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

From |
"Maarten Buis" <M.Buis@fsw.vu.nl> |

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

Subject |
st: RE: RE: Log Transformation and GLM |

Date |
Mon, 16 Jan 2006 12:05:24 +0100 |

> > Charles Goss wrote: > > I have tried several methods to analyze my zero-inflated data set (zinb, hurdle and glm). > Maarten Buis wrote: > Third, it seems you have been doing some heavy duty data snooping, and some member of the statistical police won't approve of that. You might also be interested in paragraph 7.5 of the Scott Long and Freese book "Regression models for categorical dependent variables using Stata" from Stata Press. This paragraph is about comparing different models dealing with zero inflated counts, and deciding which one to use. I have an old version of the book. I hear there is a new edition out which discusses even more of these models. Another approach I find useful is to choose before hand a number of models that you think are plausible. If the substantive conclusions from all these models are the same, than apparently that conclusion does not depend on the difference in assumptions between the different models, and you call the conclusion robust. If the substantive conclusions from the different models differ, than you call the conclusion not robust. Notice that this increases the burden of evidence, unlike "continue testing until you find a significant result", which wrongly decreases the burden of evidence. The increase in the burden of proof is justified, since conclusions drawn from a single model assume that you are absolutely sure that that model is correct. Since you are uncertain about which model is correct, you should use a more conservative criterion for drawing conclusions. If you like this reasoning than you might be interested in Edward E. Leamer (1985), "Sensitivity Analyses Would Help", Th e American Economic Review 75(3). HTH, Maarten ----------------------------------------- Maarten L. Buis Department of Social Research Methodology Vrije Universiteit Amsterdam Boelelaan 1081 1081 HV Amsterdam The Netherlands visiting adress: Buitenveldertselaan 3 (Metropolitan), room Z214 +31 20 5986715 http://home.fsw.vu.nl/m.buis/ ----------------------------------------- * * 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:
**Re: st: how to merge to datasets containing dates?** - Next by Date:
**st: Re: standard errors for 2SLS** - Previous by thread:
**st: panel data and heteroskedasticity** - Next by thread:
**st: Re: standard errors for 2SLS** - Index(es):

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