Thanks for Professor Williams' answer!
When I use -suest ml, svy- after -ivprobit- and -est store-, the error message is "option svy not allowed, r(198);". I already did -svyset [pweight=wtper2], strata(stratawr) psu(psuwrx)-. I am using Stata 9.2.
Another question: how do you choose iweight in -ivprobit-? In my case, the survey data only provides one weight variable, which is recommended by the codebook to be used as "svyset [pweight=wtper2], strata(stratawr) psu(psuwrx)". Since there is no other weight variable, in -ivprobit-, I set [iweight=wtper2]. I am not quite sure if this is where the problem comes from.
Question about -suest-: From Stata 9.2 help file, "The within-model covariance matrices computed by suest are identical to those obtained by specifying a robust or cluster() option during estimation. suest, however, also estimates the between-model covariances of parameter estimates.". Since I am using -ivprobit-, does this mean there is any between-model covariances of parameter estimates? If not, then the ways to use -suest- and to specify robust or cluster() option during estimation are identical. right?
Last question about score: it seems that Stata 9.2 considers score as old. So commands in Stata 9.2, such as -ivprobit-, do not surpport score. I do not need to specify score in -ivprobit-. Correct? By the way, I am not familiar with the score option.
Thank you for your answers!
Liye
________________________________
From: [email protected] on behalf of Richard Williams
Sent: Wed 7/12/2006 8:44 AM
To: [email protected]
Subject: RE: st: RE: svy probit/logit iv
At 08:05 AM 7/12/2006, Richard Williams wrote:
>When you say it doesn't work, what exactly is the error you are
>getting? The following works fine for me (the nhanes2f data are
>already svyset with pweight = finalwgt):
>
>webuse nhanes2f
>ivprobit diabetes female (heartatk = rural) [iw=finalwgt]
>est store m1
>suest m1, svy
Incidentally, the reason I happen to know this is because I asked
Stata Tech support a while back why svy: did not work with
-hetprob-. Here was the response:
---------------------------------
> * Why doesn't hetprob work with svy:? Is it because it would be
> statistically incorrect to do so, or is it just a programming
> problem?
There is no statistical reason; we just haven't gotten around to adding -svy-
support in -hetprob-.
This is true for most of the standard Stata estimation commands that
are not commonly mentioned in the survey literature. As you could
imagine, even though getting -svy- support for an -ml- based
estimation command usually requires little extra programming, the
documentation implications are quite large. In time, we hope to have
all of these commands support the -svy- prefix.
That being said, there is a way to get survey estimation results out of
standard estimation commands that do not currently (but could) support the
-svy- prefix. Look into the -svy- option of -suest-. For example,
svyset psu [pw=pwgt]
hetprob y x1 x2 [iw=pwgt], het(z1 z2)
est store hetprob
suest hetprob, svy
---------------------------------
Incidentally, with slight modification, this procedure seems to work
in Stata 8.2 as well. You just have to add the score option to your
estimation command, e.g.
webuse nhanes2f
hetprob diabetes female [iw=finalwgt], het(black) score(x*)
est store m1
suest m1, svy
In both Stata 8.2 and 9.2, I'm guessing that the general rule is that
the estimation command needs to support both iweights and the score
option in order to use suest this way. It would not surprise me if
many user-written commands do not support these, even if they
potentially could.
*
* 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/
<<winmail.dat>>
