquency (e.g., Flores, 2002 and Pope and Snyder, 2003) show that the rates of involvement in serious violence are much higher for blacks than for whites. In most European countries, ethnic minority boys with a non-Western background are overrepresented among juvenile offenders, such as Turks in Germany, Algerians in France, and Moroccans in Belgium (Esterle-Hedibel, 2001, Gostomski, 2003 and Put and Walgrave, 2006). This overrepresentation of ethnic minority boys among juvenile offenders can also be found in the Netherlands. Research on reported and unreported crime shows that, compared to native Dutch adolescents, non-native Dutch youngsters are more likely to commit criminal acts, especially violent offenses (De Jong, 2007, Jennissen et al., 2009, Komen, 2002 and Van der Laan and Blom, 2011). This is particularly true for Moroccan-Dutch boys, who are disproportionately represented among juvenile offenders (Lahlah et al., 2013a and Veen et al., 2011). In fact, the proportion of criminal offenses committed by Moroccan-Dutch boys is nearly four times the proportion of this group in the total population (Broekhuizen & Driessen, 2006). These ethnic differences in juvenile violent crime remain constant in temporal, regional, and gender-specific terms (Baier & Pfeiffer, 2008). Therefore, the academic and public debate has been concentrating on causes of ethnic differences in juvenile violent crime.

Results Group differences and correlation between items Characteristics of the study participants are reported in Table 1. In both the school sample and the probation office sample, Moroccan-Dutch boys reported committing more severe violent acts in the past year than their Dutch peers. These differences were statistically significant for the probation office sample only (t = −3.71, p < 0.001). As for structural factors, the social circumstances of Moroccan-Dutch boys are particularly poor in comparison with their Dutch peers: They rated their family wealth significantly lower (χ2(4) = 24.34, p < 0.001 for the school sample and χ2(4) = 29.67, p < 0.001 for the probation office sample) and the proportion of father's unemployment (χ2(1) = 36.76, p < 0.001 for the school sample and χ2(1) = 38.50, p < 0.001 for the probation office sample) and mother's unemployment (χ2(1) = 103.59, p < 0.001 for the school sample and χ2(1) = 24.92, p < 0.001 for the probation office sample) was significantly higher. Table 1. Sample characteristics. School sample Probation office sample Dutch boys (295) Moroccan-Dutch boys (69) Dutch boys (70) Moroccan-Dutch boys (43) M (SD) M (SD) M (SD) M (SD) Severe violent offending 0.07 (0.60) 0.17 (1.16) 1.17 (2.06) 3.14 (3.08) Connectedness Father's emotional warmth 60.99 (13.03) 51.94 (13.03) 54.54 (15.73) 35.79 (17.28) Mother's emotional warmth 61.77 (11.17) 55.93 (14.22) 57.80 (14.01) 39.12 (14.99) Child abuse Sexual abuse 0.19 (1.43) 0.07 (0.60) 0.65 (3.06) 0.23 (0.81) Physical assault 0.29 (1.40) 0.62 (2.12) 2.06 (4.35) 3.02 (4.18) Psychological aggression 0.14 (0.59) 0.97 (1.71) 0.64 (1.39) 2.86 (2.05) Exposure to IPV 0.32 (1.98) 1.58 (3.45) 1.80 (4.29) 2.86 (3.57) Gender based family roles 40.07 (11.03) 53.46 (11.87) 44.76 (10.15) 62.77 (14.15) % (N) % (N) % (N) % (N) Socio-economic status Family's wealth very rich 3.7% (11) 1.4% (1) 7.1% (5) 2.3% (1) quite rich 34.2% (101) 10.1% (7) 37.1% (26) 2.3% (1) medium rich 56.6% (167) 71.0% (49) 45.7% (32) 48.8% (21) not so rich 4.7% (14) 14.5% (10) 7.1% (5) 30.2% (13) not rich 0.7% (2) 2.9% (2) 2.9% (2) 16.3% (7) Paternal unemployment 5.1% (15) 29% (20) 14.3% (10) 72.1% (31) Maternal unemployment 10.2% (30) 65.2% (45) 24.3% (17) 72.1% (31) Missing data were not included in calculations of means. Table options Further, the Moroccan-Dutch boys rated significantly lower levels of paternal emotional warmth (t = 4.26, p < 0.001 