Primary Submission Category: Weighting
Towards representation learning for general weighting problems in causal inference
Authors: Oscar Clivio, Avi Feller, Chris Holmes,
Presenting Author: Oscar Clivio*
Weighting problems in treatment effect estimation can be solved by minimising an appropriate probability distance. Choosing which distance to minimise, however, can be challenging as it depends on the unknown data generating process. An alternative is to instead choose a distance that depends on a suitable representation of covariates. In this work, we give errors that quantify the bias added to a weighting estimator when using a representation, giving clear objectives to minimise when learning the representation and generalising a large body of previous work on deconfounding, prognostic, balancing and propensity scores. We further outline a method minimising such objectives, and show promising numerical results on two semi-synthetic datasets.