[R-sig-ME] gamm4 model formulation clarification

[David Duffy]

On Tue, 3 Aug 2010, Mike Lawrence wrote:

> Oops. It seems that my logic in computing the CI for the difference
> scores in the "solution" I posted was seriously flawed (see:
> http://stats.stackexchange.com/questions/1169/ci-for-a-difference-based-on-independent-cis).
> So, predict.gam(..., se.fit=T) produces predicted values and and SEs
> for those values; any suggestions on how to compute the proper CI for
> a difference between such predicted values given the SEs?

I know very little about this, but I thought the smoothing procedure in 
mgcv etc tests the rank of the GCV problem, where your hypothesis is 
something like "rank of smooth of interaction = 1".  Then the AIC can be 
plotted for different fixed values of ranks/dfs.

Cheers, David Duffy.
-- 
| 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