Bayesian variable selection for logistic regression with a differentially misclassified binary covariate.

Area of Science:

• Statistics
• Biostatistics
• Machine Learning

Background:

• Variable selection is crucial in statistical modeling, especially with complex data structures.
• Misclassified predictor variables can introduce bias and reduce model accuracy.
• Bayesian methods offer a robust framework for handling uncertainty in variable selection.

Purpose of the Study:

• To develop a Bayesian variable selection approach for statistical models incorporating a misclassified binary predictor.
• To define and integrate models for the latent predictor, its prevalence, and classifier accuracy (sensitivity and specificity).
• To optimize model performance using the developed selection method.

Main Methods:

• A Bayesian framework was employed for variable selection.
• The approach models the outcome, predictor prevalence, and classifier performance (sensitivity/specificity).
• Gibbs sampling with binary indicator variables was used for variable selection, identifying the highest posterior probability model.

Main Results:

• The developed Bayesian variable selection procedure was demonstrated through simulation studies.
• The method was applied to two real-world datasets to optimize model performance.
• The highest posterior probability model was successfully identified given the data.

Conclusions:

• The proposed Bayesian variable selection method effectively handles misclassified binary predictors.
• The approach enhances statistical model performance by selecting optimal variables.
• This method provides a valuable tool for researchers dealing with measurement error in predictor variables.