Feature Selection for Causal Effect Estimation

We define the notion of a minimal control function, on the basis of which a novel regression penalty is devised that is unbiased for average treatment effects. The development of the new approach combines insights from three distinct methodological traditions for studying causal effect estimation: potential outcomes, causal diagrams, and structural models with additive errors. It is demonstrated that traditional feature selection and/or regularization approaches to treatment effect estimation can exhibit severe bias for average and conditional average treatment effects. This is joint work with Drew Herren.