Causal inference and balance weighting

Abstract  In causal inference, estimating appropriate weights w is critical for balancing covariates and reducing bias in treatment effect estimation. We propose a general framework where w is obtained by directly minimizing a user-defined loss function that captures the desired balance criterion. By embedding treatments and covariates into a reproducing kernel Hilbert space (RKHS), our approach leverages kernel methods to flexibly approximate complex relationships without restrictive parametric assumptions.

Slides

Photos

Participants: 19