Re: [S] predict(..., type="link", se.fit=T) bug.

John Maindonald (john.maindonald@anu.edu.au)
Sat, 27 Jun 1998 16:02:07 +1000 (EST)


The following is from S-PLUS3.4, release 1, under SunOS5.3. The same
behaviour is present in S-PLUS 4.0, release 3, on the PC.

> kyph.obj_gam(Kyphosis~Age,data=kyphosis,family=binomial)
> hat1_predict(kyph.obj,type="response",se.fit=T)

> names(hat1) # This is ok
[1] "fit" "se.fit" "residual.scale" "df"

> hat2_predict(kyph.obj,type="link",se.fit=T) # This is not ok
> names(hat2)
[1] "1" "2" "3" "4" "5" "6" "7" "8" "9" "10" "11" ......
.......
> hat2[1:4]
1 2 3 4
-1.422848 -0.949581 -1.112776 -1.798197

In other words, setting se.fit=T in
predict(..., type="link", se.fit=T)
has had no effect. I get the predictions, but not the SEs. If I
replace Age by lo(Age), or glm by gam, all is well.

Why did I have Age, rather than lo(Age), when I intend finally to
include the smoothing term? I was testing, and Age invites fewer
complications than lo(Age), does it not?
-----------------------------------------------------------------------
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