for the school sample; t = 5.93, p < 0.001 for the probation office sample) and significantly lower levels of maternal emotional warmth (t = 3.19, p < 0.001 for the school sample; t = 6.70, p < 0.001 for the probation office sample) in comparison with their Dutch peers. With the exception of sexual abuse by a family member, Moroccan-Dutch boys reported significantly more exposure to different types of child abuse in comparison with their Dutch peers. In both samples, significant differences between the two groups were found only for psychological aggression (t = −4.00, p < 0.001 for the school sample; t = −6.25, p < 0.001 for the probation office sample) and exposure to IPV (t = −2.93, p = 0.004 for the school sample; t = −1.36, p = 0.02 for the probation office sample). Finally, in both samples, significant differences in gender attitudes were found (t = −8.95, p < 0.001 for the school sample; t = −7.28, p < 0.001 for the probation office sample), with Moroccan-Dutch boys having more conventionally defined roles compared to Dutch boys. Table 2 presents correlations between measured variables. The upper part of the matrix (above the diagonal) shows correlations in the school sample. The variables ‘Child abuse’ and ‘Severe violent offending’ are slightly skewed, with L-shaped distributions. Skewness of these variables is slightly lower in the probation office sample (below the diagonal). Patterns of correlations were fairly similar across both samples, although effect sizes were stronger in the probation office sample. Table 2. Correlations between measured variables, means and standard deviations. School sample above the diagonal, probation office sample below the diagonal. 1 2 3 4 5 6 7 8 9 10 11 M SD Socio-economic status 1. Family wealth .30 .19 −.14 −.11 .01 .03 .11 .05 .13 .13 2. Father's unemployment .58 .41 −.15 −.15 .01 .08 .16 .02 .20 .05 Mother's unemployment .34 .62 −.21 −.18 −.06 .03 .27 .06 .33 .04 Parental connectedness 4. Father's emotional warmth −.15 −.31 −.51 .69 −.01 −.08 −.27 −.08 −.21 −.06 59.28 14.18 5. Mother's emotional warmth −.32 −.36 −.55 .76 −.09 −.16 −.23 −.14 −.19 −.12 60.66 12.01 Child abuse 6. Sexual Abuse by a family member .17 .04 .19 −.10 −.07 .21 −.02 .42 −.02 .02 0.16 1.29 7. Physical Assault .09 .03 .02 −.08 −.06 .03 .28 .61 .07 .01 0.35 1.56 8. Psychological Aggression .31 .49 .45 −.45 −.51 −.01 .34 .27 .25 .04 0.30 0.97 9. Exposure to IPV .18 .28 .36 −.29 −.36 .04 .36 .47 .14 .04 0.55 2.38 10. Gender Attitudes .38 .63 .60 −.60 −.63 −.04 .08 .53 .21 .09 42.61 12.35 11. Violent delinquency .31 .34 .40 −.48 −.51 .03 .12 .45 .43 .39 0.15 1.01 Mean 47.41 50.69 0.50 2.42 1.49 2.20 51.61 1.92 SD 18.66 16.98 2.46 4.30 1.99 4.04 14.68 2.66 Table options Among all indicators, only Family wealth was significantly associated with severe violent offending in the school sample, while in the probation office sample all indicators, with the exception of Sexual Abuse and Physical assault were significantly associated. Structural equation modeling Tests of measurement models (latent variables) Severe violent delinquency Four indicators of the tendency to commit severe violent offending were used: robbery with assault, assault with a weapon, weapon possession, and rape. A model with a latent variable loading on all four indicators provided a close approximate fit (χ2 based on p = 0.984; RMSEA = 0.000). The measurement model was also supported when tested on the probation office sample (χ2 based on p = 0.503; RMSEA = 0.000). Socio-economic status A measurement model that applied three indicators of socio-economic status was supported (CFI = 0.988; RMSEA = 0.080). This measurement model was also supported when tested on the probation office sample (CFI = 1.000; RMSEA = 0.000). Parental connectedness A measurement model that applied 18 indicators of father's emotional