Collider Stratification Bias as a Threat to Validity in U.S.-Based Health Disparities Research

• Activity Number: 508
• Type: Topic Contributed
• Date/Time: Wednesday, August 4, 2010 : 10:30 AM to 12:20 PM
• Sponsor: Health Policy Statistics Section
• Author(s): Whitney R. Robinson*+ and Jay Kaufman
• Companies: University of Michigan and McGill University
• Keywords: regression ; causal inference ; race ; sex ; health disparities ; directed acyclic graphs

Abstract

We argue that collider-stratification bias (CSB) is a threat to validity when estimating covariate-adjusted disparities for demographic factors like sex and race. Using directed acyclic graphs (DAGs), we describe the pervasiveness of CSB, a type of induced selection bias, in disparities research. In modeling Black-White disparities, most potential covariates are (1) differentially distributed between Blacks and Whites and (2) differentially associated with unmeasured causal factors in Blacks and Whites. If adjusted for, the covariates will be colliders and induce CSB. In estimating sex disparities, fewer potential covariates are colliders: adjusting for parental and residential characteristics uncorrelated with a respondent's sex should not necessarily induce bias. Using regression-based modeling to estimate adjusted disparities may be of limited utility in health disparities research.