[S] GAM models with lo() terms; interpretation.

John Maindonald (john.maindonald@anu.edu.au)
Tue, 19 May 1998 10:33:49 +1000 (EST)

A day ago I put up the message

> I'm using a gam model to regress logweight on lo(logheight), lo(age),
> and altitude, in a modest sized data set. My concern is that
> summary.gam() and summary.lm() give what seem inconsistent views of
> the output, especialy allowing for the fact that summary.lm is likely
> to exaggerate statistical significance.

As Trevor Hastie and Maria Durban pointed out, the answer is that
summary.lm() looks only at the linear component of lo(age), even
though it prints out the term as lo(age). On the other hand
summary.gam() looks at departures from linearity. The two outputs
examine distinct components of the fit, In my example, the linear
component of age had p=0.0025, with the non-linear component very
highly significant.

In a nutshell, I was fooled! I apologise for spreading confusion!
As often with S-PLUS there is a need to check with carefully
constructed examples or by examining the code any semantics for which
one cannot find explicit documentation. This may be necessary even
when there is explicit documentation.

John Maindonald email : john.maindonald@anu.edu.au
Statistical Consulting Unit, phone : (6249)3998
c/o CMA, SMS, fax : (6249)5549
Australian National University
Canberra ACT 0200
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