[R-sig-ME] Localization of nonlinear mixed models

Jaakko Heinonen jaakko.heinonen at metla.fi
Wed Nov 4 11:03:49 CET 2009


Dear list,

My question is about the localization (or calibration) of nonlinear 
mixed models.

I have the estimates of the parameters of some hierarchical nonlinear 
mixed models that we’d like to use to predict new responses. I also have 
measurements that can be used to calibrate the models and my question is 
how to compute the conditional expected values of the random parameter 
given the new measurements.

One way could be to find the maximum of p1(u)*p2(y|u) using nlm 
procedure, where p1(u) is the density of random parameters u and p2(y|u) 
is the conditional density of the response y. But can this be done using 
nlme? In nlme the models can be formulated in a familiar and handy way, 
which would be a great advantage. I understand that if I give nlme the 
estimates of the fixed mean and variance parameters as initial values 
and prevent nlme from updating the initial values, nlme computes the 
conditional expectations (and their variances) I need. If this is true, 
how can I give the initial values and prevent updating? Or is there a 
better way to do the job?

Best regards
Jaakko Heinonen




More information about the R-sig-mixed-models mailing list