apoorvalal / covariate_balancing_propensity_scores

CBPS via exponential tilting for ATE/ATT

CBPS via empirical loss minimization

Solves ATT analogue of the covariate-balancing propensity scores problem

R implementation lightly edited version of implementation here (courtesy Wager/Svedrup)

Uses exponential tilting for exact balance by default, which is preferable in low dimensions. Regularization results in approximate balance in high dimensions.

Refs:

• Imai and Ratkovic
• Zhao
• Wang and Zubizarreta
• Hainmueller