Using Propensity Scores for Causal Inference: Pitfalls and Tips.

Area of Science:

• Epidemiology
• Biostatistics
• Health Services Research

Background:

• Propensity score (PS) methods are increasingly utilized for causal inference in observational studies.
• Understanding the nuances of different PS approaches is crucial for effective application.

Purpose of the Study:

• To provide an overview of causal inference using observational data.
• To compare and contrast major PS-based methods (matching, inverse probability weighting) with traditional regression models.
• To highlight assumptions, decision-making processes, and common pitfalls in PS method application.

Main Methods:

• Overview of causal inference principles for observational data.
• Detailed explanation of propensity score matching and inverse probability weighting techniques.
• Comparative analysis of PS methods against multivariable outcome regression.

Main Results:

• Subtle differences in causal identification assumptions exist between methods.
• Key distinctions lie in statistical modeling assumptions and the target populations for effect estimation.
• PS methods offer alternative frameworks to conventional regression for causal inference.

Conclusions:

• Optimal selection and use of PS methods require understanding their specific assumptions and target populations.
• This work offers practical guidance for researchers applying PS methods in empirical studies.
• Clarifying method distinctions aids in choosing the most appropriate analytical approach for causal inference.