[S] GAMS: residuals and spectra

Stephen Bremner (sbremner@sghms.ac.uk)
Fri, 13 Mar 1998 13:33:39 +0000 (GMT)

I am fitting GAM models with SPlus 4.0. I would
like to internally studentise the Pearson residuals
obtained and can see no inbuilt function for obtaining the
leverages. However, on issuing x <- proj(last.gam) the
matrix I obtain is not square as I would have expected and
hence y <- diag(x) has only as many dimensions as variables
in my model so I cannot obtain the leverages.
Any suggestions please?
Also, I would like to know if the Pearson residuals
obtained in Splus are different from those in SAS (they
appear to me to be so but I cannot see why). I'm
curious because on using z <- spec.pgram(pearson.resids)
and rescaling the ordinates z$spec by the transformation
10^(0.1*z$spec) to obtain a periodogram not on the dB
scale, the range of the periodogram is much smaller in
Splus than in SAS. I've also transformed the z$freq
variable by multiplying by 2*pi to obtain the angular
frequency. Not only do the ordinates of the periodogram
differ from SAS but also the positioning of the estimates
along the frequency scale. Any ideas what is happening?
Thanking you in advance,

Stephen A. Bremner
Room 6.25 Hunter Wing
Department of Public Health Sciences
St. George's Hospital Medical School
(University of London)
Cranmer Terrace

Telephone: +44 (0)181 725 0284 (Direct)
Facsimile: 3584

This message was distributed by s-news@wubios.wustl.edu. To unsubscribe
send e-mail to s-news-request@wubios.wustl.edu with the BODY of the
message: unsubscribe s-news