A Simulation-Based Causal Inference Approach Using Cross-Sectional Data

• In Event: Poster Session 15
  In Poster Session: Applied Statistical Analysis of Large-Scale Data

Abstract

When a comparison group is absent, evaluation researchers often have to choose alternative methods to estimating the impact of a program using the best available data. This study introduces a simulation-based approach to estimating the causal effect using the data from a four-year literacy intervention study. Cross-sectional data are available for two cohorts of students. The simulation procedure combines the cross-validation technique and a post-matching analytical approach to estimating the average treatment effect. This method is more robust than regression-based inference because the derived estimate is less dependent on the analytical model. The method is especially useful when matching doesn’t achieve satisfactory balance on the covariates, as it controls for biases induced by the covariates in post-matching analysis.

Author

• Ning Rui, Westat