Mutual Information Based Matching for Causal Inference with Observational Data

Authors: Lei Sun, Alexander G. Nikolaev

JMLR 2016 | Venue PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Computational experiments with mixed integer-programming formulations and four matching algorithms demonstrate the utility of MI based matching for causal inference studies.
Researcher Affiliation Academia Lei Sun EMAIL Department of Industrial and Systems Engineering University at Buffalo, Buffalo, NY 14260, USA Alexander G. Nikolaev EMAIL Department of Industrial and Systems Engineering University at Buffalo, 312 Bell Hall, Buffalo, NY 14260, USA Department of Computer Science and Information Systems University of Jyvaskyla, Jyvaskyla, FIN-40014, Finland
Pseudocode Yes Algorithm 1 MIP-based matching for MIM-Joint problem 1: Initialize the bin set {b : Cb + Tb 1} consisting of all the bins occupied by the units in T C; compute Tb and Cb; forbidden bin set BF = ....
Open Source Code No The paper does not contain any explicit statements or links indicating that source code for the described methodology is publicly available.
Open Datasets Yes This section presents the results of the computational experiments with synthetic and realworld (La Londe, 1986) data sets... M. Lichman. UCI machine learning repository, 2013. URL http://archive.ics.uci.edu/ml.
Dataset Splits Yes This synthetic data set with 25 covariates contains 100 treatment units and 10,000 control units... The target control group size was set equal to the treatment group size.
Hardware Specification Yes it took about 7 seconds on average to find solutions for the instances with 25 covariates on a desktop with an Intel Xeon E5-2420 1.9GHz CPU and 16G RAM.
Software Dependencies No The MIP models for MIM were solved using CPLEX. (No version number provided).
Experiment Setup Yes In optimizing the covariate balance, Sauppe et al. (2014) uniformly partitioned the range of the observed unit values in each covariate into 20 bins, and used Balance Optimization Subset Selection (BOSS) for control group selection... The target control group size was set equal to the treatment group size.