Stata The Stata listserver
[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]

Re: st: svyregress and single psu

From   Stas Kolenikov <>
Subject   Re: st: svyregress and single psu
Date   Tue, 16 Nov 2004 13:26:19 -0500

My guess rather is that there are PSUs with all y's missing. That's OK
for -svyheckman-, as it treats them as valid observations, but
-svyregress- takes them out, and ends up with one PSU per some strata.

Of course there is also a worst case that you've found a bug, and
-svyheckman- should have complained about single PSU, too, but it did
not. My guess would be that the -svy- commands have a common engine
for the variance estimation, and it should provide this error message
for any -svy- command when it runs into a single PSU/stratum problem.


On Tue, 16 Nov 2004 16:28:53 +0000, Rafa De Hoyos <> wrote:
> Im estimating a Heckman (1979) selection model using survey data:
> svyheckman y x [if], sel(z)
> I then estimate a model without correcting for selectivity:
> svyregress y x [if]
> In the first estimation all the results are reported, however for the
> -svyregress- model, error (460) i.e. stratum with only one PSU detected, is
> reported.
> To identify the strata's with single PSU's I did the following:
> svydes y x
> As well as:
> svydes y x z
> In both cases Stata highlighted the strata with single PSU's. I can't see
> why -svyheckman- do not face the single PSU problem and carries-on and
> reports the estimations.
> I have two guesses. The first one is that while estimating the "first-stage"
> probit in -svyheckman- some strata's are being deleted and these might be
> the ones containing a single PSU, hence when the main equation is being
> estimated, the single PSU problem is no longer there. The second guess is
> that the single-PSU problem is overcomed by the pseudo-maximum likelihood
> method used in -svyheckman-.
> Any thoughts about this?
> Many thanks,
> Rafa

Stas Kolenikov
*   For searches and help try:

© Copyright 1996–2015 StataCorp LP   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index