[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]

From |
Stas Kolenikov <[email protected]> |

To |
[email protected] |

Subject |
Re: st: Pseudo R2 for XTprobit |

Date |
Thu, 7 Oct 2004 15:48:51 -0400 |

Richard Williams pointed to the FAQ written by Bill Gould. Of course I cannot disagree with the StataBoss has to say on this :). OK, you use it in your thesis, but what exactly is the function of that statistics? How exactly are you going to use it? Would you do any model selection based on it? Would you compare your R^2 to others published? (For the latter to make sense, you would need to make sure that you use the same methodology as previous authors. That way, you may end up reporting something like five different pseudo-R^2, according to the whole variety of those available.) If I never really used it in my work, I probably wouldn't bother that much reporting it (and figuring out the right way to do so, if there is one). Oh yes, I just capitalized FIT to show that it means fit... not a statistical significance, not a correct specification, not the strength of a certain effect, but just the closeness of fitted values to the data. Fitted values on the same range as the original dependent variables are zeroes and ones that you can predict with your model. So the goodness of fit measure to me in this context is something that compares the original 0s and 1s and whatever comes out of your model. But as I said, there is no real telling for what the probabilities of positive response (1) are, as long as you don't observe the individual effects. If your area is macroeconomics, then you would have all sorts of troubles due to autocorrelation in errors. You can deal with that with dynamic panel regression models in the linear case (all Arellano-Bond story and various other estimation procedures), but I've no idea if there is anything available for the limited dependent variable context. On Thu, 07 Oct 2004 19:52:34 +0200, Uwe Berberich <[email protected]> wrote: > For my thesis. It's about how different classification schemes (IMF, de > facto, natural algorithm by Reinhart/Rogoff (2004) affect empirical > findings on the economic performance of exchange rate regimes. Probit > regressions are used to test the impact of the nominal exchange rate > regime on the probability of banking and currency crises onsets > respectively. I was just curious that there is no FIT measure reported > in my tables so far. The few emprical studies on the issue mostly > feature Pseudo-R^2. I reported R^2 for regressios where I uesd OLS and > was wondering what the equivalent for probit regressions would be. As > said before, my search turned out to be rather unseccessful until I > described to Stata-list. So, do you think it might be OK to report no > FIT measure at all? > > uwe > > > > Stas Kolenikov wrote: > > >well the only measure I can think of as really a FIT measure is how > >many observations were classified appropriately to their 0/1 > >categories. With random effects, you don't even have a predicted > >probability -- at least marginally, unless you are willing to > >integrate out the random effects. > > > >Within GLM framework, there are some reasonable goodness of fit > >measures based on deviance and residuals of different kind. Again, > >that has an i.i.d. consideration in mind. > > > >What are you going to use this measure for? > > > >On Thu, 07 Oct 2004 19:09:22 +0200, Uwe Berberich <[email protected]> wrote: > > > > > >>Stas, > >> > >>thanks for your reply. Do you have any suggestions for better goodness > >>of fit measures in probit regressions on panel data? McFadden R2 or > >>Pseudo R2 were the ones suggested in various sources and empirical > >>studies but I'm open for whatever turns out to be useful and meaningful. > >>I'm just puzzled that *xtprobit* does not return any measure at all. Is > >>there any theoretical motivation why I could report probit results > >>without any goodness of fit measure? > >> > >>Thank you very much > >>uwe > >> > >> > > > > > > > > -- > Uwe Berberich > Hemmerleinstrasse 1 > 96050 Bamberg > > home: 0049.951.2084473 > mobile: 0049.179.9020004 > > * > * 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/ > -- Stas Kolenikov http://stas.kolenikov.name * * 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/

**References**:**st: Pseudo R2 for XTprobit***From:*Uwe Berberich <[email protected]>

**Re: st: Pseudo R2 for XTprobit***From:*Stas Kolenikov <[email protected]>

**Re: st: Pseudo R2 for XTprobit***From:*Uwe Berberich <[email protected]>

**Re: st: Pseudo R2 for XTprobit***From:*Stas Kolenikov <[email protected]>

**Re: st: Pseudo R2 for XTprobit***From:*Uwe Berberich <[email protected]>

- Prev by Date:
**Re: st: A continuous DV between 0 and 1** - Next by Date:
**st: Design-based truncated regression** - Previous by thread:
**Re: st: Pseudo R2 for XTprobit** - Next by thread:
**Re: st: Pseudo R2 for XTprobit** - Index(es):

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