Offset() option in Poisson/negative binomial regression

[Giorgia Sulis]

Hello everyone,

I am working on count data and have the follow-up time in days. Do I need to log-transform the follow-up time before putting it in a Poisson or negative binomial model as an offset? To my knowledge, the offset() option performs the log-transformation automatically but, based on the output I get, I am a bit confused about this.

Thanks in advance for your help!

[Clyde Schechter]

No, you have it wrong. If you want to use follow-up time in the -offset()- option, you have to log transform it yourself. You can do that if you want.

But it is much simpler to instead put the follow-up time, without any transformation, into the -exposure()- option. Then Stata will handle it correctly by log-transforming it for you and then including the log-transformed version as an offset and constraining the coefficient of that to 1.

[Giorgia Sulis]

Very helpful! Thanks a lot!