A weighted partial likelihood approach for zero-truncated models

Area of Science:

• Statistics
• Biostatistics
• Econometrics

Background:

• Zero-truncated data, where zero counts are unobservable, presents unique statistical challenges.
• Traditional maximum likelihood estimation for these data is computationally complex and requires specialized programming.
• Existing methods often lack accessibility in standard statistical software packages.

Purpose of the Study:

• To develop a computationally efficient and accessible method for modeling zero-truncated count data.
• To introduce a weighted partial likelihood approach as an alternative to maximum likelihood estimation.
• To extend this methodology for regression modeling and variable selection in zero-truncated data.

Main Methods:

• Developed a weighted partial likelihood estimating function inspired by the Rao-Blackwell theorem.
• Demonstrated the equivalence of this function to a weighted score function for standard count models.
• Extended the approach to incorporate regression and variable selection techniques.

Main Results:

• The weighted partial likelihood method provides accurate parameter estimates for zero-truncated binomial and Poisson data.
• The proposed method demonstrates efficiency comparable to maximum likelihood estimation.
• The approach is readily implementable using standard statistical software.

Conclusions:

• The weighted partial likelihood offers a practical and efficient solution for analyzing zero-truncated count data.
• This method enhances accessibility to advanced statistical modeling for researchers across disciplines.
• The extended regression and variable selection capabilities provide robust tools for data analysis.