Comparison of subject-specific and population averaged models for count data from cluster-unit intervention trials