12-16-2013 07:58 AM
Is there an equivalent of Proc PLM or Proc Score that can be used with Proc NLMixed? I want to save the parameters from Proc NLMixed and use them at a later date without rerunning NLMixed.
12-16-2013 08:38 AM
I think the closest you could come is to append the additional observations with the dependent variable set to missing, and then use the PREDICT statement. I think the problem that will arise has to do with subjects not being in both datasets, and consequently a lack of EBE's for the variance for the new subjects.
Given that, probably the best "predictor" would be the marginal estimate obtained by plugging in the fixed effect estimators. This assumes that the random effects are normal[0,sigma2], and so would not change the point estimate. Interval estimates, well, I am stumped as I don't see how to include the random effect variance for the additional subjects.