that's exactly right, by 'base probability' i mean the 2 raw probabilities
that go into the ratio (prob of suicide given your treatment and given
your control, holding constant the other variables). what you want is
Pr(Y|x1=1,X)/Pr(Y|x1=0,X) or something like that. so what you do is to
take the simulations of each separately, and take the ratio of each pair
of simulations. that gives you simulations of the ratio. you can then
take the mean for the point estimate, and take the 2.5th and 97.5th
percentile estimates (i.e., sort them sims and take the ones 2.5% fromthe
bottom and 97.5 from the top).
Good luck with the journal!
Gary
On Mon, 11 Jul 2005, Martin.Ploederl(a)sbg.ac.at wrote:
Gary,
thanks for the quick reply!
My study is no case-control design,
but a random sample, where the dependent variable is rare (suicide attempts).
The publishers require odds ratios AND the 95% confidence intervals.
I did not find the term "base probability" in the manual.
Is it correct, that the base probabilities can be figured out by setting the predictor
varibales to the desired level with setx, and then simulate with simx?
But how can I calculate the 95% interval of the odds-ratios?
Thanks,
Martin
Dr. Martin Pl�derl
Fachbereich Psychologie
Universit�t Salzburg
Hellbrunnerstr. 34
5020 Salzburg
Tel: +0043 - 8044 - 5130
Email: martin.ploederl(a)sbg.ac.at
________________________________
From: Gary King [mailto:king@harvard.edu]
Sent: Mon 11.07.2005 20:47
To: PL�DERL, Martin
Cc: zelig(a)latte.harvard.edu
Subject: Re: [zelig] summary function - odds ratio output
I would recommend not using the odds ratios, since there is a lot more
information in the other quantities available. (On this point, you might
see Gary King and Langche Zeng. 2002. ``Estimating Risk and Rate Levels,
Ratios, and Differences in Case-Control Studies,'' Statistics in Medicine,
Vol. 21, Pp. 1409-1427.
http://gking.harvard.edu/files/abs/1s-abs.shtml).
But if you'd still like them, what you should do is to calculate the two
base probabilities, which relogit gives, and compute the risk ratio (or
relative risk) from that. The odds ratio is an approximation to the risk
ratio anyway, so you'd certainly be better off with that, even aside from
the issues discussed in the article I mentioned. Of course you can still
calculate the OR from the base probabilities too if you like.
Gary
---
Gary King
Institute for Quantitative Social Science
Harvard University, 34 Kirkland St, Cambridge, MA 02138
http://GKing.Harvard.Edu, King(a)Harvard.Edu
Direct 617-495-2027, Assistant 495-9271, eFax 812-8581
On Mon, 11 Jul 2005, Martin.Ploederl(a)sbg.ac.at wrote:
Hi,
I'm glad to have relogit available in R, because I've rare event data
(suididology). Running relogit works fine. For publishing data, I need
as an output odds-ratios plus the confidence intervals for each
predictor. How can this be realized with Zelig? (Usually I use Frank
Harrels "ols" and "lrm" - functions which give me the odds ratios) I
would appreciate helping me with regard to this. Thank you!
Martin
Dr. Martin Pl�derl
Fachbereich Psychologie
Universit�t Salzburg
Hellbrunnerstr. 34
5020 Salzburg
Tel: +0043 - 8044 - 5130
Email: martin.ploederl(a)sbg.ac.at