Re: st: Construct Null Datasets through Bootstrap Resampling

["Michael Blasnik" <michael.blasnik@verizon.net>]

Sure- here's a quick way to scramble the observations while keeping groups 
of variables together.  It will work with up to four groups.  It's not the 
most elegant code, but it works (I think) . The way to specify the groups is 
to enclose the variable name list in double quotes.

program define scramblegrp
version 9.2
args grp1 grp2 grp3 grp4
tempvar hold order rand
qui gen `hold'=.
gen long `order'=_n
foreach grp in "`grp1'"  "`grp2'"  "`grp3'" "`grp4'" {
qui gen `rand'=uniform()
 sort `rand'
   foreach var of local grp {
      qui replace `hold'=`var'
      qui replace `var'=`hold'[`order']
   }
drop `rand'
}
end

Here's a quick test using two groups:

sysuse auto
scramblegrp "price mpg" "weight length turn"

Michael Blasnik


----- Original Message ----- 
From: "Erik Ingelsson" <Erik.Ingelsson@pubcare.uu.se>
To: <statalist@hsphsun2.harvard.edu>
Sent: Friday, December 01, 2006 5:17 PM
Subject: Re: st: Construct Null Datasets through Bootstrap Resampling



> Thanks Michael,
>
> After discussing this with our senior statistician today, I will  probably 
> go with the replacement, since this is how they have done it  before in 
> SAS. However, do you think that you could easily explain for  me how to 
> keep groups together in the scramble code as well? Just if I  need to do 
> it that way on a later occasion?
>
> Best,
> Erik Ingelsson
>
>
> Quoting Michael Blasnik <michael.blasnik@verizon.net>:
>
>
> > One more difference between the approaches that you should recognize is
> > that bsample samples with replacement (a value from a given observation
> > can appear more than once) while the scramble program I wrote does not
> > -- it simply re-arranges the order.  I'm not sure which is preferred
> > for your application.
> >
> > If you end up wanting to sample without replacement (as scramble does)
> > but want to keep groups of variables together, a relatively modest
> > change to the scramble code would do the trick.
> >
> > Michael Blasnik
*
*   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/