[Rd] Problem with NAs using chisq.test() (PR#748)

Thomas Lumley thomas@biostat.washington.edu
Sun, 26 Nov 2000 13:13:55 -0800 (PST)

On Sun, 26 Nov 2000 kjetikj@astro.uio.no wrote:

> Full_Name: Kjetil Kjernsmo
> Now, it turned out that a fairly deep bug in my code occasionally produced
> NAs in the matrix, and that was the source of the confusion. Looking at
> the code, it is apparent that the test
>     if (any(x < 0) || any(is.na(x)))
>         stop("all entries of x must be nonnegative and finite")
> was designed to catch this, but it doesn't seem to, because the first
> any()-call is NA and so when using the operator ||, well, I bet you
> understand this better than I... :-)
> I figured, perhaps turn this around:
>     if (any(is.na(x)) || any(x < 0))
>         stop("all entries of x must be nonnegative and finite")
> That seems to work: 
> > chisq.test(t2)
> Error in chisq.test(t2) : all entries of x must be nonnegative and finite
> Now, I hope this was a true bug, not just one of my misunderstandings, and that
> this is a real fix too.... :-)

This is a real fix; it may be a real bug.

any(x<0) correctly returns NA with NAs in x, that is, it doesn't know
whether there are any negative entries because some are missing.

any(x<0) || any(is.na(x)) returns NA because of short-circuit
evaluation.  That is, we do sequential if()s on each element from left to
right. The first one fails because of NAs.  Reversing the expression is
more sensible because you would expect to rule out NAs before testing
for x<0.

However, it would also be reasonable for NA || B to evaluate B and return
TRUE or NA depending on the value.  This is what S-PLUS 5 does, so at
least it's an incompatibility. 

The R documentation says 
    The longer form [ie ||]evaluates left to right examining only the first
    element of each vector.  Evaluation proceeds only until the result
    is determined. 
which at least seems incomplete as NA | TRUE gives TRUE, but
NA || TRUE gives an error.


Thomas Lumley
Assistant Professor, Biostatistics
University of Washington, Seattle

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