This is a likelihood approach for the regression analysis of overdispersed correlated count data with cluster varying covariates. The approach fits a multivariate negative binomial model by maximum likelihood and provides robust estimates of the regression coefficients.