A Bayesian Approach to the Analysis of Local Average Treatment Effect for Missing and Non-normal Data in Causal...

Area of Science:

• Statistics
• Causal Inference
• Econometrics

Background:

• Selection bias is a significant challenge in observational and quasi-experimental studies.
• Non-normal data and missing values exacerbate issues related to selection bias.
• Existing methods often struggle to address these challenges concurrently.

Purpose of the Study:

• To present a novel Bayesian robust two-stage causal modeling approach.
• To introduce the ALMOND R package for implementing this method.
• To demonstrate the package's utility with empirical examples.

Main Methods:

• Developed a Bayesian robust two-stage causal model capable of handling selection bias, non-normal data, and missing data.
• Created the ALMOND R package to facilitate the application of this modeling technique.
• Utilized instrumental variables within the Bayesian framework.

Main Results:

• The Bayesian approach provides reliable parameter and standard error estimates even with missing data and outliers.
• The ALMOND package allows flexible application of the described causal modeling technique.
• Empirical examples illustrate the practical implementation and benefits.

Conclusions:

• The Bayesian robust two-stage causal modeling technique effectively addresses common data challenges in empirical research.
• The ALMOND R package offers a valuable tool for researchers in social, psychological, and behavioral sciences.
• This method enhances the reliability of causal inference in the presence of selection bias, non-normality, and missing data.