HI there, i know this is a basic question, though i need some help because this is somewhat away from my current issue, but nevertheless interesting to me... Lets assume i have some estimated probabilities, say estimated by a logit model. i know i can also state them as an odds ratio. Now i?d like to state these odds ratios as a reference to a specific outcome of my investigated variable. for example, if my covariate of interest is race and possible outcomes are white, black and hispanic, whereas the latter are minorities in my case - how can i state the odds ratio in such a way that white is the reference (always 1) and other races' odds ratio are relative to the reference. e.g. hispanics are 1.5 times more likely to ... Is creating 3 binary dummies for race the right way ? And if so how can i go on. As i said, i know this is rather basic, i am thankful for any links / references... thanks in advance !
Hi, You can state a probability p as odds p/(1-p) and vice versa. To get an odds ratio you need actually two odds. Then you can get the odds ration of being/having "a" instead of "b" by odds(a)/odds(b), where "b" is the reference level. If you fit a logistic regression model (which means that your outcome is dichotomous) then the estimated coefficients are actually log(oddsratios) - which you can transform to odds by exp() . You can use a factor-variable with three levels for race and treatment-contrasts to get odds ratios for not being white against being white - make sure, that either your factor has "white" as first level or specify the contrast with the "base" argument. If you create 3 dummy variables and involve an intercept in your model your model will be perfectly collinear - the so called "dummy variable trap" - you can use an intercept and create two dummies for the covariate levels you are actually interested in and put this in your logistic model - the result will be the same as with the treatment contrasts. hth. Bunny, lautloscrew.com schrieb:> HI there, > > i know this is a basic question, though i need some help because this > is somewhat away from my current issue, but nevertheless interesting > to me... Lets assume i have some estimated probabilities, say > estimated by a logit model. i know i can also state them as an odds > ratio. > > Now i?d like to state these odds ratios as a reference to a specific > outcome of my investigated variable. > > for example, if my covariate of interest is race and possible outcomes > are white, black and hispanic, whereas the latter are minorities in my > case - how can i state the odds ratio in such a way that white is the > reference (always 1) and other races' odds ratio are relative to the > reference. e.g. hispanics are 1.5 times more likely to ... > > Is creating 3 binary dummies for race the right way ? And if so how > can i go on. > As i said, i know this is rather basic, i am thankful for any links / > references... > > thanks in advance ! > > ______________________________________________ > R-help at r-project.org mailing list > https://stat.ethz.ch/mailman/listinfo/r-help > PLEASE do read the posting guide > http://www.R-project.org/posting-guide.html > and provide commented, minimal, self-contained, reproducible code.-- Eik Vettorazzi Institut f?r Medizinische Biometrie und Epidemiologie Universit?tsklinikum Hamburg-Eppendorf Martinistr. 52 20246 Hamburg T ++49/40/42803-8243 F ++49/40/42803-7790
You notice that your reference is for the SAS-procedure glm?!
To play around and since you provided no data example I've created some
data (by an idea from Frank Harrell jr.
http://www.biostat.wustl.edu/archives/html/s-news/2002-04/msg00103.html)
n <- 30
treat <- rep(c('a','b','c'), length.out=n)
lg<- log(1.2)*(treat=='b')+log(1.5)*(treat=='c') # here the
intercept
is assumed to be 0 which means that there is an even chance of having
y=1 or y=0 when treat is 'a'.
y <- ifelse(runif(n) <= plogis(lg), 1, 0)
trf<-as.factor(treat)
To have a first look at the data you can do
table(y,treat)
and you can fit your model via
mm<-glm(y~trf,family=binomial(link="logit"))
and to get your odds ratios
exp(coef(mm))
you may look at the model matrix to see that treatment contrasts
correspond to use two dummies for "b" and "c" and an
intercept term
(which here refers to the reference level "a")
model.matrix(mm)
hth again.
Bunny, lautloscrew.com schrieb:> Hi,
>
> i found something interesting to me:
> http://www.otago.ac.nz/sas/stat/chap39/sect27.htm
>
> and i am about to get an idea of the reference thing ( -1 ) , but
> couldn?t connect on the following example:
>
> For the GLM parameterization scheme (PARAM=GLM), the design variables
> are as follows.
