Predicted counts for negative binomial model

skavyasindhu — Thu, 28 May 2020 17:59:22 GMT

Hi everyone,

I am doing a negative binomial model and have a question predicted counts.

Outcome = #events/person-years

primary predictor - year (calender years from 2009 to 2018)

model statement:

model outcome = year age gender race/ offset=log_personyears dist=nb link=log;

I am trying to calculate expected counts per 1000 person years. How can I change the offset to predict 1000 person years rather than 1 person year?

Thanks.

Re: Predicted counts for negative binomial model

SteveDenham — Thu, 28 May 2020 18:12:31 GMT

When you calculate log_personyears in your DATA step, I assume you do something like:

log_personyears = log (personyears);

So if you wish the offset to be per 1000 personyears, change that calculation to

log_thoupersonyears = log(1000*personyers);

Let us know if that does what you need.

SteveDenham