This paper demonstrates that the unobserved heterogeneity commonly ass
umed to be the source of overdispersion in count data models has predi
ctable implications for the probability structure of such mixture mode
ls. In particular, the common observation of excess zeros is a strict
implication of unobserved heterogeneity. This result has important imp
lications for using count model estimates for predicting certain inter
esting parameters. Test statistics to detect such heterogeneity-relate
d departures from the null model are proposed and applied in a health-
care utilization example, suggesting that a null Poisson model should
be rejected in favour of a mixed alternative. (C) 1997 by John Wiley &
Sons, Ltd.