Stabilized Inverse Probability Weighting via Isotonic Calibration

Lars van der Laan, Ziming Lin, Marco Carone, Alex Luedtke
Proceedings of the Fourth Conference on Causal Learning and Reasoning, PMLR 275:139-173, 2025.

Abstract

Inverse weighting with an estimated propensity score is widely used by estimation methods in causal inference to adjust for confounding bias. However, directly inverting propensity score estimates can lead to instability, bias, and excessive variability due to large inverse weights, especially when treatment overlap is limited. In this work, we propose a post-hoc calibration algorithm for inverse propensity weights that generates well-calibrated, stabilized weights from user-supplied, cross-fitted propensity score estimates. Our approach employs a variant of isotonic regression with a loss function specifically tailored to the inverse propensity weights. Through theoretical analysis and empirical studies, we demonstrate that isotonic calibration improves the performance of doubly robust estimators of the average treatment effect.

Cite this Paper

BibTeX
@InProceedings{pmlr-v275-laan25a, title = {Stabilized Inverse Probability Weighting via Isotonic Calibration}, author = {van der Laan, Lars and Lin, Ziming and Carone, Marco and Luedtke, Alex}, booktitle = {Proceedings of the Fourth Conference on Causal Learning and Reasoning}, pages = {139--173}, year = {2025}, editor = {Huang, Biwei and Drton, Mathias}, volume = {275}, series = {Proceedings of Machine Learning Research}, month = {07--09 May}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v275/main/assets/laan25a/laan25a.pdf}, url = {https://proceedings.mlr.press/v275/laan25a.html}, abstract = {Inverse weighting with an estimated propensity score is widely used by estimation methods in causal inference to adjust for confounding bias. However, directly inverting propensity score estimates can lead to instability, bias, and excessive variability due to large inverse weights, especially when treatment overlap is limited. In this work, we propose a post-hoc calibration algorithm for inverse propensity weights that generates well-calibrated, stabilized weights from user-supplied, cross-fitted propensity score estimates. Our approach employs a variant of isotonic regression with a loss function specifically tailored to the inverse propensity weights. Through theoretical analysis and empirical studies, we demonstrate that isotonic calibration improves the performance of doubly robust estimators of the average treatment effect.} }
Endnote
%0 Conference Paper %T Stabilized Inverse Probability Weighting via Isotonic Calibration %A Lars van der Laan %A Ziming Lin %A Marco Carone %A Alex Luedtke %B Proceedings of the Fourth Conference on Causal Learning and Reasoning %C Proceedings of Machine Learning Research %D 2025 %E Biwei Huang %E Mathias Drton %F pmlr-v275-laan25a %I PMLR %P 139--173 %U https://proceedings.mlr.press/v275/laan25a.html %V 275 %X Inverse weighting with an estimated propensity score is widely used by estimation methods in causal inference to adjust for confounding bias. However, directly inverting propensity score estimates can lead to instability, bias, and excessive variability due to large inverse weights, especially when treatment overlap is limited. In this work, we propose a post-hoc calibration algorithm for inverse propensity weights that generates well-calibrated, stabilized weights from user-supplied, cross-fitted propensity score estimates. Our approach employs a variant of isotonic regression with a loss function specifically tailored to the inverse propensity weights. Through theoretical analysis and empirical studies, we demonstrate that isotonic calibration improves the performance of doubly robust estimators of the average treatment effect.
APA
van der Laan, L., Lin, Z., Carone, M. & Luedtke, A.. (2025). Stabilized Inverse Probability Weighting via Isotonic Calibration. Proceedings of the Fourth Conference on Causal Learning and Reasoning, in Proceedings of Machine Learning Research 275:139-173 Available from https://proceedings.mlr.press/v275/laan25a.html.

Related Material