The Implications of Using Multiple Data Sources for Generalization Studies

• In Event: Data Analysis and Statistics

Abstract

The generalizability of a study’s results continues to be an area of interest among policymakers and researchers. Many generalization studies are based on small samples where the average sample to population size ratio is approximately 5%. The limited sample size raises challenges for both the precision and bias of estimates, leaving open the question of whether inferences can be improved with alternative statistical methods or additional data. This study assesses the implications of combining multiple data sources on generalization. We examine the performance of two propensity score-based estimators when various subsets of covariates are used. The goal is to determine whether additional data improves the common support between the sample and population and whether this implies improved generalization inferences.

Authors

• Wendy Chan, University of Pennsylvania
  Presenting Author

• Jimin Oh, University of Pennsylvania
  Non-Presenting Author

• Yeran Tong, University of Pennsylvania
  Non-Presenting Author

• Jiexuan Huang, University of Pennsylvania
  Non-Presenting Author

• Chen Li, University of Pennsylvania
  Non-Presenting Author