Propensity Score Estimation Using Multilevel Classification and Regression Trees

• In Event: Poster Session 14
  In Poster Session: Advanced Issues in Multilevel Factor Analysis

Abstract

Logistic regressions have been used in propensity score analysis (PSA) for a long time without much development to draw casual inferences. As data mining technique CART becomes a promising alternative for logistic regressions when conducting PSA under single-level settings. No study extends CART into multilevel PSA. To fill in this gap, we compared the performance of employing multilevel CART and multilevel logits in multilevel PSA across various simulated conditions including different sample sizes, intra-class correlations, and relationships between treatment assignments (T) and pre-intervention covariates (X). Preliminary results indicated that multilevel CART outperformed multilevel logistic regressions in terms of balancing X and estimating treatment effects, especially when the relationship between T and X was non-additive and/or non-linear.

Authors

• Shuqiong Lin, Texas A&M University - College Station

• Wen Luo, Texas A&M University - College Station