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

From |
"James Hardin" <jhardin@stat.tamu.edu> |

To |
<statalist@hsphsun2.harvard.edu> |

Subject |
st: Re: panels' quasiscores after -xtgee- |

Date |
Tue, 17 Sep 2002 10:35:06 -0500 |

Joseph Coveney <jcoveney@bigplanet.com> wrote: > Along the way, I came across the following passage from the > _Stata Reference Manual_ (Release 7, Volume 4, Page 327) > under the entry for -score-, "Note that the scores for the > independent panels can be obtained by . . . -egen > ui=sum(uit), by(t)- . . ." > > Is that a typographical error? It seems as if it ought to > be -egen ui=sum(uit), by(i)-. I didn't see anything about > it in the online errata list. Yes. > A more general question: from what I've seen, users of > -proc genmod . . . repeated- nearly reflexively use score > tests in lieu of Wald tests, and I was led to believe that a > score test has somewhat better properties than a Wald test. > Yet Hardin & Hilbe don't go into much discussion or detail > about score tests, and Stata doesn't have it as a handy > option in -xtgee-. Is there a disadvantage to score tests > vis à vis Wald tests in the context of PA-GEE that I'm not > aware of? There are no disadvantages of which I am aware. The paper to which you referred outlines three different approaches to building test statistics for various hypotheses on the coefficients of the model. It does not offer critical comparisons of the three. In preparing the chapter to which you referred, I searched for but did not find any treatment in the literature specific to GEE models, but I will keep looking. The three tests are asymptotically equivalent. We can't make unrestricted statements of comparison of the tests; i.e., we can't say that the LM test has better properties than the Wald test. We can say, the LM test statistic more closely follows its prescribed distribution than does the Wald test in certain circumstances. An example is a null hypothesis on a single coefficient where the value of the null is "far" from the value found in the unrestricted model. Current practice indicates that the LR test is preferred wherever possible. There is no clear rule for choosing between the Wald and LM tests even in the case of simpler models like multiple regression. For example, see Berndt and Savin (1977) "Conflict among criteria for testing hypotheses in the multivariate linear regression model." Econometrica 45 pages 1263-1277. Best, James ---------------------------------------------------------------------- James W. Hardin, Ph.D., Lecturer jhardin@stat.tamu.edu Department of Statistics, Blocker 416G 979-845-3141 (phone) Texas A&M University Mail Stop-3143 979-845-3144 (fax) College Station, TX 77843-3143 http://stat.tamu.edu/~jhardin ---------------------------------------------------------------------- * * 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/

- Prev by Date:
**st: Re: Random Effects Probit** - Next by Date:
**Re: st: problems starting xstata** - Previous by thread:
**st: RE: repeated ANOVA** - Next by thread:
**st: compiler for likelihood routines?** - Index(es):

© Copyright 1996–2015 StataCorp LP | Terms of use | Privacy | Contact us | What's new | Site index |