Bookmark and Share

Notice: On March 31, it was announced that Statalist is moving from an email list to a forum. The old list will shut down at the end of May, and its replacement, statalist.org is already up and running.


[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: st: biprobit gof test, partial observability, survey data (svy)


From   J Gonzalez <jgonzalez.1981@yahoo.com>
To   "statalist@hsphsun2.harvard.edu" <statalist@hsphsun2.harvard.edu>
Subject   Re: st: biprobit gof test, partial observability, survey data (svy)
Date   Sat, 8 Oct 2011 14:45:32 +0100 (BST)

Thank you for your answer.

Regarding your idea of running the goodness of fit tests on the model without the svy, I have not found support in the literature to do this. However, I found this book by Heeringa et al. (2010), where they say:

"If the logistic regression program for complex sample survey data in the chosen software system does not provide capabilities to generate summary goodness-of-fit measures, reestimate the model using the sampling weights in the system’s standard logistic regression program. The weighted estimates of parameters and predicted probabilities will be identical and serious lack of fit should be quantifiable even though the standard program tools do not correctly reflect the variances and covariances of the parameter estimates given the complex sample design". (Heeringa et al., 2010, pp. 244)

I really did not fully understand that recommendation, but now that you mention it, I am wondering if they are actually proposing just what you did in your answer (testing models without survey design adjustment). Do you think that is what they mean???
I am not sure. because they suggest to reestimate the model using "standard logistic regression program", but "using the sampling weights". So I am confused. If what they mean is running the model just with [pw=pwvar], that does not solve the problem because with such weights it is also not possible to run the gof tests either.

Any ideas?

JG

References:


CRC Press.Heeringa, Steven G., West, Brady T., & Berglund, Patricia A. (2010). Applied Survey Data Analysis. Boca Raton, FL: Taylor & Francis Group




----- Original Message -----
From: Patrik Morgetz <pmorgetz@gmail.com>
To: statalist@hsphsun2.harvard.edu
Cc: 
Sent: Thursday, 6 October 2011, 1:51
Subject: Re: st: biprobit gof test, partial observability, survey data (svy)

I've experienced similar difficulties, not with this biprobit model,
but with other models when I use svy. I mean, a lot of  postestimation
commands are not appropiate after svy.

For the sake of discussion, and for the experts to debate (I am not),
what if you fit the model and run the tests without the svy???, and
use svy just for the final results?? Does it make any sense at all?

I think it may give, at least, some indications of what you want to
test, clearly it is not a first best, but it may be better to nothing
(assuming there is really not suitable test in your context, which I'm
not sure).

Best!

Patrik



On 10/5/11, J Gonzalez <jgonzalez.1981@yahoo.com> wrote:
> Hello stata list members
>
> I am working on a bivariate probit model (biprobit) with partial
> observability that looks like this.
>
> y1 = x1 + x2 + x3 + u
> y2 = x1 + x2 + v
> But I only have available y=y1*y2 (hence, there is partial observability).
>
> And I am fitting the model with survey data (using svyset, with pweight and
> strata), so I run something like:
>
>
> svy:biprobit (y1po = x1 x2 x3) (y2po = x1 x2), partial
> where y1po=y2po=y=y1*y2
>
> However, I do not have clear how to test this model (gof test, normality
> assumption) and how to decide among competing specifications.
>
> I have been reading (Cameron & Trivedi, 2005; Maddala, 1983) and also the
> stata list archives, and althought I have indeed found several tests I am
> still not sure how to proceed because of the svy complication and the
> partial observability issue.
>
> I have found the likelihood ratio testof rho=0,  the biprobit command
> usually estimates, but according to the stata manual likelihood ratio tests
> is not appropriate with svy, so I cannot use it for my model (and actually
> the svy:biprobit does not produce it).
> I also found Murphy's score test (*), but according to the help file the
> test does not work when the partial option is used, hence, I neither can use
> this test.
> Similarly, the estat ic command for the information criteria is also not
> suitable after svy estimation.
>
> So, which tests may and should I use?
> How to test if rho=0?
> Which criteria may I use to decide among competing specifications?
>
>
> Thanks for your help.
>
> Jesus González
>
> * https://webspace.utexas.edu/rcc485/www/papers/murphycomment.pdf
> * https://webspace.utexas.edu/rcc485/www/code.html
>
>
> Cameron, A. C., & Trivedi, P. K. (2005). Microeconometrics. Methods and
> Applications. New York: Cambridge University Press.
> Maddala, G.S. (1983). Limited-dependent and qualitative variables in
> econometrics (Vol. 3): Cambridge Univ Press.
>
>
> *
> *   For searches and help try:
> *  http://www.stata.com/help.cgi?search
> *  http://www.stata.com/support/statalist/faq
> *  http://www.ats.ucla.edu/stat/stata/
>

-- 
Sent from my mobile device

*
*   For searches and help try:
*  http://www.stata.com/help.cgi?searchhttp://www.stata.com/support/statalist/faqhttp://www.ats.ucla.edu/stat/stata/


*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/


© Copyright 1996–2014 StataCorp LP   |   Terms of use   |   Privacy   |   Contact us   |   Site index