Message-ID: <x2n1iouavw.fsf@blueberry.kubism.ku.dk>
Date: 2000-08-07T19:30:43Z
From: Peter Dalgaard
Subject: predict.lm is broken in 1.1.0-patched (2000-August-7) (PR#626)
In-Reply-To: ripley@stats.ox.ac.uk's message of "Mon, 7 Aug 2000 20:43:38 +0200 (MET DST)"
ripley@stats.ox.ac.uk writes:
> predict.lm has been broken by recent changes to the patched branch.
>
> It fails for all singular fits. An example:
>
> library(MASS)
> data(quine)
> quine.hi <- aov(log(Days + 2.5) ~ .^4, quine)
> quine.nxt <- update(quine.hi, . ~ . - Eth:Sex:Age:Lrn)
> predict(quine.nxt)
> 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
> NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA
...
> I believe this is a lot worse than that which the changes were intended to
> fix. (Are we not supposed to be ultra-cautious about not making things
> worse in that branch?)
Sometimes the fix for one bug uncovers another. We have
> coef.aov
function (object, ...)
{
z <- object$coef
z[!is.na(z)]
}
> coef.lm
function (object, ...)
object$coefficients
and both of these cannot be right when predict.lm calls coef(object)
and object$coefficients contains NA's. Using object$coefficients
inside predict.lm fixes things, but I'm still not happy with the two
different conventions for coef...
--
O__ ---- Peter Dalgaard Blegdamsvej 3
c/ /'_ --- Dept. of Biostatistics 2200 Cph. N
(*) \(*) -- University of Copenhagen Denmark Ph: (+45) 35327918
~~~~~~~~~~ - (p.dalgaard@biostat.ku.dk) FAX: (+45) 35327907
-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-
r-devel mailing list -- Read http://www.ci.tuwien.ac.at/~hornik/R/R-FAQ.html
Send "info", "help", or "[un]subscribe"
(in the "body", not the subject !) To: r-devel-request@stat.math.ethz.ch
_._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._