As discussed above, multivariate analyses can make a strong statement about an association between two variables due to their ability to reduce the likelihood of reporting spurious associations. In an ideal study, longitudinal associations with good measures of warmth or rejection in childhood and later violence, controlling for demographics, possible confounds such as parental supervision, severe abuse, and perhaps attachment would be included. In the present set of studies, we have no such “ideal” studies, but we have several whose methodology sheds the most light on our research question.
Overall, a comparison of findings from multivariate models to those from simple correlational tests merits just a few observations. First, there were far fewer multivariate analyses of parental warmth and violence than we would need to draw firm conclusions, given the likelihood of confounds. Second, among those that reported both correlations and multivariate coefficients, we see a few cases where the association was attenuated in models adding control variables. For example, the association between parental warmth and violence was significant in a bivariate model but null in the multivariate model reported by Barnow, Lucht, and Freyberger (2005). Unfortunately, Barnow et al. included parental rejection in the same model, which is probably overly conservative for our purposes4. This also occurred in Stacks’s analysis where the bivariate correlation was significant but three multivariate models were not. In this case, the authors were mainly interested in the moderating effect of parental warmth on the association between spanking and aggression. Thus, their models included spanking, and the interaction between spanking and warmth, as well as a very conservative control for aggressive behavior one year earlier (Stacks, Oshio, Gerard, & Roe, 2009). Scholte et al. (2007) also employed a conservative model, controlling for previous wave delinquency, and found no association between maternal warmth and boys or girls’ violent behavior in this teenage sample. Because previous wave delinquency is likely to have been influenced by parental warmth, this model may be overly conservative. The same could be said about findings reported by Brendgen et al. (2001). We note that studies which simultaneously included measures of both maternal and paternal warmth/rejection have uniformly reported non-significant partial coefficients, which supports our contention that model overspecification4 is a problem in this line of research.
Among studies with the best models for our purposes, two examined parental warmth, and the findings were divided. Shekarkhar and Gibson (2011) include controls for gender, age, SES, foreign born, family size, “Puerto Rican,” “other Latino,” parent supervision, and self-control. They did not find a significant association between parental warmth and violent offending. Tyler, Brownridge, and Melander (2011) looked at dating violence using a structural model which included controls for sex, grade in school, race, physical abuse, sexual abuse, neglect, and mediating factors such as delinquency and substance use and report that low parental warmth had a significant direct effect on dating violence. Hence, we conclude that our understanding of the association between parental warmth and violence could be greatly enhanced by more studies with carefully specified multivariate models.
By contrast, researchers have uncovered significant associations between parental rejection and physical aggression or violence in multivariate analyses of data from several countries including Germany, Canada, and the Netherlands, and spanning age groups (toddlers, adolescents and adults). Two studies that employed models that we deem to be the most appropriate were published by Day et al. and Huijbregts et al. In their analysis of prenatal tobacco use, Day et al. use a fairly large sample of very young children (n = 672) and regress age 3 aggression on maternal hostility including many control variables in the model: gender, age, maternal smoking, man in the house, “home environment,” life events, mother’s education, mother’s depression and current cocaine use (Day, Richardson, Goldschmidt, & Cornelius, 2000). They report that maternal hostility was still significantly associated with physical aggression. In a study of Quebecois children between ages 17 and 42 months, Huijbregts et al. (2008) found, likewise, that maternal hostile-reactive parenting was positively associated with 3yearold “high and rising” physical aggression, controlling for many factors such as maternal smoking, parent antisocial, maternal education, income, maternal depression, maternal age, drug use etc.