topic Re: How can I run log normal distribution from proc genmod. There is no default option for log norma in SAS Programming

How can I run log normal distribution from proc genmod. There is no default option for log normal

Vitthal — Fri, 06 Oct 2023 04:33:12 GMT

I have been trying to run a GLM on my data. My data seems to be a good fit for log normal distribution. However, there is no default distribution of log normal in proc genmod. Please suggest how can I do so.

Re: How can I run log normal distribution from proc genmod. There is no default option for log norma

Ksharp — Fri, 06 Oct 2023 04:58:05 GMT

Try FMM procedure, @Rick_SAS 's blog mention it, or LOG(Y) firstly and model it by normal distribution.
Check Rick's blog:
https://blogs.sas.com/content/iml/2015/09/16/plot-distrib-exp.html

Re: How can I run log normal distribution from proc genmod. There is no default option for log norma

Vitthal — Mon, 09 Oct 2023 12:53:48 GMT

How would I calculate the relativities basis the estimates? It shouldnt be exp(Estimates)

Re: How can I run log normal distribution from proc genmod. There is no default option for log norma

Ksharp — Tue, 10 Oct 2023 11:23:07 GMT

Sorry. I have no clue. Maybe @StatDave @Rick_SAS could give you a hint.

Did you check PROC FMM ?

Re: How can I run log normal distribution from proc genmod. There is no default option for log norma

Rick_SAS — Tue, 10 Oct 2023 13:03:10 GMT

KSharp linked to an article that shows two different models. I assume you want the first model, which is Y = exp(a + b*X) + eps, where eps ~ N(0,sigma) iid.

title "Generalized Linear Model of Y with log link";
proc genmod data=MyData;
   model y = x / dist=normal link=log;
run;

If that is not what you want, please provide more information.

Your most recent question is

> How would I calculate the relativities basis the estimates? It shouldnt be exp(Estimates)

I don't understand the question. If the model is Y = exp(a + b*X) + eps, then the procedure gives you the parameter estimates. They do not need to be transformed. If you prefer an equivalent multiplicative model, you can use the predictor exp(a)*exp(b*x).

The article shows how to transform the predicted values by applying the inverse link function.