B0561
Title: Flexible copula models for mixed binary-continuous data
Authors: Rosalba Radice - Cass Business School (United Kingdom) [presenting]
Giampiero Marra - University College London (United Kingdom)
Abstract: The focus is on regression models for associated mixed binary and continuous outcomes constructed using copulae. The approach entails specifying marginal regression models for the outcomes, and combining them via a copula function to form a joint model. Specifically, the framework allows for Gaussian and non Gaussian dependencies and for the mean, higher order moments and association parameters to be heterogeneous by incorporating flexible linear predictor structures. The utilization of penalized regression splines and Gaussian Markov random fields allows one to account for non-linear covariate effects and for geographic clustering. A ridge penalty avoids convergence failures, even when the parameters of a highly collinear variable are not fully identified. The theoretical background and software for straightforward implementation of this approach are provided. The approach is illustrated by fitting interpretable models of different complexity on different data-sets.
