ABSTRACT
Prior research and theorizing have suggested that the relationship between child maltreatment and delinquent outcomes is gendered. However, research to date that has sought to test this hypothesis has been incomplete due to issues such as inconsistency in findings, focusing on a single type of maltreatment or delinquent outcome within a given study, and an overreliance on nonprobability samples. The authors sought to expand on this literature by examining the effects of several different types of child maltreatment on six different delinquent behaviors during middle adolescence in a nationally representative probability sample of American adolescents. Results suggest that in the case of physical abuse and alcohol use, child maltreatment shapes delinquent behaviors differently for girls and boys, with several other types of delinquent behaviors being similarly shaped by child maltreatment among girls and boys. Implications of the findings are discussed.
Disclosure statement
No potential conflict of interest was reported by the authors.
Notes
1. Missing data on the dependent variables were treated as missing, while missing values on the independent and control variables were imputed using multiple imputation in Stata 13.1 (StataCorp, College Station, TX).
2. While these variables could also mediate the child maltreatment-delinquent outcomes relationship, checking for such effects would not be appropriate in the present analysis given that they were recorded during and cover the same time frame as the dependent variables, resulting in issues for claims of causality. Interaction terms were examined between the maltreatment items and these measures, none of which reached statistical significance.
3. These variables, and the guidance on how to appropriately use them, are provided by Add Health. Specifically, Add Health respondents have an individual weighting variable, a cluster variable based on their school, and a stratification variable based on their census region. These variables are meant to correct for the unequal probability of selection in the Add Health sample, and correct coefficients such that results from analyses are generalizable to the entire target population.
4. For all predictor variables, in all models presented, variance inflation factor < 4.
5. When data are “svyset” in Stata 13.1, pseudo-R2 values cannot be produced when running negative binomial regression models for statistical reasons.
6. Diagnostic tests confirm that for all models in the NB model is more appropriate with this dependent variable than the Poisson model (for all alpha statistics, p < .05).
7. When data are “svyset” in Stata 13.1, McKelvey & Zavoina’s R2 is the only one of the numerous pseudo-R2 values for use with logistic regression models that is produced. This same pseudo-R2 value is presented in conjunction with running away in Table 4.
8. When data are “svyset” in Stata 13.1, standardized (beta) coefficients cannot be obtained for OLS regression models for statistical reasons.
9. For the sake of parsimony, this is the only model for which we present results. Results from the models that showed no significant gender-maltreatment interaction are available upon request.
10. The number of regression models run introduces the potential for the multiple testing (i.e., multiple comparisons) problem. The very conservative Bonferroni correction would argue for a significance cutoff of α/n, in this case .05/36, or .00139. We will note that in 11 of 20 models where the maltreatment effect was significant at the .05 level, the p value was larger than this very strict cutoff. However, in only 6 of these 11 cases was alpha > .01. We would argue strongly that the results be taken at face value. Given the clear record of maltreatment having effects on delinquent outcomes in the research literature, both in general and specifically within this dataset, these significant effects do not seem coincidental, and the risk is high here for accepting false negatives based on a very strict cutoff for alpha. Still, we provide this information in an effort to be completely forthright with our findings.
11. An alternative modeling strategy considered was to employ multiple group models to examine whether the relationships worked differently for girls and boys. The advantage of multiple group models is to be able to test whether any visible differences between girls and boys are significantly different. Our initial models tested for any gender differences using interactions (Gender × Maltreatment), and largely found no differences between gender groups (no difference in 17 of 18 models). Given the absence of any differences in the vast majority of these initial models, we did not proceed with multiple group models.