Re: st: sortpreserve option

[David Kantor <kantor.d@att.net>]

Disregard the prior message from me. It got sent prematurely.

At 06:29 PM 8/14/2013, you wrote:
> On Wed, Aug 14, 2013 at 5:15 PM, David Kantor <kantor.d@att.net> wrote:
> > At 03:14 PM 8/14/2013, Sergiy Radyakin wrote:
> >>
> >> What is the effect of inclusion of the -sortpreserve- option in the
> >> declaration of a program, which might change the number of
> >> observations? (add or remove)
> >
> >
> > My understanding is that it should have no effect on the number of
> > observations.
> Dear David, just to make sure we are talking about the same thing. The
> program _is expected_ to make changes to data by adding/removing
> observations. Can I still declare it -sortpreserve-?
> Specifically, what will happen to the newly added observations? will
> they end up in the end of the restored-order-dataset? in the beginning
> of it? (elsewhere if other possibilities exist?)

Interesting question. I don't know the answer; I never considered 
that, but I can imagine some possibilities. It may pay attention to 
the sorted-by attribute, or it may pay attention to the actual order. 
In the former case, you may ask, what happens if one of the sort 
variables is dropped -- or altered? For the latter possibility, other 
questions arise; where do the new observations go? And does it matter 
how they were created?
This is where, if I needed to know the answers, I would run some experiments.

I think you can always declare a program as sortpreserve. I don't 
think it will crash. But again, I'd experiment.


> > Its effect is to insure that when the program exits, the data are left in
> > the same sort order as before.
>
> Yes, that's by the book. But we now have more data, so how would Stata
> know where to put the new observations which it hasn't seen before?

Yes, I gave you the book answer, which you already knew. My own 
inclination is that sortpreserve is appropriate only when the set of 
observations is not altered by the program. But that is evading your question.
I'll be curious to hear about what you find.

> > That may or may not be desirable, depending on your needs.
> If there is reasonable/predictable/logical behavior of sortpreserve in
> this case I definitely like to know it. Whether it is desirable,
> depends on what it is.

I just meant that, based on my own programs (funcdep, in particular) 
and how they are used, there are times when either...
a, you don't want Stata to do the extra work of putting the data back 
into its original order, especially if the set is large, and if you 
are just going to do another sorting afterward, or you just don't 
need it in any particular order.
b, you actually want the dataset to be left in the order that the 
program imposed; it would be counterproductive to undo that sorting.


> Best, Sergiy Radyakin
>
> > It may take a little extra run-time.
> >
> > I have written one program where I made it an option -- whether 
> to run it as
> > sortpreserve. There was a little trick to make that happen, which I can
> > explain if you want.
>
> I envision this as:
> http://radyakin.org/statalist/2013081401/myprog.ado
> with any sort-destructive actions to be taken in the common part.
> Perhaps it can be more elegant, but this is not what I am looking for now.
>
> Thank you for your help
> Sergiy

Indeed, your example is like what I had in mind. I believe that it 
can be abbreviated a bit, though.

Best,
--David

*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/faqs/resources/statalist-faq/
*   http://www.ats.ucla.edu/stat/stata/