After further experimentation, I've found that my
earlier problem with -swor- is due to the fact that
-sort- is not doing the same sort every time on my
data. Thus when -swor- sorts the data (or I sort with
-sort- before using -sample-), I get different draws
because the data is in a different order.
Specifically, consider the following:
use temp, clear
sort cluster
li
This produces different results each time. Although
the dataset is sorted by cluster each time, the order
within cluster is different.
Is this a general phenomenon, or is this somehow
particular to my data? This is an older dataset, with
variables originally created in version 6.
I was able to get around the problem by creating a
variable equal to the original observation number and
subsorting on that.
use temp, clear
gen counter=_n
sort cluster counter
li
Thanks,
Gabriel
--- Gabriel Demombynes <[email protected]> wrote:
> Oops, on closer consideration, I remembered that the
> swor and by problem was in the 2nd stage of my two
> 2-stage sampling routine, so I can leave the first
> swor statement (with no by) as is.
>
> For the second stage, I can replace
>
> swor 10, by(cluster)
>
> with
>
> by cluster: sample 10, count
>
> Thanks for the help! I'd still be interested to know
> how one might handle the general problem of random
> sampling with varying probabilities of selection.
> Isn't this a common issue for survey design?
>
> Gabriel
>
>
> --- n j cox <[email protected]> wrote:
>
> > -swor- was published in STB-59.
> > I would have sworn that you don't
> > now need it for your purpose,
> > as -sample- has since been modified
> > to support a -count- option.
> >
> > However, I cannot reproduce your problem.
> > I can get -swor- to select the same sample
> > again and again.
> >
> > . forval i = 1/10 {
> > 2. qui sysuse auto, clear
> > 3. set seed 28031952
> > 4. qui swor 1, gen(select) by(foreign)
> > 5. levels make, clean
> > 6. }
> > Buick Riviera VW Rabbit
> > Buick Riviera VW Rabbit
> > Buick Riviera VW Rabbit
> > Buick Riviera VW Rabbit
> > Buick Riviera VW Rabbit
> > Buick Riviera VW Rabbit
> > Buick Riviera VW Rabbit
> > Buick Riviera VW Rabbit
> > Buick Riviera VW Rabbit
> > Buick Riviera VW Rabbit
> >
> > Try as I can, the Rabbit keeps coming
> > out of the hat.
> >
> > Nick
> > [email protected]
> >
> > Gabriel Demombynes wrote
> >
> > I'm using the user-written command swor to draw
> > random
> > samples with
> > replacement. I've discovered that even when I've
> set
> > the seed, the
> > command does not always draw the same samples in
> > repeated runs when
> > using by. It does not have this problem when not
> > using
> > by.
> >
> > When I do the following, I get different results
> > each
> > time I run the
> > program:
> >
> > set seed 6662
> > use alldata, clear
> > swor 1, gen(select) by(sex)
> > li
> >
> > The problem seems to be particular to the
> variables
> > in
> > my dataset.
> > When I create a new group variable, e.g. assigning
> > group=1 for the
> > first half of the observations, and group=2 for
> the
> > second half, and
> > then run the analysis by(group) there's no
> problem;
> > I
> > can reproduce the same results on multiple draws.
> > But the problem
> > persists even when I create a new group variable
> > based
> > on a pre-
> > existing variable and use that, e.g.
> >
> > set seed 6662
> > use alldata, clear
> > gen group=1 if sex==1
> > replace group=2 if sex==2
> > swor 1, gen(select) by(group)
> > li
> >
> > I've looked at the swor ado file, but I can't
> figure
> > out what might be wrong. Any suggestions?
> > *
> > * 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/
> >
>
>
> __________________________________________________
> Do You Yahoo!?
> Tired of spam? Yahoo! Mail has the best spam
> protection around
> http://mail.yahoo.com
> *
> * 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/
>
__________________________________________________
Do You Yahoo!?
Tired of spam? Yahoo! Mail has the best spam protection around
http://mail.yahoo.com
*
* 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/
