Bayesian variable selection in sample selection models using spike-and-slab priors.

Area of Science:

• Statistics
• Econometrics
• Biostatistics

Background:

• Sample selection models address bias in data missing not at random.
• Current methods rely on expert knowledge for variable specification, risking errors.
• Exclusion restrictions are often imposed to prevent inferential issues.

Purpose of the Study:

• To propose novel Bayesian spike-and-slab priors for variable selection in sample selection models.
• To offer a scalable and robust methodology for bias correction.
• To compare the proposed method against adaptive LASSO and stepwise selection.

Main Methods:

• Development of two families of spike-and-slab priors.
• Construction of a Gibbs sampler with tractable conditionals.
• Bayesian variable selection for outcome and selection equations.

Main Results:

• The proposed Bayesian approach demonstrates effective variable selection.
• The methodology is scalable to high-dimensional data.
• Simulations and real-data applications show competitive performance.

Conclusions:

• Bayesian spike-and-slab priors provide a powerful tool for sample selection models.
• This approach enhances bias correction in missing data scenarios.
• The method offers a viable alternative to traditional techniques like adaptive LASSO.