The spike-and-slab lasso and scalable algorithm to accommodate multinomial outcomes in variable selection problems.

Area of Science:

• Bayesian statistics
• Statistical modeling
• Machine learning

Background:

• Spike-and-slab priors are effective for variable selection in Bayesian regression with numerous predictors.
• Existing spike-and-slab lasso methods primarily address continuous, count, or binary outcomes.
• Standard generalized linear models require modifications for multinomial outcomes, especially in Bayesian settings.

Purpose of the Study:

• To generalize the spike-and-slab lasso for variable selection with multinomial outcomes.
• To establish the theoretical foundation for this generalized model.
• To develop a computational algorithm for fitting the model.

Main Methods:

• Extension of the spike-and-slab lasso framework to accommodate multinomial outcome variables.
• Development of the theoretical underpinnings for Bayesian generalized linear models with multinomial outcomes.
• Implementation of an expectation-maximization algorithm for model parameter estimation.

Main Results:

• Successful generalization of the spike-and-slab lasso to handle multinomial outcomes.
• Provided a robust theoretical basis for the extended methodology.
• Developed a functional expectation-maximization algorithm for practical application.

Conclusions:

• This work represents the first generalization of the spike-and-slab lasso for multinomial outcomes.
• The developed methods offer a novel approach for variable selection in complex Bayesian models.
• The findings facilitate more sophisticated analysis of categorical data with a large number of predictors.