FLAME: A Fast Large-scale Almost Matching Exactly Approach to Causal Inference

A classical problem in causal inference is that of matching, where treatment units need to be matched to control units based on covariate information. In this work, we propose a method that computes high quality almost-exact matches for high-dimensional categorical datasets. This method, called FLAME (Fast Large-scale Almost Matching Exactly), learns a distance metric for matching using a hold-out training data set. In order to perform matching efficiently for large datasets, FLAME leverages techniques that are natural for query processing in the area of database management, and two implementations of FLAME are provided: the first uses SQL queries and the second uses bit-vector techniques. The algorithm starts by constructing matches of the highest quality (exact matches on all covariates), and successively eliminates variables in order to match exactly on as many variables as possible, while still maintaining interpretable high-quality matches and balance between treatment and control groups. We leverage these high quality matches to estimate conditional average treatment effects (CATEs). Our experiments show that FLAME scales to huge datasets with millions of observations where existing state-of-the-art methods fail, and that it achieves significantly better performance than other matching methods.

[1]  Oscar Kempthorne,et al.  Experimental Designs in Sociological Research. , 1949 .

[2]  H. Chipman,et al.  BART: Bayesian Additive Regression Trees , 2008, 0806.3286.

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

[4]  Guido W. Imbens,et al.  A Martingale Representation for Matching Estimators , 2009, SSRN Electronic Journal.

[5]  Jinyong Hahn,et al.  Functional Restriction and Efficiency in Causal Inference , 2004, Review of Economics and Statistics.

[6]  J. Zubizarreta Matching for Balance , 2019 .

[7]  Sophie Papst Experimental Sociology A Study In Method , 2016 .

[8]  J. Zubizarreta,et al.  Evaluation of subset matching methods and forms of covariate balance , 2016, Statistics in medicine.

[9]  Jennifer Hill,et al.  Automated versus Do-It-Yourself Methods for Causal Inference: Lessons Learned from a Data Analysis Competition , 2017, Statistical Science.

[10]  Venkat Reddy Konasani,et al.  Multiple Regression Analysis , 2015 .

[11]  Jason Abrevaya,et al.  Estimating the effect of smoking on birth outcomes using a matched panel data approach , 2006 .

[12]  J. Sekhon,et al.  Genetic Matching for Estimating Causal Effects: A General Multivariate Matching Method for Achieving Balance in Observational Studies , 2006, Review of Economics and Statistics.

[13]  R. Stolzenberg,et al.  Multiple Regression Analysis , 2004 .

[14]  M. Feinleib National Center for Health Statistics (NCHS) , 2005 .

[15]  David M. Rubin,et al.  Practice of Epidemiology Propensity Score Methods for Analyzing Observational Data Like Randomized Experiments: Challenges and Solutions for Rare Outcomes and Exposures , 2015 .

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

[17]  Cynthia Rudin,et al.  Almost-Matching-Exactly for Treatment Effect Estimation under Network Interference , 2020, AISTATS.

[18]  Anthony J. Kondracki,et al.  Prevalence and patterns of cigarette smoking before and during early and late pregnancy according to maternal characteristics: the first national data based on the 2003 birth certificate revision, United States, 2016 , 2019, Reproductive Health.

[19]  J. Zubizarreta Journal of the American Statistical Association Using Mixed Integer Programming for Matching in an Observational Study of Kidney Failure after Surgery Using Mixed Integer Programming for Matching in an Observational Study of Kidney Failure after Surgery , 2022 .

[20]  Sanjog Misra,et al.  Deep Neural Networks for Estimation and Inference , 2018, Econometrica.

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

[22]  D B Rubin,et al.  Matching using estimated propensity scores: relating theory to practice. , 1996, Biometrics.

[23]  H. Chipman,et al.  Bayesian Additive Regression Trees , 2006 .

[24]  Max H. Farrell,et al.  Efficient Estimation of the Dose–Response Function Under Ignorability Using Subclassification on the Covariates , 2011 .

[25]  Stefan Wager,et al.  Estimation and Inference of Heterogeneous Treatment Effects using Random Forests , 2015, Journal of the American Statistical Association.

