Re: st: Biprobit and clustering standard errors

[Lina C <linacs81@gmail.com>]

Dear Stas,
Thank you. I have around 90 clusters. The thing is that biprobit uses
the whole sum of "X" of both probits to comput the covariance matriz.
It cannot calculate the Chi2, but according to the manual, the
standard errors calculated using clustering are valid for 1
restriction (i.e., for the t-test of each coefficient)...

Thanks again,

2011/9/7 Stas Kolenikov <skolenik@gmail.com>:
> Yes. You will have (at least) two issues.
>
> 1. Your variance-covariance matrix, -vce-, will not be of full rank.
> Hence, you won't be able to estimate variances of certain combinations
> of parameters (there is no telling which combinations will be affected
> though).
>
> 2. If you have but few clusters, the assumptions of the asymptotic
> behavior may not be satisfied. The standard errors will suffer from
> small sample biases, and the test statistics (z-statistics or
> likelihood ratios) will have distributions different from their
> asymptotic targets (normal or chi-squared distributions,
> respectively).
>
> As a background, Stata (or any other statistical software) needs to
> compute the likelihood scores, i.e., the derivatives of the likelihood
> wrt to the parameters of the model. For variance estimation purposes,
> you would need to have as many scores (represented by temporary
> variables used in -robust- or -cluster- calculations) as you have
> parameters. So this is not the number of variables, really, but the
> number of parameters that matters.
>
> On Wed, Sep 7, 2011 at 5:20 AM, Lina C <linacs81@gmail.com> wrote:
>> Hello everybody.
>>
>> I'm running a biprobit clustering the standard errors as follows:
>>
>> biprobit ( y1 = y2 x ) ( y2 = z x), robust cluster(area)
>>
>> The "x" vector of regressors is much below the number of clusters
>> (areas), however Stata cannot calculate the chi_2. What I have noticed
>> is that STATA use the sum of the X in both equations as the total
>> number of regressors, and in this way the "x" of the first probit and
>> the "x" of the second probit sum up a number that is above the number
>> of clusters. Once I reduced the X to be, the sum in the first probit
>> and in the second probit, below the number of clusters, the chi2
>> appears..
>>
>> The problem is that I need to use more regressors..Is there a problem
>> if I rely on that estimation with the missing estimation of the chi2?
>>
>> Thank you.
>> Lina.
>> *
>> *   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/
>>
>
>
>
> --
> Stas Kolenikov, also found at http://stas.kolenikov.name
> Small print: I use this email account for mailing lists only.
>
> *
> *   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/
>

*
*   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/