In addition, an odds ratio is only approximately a relative risk when the
dependent variable is highly imbalanced (i.e., rare events).  Langche Zeng
and I wrote about this strange situation: despite the fact that the whole
medical community communicates in ORs, essentially no one is more
comfortable using them than relative risks, first differences, base
probabilities, etc.  Since most social scientists have avoided this
strange path, I don't see any reason we should start down it now.  Its
easy to compute any real quantity of interest from your multinomial logit
model, and so that's what I'd suggest you do.

Gary

P.S. more at http://gking.harvard.edu/projects/rareevents.shtml


On Fri, 5 May 2006, Omar Lizardo wrote:

> Bill,
>
> There was a fairly long discussion of this issue on the Stata listserv. As
> far as I could tell, the short answer is that the RRR is not the same as the
> OR, because the coefficients from a logistic regression are not the same
> (they are a special case) as the coefficients from a multinomial logit
> (however if you run a multinomial logit on a binary dependent variable
> OR=RRR).  But you can follow the thread for yourself from here:
>
> http://www.stata.com/statalist/archive/2005-06/msg00063.html
>
> Omar.
>
>
> ----- Original Message -----
> From: "William Ford, University of Chicago" <[log in to unmask]>
> To: <[log in to unmask]>
> Sent: Thursday, May 04, 2006 8:22 PM
> Subject: [POLMETH] Stata: mlogit & RRRs
>
>
>> I hope someone might be able to clear up what should be a
>> simple question, though I cannot find the simple answer.
>>
>> Stata's mlogit command reports logit coefficients by default.
>> By using the RRR option, Stata will supposedly report relative
>> risk ratios instead. But Stata's method of generating these
>> "relative risk ratios" is simply to exponentiate the
>> coefficients. Stata says, "rrr reports the estimated
>> coefficients transformed to relative risk ratios, i.e., exp(b)
>> rather than b."
>>
>> But exponentiating the coefficients is how one converts logit
>> coefficients to odds ratios. Since odds ratios and relative
>> risk ratios are not the same thing (though in some cases they
>> may be almost equal), Statas description of these
>> exponentiated coefficients as relative risk ratios makes no
>> sense to me. Shouldnt Stata just call these numbers odds
>> ratios? And shouldn't I just call them odds ratios too?
>>
>> Thanks for any guidance.
>>
>> Bill
>>
>> **********************************************************
>>             Political Methodology E-Mail List
>>        Editor: Karen Long Jusko <[log in to unmask]>
>> **********************************************************
>>        Send messages to [log in to unmask]
>>  To join the list, cancel your subscription, or modify
>>           your subscription settings visit:
>>
>>          http://polmeth.wustl.edu/polmeth.php
>>
>> **********************************************************
>>
>
> **********************************************************
>            Political Methodology E-Mail List
>       Editor: Karen Long Jusko <[log in to unmask]>
> **********************************************************
>       Send messages to [log in to unmask]
> To join the list, cancel your subscription, or modify
>          your subscription settings visit:
>
>         http://polmeth.wustl.edu/polmeth.php
>
> **********************************************************

**********************************************************
             Political Methodology E-Mail List
        Editor: Karen Long Jusko <[log in to unmask]>
**********************************************************
        Send messages to [log in to unmask]
  To join the list, cancel your subscription, or modify
           your subscription settings visit:

          http://polmeth.wustl.edu/polmeth.php

**********************************************************