Unreliable Continuous Treatments in Propensity Score Analyses

• In Event: Division D Graduate Student in Progress Research Gala

Abstract

Propensity score analyses using continuous treatments are often comprised of multi-indicator composites and their composite reliability is unreported. Latent variables or factor scores are seldom used as alternatives. A Monte Carlo simulation study was conducted varying composite reliability, continuous treatment representation, variability of factor loadings, sample size, and number of treatment indicators to assess whether Average Treatment Effect estimates differed in their relative bias, root mean squared error, and power to detect the ATE. Preliminary results suggest higher composite reliability is associated with higher statistical power, but variance inflation in ATE estimates from IPW weighting currently limits the study’s generalizability.

Author

• Gail Fish, University of Florida