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

From |
David Bell <dcbell@iupui.edu> |

To |
statalist@hsphsun2.harvard.edu |

Subject |
Re: st: Binomial regression |

Date |
Fri, 3 Aug 2007 17:35:45 -0400 |

If you use an identity link, you are assuming that, given a difference in the independent variables that produces a change from . 50 to .60 in the probability represented by your dependent variable, the same difference will also produce a change from .90 to 1.00 or from .95 to 1.05. I can't imagine many real world processes that would fit that model.

====================================

David C. Bell

Professor of Sociology

Indiana University Purdue University Indianapolis (IUPUI)

(317) 278-1336

====================================

On Aug 2, 2007, at 4:28 PM, Constantine Daskalakis wrote:

No argument about logistic regression. But that gives you odds ratios. What if you want risk differences instead?

There are several good reasons why we might want binomial regression (RD) instead of logistic regression (OR):

(i) Easier conceptual interpretation (RD vs. OR).

(ii) Causal effects interpretation (the RD can have it, but not the OR).

(iii) Effects of independent variables may be more additive on the original risk scale rather than on the log-odds scale (thus, logistic regression would need lots of interaction terms to get a good fit).

(iv) Because the reviewer/editor/boss fancies it.

By the way, there is no inherently "more appropriate" link function for a particular type of outcome. It's just that some are technically easier than others.

Best,

CD

On 8/2/2007 3:07 PM, Maarten buis wrote:

This may be a silly question, but why are you using the identity link?

It is not very appropriate for a binary dependent variable since it

will eventually lead to prediction outside the allowable range, and

apperently there are also problems with getting the model to converge.

Logit and probit links will converge in no time in Stata (probably also

in SAS), and are more appropriate for binary dependent variables.

Maarten

-----------------------------------------

Maarten L. Buis

Department of Social Research Methodology

Vrije Universiteit Amsterdam

Boelelaan 1081

1081 HV Amsterdam

The Netherlands

visiting address:

Buitenveldertselaan 3 (Metropolitan), room Z434

+31 20 5986715

http://home.fsw.vu.nl/m.buis/

-----------------------------------------

-- The documents accompanying this transmission may contain confidential health or business information. This information is intended for the use of the individual or entity named above. If you have received this information in error, please notify the sender immediately and arrange for the return or destruction of these documents. Constantine Daskalakis, ScD Assistant Professor, Thomas Jefferson University, Division of Biostatistics 1015 Chestnut St., Suite M100, Philadelphia, PA 19107 Tel: 215-955-5695 Fax: 215-503-3804 Email: c_daskalakis@mail.jci.tju.edu Webpage: http://www.jefferson.edu/clinpharm/biostatistics/ * * 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/

* * 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**:**Re: st: Binomial regression***From:*Maarten buis <maartenbuis@yahoo.co.uk>

**Re: st: Binomial regression***From:*Constantine Daskalakis <C_Daskalakis@mail.jci.tju.edu>

- Prev by Date:
**st: xtivreg** - Next by Date:
**Re: st: Binomial regression** - Previous by thread:
**RE: st: Binomial regression** - Next by thread:
**Re: st: Number Needed to be Treated (NNT)** - Index(es):

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