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

From |
"Stas Kolenikov" <skolenik@gmail.com> |

To |
statalist@hsphsun2.harvard.edu |

Subject |
Re: st: Negative Hausman |

Date |
Wed, 24 Sep 2008 15:46:09 -0500 |

On 9/24/08, I. HESSELINK <I.Hesselink@uvt.nl> wrote: > For my master thesis I am analyzing a sample of panel data (y = whether a company diversified in year t, x = set of company characteristics). Because of the binary dependent variable I use xtlogit, and now want to apply a Hausman test to see whether fixed or random effects models are appropriate. > > xtlogit y x, fe > est store fixed > xtlogit y x, re > est store random > hausman fixed random > I am not quite sure those models are directly comparable. Your econometrics teacher should have explained that in binary dependent variable models, only beta/sigma is identified. With fixed effects logit (aka conditional logit aka McFadden's model), you don't really talk about panel-level variance, so your total variance should probably thought of as _pi^2/6 (the variance of the logistic variable). In the random effects model, you explicitly allow your sigma to go up by the value of the panel-level variance u_i, and hence your coefficients get scaled by sqrt( (_pi^2/6 + Var[u])/(_pi^2/6) ). See this: http://www.citeulike.org/user/ctacmo/article/3057661 Even if those models are comparable, I don't think their comparison can have the same interpretation as in the linear models, that of correlatedness of u_i with regressors. In linear models, those are moment conditions, and linear models are all about moments. In binary outcomes models, it's more of the likelihood and conditioning, and I don't really know if there's much place for E[ux]=0 or !=0 in those. Even further, biostatisticians would argue that sigma in those models is a figment of econmetricians' imagination. We are talking about probabilities, and they don't have to have the utility maximization and binary choice foundation as econometricians like to think of. So much for sigmaless/sigmamore story... -- 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/

**References**:**st: Negative Hausman***From:*"I. HESSELINK" <I.Hesselink@uvt.nl>

- Prev by Date:
**Re: st: RE: Modify large data base without opening it** - Next by Date:
**RE: st: Find all subsets of variables** - Previous by thread:
**st: RE: Negative Hausman** - Next by thread:
**st: Questions related to -predict-, -adjust-, and predictive margins** - Index(es):

© Copyright 1996–2016 StataCorp LP | Terms of use | Privacy | Contact us | What's new | Site index |