warmth provided a reasonable fit, though it had a relatively high RMSEA in both samples (CFI = 0.920; RMSEA = 0.082 on the school sample; CFI = 0.946; RMSEA = 0.082 on the probation office sample). Further, a measurement model that applied 18 indicators of mother's emotional warmth provided a reasonable fit, though it had a relatively high RMSEA (CFI = 0.893; RMSEA = 0.084). On the probation office sample, the measurement model resulted in a relatively high RMSEA as well (CFI = 0.920; RMSEA = 0.082 on the probation office sample). Child abuse A measurement model with a latent variable loading on all four indicators provided a close fit (χ2 based on p = 0.286; RMSEA = 0.020) when two theoretically reasonable correlations between residual variables were included: (1) a correlation between the residual variables for sexual abuse by a family member and psychological assault and (2) a correlation between the residual variables for physical assault and psychological assault. The measurement model was also supported when tested on the probation office sample (p = 0.648; RMSEA = 0.000). Gender attitudes A measurement model that applied ten indicators of gender-based family roles was supported (CFI = 0.966; RMSEA = 0.055). This measurement model was also supported when tested on the probation office sample (CFI = 0.967; RMSEA = 0.062). Predicting violent offending Fig. 2 presents the results for a model seeing severe violent delinquency as dependent on socio-economic status, parental connectedness, child abuse and gender attitudes, analyzed with full information maximum likelihood estimation using the full sample (The figure uses standardized coefficients). The model resulted in satisfying fit measures (CFI = 0.842; RMSEA = 0.074) and could explain a moderate percentage of the variance of the latent variable severe violent offending (R2 = 0.14). Child abuse (beta = 0.28) and connectedness (beta = −0.16) were estimated to be more closely related to violent offending than socio-economic status (beta = 0.01) and gender attitudes (beta = 0.05). On the other hand, if socio-economic status was estimated as the sole predictor of severe violent offending, it demonstrated a beta = 0.17. Similarly, using connectedness as the sole indicator gave a beta = −0.30. Child abuse as the sole indicator gave a beta = 0.31. Lastly, if gender attitudes were estimated as the sole predictor of severe violent offending, it demonstrated a beta = 0.13. Severe violent offending seen as dependent on socio-economic status, parental ... Fig. 2. Severe violent offending seen as dependent on socio-economic status, parental connectedness, child abuse and gender attitudes, full information maximum likelihood estimation using the whole sample including missing data. Standardized estimated. Figure options The results obtained with the school sample were compared with an analysis of the probation office sample (Table 3). It was necessary to use measurement variance between the school sample and the probation office sample, since identical unstandardized factor loadings for the school and probation office were not supported by the data. Full information maximum likelihood estimation then provided a close fit for the school sample (CFI = 0.966; RMSEA = 0.055) and a moderate fit for the probation office sample (CFI = 0.800; RMSEA = 0.094). The regression coefficient for ‘socio-economic status’ loading on ‘severe violent offending’ was similar for the school sample (b = 0.08) and the probation office sample (b = 0.04). The regression weight for ‘connectedness’ loading on severe violent offending’ was similar as well (b = −0.00 for the school sample; b = −0.00 for the probation office sample), while the regression weight for ‘child abuse’ on ‘severe violent offending’ became statistically significant in the probation sample (b = 