st: xtile with survey weights

[Mark Orr <mo2259@columbia.edu>]

Hello all,

I'm using a national dataset for which the weights adjust for 
oversampling and some other factors.  My goal is to make deciles of a 
variable, say var_x, but with taking the weighted empirical distribution 
as the basis for making the deciles.  I've computed this three ways, but 
all with slightly different results.  The code for the three different 
decile computations were as follows:

1. 
xtile var_x [wt=my_weight], nq(10)

2.
tab var_x [wt=my_weight]
THEN COMPUTE THE DECILES BY HAND BASED ON THE WGTED EMPIRICAL CDF

3.
cumul var_x [wt=my_weight], gen(cdist)
gen var_x_decile = ceil(10 * cdist)

So, my method for comparing these was to compute the mean for each 
decile.  I get three different (but) similar means for each computation 
type.

Any thoughts.  I'm stumped.

Thanks very kindly. (i've checked the archives pretty well, but nothing 
that really explains what is going on here)

-Mark Orr

--
***********************************************
Mark G. Orr, PhD
Heilbrunn Dept. of Population and Family Health
Columbia University
60 Haven Ave., B-2
New York, NY 10032

Tele: 212-304-7823
Fax:  212-305-7024

www.columbia.edu/~mo2259


begin:vcard
fn:Mark G. Orr
n:Orr;Mark G.
org:Columbia University;Heilbrunn Department of Population and Family Health
adr;dom:;;60 Have Ave., B2;New York;NY;10032
email;internet:mo2259@columbia.edu
title:Ph.D.
tel;work:212-304-7823
tel;fax:212-305-7024
tel;home:718-414-5151
url:http://www.columbia.edu/~mo2259
version:2.1
end:vcard