Comparing Coefficients for Viofent and Nonviofent Offending
Even if attachment is significantly associated with nonviolent offending, if the coefficients representing its association with violence are larger than the association with nonviolent-only offending, this would also provide evidence of a differential effect. Unfortunately, few studies provide adequate information to compute a statistical test, but some report correlation coefficients, which is a function of proportional reduction of error in prediction, or standardized Beta coefficients, which can give us some indication of the relative size of effects. (When standardized, the regression coefficient expresses the change in a dependent variable associated with an increase or decrease of one standard deviation in an independent variable; e.g., Vogt, 1999.) The reader should be cognizant of the fact that comparing standardized Beta weights across two different dependent variables is risky; in particular if one of them is measured with more error than the other.
Five studies provide separate analyses for relationships between attachment and both violent and nonviolent offending. Rebellon’s (2002) analyses indicate that the association between a score on an attachment scale and violence is more consistently in the right direction than the association between attachment and nonviolent offending. Findings by Heck and Walsh (2000) were consistent with our predictions; parental death was significantly, negatively associated with violent offending but slightly positively associated with nonviolent offending. The estimated standardized relationship between violent arrest and “no mother” was barely larger than the relationship with nonviolent arrest in a study by Cusick et al. (2012). There was only one study using attachment categories that used both violent (physical aggression) and nonviolent outcomes. Lewis et al. reported that attachment security was positively associated with physical aggression and nonviolent antisocial behavior at age 6 among girls, and negatively associated (as predicted) for boys (Lewis, Feirling, McGuffog, & Jaskir, 1984). On the other hand, coefficients estimated by Pogarsky et al. (2003) on the associations between separation from parents and violent and nonviolent delinquency are not different (not statistically significant in any model). Overall, in this piecemeal set of studies, associations between measures of attachment and nonviolent antisocial behavior were weaker than they were for violent behavior which is consistent with our hypothesis. Thus, there is a small body of evidence supplying weak support for our differential etiology of violence thesis, at least when measures of parental loss are used.