Notice: On April 23, 2014, Statalist moved from an email list to a forum, based at statalist.org.
From | Richard Williams <richardwilliams.ndu@gmail.com> |
To | "statalist@hsphsun2.harvard.edu" <statalist@hsphsun2.harvard.edu> |
Subject | Re: st: heteroskedasticity in ordered probit model |
Date | Sun, 23 Dec 2012 12:04:15 -0500 |
Ologit/oprobit models fare a bit better because there is more info in an ordinal variable. Nonetheless I agree that the appearance of hetero is often caused by a mis-specified model, e.g important variables are omitted. There are all sorts of issues with a hetero model (including the possibility of radically different interpretations of the results, which the previously mentioned SJ articles also discusses). So, if you can find a way to not use one, you will often be better off. Sent from my iPad On Dec 23, 2012, at 11:31 AM, "JVerkuilen (Gmail)" <jvverkuilen@gmail.com> wrote: > Rich Williams' good advice is apropos, but the heteroscedastic probit > model is kind of fragile and there's an inherent ill-conditioning to > it. You may well be running it too lean in terms of N. I say this > speaking as someone who has advocated for the interpretation of > heteroscedasticity terms in models as often being substantively > important, so if it's important to you then it may be worth modeling, > but often heteroscedasticity is "apparent", caused primarily by an > omitted variable, a poor choice of link function (i.e., maybe you > should have fit logit or even cauchit instead), or something else. > > You may also be better off simply using bootstrapping for your standard errors. > > Just some things to think about. > * > * For searches and help try: > * http://www.stata.com/help.cgi?search > * http://www.stata.com/support/faqs/resources/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/faqs/resources/statalist-faq/ * http://www.ats.ucla.edu/stat/stata/