Modelling ‘crime-proneness’ : a comparison of models for repeated count outcomes