Full Citation

Title: Probabilistic Matching: Incorporating Uncertainty to Correct for Selection Bias

Citation Type: Conference Paper

Publication Year: 2016

ISBN:

ISSN:

DOI:

NSFID:

PMCID:

PMID:

Abstract: Matching methods such as propensity score matching are commonly used to construct artificial treatment and control groups from observational data, to determine the causal effect of treatment. However, propensity scores, once estimated, are frequently treated as known, and the uncertainty inherent in their estimation is ignored. We introduce probabilistic matching, an improvement on propensity score matching, that incorporates the uncertainty of the estimated propensity score into the subsequent matching process by weighting matches by the estimated probability of matching. Notably, this is equivalent to averaging the estimated treatment effect over the propensity score distribution, given the data. Preliminary results demonstrate that our approach achieves comparable or lower bias and lower variance, when compared to vanilla propensity score matching. While we focus on matching in this paper, the idea of incorporating uncertainty can also be brought into other ways of utilizing estimated propensity scores, such as weighing and substratification.

Url: http://www.homepages.ucl.ac.uk/~ucgtrbd/whatif/Paper23.pdf

User Submitted?: No

Authors: Tan, Hui Fen; Hooker, Giles J; Wells, Martin T

Conference Name: NIPS 2016 Workshop: What If? Inference and Learning of Hypothetical and Counterfactual Interventions in Complex Systems

Publisher Location: Barcelona, Spain

Data Collections: IPUMS CPS

Topics: Other

Countries: