On Auto-regression Type Dynamic Mixed Models For Binary Panel Data

摘要

Binary panel data studies are common in many biomedical and socioeco-nomic research. In this set up, binary responses along with a set of multidimensional covariates are collected repeatedly over a small period of time from a large number of independent individuals. These repeated binary responses of an individual become stochastically correlated as well as they may be influenced by the individual's unobserved random effect. The efficient estimation of the effects of the covariates on the responses requires the use of the underlying correlation structure of the data. There is however no unique correlation structure for such binary panel data. For example, a correlation structure, based on a dynamic observations-driven model, may be quite different than the so-called latent process based correlation structures. Note that the observations-driven correlation structures are in general simpler than the latent process based correlation structures. In this paper, conditional on the individual random effect, we generalize Kanter's (1975, J. of Appl. Probab., 371-375) observations-driven binary dynamic stationary model to the non-stationary case. We also demonstrate how to estimate the parameters of this generalized dynamic binary mixed model.