[R] gamm with correlation structure question

Kel Lam lamkelj at yahoo.com
Fri Jan 21 17:31:31 CET 2005


Dear group,

I am trying to use gamm() in mgcv.  Here's the
scenario.  The data frame has approx. 110K
observations with information on paediatric
readmission binary outcome (Y/N) and total volume of
their most responsible physician as the covariate. 
Since any physician can have multiple patients, the
data contains clustering structure which I am trying
to account for.  My original formula is

a <- gamm(readmission~s(volume,fx=F),
correlation=corCompSymm(0.1,form=~1|physician),
family=binomial)

During the first iteration, I received the warning "
Incompatible formulas for groups in "random" and
"correlation"".  Frankly I don't understand what it
means.  So I looked up the archive and saw one posted
by Prof. Thomas Lumley on lme as follows:

http://tolstoy.newcastle.edu.au/R/help/00a/0913.html

So I changed my formula as this:

a <- gamm(readmission~s(volume,fx=F),
random=list(physician=~1),family=binomial)

However I still get an error message saying "Lapack
routine dgesv : system is exactly singular".  

Can anyone suggest a better way to deal with my
problem.  Please let me know if you need further
details.  Thank you so much.

Regards,
Kel




More information about the R-help mailing list