Primary Submission Category: Matching, Weighting
High-Dimensional Matching with Genetic Algorithms
Authors: Hajoung Lee, Kwonsang Lee,
Presenting Author: Hajoung Lee*
Matching in observational studies is widely used to estimate causal effects by obtaining treated and control groups with similar covariate distributions. Traditional matching methods rely on distances between observations to form pairs. However, this process often faces challenges in high-dimensional and low-sample size settings due to the curse of dimensionality, where the concentration of distances makes it difficult to distinguish between observations. To address this issue, we propose a novel matching method using genetic algorithms, shifting the focus from individual-level to group-level distances. By optimizing the similarity of the high-dimensional joint distributions of covariates between treated and control groups, our method improves causal effect estimation. This approach has key advantages: (1) it avoids dimension reduction, preserving the full scope of high-dimensional information without additional modeling, and (2) it maintains transparency by not relying on outcomes, akin to traditional matching, and (3) it performs robustly in low-sample size settings, where traditional methods may struggle. Furthermore, our results show that the proposed method is competitive with existing approaches even in low-dimensional settings. Through extensive simulations and real data applications, we validate the performance and provide practical guidance for the method, highlighting its potential as a powerful tool for causal inference in both high- and low-dimensional scenarios.