Sampling Weights in the Estimation of Causal Effects in Multilevel Observational Studies: A Monte Carlo Study

• In Event: Statistics in the Field: Innovative Applications and Investigations

Abstract

The use of Propensity Score methods in observational studies for the purpose of causal inference is well documented, as the use of sampling weights in multi-level analyses of probability sample survey. Literature focusing on merging sampling information and propensity score methods is still sparse.
A Monte-Carlo study was conducted investigating the robustness of Regression Analysis, Inverse Probability of Treatment Weighting and Marginal Mean Weighting through Stratification, in conjunction with sampling weights. A two-level designed was used, varying the proportion of units in the treatment condition within cluster, the level-two treatment effect and the relation of the treatment effect to sampling weights and proportion of units in the treatment condition.
Limitations of each method and recommendations for typical scenarios are provided.

Authors

• Tiago A. Caliço, University of Maryland

• Laura M. Stapleton, University of Maryland