The analysis of count data is important in agricultural science. If more than one observation per object exists then correlations of the observations must be taken into account. The correlations are considered differently by marginal or subject-specific models within the generalized linear model. The experimental design and the estimated model parameters based on two trials were used to simulate count data. Marginal as well as subject-specific models were included. The estimation methods for the model parameters were maximum likelihood, pseudo-likelihood, and generalized estimating equations. The models and estimation methods were evaluated by convergence properties, Bias, and realizing the nominal type one error of the statistical hypothesis testing. In comparison, the subject-specific models and maximum likelihood showed the best results depending on the predetermined parameters.