I'd like to thank everyone for their responses regarding my multiple-NA
problem. The responses universally pointed to the well-known problem in
S-PLUS regarding confusing NaN's with NA's in certain situations.
The problem remains a bit spooky (despite some assertions to the contrary
;-), because
1. I had checked that possibility already, but neglected to mention it
(always check the obvious cases first),
2. is.nan() returned F for every component of said vector, and
3. The vector in question was not the result of any math operation, but
rather came from reading in a table.
Along those lines, Bill Dunlop may have hit upon a winner when he mentioned
in a subsequent email that there are also bugs when using import.data() with
databases that can cause NA's with different bit patterns to be imported. As
this data came in via that route (ODBC to Sybase), that's one likely cause.
On the other hand, ODBC to Excel to Sybase produced the same effect, so
maybe it's a bit subtler than that.
In any event, a binary copy of the offending vector has been sent off to
MathSoft (oops, Insight), and hopefully they can examine the bits and come
up with an answer.
Thanks for all the help.
David O Nelson, Ph.D. (daven@llnl.gov)
Lawrence Livermore National Laboratory
Box 808, L-441
Livermore CA 94551
ph: +1.925.423.8898
fax: +1.925.422.2282
|