Marginal mean weighting through stratification: a generalized method for evaluating multivalued and multiple...

Area of Science:

• Social Sciences
• Statistics
• Psychological Research

Background:

• Propensity score methods are widely used for covariate adjustment in nonexperimental studies.
• Existing methods like propensity score matching/stratification have limitations with multi-valued/multiple treatments.
• Inverse-probability-of-treatment weighting (IPTW) can be unstable with partial support or misspecified models.

Purpose of the Study:

• To introduce and demonstrate the marginal mean weighting through stratification (MMW-S) method.
• To provide a nonparametric solution for causal inference with multi-valued and multiple treatments.
• To illustrate the application of MMW-S in psychological and educational research.

Main Methods:

• MMW-S computes weights based on stratified propensity scores to balance pretreatment characteristics.
• It assumes unmeasured covariates do not confound treatment effects given observed covariates.
• Causal effects are estimated by comparing average potential outcomes within an analysis of variance framework on weighted data.

Main Results:

• MMW-S provides a viable, nonparametric approach for causal inference in complex treatment scenarios.
• The method is applicable to binary, ordinal, and nominal treatment variables.
• It can approximate randomized experiments, factorial designs, and randomized block designs.

Conclusions:

• MMW-S enhances the robustness and applicability of propensity score methods for causal inference.
• The method is demonstrated effectively using data on educational services for English language learners.
• MMW-S offers a valuable tool for researchers evaluating complex interventions in nonexperimental settings.