Matching (statistics)

Matching emulates the randomized experiment by identifying similar (matched) data points. The main limitation is that the matching can be done only on the observable covariates.

¹

Matching can reduce model dependence. See Ho2007matching.

Methods

• Mahalanobis distance matching
• Propensity score matching
• Exact matching and Coarsened exact matching
• Matching after learning to stretch
• Panel matching

Packages

• https://kosukeimai.github.io/MatchIt/index.html by Daniel Ho, Kosuke Imai, Gary King, Elizabeth Stuart, Noah Greifer

Reading list from Gary King‘s course

• Iacus2011causal: Causal Inference without Balance Checking: Coarsened Exact Matching
• King2019why: Why Propensity Scores Should Not Be Used for Matching
• King2016balance: The balance‐sample size frontier in matching methods for causal inference
• Iacus2019theory: A Theory of Statistical Inference for Matching Methods in Causal Research