The Impact of Omitted Variable Bias on the Links between Childhood Mistreatment, PTSD, and Substance Use in a National Sample of Latino Americans

Background: Substance misuse has considerable negative impacts on minority and poor communities because of persistent behavioral health and health care disparities (DHH, 2010; Marrast, Himmelstein, & Woolhandler, 2016). Latino Americans (Latinos) currently represent the largest minority community in the United States, constituting 17.8 % (1/6+) of the nation’s total population (U.S. Census Bureau, 2017). The literature has established interrelationships among early trauma, such as childhood mistreatment/victimization (CM) with adulthood post-traumatic stress disorder (PTSD) and substance use disorders (SUD) in the general population. However, few studies have examined the long-term impact of childhood mistreatment on current diagnoses of SUD.  Based on the Self-medication Hypothesis (Brown & Wolfe, 1994; Khantzian, 1985, 1997) and the Hispanic-paradox assumption (Vega et al., 1998), we examined (1) a hypothetical model in which PTSD was expected to explain the CM-SUD association, and (2) the impact of key contextual variables on the potential influence of omitted-variable bias (OVB).

Method: OVB is a statistical phenomenon whereby variable associations are confounded by covariates that were not included in the original analysis (Harris & Robinson, 2007). Using a national sample, we examined a model for all Latinos that indicated only indirect effects of both childhood physical mistreatment (CPM) and childhood sexual mistreatment (CSM) on SUD, mediated through PTSD. To test the OVB assumption, we also investigated whether  subgroup differences emerged in the model examining U.S.-born and immigrant subgroup Latinos, with and without contextual- and acculturation-related factors. 

Results: The final solution lent support for our theory-driven mediation model, in which CPM showed a significant indirect effect on SUD, mediated through PTSD. Discrimination had a direct effect on SUD  among all Latinos. Subgroup analyses yielded essentially the same models, when including discrimination and other contextual variables (socioeconomic status and acculturation factors).  Leaving OVB-testing variables as “unobserved,” however, resulted in a direct effect of CPM and a between-group difference regarding both direct and indirect effects of CSM in multi-group analyses. Thus, nativity differences shown in the model with and without the contextual and acculturation-related variables can be attributed  to the explanatory power of discrimination, based on the concept of OVB. 

Conclusion: The present study may be among the first to provide empirical support for two evidence-based frameworks, the Self-medication Hypothesis and the Hispanic Paradox in a Latino national sample. The first noteworthy finding emerged from the whole-group model (i.e., all Latinos), indicating that the hypothetical model adequately fit the Latino representative sample. The second noteworthy result emerged from the sequential multi-group analyses, which lent support to both the Hispanic Paradox assumption and the OVB phenomenon. The likelihood of having a recent SUD diagnosis among U.S.-born Latinos was nearly three times the rate of their immigrant counterparts, but the tested model remained identical in both the U.S.-born and foreign born Latino subgroups. To wit, this potentially “unobserved” factor, (i.e., discrimination) may explain substantive differences between U.S.- and foreign-born Latinos, with respect to vulnerability to adulthood SUD among those who reported CM, and CPM, in particular.