We find most of the relevant studies for our research question among those that employ scales of attachment, usually self-reported by an adolescent or parent.
There were 45 such studies, and 18 of them reported a PoC that was negative and significant between measures of attachment and violent or physically aggressive behavior.
Many of the studies employed multivariate designs in this category, and in many cases, associations between attachment and violence were in the right direction, but no longer significant when control variables were applied. This raises the question of whether attachment is associated with violence at all. Our concerns about this attenuation are mollified to some degree by the fact that some of these studies included redundant variables in their models. For example, Pogarsky et al. (2003) included a variable measuring “ineffective parenting" while other authors controlled for “low parental support” (Saner & Ellickson, 1996) and attachment to teachers (Gardner & Shoemaker, 1989). Some authors include two separate measures of attachment to mother and father in the same model (Craft, 1996; Kerpelman & Smith-Adcock, 2005; Li, 2007), or subscales of attachment in the same model (Craft, 1996; Lyons-Ruth et al., 1993), which we believe is likely to seriously undermine the ability of the model to accurately estimate the influence of attachment as a whole. The upshot of this type of modeling problem is that the authors may commit type II errors (missing the effect of attachment).
We selected some of the “better” models, for our purposes: models with appropriate controls that were not overly conservative due to the inclusion of too many overlapping variables. These studies are mainly supportive of an attachment- violence relationship. Benda (2005) controlled for gender; age; race; urban residence; attachment; caregiver monitoring; psychological factors (self-efficacy, ego identity, resilience); emotional abuse; sexual abuse; physical abuse; depression; aggression; stress; suicidal thoughts; school satisfaction; spiritual well-being; delinquent peers; and low behavioral control; nonetheless, he reports a statistically significant association between attachment and violence. Others report statistically significant associations in the predicted direction controlling for disorganization, affluence, race/ethnicity, parent education, age, peer influence, school attachment/commitment, involvement, belief (Gottfredson et al., 1991), race, grades, age, sex, school attachment, school involvement, self-esteem, and self-control (Owens-Sabir, 2007). In a Turkish study, the association between attachment and violence was in the right direction but not significant, controlling for a host of variables (Ozbay & Ozcan, 2006); a reanalysis of the same data suggests that the association was significant for girls but not boys (Ozbay & Ozcan, 2008). Cheung and Cheung (2008) controlled for self-control, attachment to school and other social bonds, strain variables, including coercive parenting and parent labeling, and they report several coefficients that are in the right direction, but not statistically significant.
In some studies, the authors control for recent wave aggression or violence. While this may help alleviate concerns about temporal order, it changes the interpretation of the attachment coefficient. Because previous wave aggression may have been influenced by earlier attachment relations, the proper interpretation of these coefficients becomes “the effect of current attachment on recent change in offending.” Several studies in our review may have missed an effect of attachment on violent behavior due to this type of model specification (Alink et al., 2009; Rebellon, 2002; Resnick et al., 2004), though there are studies that control for previous aggression or violence and still estimate statistically significant coefficients (Brezina, 1999; Liska & Reed, 1985; McNulty & Bellair, 2003; Rebellon & van Gundy, 2005).
Some of the authors provided sex-disaggregated analyses and though, at times, the findings differed between males and females within a study, on the whole we see no pattern favoring differences in associations by sex.
Similarly, while most studies report some indicator of “parent attachment,” a few distinguish maternal attachment and paternal attachment. These comparisons sometimes have weak statistical power, as the authors generally included both maternal attachment and paternal attachment in the models together. Most provided only partial support for associations between maternal and/or paternal attachment and violence, but no pattern emerged from which we could discern any differences in effects of maternal versus paternal attachment (Benda & Corwyn, 2002; Craft, 1996; Cusick et al., 2012; Foshee, Bauman, & Linder, 1999; Kesner et al., 1997; Li, 2007; Marcus & Betzer, 1996; Sarracino, Presaghi, Degni, & Innamorati, 2011). Many studies have suggested that children who are provided care by nonmaternal figures do not differ from other children on many dimensions, including attachment (Erel, Oberman, & Yirmiya, 2000).