Exploring the Temporal Relationship between Substance Abuse and Domestic Violence Victimization

Although a correlation between substance abuse (SA) and domestic violence (DV) is well established, very little is known about the temporal relationship of SA and DV victimization. We will explore the relationship between SA and DV victimization using four waves of a longitudinal study of 949 women receiving TANF prior to the first wave of the study. In particular, we examine the relative merits of a self-medication hypothesis (SA follows DV) and a vulnerability hypothesis (SA precedes DV).

Method. This study is a secondary analysis of data collected in the Illinois Family Study. Women who received TANF in 1998 were interviewed at all four waves of the study. DV is measured using an 8-item scale. SA is indicated by self-reported treatment, use of illegal drugs, binge drinking, quantity/frequency of alcohol use, and the CAGE. Standard measures of social support, self-efficacy, depression, financial stress, and neighborhood/housing conditions are used as control variables. In the current report, we will treat SA and DV as binary variables and examine their temporal relationship using conditional logistic regression models, controlling for the effects of social support, self-efficacy, depression, financial stress, and neighborhood/housing conditions. The main strength of using conditional logistic regression or fixed effects models to analyze longitudinal data with categorical responses is the ability to control for both observed and unobserved time-invariant variables.

Results. At wave 1, the lifetime prevalence of DV among participants was 34.2% and 8.8% had been victimized in the past 12 months. Nearly one in 10 (9.2%) were assessed as likely substance abusers at wave 1. Controlling for the time-invariant variables, our first conditional logistic regression model suggests DV is a significant predictor of SA (OR = 1.93) and SA is a significant predictor of DV (OR =1.97). That is, both the self-medication and vulnerability hypotheses should be accepted and neither of the models fits the data significantly better than the other. In the second model we added social support and self-efficacy variables in addition to depression, financial stress, and neighborhood/housing conditions and other time-invariant variables. In this model, DV was a significant predictor of SA (OR = 2.31), but SA was not a significant predictor of DV. In the second model, the self-medication hypothesis seems more plausible than the vulnerability hypothesis.

Discussion. Conditional logistic regression models of the relationship between SA and DV provide some support for both hypotheses we considered: (1) women use alcohol and drugs to ameliorate the effects of trauma (self-medication hypothesis) and (2) women who abuse alcohol and drugs are subsequently more likely to be abused by their partners (vulnerability hypothesis). Taken together our findings support a reciprocity hypothesis (Kilpatrick, et al., 1997): either SA or DV increases the risk for the other, and the relationship is often cyclic. Applications of the reciprocity hypothesis to practice and policy are discussed.