Rather,foraspecific segment,covariance is an increasing
function of the product of the expected values λji; λki. The rate of increase is not necessarily unrestricted;increases in the
expected numbers of any two crash types are offset by the degree of heterogeneity (the overdispersion parameter) present in the data. In addition, it eliminates the strict upperbound to the correlation that was observed for the other types of multivariate count models.