This is now a matter of taste to be
argued out with Patrick Royston,
the author of -uvis-, not myself.
I am copying this to him, as he
is not a member of Statalist.
That said, I think his decision
is correct, as it makes a distinction
between imputation and whatever else one may
decide is sensible, and it follows
general Stata philosophy of doing just
one thing properly. It's also closer
to the philosophy of -predict-, which
won't make a prediction when not instructed
to, even if there is a possible
non-missing prediction.
Otherwise said, the job of an imputation
program is to impute, not to remind
you of what you may happen to know.
Nick
[email protected]
> -----Original Message-----
> From: [email protected]
> [mailto:[email protected]]On Behalf Of Richard
> Williams
> Sent: 25 February 2005 01:00
> To: [email protected]
> Subject: RE: st: RE: Imputed Missing Values with Uvis
>
>
> At 11:36 PM 2/24/2005 +0000, Nick Cox wrote:
> >No doubt this is what is often
> >wanted. But the help for -uvis-
> >is explicit:
> >
> >Note that uvis will not impute observations for which
> >a value of a variable in xvarlist is
> >missing. Only complete cases within xvarlist are used.
>
> I still find that unclear and not what I would expect! I
> agree that it
> cannot and should not impute observations when one or more
> x's have missing
> values. But, it should not be imputing in the first place if
> y is not
> missing, it should just plug in the observed value of y. As
> it stands, the
> program can actually create more missing data than was there
> in the first
> place. I suspect this is an unintended design flaw; at least
> I cannot
> think of any reason why you would want it to behave this way.
*
* For searches and help try:
* http://www.stata.com/support/faqs/res/findit.html
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
