Efficient and robust estimation for longitudinal mixed models for binary data

This paper proposes a longitudinal mixed model for binary data. The model extends the classical Poisson trick, in which a binomial regression is fitted by switching to a Poisson framework. A recent estimating equations method for generalized linear longitudinal mixed models, called GEEP, is used as a vehicle for fitting the conditional Poisson regressions, given a latent process of serial correlated Tweedie variables.

The regression parameters are estimated using a quasi-score method, whereas the dispersion and correlation parameters are estimated by use of bias-corrected Pearson-type estimating equations, using second moments only. Random effects are predicted by BLUPs. The method provides a computationally efficient and robust approach to the estimation of longitudinal clustered binary data and accommodates linear and non-linear models.

A simulation study is used for validation and finally the method is applied to some fishing gear selectivity data.