topic Re: How to use transformed data in proc mixed? in SAS Procedures

How to use transformed data in proc mixed?

Lantern — Sat, 19 Jul 2014 20:48:44 GMT

Hi,

I am relatively new to SAS. I used proc transreg to Box Cox transform my data, but I am at a loss for how to proceed from here. Do I need to output the transformed data and use that new data file in the proc mixed model statement, or can I incorporate the transformation statement into the proc mixed code? I've very sorry for the simple question, but I have searched various SAS/statistics help forums, Googled it, and I cannot find an answer. Details are appreciated.

Thanks so much!

Lan

Re: How to use transformed data in proc mixed?

SteveDenham — Tue, 22 Jul 2014 17:16:04 GMT

Using the transformed dependent data is fine. Remember that all lsmeans and their standard errors are on the transformed scale. If you happened to find a "convenient" transformation through Box-Cox, you may wish to consider using GLIMMIX, provided an appropriate distribution can be specified.

Steve Denham

Re: How to use transformed data in proc mixed?

lvm — Tue, 22 Jul 2014 20:11:36 GMT

One would typically just transform the response variable in a data step, and then use this in MIXED. Now here is something if you are more ambitious. If you are using MIXED, you probably have random effects in your model, or repeated measures. But TRANSREG does not use the random effects in determining the power coefficient for Box-Cox. Hans-Peter Piepho wrote a macro to determine lambda power coefficient for a mixed model. It uses profile likelihoods by calling MIXED multiple times. (Piepho. 2009. Agronomy Journal 101(4): 865-869). But someone new to SAS could struggle with this. I mostly point this out for the more experienced SAS users.

Re: How to use transformed data in proc mixed?

Lantern — Wed, 23 Jul 2014 01:29:20 GMT

Thanks! You're right, I do have random effects and repeated measures. The macro you mentioned sounds a bit over my head, but I will have a look at the paper and see if I can work it out. Thanks so much!