xtpoisson with exposure(varname): how to select the appropriate varname?

[Bruno Hoepers]

I'm working on a paper with count data and am using a panel Poisson regression model (xtpoisson).

I've avoided using the exposure(varname) option in my models so far and wonder whether this decision is correct.

Therefore,

a) When should I use the exposure(varname) option?

b) And if I have to use it, how can I know which varname to use for exposure?


Thank you

[Clyde Schechter]

Think of your Poisson model as modeling a rate. That rate is a fraction: its numerator is the number of events that occurred in the observation, and the denominator is some extent to which the entity in that observation was in a state where such an event could occur. So it might be motor vehicle crashes per 10,000 miles traveled, or defective products produced per month of manufacturing, or fabric tears per square meter of cloth, or something like that. The exposure variable is the one that tells you the denominator: the distance traveled, or the number of months of manufacturing, or the number of square meters of cloth.

If the extent of exposure is the same for all your observations, and if all you are trying to get from your analysis is incident rate ratios, then you don't need to use -exposure()-. This might happen in a study where, for example, every entity in the study is observed for exactly the same period of time (and assuming the outcome events per unit of time is the rate in question). But if different entities have different extents of exposure, then the use of the -exposure()- variable is required in order to get unbiased estimates.