0.07, p = 0.03). For ‘gender attitudes’, the regression coefficient was similar for both samples (b = 0.02 for the school sample; b = 0.07 for the probation office sample). Table 3. The model in Fig. 1 used with the school sample and the probation office sample. Full information likelihood estimations with unstandardized estimates. School sample Probation-office sample (N = 364) (N = 113) Socio-economic status Family's wealth 1.00a 1.00a Father's unemployment 0.72*** 0.83*** Mother's unemployment 1.05*** 0.56*** Parental connectedness Father's emotional warmth 1.00a 1.00a Mother's emotional warmth 1.45** 1.00*** Child abuse Exposure to IPV 1.00a 1.00a Sexual abuse by a family member 0.24*** 0.08 Physical assault 0.44*** 0.55** Psychological aggression 0.12*** 0.72*** Gender attitudes GAI1 1.00a 1.00a GAI 2 1.13*** 0.91*** GAI 3 2.66** 1.57*** GAI 4 4.18** 1.50*** GAI 5 3.53** 1.48*** GAI 6 3.12** 1.63*** GAI 7 3.88** 1.74*** GAI 8 0.26 0.83** GAI 9 3.52** 1.48*** GAI 10 1.98** 0.94** Severe violent offending Robbery with assault 1.00a 1.00a Assault with a weapon 1.75*** 1.00* Weapon possession 0.03 1.67*** Rape 1.74*** 0.61** SES → Severe violent offending 0.08 0.04 Connectedness → Severe violent offending −0.00 −0.00 Child abuse → Severe violent offending 0.00 0.07* Gender attitudes → Severe violent offending 0.02 0.07 SES → Child abuse 0.15 1.50** SES → Gender attitudes 0.64* 0.70** Child abuse → Gender attitudes 0.01 0.166* Connectedness → Child abuse −0.04** −0.08*** χ2 456.073 437.988 df 220 220 p 0.000 0.000 Normed chi-square (NC) 2.073 1.991 Comparative fit index (CFI) 0.925 0.800 Root mean square error of approximation (RMSEA) 0.054 0.094 RMSEA conf. interval, lower bound 0.047 0.081 RMSEA conf. interval, upper bound 0.061 0.107 a Fixed to unstandardized value of 1 to identify the model (which implies that no significance test of this individual parameter is provided). * p < 0.05. ** p < 0.01. *** p < 0.001. Table options Ethnicity predicting severe violent offending A separate part of the analyses explored whether ethnicity could account for differences in severe violent offending, using an alternative model which considered indirect paths, with ethnicity as the only exogenous variable (see Fig. 3), thus testing ethnicity as a predictor of socio-economic status, connectedness, child abuse, and gender attitudes, while all these five variables were used to predict severe violent offending. Severe violent offending seen as dependent on ethnicity, socio-economic status, ... Fig. 3. Severe violent offending seen as dependent on ethnicity, socio-economic status, parental connectedness, child abuse and gender attitudes. Figure options In the school sample, the SEM-based analysis with only ethnicity (Dutch = 1) predicting the latent variable ‘severe violent offending’ found a small association: beta = 0.05 (see Table 4). The estimated weight of ethnicity was reduced when socio-economic status and connectedness was accounted for, beta = 0.04; extending the model further by also including child abuse did not improve the explanation of ethnic differences, beta = 0.02. However, ethnicity did have a significant effect on all remaining predictor variables: socio-economic status (beta = 0.69, p < 0.001); connectedness (beta = −0.27, p < 0.001); child abuse (beta = 0.31, p < 0.001), and gender attitudes (beta = 0.29, p = 0.03). Since the complete model ( Fig. 2) accounted for only 1% of the variance of severe violent offending, no further analyses were performed on the school sample. Table 4. SEM-models testing the impact of ethnicity on a latent variable of violent offending, with full information maximum likelihood estimations (standardized estimates): school sample. Model 1 Model 2 Model 3 Model 4 Model 5 Ethnicity (Dutch = 1) 0.05 0.05 0.04 0.04 0.03 Socio-economic status 0.08 0.10 