Converting monthly Incidence rate ratios from negative binomial regression

[Omar Okasha]

Dear All

I have rather basic statistical question. I have a monthly data with a count outcome (n), for which I fitted the followIng model:

nbreg n var1 var2, offset(logcohort) irr

The model is controlled for sex (var1), and indicator for certain calendar months to control for seasonality (var2).

My question: this model estimates monthly-based incidence rate ratios, which are not so informative given the slight changes reported. To make them more informative, I would like to report the annual incidence rate ratios, without having to convert my data from monthly to annual by collapsing it, not to lose the ability to adjust for monthly seasonal patterns. I am not sure as it sounds oversimplified, but should I just multiply 1-IRR by 12 to obtain the annual change?

Many thanks
Omar

[Omar Okasha]

In case someone is interested:

I consulted with a statistician and the answer is I should not multiply by 12 if the estimates are derived form monthly data, only I could do that with emphasis on that step being mere approximation which might not be entirely accurate.

Note: this is probably my third post on Statalist that goes completely unnoticed, which I did not experience before since I started using the forum. I usually get a response even if I posted a silly question; naturally the response in that case is that my question is silly, which I would appreciate much more than nothing.

[Clyde Schechter]

There was nothing silly about this post. There are many reasons why a post can fail to draw any responses.

In this case, for my part, I had looked at it, and pondered the question. While I recognized that multiplying by 12 was not correct, I couldn't figure out an approach that would work. But since I wasn't sure that there is no way to do this that I just don't happen to know, I didn't want to post a nihilistic response. I suspect others also considered the question, and coming up with nothing to suggest, passed it over.

[Carlo Lazzaro]

Omar:
I share Clyde's thoughts altogether.
I read with interest your query (which is far from being silly) and spotted an old Stata thread on a similar topic (I share the previous expressed belief that multiplying by 12 is notb the way to go):
http://www.stata.com/statalist/archi.../msg00413.html