Re: [S] Change to NLME3 sooner rather than later?

Douglas Bates (
19 Jun 1998 07:34:08 -0500

"Rene Holst" <> writes:

> Will it be possible to use NLME3, if say, the within-subject covariances
> were known ?.

Probably yes. I would need to know a bit more detail about what you
wanted to do to be able to give a definite answer.

The within-subject covariances are described by a weight function
and/or a correlation function. For some classes of these functions
you can use a fixed parameter. For example, you can establish the
weights as proportional to a fixed power of the fitted values or you
could establish an AR(1)-type correlation with a fixed rho.

You can also add a class of weight functions or a class of correlation
functions. Adding such functions that do not depend on parameters to
be estimated is more straightforward than in NLME2. The interface for
such classes is more cleanly defined in NLME3.

Douglas Bates                  
Statistics Department                    608/262-2598
University of Wisconsin - Madison
This message was distributed by  To unsubscribe
send e-mail to with the BODY of the
message:  unsubscribe s-news