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

[27]  Donald B. Rubin,et al.  MULTIVARIATE MATCHING METHODS THAT ARE EQUAL PERCENT BIAS REDUCING, I: SOME EXAMPLES , 1974 .

[28]  L. Paulozzi,et al.  Maternal smoking and birth defects: validity of birth certificate data for effect estimation. , 2001, Public health reports.

[29]  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.

[30]  A. Baron Experimental Designs , 1990, The Behavior analyst.

[31]  M. J. van der Laan,et al.  The International Journal of Biostatistics Targeted Maximum Likelihood Learning , 2011 .

[32]  Mariana Caricati Kataoka,et al.  Smoking during pregnancy and harm reduction in birth weight: a cross-sectional study , 2018, BMC Pregnancy and Childbirth.

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

[34]  Cynthia Rudin,et al.  Interpretable Almost-Exact Matching for Causal Inference , 2019, AISTATS.

[35]  Luke Keele,et al.  Optimal Multilevel Matching in Clustered Observational Studies: A Case Study of the School Voucher System in Chile , 2014 .

[36]  D. Rubin Causal Inference Using Potential Outcomes , 2005 .

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

[38]  Cynthia Rudin,et al.  dame-flame: A Python Library Providing Fast Interpretable Matching for Causal Inference , 2021, ArXiv.

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

[40]  Robert L. Mccornack A comparison of three predictor selection techniques in multiple regression , 1970 .

[41]  D. Rubin,et al.  Causal Inference for Statistics, Social, and Biomedical Sciences: A General Method for Estimating Sampling Variances for Standard Estimators for Average Causal Effects , 2015 .

[42]  Elizabeth A Stuart,et al.  Matching methods for causal inference: A review and a look forward. , 2010, Statistical science : a review journal of the Institute of Mathematical Statistics.

[43]  Paul R. Rosenbaum,et al.  Matching for Balance, Pairing for Heterogeneity in an Observational Study of the Effectiveness of For-Profit and Not-For-Profit High Schools in Chile , 2014, 1404.3584.

[44]  Luke Keele,et al.  Optimal Multilevel Matching in Clustered Observational Studies: A Case Study of the Effectiveness of Private Schools Under a Large-Scale Voucher System , 2014, 1409.8597.

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

[46]  Cynthia Rudin,et al.  Adaptive Hyper-box Matching for Interpretable Individualized Treatment Effect Estimation , 2020, UAI.

[47]  J. Avorn,et al.  Variable selection for propensity score models. , 2006, American journal of epidemiology.

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

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

[50]  Sanjog Misra,et al.  Heterogeneous Treatment Effects and Optimal Targeting Policy Evaluation , 2018 .

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

[52]  Cynthia Rudin,et al.  MALTS: Matching After Learning to Stretch , 2018, J. Mach. Learn. Res..

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

[54]  Cynthia Rudin,et al.  Stop explaining black box machine learning models for high stakes decisions and use interpretable models instead , 2018, Nature Machine Intelligence.

[55]  S. Hohmann,et al.  Predicting NICU Admissions in Near-Term and Term Infants with Low Illness Acuity , 2020, Journal of Perinatology.

[56]  Cynthia Rudin,et al.  Interpretable Almost Matching Exactly With Instrumental Variables , 2019, UAI.

[57]  D. Rubin Matched Sampling for Causal Effects , 2006 .

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

[59]  J. Zagorsky Marriage and divorce’s impact on wealth , 2005 .

[60]  Md. Noor-E-Alam,et al.  Hypothesis Tests That Are Robust to Choice of Matching Method , 2018 .

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

[62]  M. Farrell Robust Inference on Average Treatment Effects with Possibly More Covariates than Observations , 2013, 1309.4686.

[63]  E. K. Adams,et al.  Neonatal health care costs related to smoking during pregnancy. , 2002, Health economics.

[64]  Sanjog Misra,et al.  Deep Neural Networks for Estimation and Inference: Application to Causal Effects and Other Semiparametric Estimands , 2018, Econometrica.

[65]  Anthony J. Kondracki,et al.  Low birthweight in term singletons mediates the association between maternal smoking intensity exposure status and immediate neonatal intensive care unit admission: the E-value assessment , 2020, BMC Pregnancy and Childbirth.