topic Re: Cross Validation possible with Proc Mixed? in Statistical Procedures

Cross Validation possible with Proc Mixed?

Daisy2 — Tue, 11 Aug 2020 16:37:00 GMT

I'm using Proc Mixed to build a model (see code below) and I would like to split the data into training and validation so I can see how well the model works on predicting data it has not seen. I found that proc glm allows data to be split into different fractions, but I wonder how to use both sets of code? Does proc glm select just split the data and then I run proc mixed on the different fractions separately? Thanks for the help.

proc mixed data=ttnc_subsetsort PLOTS=RESIDUALPANEL(UNPACK) RATIO;
  class hyb ntrt pd loc pass rep;
  model tncV12log =hyb ntrt pd ntrt*hyb pd*hyb ntrt*pd / residual;
  random loc rep(loc) loc*ntrt loc*hyb ;
  lsmeans hyb ntrt pd hyb*ntrt / pdiff=all cl adjust=tukey;
  title1 'Fit TNC V12 Model with Factors';
  run;

Re: Cross Validation possible with Proc Mixed?

PaigeMiller — Tue, 11 Aug 2020 16:58:44 GMT

This will give you a great start on doing resampling (from @Rick_SAS )

https://blogs.sas.com/content/iml/tag/bootstrap-and-resampling

and señor @Ksharp wrote this which is for Logistic regression but ought to work for PROC MIXED as well with minor modifications.

https://communities.sas.com/t5/SAS-Programming/10-FOLD-CROSS-VALIDATION-amp-BOOTSTRAPPING/m-p/448485#M112822

Re: Cross Validation possible with Proc Mixed?

Daisy2 — Fri, 14 Aug 2020 21:09:12 GMT

THanks. I'll have to dig into these -I'm a new SAS user so not familiar with macros.