i have a fitted a classical Poisson regression model to my 584 claim frequency data set for a period of 5 years (panel data)i have found out that predictors area,age and make of vehicle are siginificant (preferred model).Other predictors are age and gender of insured and cc of car. Now i want to fit a Poisson random effect and compare it to the preferred model fitted by classical Poisson regression. What can i include as my random effects?
It makes sense to use a variable as a random effect when you suspect that it may represent a source of hidden variance in your model and the data to which your model is fitted only includes a sample of the full range of this variable.
So, for example, if your data comes from a sample of geographic locations and you want your model to explain the influence of other variables, such as age and car model, across all locations, it makes sense to use location as a random effect.