Fast forward selection for generalized estimating equations with a large number of predictor variables

Area of Science:

• Statistics
• Computational Biology
• Ecology

Background:

• Variable selection is crucial in statistical modeling, especially with numerous predictors.
• Existing methods can be computationally intensive, particularly with large datasets and interaction terms.
• Correlated response data presents unique challenges for model selection.

Purpose of the Study:

• To introduce a novel, computationally efficient variable selection criterion: the score information criterion (SIC).
• To evaluate the performance of SIC in terms of selection accuracy, prediction, and computational speed.
• To demonstrate the utility of SIC in ecological studies with complex interaction terms.

Main Methods:

• Developed the score information criterion (SIC) based on score statistics for correlated response data.
• Utilized forward selection algorithms for variable selection.
• Conducted simulation studies to compare SIC with existing criteria.
• Applied SIC to a real-world ecological dataset involving arthropod species traits and environmental covariates.

Main Results:

• SIC demonstrates faster computation times compared to existing criteria, especially for models with many predictor variables.
• SIC incorporates variable correlations into its quasi-likelihood, yielding improved selection and prediction properties.
• Simulation studies confirm the consistency and desirable performance of SIC.
• Application to arthropod data highlights SIC's effectiveness in handling numerous interaction terms.

Conclusions:

• The score information criterion (SIC) offers a computationally efficient and statistically sound approach to variable selection.
• SIC is particularly advantageous for large-scale analyses involving complex interactions and correlated data.
• SIC provides a valuable tool for ecological and biological research requiring robust variable selection.