Re: FO vs FOCE (Any test?)
From:Leonid Gibiansky
Subject:Re: [NMusers] FO vs FOCE (Any test?)
Date:Wed, 29 Jan 2003 10:35:49 -0500
Hi Atul,
Here is the summary of my experience: You can plot PRED vs PRED and IPRED vs IPRED to
compare models fitted by FO and FOCE methods. If the points are clustered around the x=y
diagonal then you can use any of the model. If not, prefer FOCE. If you need your model
for simulations, you should care about variances of the random effects and error terms:
they are more precisely estimated with FOCE. If both the predictions and the variances are similar
between methods, FO can be used. For covariate model development, difference in OF between
the models can be larger for FO. Therefore, after you complete your step-down procedure with
FO, you may test whether additional covariate can be excluded using FOCE fit. With exponential
error model, make sure to use INTERACTION option that does not add time but adds precision. In
general, if you can - then use FOCE; if you cannot (too long runs, does
not converge, etc.), use FO and test the final model with FOCE.
Good luck,
Leonid