0.10 0.10 Parental connectedness −0.08 −0.09 −0.09 Child abuse 0.01 0.02 Gender attitudes 0.02 R2 0.00 0.00 0.01 0.01 0.01 Ethnicity → SES 0.69*** Ethnicity → Connectedness −0.27*** Ethnicity → Child abuse 0.31*** Ethnicity → Gender attitudes 0.29* χ2 4.263 42.104 58.633 173.104 482.73 df 5 18 31 69 239 p 0.512 0.001 0.002 0.000 0.000 Normed chi-square (NC) 0.853 2.339 1.891 2.509 2.020 CFI 1.000 0.984 0.984 0.951 0.927 RMSEA 0.000 0.061 0.050 0.064 0.053 RMSEA conf. interval, lower bound 0.000 0.037 0.030 0.053 0.046 RMSEA conf. interval, upper bound 0.067 0.085 0.069 0.077 0.060 **p < 0.01. * p < 0.05. *** p < 0.001. Table options In the probation office sample, the SEM-based analysis with only ethnicity predicting severe violent offending found a strong association: beta = 0.38 (see Table 5). The estimated weight of ethnicity in the probation office sample was reduced when socio-economic status (beta = 0.13), connectedness (beta = 0.13), child abuse (beta = 0.17), and gender attitudes (beta = 0.18) was accounted for. In addition, ethnicity had a significant effect on the predictor variables: socio-economic status (beta = 0.67, p < 0.001) and connectedness (beta = −0.59, p < 0.001). Furthermore, socio-economic status had a significant effect on both child abuse (beta = 0.35, p = 0.02) and gender attitudes (beta = 0.68, p < 0.001). Connectedness had a significant effect on child abuse as well (beta = −0.34, p = 0.01). Explained variance of severe violent offending was high (R2 = 0.61). Table 5. SEM-models testing the impact of ethnicity on a latent variable of violent offending, with full information maximum likelihood estimations (standardized estimates): probation office sample. Model 1 Model 2 Model 3 Model 4 Model 5 Ethnicity (Dutch = 1) 0.38* 0.13 0.13 0.17 0.18 Socio-economic status 0.38* 0.28 0.16 0.34 Parental connectedness −0.58** −0.46** −0.53** Child abuse 0.40* 0.38** Gender attitudes 0.26 R2 0.15 0.23 0.42 0.56 0.61 Ethnicity → SES 0.67*** Ethnicity → Connectedness −0.59*** Ethnicity → Child abuse 0.13 Ethnicity → Gender attitudes 0.12 χ2 6.564 28.482 67.210 161.02 455.015 df 5 18 31 69 239 p 0.255 0.055 0.000 0.000 0.000 Normed chi-square (NC) 1.313 1.582 2.178 2.334 1.904 CFI 0.967 0.950 0.907 0.831 0.810 RMSEA 0.053 0.072 0.083 0.109 0.089 RMSEA conf. interval, lower bound 0.000 0.000 0.069 0.087 0.077 RMSEA conf. interval, upper bound 0.149 0.120 0.130 0.131 0.102 * p < 0.05. ** p < 0.01. *** p < 0.001. Table options Alternative models were tested by hierarchical χ2. Both the direct path from ethnicity on severe violent offending (Δχ2 = 1.59, p > 0.05) and the direct path from socio-economic status on severe violent offending could be released (Δχ2 = 2.25, p > 0.05). However, neither the direct path from connectedness on severe violent offending, nor the path from child abuse on severe violent offending could be released (p < 0.001 in both cases). As for the indirect paths, both the indirect path from ethnicity on severe violent offending through child abuse (Δχ2 = 0.932, p > 0.05) as well as the indirect path from ethnicity through gender attitudes (Δχ2 = 0.828, p > 0.05) could be released. Likewise, the indirect path from child abuse through gender attitudes could be released (Δχ2 = 2.39, p > 0.05). All other indirect paths could not be released (p < 0.001 in both cases). SEM found that the alternative model (see Fig. 4) provided a reasonable fit (NC = 1.889; CFI = 0.810; RMSEA = 0.089) and explained 63% of the variance of severe violent offending. Severe violent offending seen as dependent on socio-economic status, parental ... Fig. 4. Severe violent offending seen as dependent on socio-economic status, parental connectedness, child abuse and gender attitudes, full information maximum likelihood estimation using the probation office sample including missing data.