Hello,

I am running negative binomial regression on panel data of counts (panel = county, time = date) and running into issues with postestimation.

After my regression:

xtnbreg cases i.exposure [covariates] i.calendar_week, exposure(county_population) irr fe

...I would like to predict incidence (cases per county population) over exposure. I use the command

margins i.exposure, predict(iru0)

However, the predicted margins are more than 10 times smaller than typical incidence rates in my sample (e.g. the margins command predicts a marginal incidence of 0.2 cases per 100,000 population, whereas the mean incidence in my sample is 8.2 cases per 100,000 population.)

Any suggestions as to what's going on would be much appreciated. Thank you!