A general averaging method for count data with overdispersion and/or excess zeros in biomedicine

Area of Science:

• Statistical modeling
• Econometrics
• Biostatistics

Background:

• Count data often exhibit overdispersion and excess zeros, posing challenges for standard statistical models.
• Existing methods for zero-inflated models may lack robustness or computational efficiency.

Purpose of the Study:

• To develop a novel optimal weighting estimation method for zero-inflated negative binomial models.
• To enhance estimation accuracy for count data with overdispersion and/or excess zeros.

Main Methods:

• Developed an optimal weighting estimation method based on cross-validation for zero-inflated models.
• Utilized k-fold cross-validation for optimal weight vector selection.
• Employed group observation deletion for computational efficiency, differing from single observation deletion methods.

Main Results:

• The proposed optimal weighting method demonstrates asymptotic optimality.
• Simulation studies and empirical applications show its superiority over information-based model selection and averaging methods.
• The method effectively handles count data with overdispersion and excess zeros.

Conclusions:

• The optimal weighting based on cross-validation is a powerful and computationally efficient tool for analyzing complex count data.
• This novel method provides a more reliable alternative to existing approaches for zero-inflated models.