Findings: School Problems
Few studies estimated associations between school problems and violence. In these studies, school problems were operationalized as truancy (Farrington, 1989; Farrington, Loeber, & Berg, 2012; Loeber et al., 2005), school disciplinary problems as noted in school records (Piquero, 2000), problem school behavior including going to the principal’s office and being suspended or expelled (Salts et al., 1995), and “more than one suspension” (Ybarra et al., 2008). The findings consistently indicate that school problems are associated with violent behavior. The PoC was statistically significant in all of these studies. There were no studies examining associations between school problems and strictly nonviolent criminal behavior, however.
One important concern about these findings is that the causal order between this particular independent variable and violence is questionable. It is obvious that violent behavior in school could easily cause suspension and expulsion. Findings on other operationalizations such as truancy (which is not so obviously caused by violent behavior) assuage that concern.
There were 3 studies that compared the school problems of violent and nonviolent offenders, and the findings lean in the expected direction (at least 2 out of 3) (Hill-Smith, Hugo, Hughes, Fonagy, & Hartman, 2002; Loeber et al., 2005). Loeber et al. (2005) report a series of models which suggest that violent offenders in the large-scale Pittsburgh youth study were more truant than nonviolent offenders, and this effect was robust across multivariate models. Piquero (2000) found that, while violent offenders had more school problems than nonviolent offenders, the difference was not statistically significant (n = 91). In addition, Piquero compared frequent violent to frequent nonviolent offenders, and the relationship was not in the predicted direction, so, following our established decision rules, we coded his study as “ambiguous” in spite of the partially supportive evidence.
We decided to have a quick look at the associations in the Add Health data set which includes self-report measures of suspension and expulsion. We also created a school problems index from Add Health data (wave 1), including 4 items that indicate getting along with teachers, trouble paying attention in school, trouble getting homework done, and trouble with other students.
We first looked at means, comparing those for violent offenders to nonviolent- only offenders and then between violent offenders and chronic nonviolent-only offenders. In all comparisons, the means for indicators of school problems (our index, and the rates of expulsion and suspension) are higher among the subjects reporting violence compared to those reporting nonviolent-only offending. Simple correlations suggest that these indicators are all strongly associated with violence (p < .001) with the exception of the difference between violent and frequent-nonviolent-only offenders, which was not significantly different. When we ran a multivariate logistic regression model, we found that school problems significantly predicted violent, compared to nonviolent-only offending controlling for myriad factors (demographics, income, intact family, peer drug and alcohol use, subject alcohol use, and even parent education and frequency of nonviolent offending), but the association was no longer significant when we added a control for GPA, suggesting that school problems and academic achievement are linked. Although we find that school problems are highly predictive of violent, versus nonviolent offending, we must hypothesize that the association may really be due to poor academics and that chronic nonviolent offenders may have equivalent school problems with violent ones.