[R-sig-ME] problem fitting a mixed model due to between subjectvariation

[David Duffy]

On Wed, 28 Jul 2010, R.Donders at ebh.umcn.nl wrote:

> ... These are not my own
> data and the people who provided me with the data gave me a choice:
> either I could use their data but then it should be impossible to trace
> back to any content matter [...]

Hmph.  So, should one expect autoregressive or order effects within each 
subject across values of pred?

>
> Indeed from a graphical inspection (not as nice as yours) I also
> concluded that there really should not be any effect of pred for
> individual subjects. I thought that conclusion was even valid for
> clusters of all sizes.

Yes.  I did this so there would be enough within-cluster data to 
look at nonlinear effects.  Specifically whether the curviness of loess 
fits was systematic.  I was thinking of the textbook examples of a 
completely deterministic relationship with zero linear correlation.

> dropping unused levels (still don't know how to do that in one step)

[drop=T]

Just 2c, David.


| David Duffy (MBBS PhD)                                         ,-_|\
| email: davidD at qimr.edu.au  ph: INT+61+7+3362-0217 fax: -0101  /     *
| Epidemiology Unit, Queensland Institute of Medical Research   \_,-._/
| 300 Herston Rd, Brisbane, Queensland 4029, Australia  GPG 4D0B994A v