Primary Submission Category: Causal Inference and Bias/Discrimination
Causal estimands for equity
Authors: Laura Hatfield,
Presenting Author: Laura Hatfield*
To estimate the impacts of programs and policies on health equity, we must first define the causal target estimand. This initial step structures everything that follows in the analysis, from defining the population and outcomes to selecting comparison groups and potential confounders. Yet many health equity evaluations fail to engage seriously with this initial process. They may simply estimate different program effects in each group (e.g., Black and white patients) and informally compare the estimated treatment effects across subgroups. These analyses may conclude that equity has been improved if the effects are more beneficial in Black patients. However, this does not correspond to a principled causal analysis. In this talk, I detail five possible target estimands, the causal assumptions that would be required to identify them, and potential estimation strategies. Among these is a proposed novel estimand that puts an adjusted measure of equity on the left-hand side (using rank-and-replace methods). I discuss the causal assumptions implied by incorporating covariates directly in the outcome measure (compared to say, regression adjustment or propensity score methods). I compare these estimands’ ability to yield informative conclusions about the effects of health policies and programs on equity.