Date:         Tue, 31 Oct 2006 01:00:08 -0800
Reply-To:     Albert-jan Roskam <fomcl@YAHOO.COM>
Sender:       "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From:         Albert-jan Roskam <fomcl@YAHOO.COM>
Subject:      Re: OR >999.999
Comments: To: Johanna LEPEULE <lepeule@VET-NANTES.FR>
In-Reply-To:  <6.0.0.22.0.20061031083638.01b39570@sun.vet-nantes.fr>
Content-Type: text/plain; charset=iso-8859-1

Hi Johanna,

The OR expresses the odds that an event occurs divided
by (--> ratio) the odds that this event does not
occur. If the odds are very high, for instance the
odds of people having two legs, your OR also becomes
very high. Let's say this likelihood is 0.999.
OR = p(have two legs) / p(have not two legs) =
OR = 0.999 / (1 - 0.999) = 999.

The phenomenon you're investigating is way too common
to meet the so-called rare disease assumption. A rule
of thumb is that the prevalence should be <10% for the
OR to be a good estimate of the relative risk. Mr
Bayes proved that long time ago.

HTH,
Albert-Jan

--- Johanna LEPEULE <lepeule@VET-NANTES.FR> wrote:

> Hi
>
> I use a logistic model with one variable and I
> obtain an OR and
> 95%IC >999.999 without error message.
> Does anyone have suggestions ?
>
> Thanks
> Johanna Lepeule
>
>
>
******************************************************
>  Ce message a ete verifie par le systeme anti-virus
>     de l'Ecole Nationale Veterinaire de NANTES
>
>
******************************************************
>




____________________________________________________________________________________
Everyone is raving about the all-new Yahoo! Mail
(http://advision.webevents.yahoo.com/mailbeta/)