Penalized likelihood estimation of a trivariate additive probit model

Area of Science:

• Biostatistics
• Econometrics
• Spatial Statistics

Background:

• Accurate modeling of multiple, correlated binary outcomes is crucial in various fields, including public health and economics.
• Existing methods often struggle with estimating dependence structures, especially when incorporating complex covariate effects and spatial correlations.
• Trivariate probit models are valuable but present estimation challenges, particularly for correlation coefficients.

Purpose of the Study:

• To propose a novel penalized likelihood method for estimating trivariate probit models.
• To effectively handle diverse covariate effects (linear, nonlinear, random, spatial) and error correlations.
• To improve the accurate estimation of correlation coefficients that define conditional dependence.

Main Methods:

• Utilized a penalized likelihood framework for parameter estimation.
• Employed a trust region algorithm with automatic multiple smoothing parameter selection.
• Developed the `SemiParTRIV()` function in R for accessible numerical computation.

Main Results:

• The proposed method successfully estimates trivariate probit models with complex covariate structures.
• Demonstrated improved accuracy in estimating correlation coefficients, crucial for understanding response dependencies.
• The R package provides a practical tool for implementing the advanced statistical methodology.

Conclusions:

• The penalized likelihood approach offers a robust solution for estimating trivariate probit models.
• This method enhances the analysis of jointly occurring binary outcomes, considering intricate covariate and correlation patterns.
• The study provides a valuable, computationally efficient tool for researchers analyzing complex binary data.