Re: NaN and missing values

[Mike Miller]

On Thu, 23 Sep 1999, John W. Eaton wrote:

> Overloading NaN to mean `missing data' seems wrong to me, because NaNs
> can be generated for other reasons too, and if you just see a NaN,
> there would be no way to tell if it came from a calculation or because
> it was intended to fill the location of a missting datum.

That's a fair question, but I wonder how important it turns out to be in
most cases.  As was pointed out in some of the messages that followed the
one above, NaN has the right kind of behavior, e.g., f(NaN) = NaN.  This
is how missing values behave in most stats packages.  In SAS it is
possible to use several different missing data types to encode, for
example, *why* the data are missing (death, refusal to participate, etc.)
This sort of thing could be handled in octave by having (as John
suggested) a second matrix to encode information about missingness.

So, can't we deal with everything using NaN and addtional codes stored in
another matrix?  I think maybe we can.  People who need more might than
that might be doing things that are not well-suited to an octave-style
computing environment and they might want to actually use SAS or similar
software.

Regards,

Mike



---------------------------------------------------------------------
Octave is freely available under the terms of the GNU GPL.  To ensure
that development continues, see www.che.wisc.edu/octave/giftform.html
Instructions for unsubscribing: www.che.wisc.edu/octave/archive.html
---------------------------------------------------------------------