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

From |
grsanta@unict.it |

To |
statalist@hsphsun2.harvard.edu |

Subject |
Re: st: RE: negative binomial |

Date |
Thu, 6 Feb 2003 16:14:39 +0100 |

Nick, many thanks for answering. I am not sure I understood what you said. I tell you what I have done: 1. I have run a poisson regression model where I estimated a count data dependent variable as a function of 13 variables calculated across 90 observations + a costant, using the "poisson" command. 2. by using the "poisgof" command I then evaluated the goodness of fit, which gives me a chi2 prob = 0.9. 3. I have also run a negative binomial regression model on the same data, by using the command "nbreg" as well as the command "nbreg [var], dispersion (constant)". In the first case, the alfa was not significant (chi2=0.5). In the second case, the alfa was also not significant (although the chi2 prob=0.25). 4.I have then use the command "summarize [var], detail" in order to have the mean and variance of the dependent variable. The result is that the difference between the two measures is about 0.40. 5. Following a procedure suggested in the FAQs STATA web page, I have then used the "nbvarg" command in order to estimate the theoretical probability for the poisson and negative binomial distributions as well as for the observed. But, if I have understood you correctly, the two theoretical distributions cannot be compared. QUESTIONS 1. Do you think I can adopt the poisson regression model given these results? 2. What do you mean by estimating the nbreg with or without covariates? 3. What do you mean by fitting the poisson with one parameter and fitting the negative binomial with two? Sorry to bother you, but I really need some help on this topic. Many thanks in advance Grazia Scrive Nick Cox <n.j.cox@durham.ac.uk>: > grsanta@unict.it > > > I am trying to run a regression model having as depedent > > variable a count > > variable. The mean of this variable is 0.344 and the > > variance 0.745. I have > > tried to fit a negative binomial model to the data, but the > > alfa test is not > > significant, while the goodness of fit of the poisson model > > is much better. > > However, when I used the nbvargr command to obtain the > > parameter estimates and > > the relative graph the negative binomial distribution seems > > to fit better the > > data. Am I missing something? > > I am not clear that you are comparing like with > like. > > -nbvargr- does univariate fits without > covariates, whereas -nbreg- (is that what you are > using?) allows you to fit with or without covariates. Are you > fitting with covariates? > > Setting that on one side, the tone of your report > does not seem that surprising, as the Poisson > is a limiting form of the negative binomial, and > -- for univariate fits, no covariates -- > the comparison is thus between fitting the Poisson > with one parameter and fitting the negative binomial > with two. I can imagine cases in which the fit > of the more general distribution is better, but > various criteria of goodness of fit appear > to contradict that. > > Nick > n.j.cox@durham.ac.uk > * > * 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/ > ------------------------------------------------- Università di Catania - C.E.A. Servizio di WebMail http://www.cea.unict.it * * 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/

**Follow-Ups**:**RE: st: RE: negative binomial***From:*"Nick Cox" <n.j.cox@durham.ac.uk>

**st: St: Chamberlain's PPA***From:*<chris@cmcintyre.info>

**References**:**st: RE: negative binomial***From:*"Nick Cox" <n.j.cox@durham.ac.uk>

- Prev by Date:
**st: RE: postfile problem** - Next by Date:
**RE: st: e(depvar) in xtabond** - Previous by thread:
**st: RE: negative binomial** - Next by thread:
**st: St: Chamberlain's PPA** - Index(es):

© Copyright 1996–2014 StataCorp LP | Terms of use | Privacy | Contact us | What's new | Site index |