Collapsing-Fast-Large-Almost-Matching-Exactly: A Matching Method for Causal Inference

We aim to create the highest possible quality of treatment-control matches for categorical data in the potential outcomes framework. Matching methods are heavily used in the social sciences due to their interpretability, but most matching methods in the past do not pass basic sanity checks in that they fail when irrelevant variables are introduced. Also, past methods tend to be either computationally slow or produce poor matches. The method proposed in this work aims to match units on a weighted Hamming distance, taking into account the relative importance of the covariates; the algorithm aims to match units on as many relevant variables as possible. To do this, the algorithm creates a hierarchy of covariate combinations on which to match (similar to downward closure), in the process solving an optimization problem for each unit in order to construct the optimal matches. The algorithm uses a single dynamic program to solve all of optimization problems simultaneously. Notable advantages of our method over existing matching procedures are its high-quality matches, versatility in handling different data distributions that may have irrelevant variables, and ability to handle missing data by matching on as many available covariates as possible

[1]  Sebastian Schneeweiss,et al.  Using high‐dimensional propensity scores to automate confounding control in a distributed medical product safety surveillance system , 2012, Pharmacoepidemiology and drug safety.

[2]  Cynthia Rudin,et al.  FLAME: A Fast Large-scale Almost Matching Exactly Approach to Causal Inference , 2017, J. Mach. Learn. Res..

[3]  W. G. Cochran,et al.  Controlling Bias in Observational Studies: A Review. , 1974 .

[4]  Cynthia Rudin,et al.  A Minimax Surrogate Loss Approach to Conditional Difference Estimation , 2018, ArXiv.

[5]  J. Avorn,et al.  High-dimensional Propensity Score Adjustment in Studies of Treatment Effects Using Health Care Claims Data , 2009, Epidemiology.

[6]  Dan Suciu,et al.  ZaliQL: Causal Inference from Observational Data at Scale , 2017, Proc. VLDB Endow..

[7]  Paul R. Rosenbaum,et al.  Imposing Minimax and Quantile Constraints on Optimal Matching in Observational Studies , 2017 .

[8]  D. Rubin.  Matched Sampling for Causal Effects: The Use of Matched Sampling and Regression Adjustment to Remove Bias in Observational Studies , 1973 .

[9]  D. Rubin.  Matched Sampling for Causal Effects: Matching to Remove Bias in Observational Studies , 1973 .

[10]  A. Belloni,et al.  Inference on Treatment Effects after Selection Amongst High-Dimensional Controls , 2011, 1201.0224.

[11]  D. Rubin.  For objective causal inference, design trumps analysis , 2008, 0811.1640.

[12]  S. Athey,et al.  Generalized random forests , 2016, The Annals of Statistics.

[13]  Donald B. Rubin,et al.  Multivariate matching methods that are equal percent bias reducing , 1974 .

[14]  G. King,et al.  Causal Inference without Balance Checking: Coarsened Exact Matching , 2012, Political Analysis.

[15]  M. Farrell.  Robust Inference on Average Treatment Effects with Possibly More Covariates than Observations , 2015 .

[16]  Ramakrishnan Srikant,et al.  Fast Algorithms for Mining Association Rules in Large Databases , 1994, VLDB.

[17]  G. King,et al.  Multivariate Matching Methods That Are Monotonic Imbalance Bounding , 2011 .

[18]  Gary King,et al.  MatchIt: Nonparametric Preprocessing for Parametric Causal Inference , 2011 .

[19]  D. Rubin,et al.  The central role of the propensity score in observational studies for causal effects , 1983 .

[20]  David A. Lane.  Randomization Analysis of Experimental Data: The Fisher Randomization Test Comment , 1980 .