Greetings,
I was hoping to get some advice on dealing with data that are overdispersed. I'm working with a sample of insurance claims data, modeling it as a function of the severity of the injury on which the claim was paid. In particular I'm interested in comparing the predictive performance of a simple model that can account for overdispersion, say a gamma regression, versus that of a linear regression. I'd appreciate any insights about comparing these models. As an aside, I was also wondering the effects that overdispersion exert vis-à-vis sample size. That is, for highly overdispersed data, what is the relationship between the width of a given confidence interval and sample size?
Thanks very much.
-Mark
|