Model selection in estimating equations

Area of Science:

• Statistics
• Biostatistics
• Regression Analysis

Background:

• Model selection is crucial in regression analysis.
• Existing model selection techniques for methods like generalized estimating equations (GEE) are limited.
• Estimating equations are widely used in statistical modeling.

Purpose of the Study:

• To propose a novel model selection criterion for estimating equation methods.
• To address the lack of well-studied model selection techniques for GEE.
• To minimize the expected predictive bias (EPB) in regression models.

Main Methods:

• Development of a new model selection criterion based on expected predictive bias (EPB).
• Utilizing bootstrap smoothed cross-validation (BCV) to estimate EPB.
• Simulation studies to assess the performance of the proposed criterion for overdispersed generalized linear models.

Main Results:

• The proposed bootstrap smoothed cross-validation (BCV) method provides a reliable estimate of expected predictive bias (EPB).
• The new model selection criterion demonstrates good performance in simulations for overdispersed generalized linear models.
• The method is successfully applied to a real-world dataset from ewe embryo development.

Conclusions:

• The proposed EPB minimization criterion offers a valuable tool for model selection in GEE and similar methods.
• The BCV estimation technique is effective for assessing model performance in regression analyses.
• This approach enhances the reliability of statistical modeling in biological and other scientific fields.