Statalist


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

Re: st: RE: RE: Testing for heteroscedasticity in probit models


From   Austin Nichols <[email protected]>
To   [email protected]
Subject   Re: st: RE: RE: Testing for heteroscedasticity in probit models
Date   Mon, 9 Nov 2009 12:19:31 -0500

Gisele <[email protected]>:
I think you are not interested in tests of non-normality and unequal
variances for observed variables, since the errors in a probit are
unobservable. I'm not sure what test you have in mind, since you don't
give a page number in Davidson and MacKinnon, but I refer you to
section 16.6 (page 564) of that book on what to expect from such a
test--you are also testing the model specification and it is likely
that "tests in skedastic directions will be much less powerful than
tests in appropriate regression directions when it is the regression
function that is misspecified."  But there is a test of
heteroskedasticity in the -hetprob- output already.  You could run a
simulation to see how it performs in data like yours.

On Mon, Nov 9, 2009 at 11:42 AM, Nick Cox <[email protected]> wrote:
> G. E. P. Box.
> Non-normality and tests on variances.
> Biometrika 1953 40: 318-335; doi:10.1093/biomet/40.3-4.318
>
> Nick
> [email protected]
>
> Lachenbruch, Peter
>
> Generally, testing for heteroscedasticity is fraught with danger:  in
> particular something like Bartlett's test depends on normality so that a
> rejection may be as much a function of non-normality as it is of unequal
> variances.
>
> George Box once wrote (around 1952, but don't rely on my memory)
> "Testing for heteroscedasticity is like putting to sea in a rowboat to
> see if it's safe for the Queen Mary to sail."
>
> [email protected]
>
> I am working with a probit model and I suspect heteroscedasticity.
> Before using the command -hetprob-, I would like to perform the
> Davidson and MacKinnon test for heteroscedasticity in probit models
> (Reference: Estimation and inference in Econometrics, 1993, Davidson
> and MacKinnon). Does such a test already exist in Stata? If this is
> not the case, how can I compute it?
>

*
*   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 1996–2024 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index