Solved
Contributor
Posts: 45

# Can I compare the goodness of fit between a NLMIXED model and a MIXED model using Information criteria?

I need to compare between NLMIXED and MIXED, and I am not sure if I can use AIC, BIC, or just look the residuals, or even fit the linear model on the NLMIXED. I would appreciate any thoughts you may have.

Thank you

Marcio

Accepted Solutions
Solution
‎11-19-2013 10:38 AM
Valued Guide
Posts: 684

## Re: Can I compare the goodness of fit between a NLMIXED model and a MIXED model using Information criteria?

NLMIXED does maximum likelihood only (ML). MIXED does REML as the default; you can't compare likelihoods or AIC, etc., between ML and REML. So if you want to compare the output of the two procedures, you must use METHOD=ML statement option in MIXED.

Also, you should use TECH=NRRIDG for the optimization technique in NLMIXED in order to match the optimization method used in MIXED.

All Replies
Posts: 2,655

## Re: Can I compare the goodness of fit between a NLMIXED model and a MIXED model using Information criteria?

You can fit linear models in NLMIXED, and with proper definition, get information criteria that exactly match.

The hard part of what you are asking comes down to making sure that your optimization methods are the same, fixed effects similarly defined, and DV measures are the same.  What do you have in mind to compare between the two methods?

Steve Denham

Solution
‎11-19-2013 10:38 AM
Valued Guide
Posts: 684

## Re: Can I compare the goodness of fit between a NLMIXED model and a MIXED model using Information criteria?

NLMIXED does maximum likelihood only (ML). MIXED does REML as the default; you can't compare likelihoods or AIC, etc., between ML and REML. So if you want to compare the output of the two procedures, you must use METHOD=ML statement option in MIXED.

Also, you should use TECH=NRRIDG for the optimization technique in NLMIXED in order to match the optimization method used in MIXED.

Posts: 2,655

## Re: Can I compare the goodness of fit between a NLMIXED model and a MIXED model using Information criteria?

has set out what is needed when I said "proper definition".  Information criteria are better defined under ML, I think.

Steve Denham

🔒 This topic is solved and locked.