topic Re: Akaike weights in GLIMMIX? in Statistical Procedures

Akaike weights in GLIMMIX?

FranAstorga — Tue, 22 Jul 2014 21:12:53 GMT

I'm running a GLIMMIX procedure, with laplace method. I get the AIC, but not the akaike w weights, which would let me to compare more clearly between models.... how could I get the value for each model??

Thanks!!

Re: Akaike weights in GLIMMIX?

lvm — Tue, 22 Jul 2014 21:41:05 GMT

You will have to be more specific. One can weight an individual model by a scaled version of its AIC, which is used in model averaging or in determining posterior probabilities of models being the best. But the weighting only occurs when you have several models fitted to the same data. There is no weighted AIC for a single model fit. These weights are only defined after you fit several different models to the data (the denominator is based on all of the models considered). If you want to weight different models after fitting, then you need to do this by hand, or in a data step.

Re: Akaike weights in GLIMMIX?

FranAstorga — Wed, 23 Jul 2014 02:25:58 GMT

Hi! Thanks for the response. I need exactly what you described: an specific weight for each model that was performed from the same data. I have 10 models from a set of data, and I need to select the best models. I'll use AIC, but also the weights of each one, as the probabilites of each one to be the best of all. How could I get those values? You mentioned "by hand" or "in a data step"... how are those processes? Thanks again!

Re: Akaike weights in GLIMMIX?

lvm — Wed, 23 Jul 2014 02:32:09 GMT

There are several sources on-line. Check out:

Lecture 17—Friday, February 10, 2006

for a detailed example of what you want. Or read:

http://www.turkupetcentre.net/reports/tpcmod0016.pdf

Good luck.