Quantitative Issues in Cancer Research Working Seminar

Eric DunipaceDoctoral Student, Department of Biostatistics, Harvard University”Optimal Transport Weights for Causal Inference” ABSTRACT: Weighting methods are an increasingly popular way to perform causal inference, especially ones that weight by the inverse propensity score. However, such weights suffer from increased variance and bias when the outcome and weighting models are both incorrect. To correct for this deficit, some more recent methods use weights that target covariate balance between treated and control units rather than rely on a particular regression model. But these newer techniques may in turn rely too heavily on the means of the covariates as being sufficient to balance distributions between experimental groups. Instead, we propose a more robust method in terms of mean-squared error that relies on an interpolation between matching and weighting by using optimal transport weights. We directly target convergence in covariate distributions by minimizing the Wasserstein distances between treated and non-treated units and demonstrate reasonable performance both in settings where the outcome model is well-specified and mis-specified.