Understanding Causal Distributional and Subgroup Effects With the Instrumental Propensity Score

Area of Science:

• Biostatistics
• Econometrics
• Epidemiology

Background:

• Observational studies face challenges with measured and unmeasured confounding.
• Existing methods for treatment effect evaluation may be limited in flexibility.

Purpose of the Study:

• To develop a unified and flexible approach for instrumental variable analysis in observational studies.
• To evaluate treatment effects while controlling for unmeasured confounders.

Main Methods:

• Developed an instrumental propensity score (IPS) conditional on baseline variables.
• Incorporated IPS into various methods: matching, regression, subclassification, and weighting.
• Utilized semiparametric density ratio models and empirical likelihood for outcome effect evaluations.

Main Results:

• The instrumental propensity score approach allows flexible outcome effect evaluations beyond standard 2-stage least squares.
• The method successfully evaluated distributional and subgroup treatment effects.
• Simulation studies confirmed the method's efficacy.

Conclusions:

• The proposed instrumental propensity score method offers a robust way to handle confounding in observational research.
• Applied to education and wages, the method revealed significant positive wage effects of Catholic school attendance for a specific subgroup.