Re: How to calculate OR from estimated coefficients of a multinomial log regression?
- From: Bruce Weaver <bweaver@xxxxxxxxxxxx>
- Date: Tue, 2 Sep 2008 11:20:36 -0700 (PDT)
On Sep 2, 2:15 pm, Bruce Weaver <bwea...@xxxxxxxxxxxx> wrote:
On Aug 30, 11:20 pm, Bruce Weaver <bwea...@xxxxxxxxxxxx> wrote:
gothamjac...@xxxxxxxxx wrote:
Hi all
I am a total novice with regard to more complex statistics.
I was wondering if there is a way to calculate adjusted OR and their
95%CI from the estimated coefficients of a multinomial logistic
regression (by use of STATA). There's a publication where this has
been
done (Lancet. 2004 Sep 11-17;364(9438):937-52). However, this study
was a case-control study.
Preferring OR over RRR for interpretation, I would like to do the same
for a retrospective cohort study for which I have run a multinomial
log regression model (since I have a multilevel dependant variable).
But I am not sure if this is possible and correct to do at all,
respectively.
Thanks for any help!
I'm not very well versed in Stata, but I suspect you're using themlogitprocedure, right? Here is its help file:
http://www.stata.com/help.cgi?mlogit
And here are some examples of its use:
http://www.ats.ucla.edu/stat/stata/examples/alr/alrstata8.htm
According to the help file, "mlogit, rr" generates a report of
"relative-risk ratios". But the "relative-risk ratios" in those
examples are equal to Exp(b) from the preceding output. That
seems a bit odd to me, because Exp(b) in a logistic regression is
usually described as giving the odds ratio.
This has me wondering if you're running into a terminological
issue here. I know that some epidemiology books when discussing
case-control studies give the formula for the odds ratio
((ad)/(bc)), but call the result a relative risk (presumably
because in a case-control study, the OR provides a good estimate
of the RR). Could it be that the Stata documentation is doing the
same thing?
HTH.
Pardon me for responding to my own post, but I just wanted to follow
up on the possibility that the RRR ("relative-risk reduction) in Stata
is actually the odds ratio. Here is another example from the UCLA
website, this time multinomial logistic regression done with the
NOMREG procedure in SPSS:
http://www.ats.ucla.edu/stat/spss/output/mlogit.htm
Notice point S in the annotated output: "Exp(B) - These are the odds
ratios for the predictors."
Notice too that the data file is available (in SPSS format) at the top
of the page. I downloaded it, and saved it as a Stata data set. Then
I opened it with Stata, and ran these two commands:
mlogit ice_cream video puzzle female, baseoutcome(2)
mlogit, rrr
The values in the RRR (relative-risk reduction) column of the Stata
output match exactly the Exp(B) values in the SPSS output, which, you
will recall, are called odds ratios in the annotated output.
CORRECTION: RRR = relative-risk ratio, not relative risk reduction.
--
Bruce Weaver
bweaver@xxxxxxxxxxxx
www.angelfire.com/wv/bwhomedir
"When all else fails, RTFM."
.
- Prev by Date: Re: detecting change in datasets
- Next by Date: random permutation in R
- Previous by thread: Re: How to calculate OR from estimated coefficients of a multinomial log regression?
- Next by thread: Covariance of two proportions?
- Index(es):
Relevant Pages
|
