R help - Aug 2008 - (Un-)intentional change in drop1() "Chisq" behaviour?

Dear List,
recently tried to reproduce the results of some custom model selection
function after updating R, which unfortunately failed. However, I
ultimately found the issue to be that testing with pchisq() in drop1()
seems to have changed. In the below example, earlier versions (e.g. R
2.4.1) produce a missing P-value for the variable x, while newer
versions (e.g. R 2.7.1) produce 0 (2.2e-16).

I would assume that the former is more appropriate, so I was just
curious if this is an intentional change.
Kind regards-
Lutz

y<-rbinom(100,1,0.5)
x<-rep(0,100)
m<-glm(y~x, family=binomial)
summary(m)
drop1(m, test="Chisq")
#################
R241> drop1(m, test="Chisq")
Single term deletions

Model:
y ~ x
       Df Deviance    AIC    LRT Pr(Chi)
<none>      138.59 140.59
x       0   138.59 140.59   0.00
Warning message:
NaNs produced in: pchisq(q, df, lower.tail, log.p)
##################
R271> drop1(m, test="Chisq")
Single term deletions

Model:
y ~ x
       Df Deviance    AIC    LRT   Pr(Chi)
<none>      137.99 139.99
x       0   137.99 139.99   0.00 < 2.2e-16 ***
---
Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05
'.' 0.1 ' ' 1
###################

The change is in pchisq:

2.4.1:
> pchisq(3, 0, lower=F)
[1] NaN
2.7.1:
> pchisq(3, 0, lower=F)
[1] 0

And that is correct according to the definition.

On Mon, 11 Aug 2008, Lutz Ph. Breitling wrote:
> Dear List,
> recently tried to reproduce the results of some custom model selection
> function after updating R, which unfortunately failed. However, I
> ultimately found the issue to be that testing with pchisq() in drop1()
> seems to have changed. In the below example, earlier versions (e.g. R
> 2.4.1) produce a missing P-value for the variable x,
Are you sure it was not NaN?
> while newer versions (e.g. R 2.7.1) produce 0 (2.2e-16).
>
> I would assume that the former is more appropriate, so I was just
> curious if this is an intentional change.
In this case the computed value ought to be zero, and
hence
> pchisq(0, 0, lower=F)
[1] 0

Given that the computed value will be unreliable, NA would be more 
appropriate, so will change.
> Kind regards-
> Lutz
>
> y<-rbinom(100,1,0.5)
> x<-rep(0,100)
> m<-glm(y~x, family=binomial)
> summary(m)
> drop1(m, test="Chisq")
> #################
> R241> drop1(m, test="Chisq")
> Single term deletions
>
> Model:
> y ~ x
>       Df Deviance    AIC    LRT Pr(Chi)
> <none>      138.59 140.59
> x       0   138.59 140.59   0.00
> Warning message:
> NaNs produced in: pchisq(q, df, lower.tail, log.p)
> ##################
> R271> drop1(m, test="Chisq")
> Single term deletions
>
> Model:
> y ~ x
>       Df Deviance    AIC    LRT   Pr(Chi)
> <none>      137.99 139.99
> x       0   137.99 139.99   0.00 < 2.2e-16 ***
> ---
> Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05
'.' 0.1 ' ' 1
> ###################
>
> ______________________________________________
> R-help at r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-help
> PLEASE do read the posting guide
http://www.R-project.org/posting-guide.html
> and provide commented, minimal, self-contained, reproducible code.
>
-- 
Brian D. Ripley,                  ripley at stats.ox.ac.uk
Professor of Applied Statistics,  http://www.stats.ox.ac.uk/~ripley/
University of Oxford,             Tel:  +44 1865 272861 (self)
1 South Parks Road,                     +44 1865 272866 (PA)
Oxford OX1 3TG, UK                Fax:  +44 1865 272595