Analysis of longitudinal binary data is often accomplished by using GEE methodology to estimate the marginal model parameters. Most of current goodness-of-fit tests for GEE models have been studied in parametric situations. In this article, we consider to develop an alternative assessment for GEE models utilizing nonparametric technique. The proposed test avoided the explosion of a large number of additional parameters and dependence on partition of covariate space. Even though exact expectation and variance of the proposed test statistic are analytically and computationally infeasible, approximated values based on bootstrap data are employed. The asymptotic distribution of the proposed test statistic in terms of a scaled chi-squared distribution, and comparison of the proposed test and the current methods with respect to power are discussed by simulation studies. In addition, the testing procedure is illustrated by a medical study from Koch et al. [12].
Relation:
International Journal of Intelligent Technology ; Applied Statistics 2, pp.77-88
