Towards a Better Understanding of When to Apply Propensity Scoring: A Comparison with Conventional Regression in Ethnic Disparities Research

In the study of ethnic disparities in health outcomes, methods incorporating Propensity Scores (PS) have been increasingly utilized. However, many researchers lack clear understanding of when PS methods should be deployed in place of conventional regression models. One such scenario is presented here: where the relationship between ethnicity and primary care utilization is confounded with and modified by socioeconomic status (SES). Here, standard regression can only produce estimates of disparities at specific values of SES, a problem that PS methods do not share. PS methods require the choice of a reference sample (RS) to which the effect estimate is generalized. Using data from the National Alcohol Surveys, racial/ethnic disparities between whites and Hispanics in access to primary care intervention were estimated applying PS stratification and weighting, a modified regression-based approach, O-B decomposition, and conventional multivariable regression. Using whites as the RS, two strategies utilizing PS and O-B decomposition achieved similar results, quite different from that from multivariate regression. Using Hispanics as the RS, PS and O-B estimates were similar but quite different from those using whites as the RS. Unlike standard regression, PS or O-B methods can provide estimates that account for effect modification and incorporate a-priori hypotheses that guide selection of reference samples.