An approximate-copula distribution for statistical modeling

Area of Science:

• Statistics
• Biostatistics
• Genetics

Background:

• Generalized estimating equations (GEE), generalized linear mixed models (GLMM), and copulas are used for correlated, non-normal grouped data.
• Parameter estimation remains a significant challenge in these statistical frameworks.

Purpose of the Study:

• To derive a novel class of probability density functions for improved parameter estimation.
• To demonstrate flexible modeling of longitudinal, non-Gaussian data.
• To showcase the utility in multivariate genome-wide association analysis.

Main Methods:

• Derivation of a new probability density function family allowing explicit moment and distribution calculations.
• Application of maximum likelihood estimation using derived score and observed information.
• Tri-variate genome-wide association analysis on UK-Biobank data (blood pressure, BMI).

Main Results:

• The new distributional family facilitates explicit calculation of moments, marginal, and conditional distributions.
• The proposed method effectively models longitudinal data with non-Gaussian distributions.
• Successful application in a genome-wide association study highlights computational scalability and modeling power.

Conclusions:

• The novel distributional family offers a robust solution for parameter estimation challenges in correlated, non-normal data.
• This approach provides a flexible and computationally scalable tool for analyzing complex longitudinal and genetic datasets.