Using Omitted Variable Bias to Assess Uncertainty in the Estimation of an AIDS Education Treatment Effect

Abstract

In a comparison of a treatment group and a control group, a difference in results can correspond to the effect of the treatment; however, the difference might also be, at least in part, a reflection of pretreatment differences between the two populations. Covariance adjustment can reduce bias in the estimate of the treatment effect (Cochran & Rubin, 1973); however, baseline group differences with respect to unobserved covariates which cannot be controlled can lead to hidden bias. This article presents a simple method which uses omitted variable bias to assess the uncertainty of the hidden bias by describing what scenarios regarding the unobserved covariate can lead to a given level of hidden bias. A comparison of a culturally sensitive AIDS video education program and a standard AIDS video education program is used as an illustration.