topic PROC MIXED: Model Selection in Statistical Procedures

PROC MIXED: Model Selection

AnalytX — Tue, 08 Jan 2013 16:50:10 GMT

Dear All,

Is it possible to automatize the model selection (based on variable selection) in PROC MIXED (such as "selection" option in PROC REG for instance)?

Thnaking you in advance for your answer.

Best,

Re: PROC MIXED: Model Selection

AncaTilea — Tue, 08 Jan 2013 17:01:11 GMT

Hi,

I don't think there is a selection = .... option in the model statement of the PROC MIXED.

Below is a paper that provides some macro that would do model selection - but be cautious.

http://www2.sas.com/proceedings/forum2007/191-2007.pdf

And another discussion in this forum, related:

Best of luck,

Anca.

Re: PROC MIXED: Model Selection

SteveDenham — Tue, 08 Jan 2013 17:51:48 GMT

It is not possible without going the macro route. There is a reason for that. Please review the article by Peter Flom and David Cassell at http://www.nesug.org/proceedings/nesug07/sa/sa07.pdf , regarding the dangers of stepwise/forward/backward/all possible subset variable selection methods.

Now, there is GLMSELECT to work on selecting fixed effects using LASSO or LAR methods. Also, check the discussion referenced by regarding selection of covariance structures and ranking of models using information criteria.

Steve Denham

Re: PROC MIXED: Model Selection

AnalytX — Wed, 09 Jan 2013 08:29:49 GMT

Thanks a lot for your answers and advises. I will try you proposals.

Regards,