RE: st: RE: survey completion "flag"

[David Kantor <dkantor@jhu.edu>]

At 05:21 PM 2/18/2005 +0000, Nick Cox wrote:
> No. And in fact that offers a much cleaner
> solution to the
>
> missing if any missing
>
> problem.
>
> gen max = <safe value>
> foreach v of local varlist {
>         replace max = cond(`v' > max, `v', max)
> }
>
> Nick
> n.j.cox@durham.ac.uk
>
> Nick Winter
>
> > Ah yes, I'd forgotten that.
> >
> > So max(a, b) is NOT the same as cond(a>b,a,b).
>
> > At 05:01 PM 2/18/2005 +0000, you wrote:
> > >The second problem is well identified.
> > >Thanks.
> > >
> > >The first problem is no problem.
> > >
> > >max(42, .) is 42, so initialising to
> > >missing is safe.
> > >
> > >This perhaps surprising feature can
> > >be rationalised as follows. Given
> > >arguments of
> > >
> > >-1, 0, 1, 2.71828, 3.14159, 42, .
> > >
> > >would you really want a -max()- function
> > >to return missing? It depends on the problem,
> > >but it can be forced the other way:
> > >
> > >gen max = <safe value>
> > >foreach v of local varlist {
> > >         replace max = cond(`v' == ., ., max(`v', `max')) if !mi(max)
> > >}
> > >
> > >Nick
> > >n.j.cox@durham.ac.uk
I'd like to add to this by offering some conceptual tools.

The max of...
 -1, 0, 1, 2.71828, 3.14159, 42, .
(that is, the general interpretation, not necessarily the value of Stata's 
max function) depends on what you mean by "missing". If the missing value 
represents some unknown value, then the max is unknown as well. On the 
other hand, the missing value in that list may also be interpreted as 
"vacuous" -- there's really nothing there. That is the interpretation that 
Stata's max function takes. Similarly, the difference between,
 gen t = x + y + z
and
 egen u = rsum(x y z)

is that, in the former, missing values are taken as unknown values, and in 
the latter, they are vacuous. (Actually, they are taken as zeroes, but 
"vacuous" can be thought of as a universal identity element in operations.)

Unfortunately, Stata is full of these kinds of inconsistencies, which we 
must be aware of. My though is that it might have been useful to have 
distinct missing values to distinctly represent unknown and vacuous, and to 
have all functions and expression evaluations regard these values 
accordingly. Thus, for example, if we suppose that .v stands for vacuous, 
then I would want
 max( -1, 0, 1, 2.71828, 3.14159, 42, .v)
to be 42, but
 max( -1, 0, 1, 2.71828, 3.14159, 42, .)
to be sysmiss (.).

I hope some of you find this useful or interesting.
-- David

David Kantor
Institute for Policy Studies
Johns Hopkins University
dkantor@jhu.edu
410-516-5404
*
*   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/