R help - Dec 2003 - p-value from chisq.test working strangely on 1.8.1

Hello everybody,

I'm seeing some strange behavior on R 1.8.1 on Intel/Linux compiled  
with gcc 3.2.2.  The p-value calculated from the chisq.test function is  
incorrect for some input values:


 > chisq.test(matrix(c(0, 1, 1, 12555), 2, 2), simulate.p.value=TRUE)

         Pearson's Chi-squared test with simulated p-value (based on 2000
         replicates)

data:  matrix(c(0, 1, 1, 12555), 2, 2)
X-squared = 1e-04, df = NA, p-value = 1

 > chisq.test(matrix(c(0, 1, 1, 12556), 2, 2), simulate.p.value=TRUE)
[...]
data:  matrix(c(0, 1, 1, 12556), 2, 2)
X-squared = 1e-04, df = NA, p-value = < 2.2e-16

 > chisq.test(matrix(c(0, 1, 1, 12557), 2, 2), simulate.p.value=TRUE)
[...]
data:  matrix(c(0, 1, 1, 12557), 2, 2)
X-squared = 1e-04, df = NA, p-value = 1


In these three calls to chisq.test, I'm varying the input matrix by  
only 1 observation, but the p-value changes by 16 orders of magnitude.   
This is reproducible on my system.  Please let me know if any other  
information would be useful.

chisq.test works properly for these inputs on Mac OS X 10.3.1 with R  
1.8.0.  I don't know if the problem is with Linux or 1.8.1.

This bug looks very similar to bug 4718, which was reported in R 1.8.0  
and fixed in R 1.8.1.  They may be related.
http://r-bugs.biostat.ku.dk/cgi-bin/R/Analyses-fixed?id=4718; 
user=guest;selectid=4718

Jeff

> Hello everybody,
>
> I'm seeing some strange behavior on R 1.8.1 on Intel/Linux compiled
> with gcc 3.2.2.  The p-value calculated from the chisq.test function is
> incorrect for some input values:
>
>
>  > chisq.test(matrix(c(0, 1, 1, 12555), 2, 2), simulate.p.value=TRUE)
>
>          Pearson's Chi-squared test with simulated p-value (based on
2000
>          replicates)
>
> data:  matrix(c(0, 1, 1, 12555), 2, 2)
> X-squared = 1e-04, df = NA, p-value = 1
>
>  > chisq.test(matrix(c(0, 1, 1, 12556), 2, 2), simulate.p.value=TRUE)
> [...]
> data:  matrix(c(0, 1, 1, 12556), 2, 2)
> X-squared = 1e-04, df = NA, p-value = < 2.2e-16
>
>  > chisq.test(matrix(c(0, 1, 1, 12557), 2, 2), simulate.p.value=TRUE)
> [...]
> data:  matrix(c(0, 1, 1, 12557), 2, 2)
> X-squared = 1e-04, df = NA, p-value = 1
>
this does not happen with R-1.8.1 and gcc 2.95.4 on Debian stable:

R> chisq.test(matrix(c(0, 1, 1, 12555), 2, 2),
simulate.p.value=TRUE)$p.value
[1] 1
R> chisq.test(matrix(c(0, 1, 1, 12556), 2, 2),
simulate.p.value=TRUE)$p.value
[1] 1
R> chisq.test(matrix(c(0, 1, 1, 12557), 2, 2),
simulate.p.value=TRUE)$p.value
[1] 1

neither with R-1.9.0 (unstable). Is this reproducible without using
`set.seed' on your system?

Best,

Torsten
>
> In these three calls to chisq.test, I'm varying the input matrix by
> only 1 observation, but the p-value changes by 16 orders of magnitude.
> This is reproducible on my system.  Please let me know if any other
> information would be useful.
>
> chisq.test works properly for these inputs on Mac OS X 10.3.1 with R
> 1.8.0.  I don't know if the problem is with Linux or 1.8.1.
>
> This bug looks very similar to bug 4718, which was reported in R 1.8.0
> and fixed in R 1.8.1.  They may be related.
> http://r-bugs.biostat.ku.dk/cgi-bin/R/Analyses-fixed?id=4718;
> user=guest;selectid=4718
>
> Jeff
>
> ______________________________________________
> R-help at stat.math.ethz.ch mailing list
> https://www.stat.math.ethz.ch/mailman/listinfo/r-help
>
>

Mostly for the record, it happens also on Mandrake 9.1 with R 1.8.1
installed from rpm.


>chisq.test(matrix(c(0, 1, 1, 12556), 2, 2), simulate.p.value=TRUE)
 
       Pearson's Chi-squared test with simulated p-value (based on 2000
        replicates)
data:  matrix(c(0, 1, 1, 12556), 2, 2)
X-squared = 1e-04, df = NA, p-value = 5e-04

Federico Calboli

-- 



================================
Federico C. F. Calboli

PLEASE NOTE NEW ADDRESS

Dipartimento di Biologia
Via Selmi 3
40126 Bologna
Italy

tel (+39) 051 209 4187
fax (+39) 051 251 208

f.calboli at ucl.ac.uk