topic Re: proc genmod dispersion parameter in Statistical Procedures

proc genmod dispersion parameter

ayaaj1 — Tue, 15 Feb 2022 08:59:55 GMT

Hello,

I am using proc genmod with the negative binomial distribution and a repeated statement.

it appears that with a repeated statement, the dispersion parameter is not provided.

Is there any way to retrieve the dispersion parameter while using the repeated statement ?

Thank you in advance!

Aya

Re: proc genmod dispersion parameter

sbxkoenk — Tue, 15 Feb 2022 09:50:43 GMT

Hello Aya ,

Use the PRINTMLE option on the REPEATED statement.

PRINTMLE --> Displays an analysis of maximum likelihood parameter estimates table

There's a starting value estimated for the negative binomial dispersion parameter (as for all other parameters), by doing a maximum likelihood (ML) estimation of the model ignoring the REPEATED statement.

You can see that estimated starting value of the negative binomial dispersion parameter by using PRINTMLE.

After initial estimation , I think this NB dispersion parm is further treated as a constant!

Is that correct @StatDave ?

Thanks,

Koen

Re: proc genmod dispersion parameter

StatDave — Tue, 15 Feb 2022 16:50:27 GMT

Yes, the initial model estimated by maximum likelihood (displayable with the PRINTMLE option) estimates the negative binomial dispersion parameter and that is then held fixed during the GEE estimation phase.

Re: proc genmod dispersion parameter

ayaaj1 — Thu, 17 Feb 2022 09:52:37 GMT

Thank you very much for this clear explanation!