>
> Maybe you?ll find some time, to take another look.
>
> thanks in advance
>
>
>
> Am 23.09.2008 um 11:36 schrieb Eik Vettorazzi:
>
>> Hi,
>> You can state a probability p as odds p/(1-p) and vice versa. To get
>> an odds ratio you need actually two odds. Then you can get the odds
>> ration of being/having "a" instead of "b" by
odds(a)/odds(b), where
>> "b" is the reference level.
>> If you fit a logistic regression model (which means that your outcome
>> is dichotomous) then the estimated coefficients are actually
>> log(oddsratios) - which you can transform to odds by exp() .
>> You can use a factor-variable with three levels for race and
>> treatment-contrasts to get odds ratios for not being white against
>> being white - make sure, that either your factor has "white"
as first
>> level or specify the contrast with the "base" argument.
>> If you create 3 dummy variables and involve an intercept in your
>> model your model will be perfectly collinear - the so called
"dummy
>> variable trap" - you can use an intercept and create two dummies
for
>> the covariate levels you are actually interested in and put this in
>> your logistic model - the result will be the same as with the
>> treatment contrasts.
>>
>> hth.
>>
>> Bunny, lautloscrew.com schrieb:
>>> HI there,
>>>
>>> i know this is a basic question, though i need some help because
>>> this is somewhat away from my current issue, but nevertheless
>>> interesting to me... Lets assume i have some estimated
>>> probabilities, say estimated by a logit model. i know i can also
>>> state them as an odds ratio.
>>>
>>> Now i?d like to state these odds ratios as a reference to a
specific
>>> outcome of my investigated variable.
>>>
>>> for example, if my covariate of interest is race and possible
>>> outcomes are white, black and hispanic, whereas the latter are
>>> minorities in my case - how can i state the odds ratio in such a
way
>>> that white is the reference (always 1) and other races' odds
ratio
>>> are relative to the reference. e.g. hispanics are 1.5 times more
>>> likely to ...
>>>
>>> Is creating 3 binary dummies for race the right way ? And if so
how
>>> can i go on.
>>> As i said, i know this is rather basic, i am thankful for any links
>>> / references...
>>>
>>> thanks in advance !
>>>
>>> ______________________________________________
>>> R-help at r-project.org mailing list
>>> https://stat.ethz.ch/mailman/listinfo/r-help
>>> PLEASE do read the posting guide
>>> http://www.R-project.org/posting-guide.html
>>> and provide commented, minimal, self-contained, reproducible code.
>>
>> --
>> Eik Vettorazzi
>> Institut f?r Medizinische Biometrie und Epidemiologie
>> Universit?tsklinikum Hamburg-Eppendorf
>>
>> Martinistr. 52
>> 20246 Hamburg
>>
>> T ++49/40/42803-8243
>> F ++49/40/42803-7790
>>
>
--
Eik Vettorazzi
Institut f?r Medizinische Biometrie und Epidemiologie
Universit?tsklinikum Hamburg-Eppendorf
Martinistr. 52
20246 Hamburg
T ++49/40/42803-8243
F ++49/40/42803-7790
To select another reference level from the default that R chooses you can use
the "relevel" function
example;
Reorder Levels of Factorwarpbreaks$tension <- relevel(warpbreaks$tension,
ref="M")
summary(lm(breaks ~ wool + tension, data=warpbreaks))
function findlink(pkg, fn) {
var Y, link;
Y = location.href.lastIndexOf("\\") + 1;
link = location.href.substring(0, Y);
link = link + "../../" + pkg + "/chtml/" + pkg +
".chm::/" + fn;
location.href = link;
}
This sets "M" as the reference level for the analysis. At the prompt
on the R workspace, type
?relevel
A Smile costs Nothing
But Rewards Everything
Happiness is not perfected until it is shared
-Jane Porter
--- On Tue, 9/23/08, Bunny, lautloscrew.com <bunny@lautloscrew.com> wrote:
From: Bunny, lautloscrew.com <bunny@lautloscrew.com>
Subject: [R] odds ratio: how to create reference
To: r-help@r-project.org
Date: Tuesday, September 23, 2008, 4:06 AM
HI there,
i know this is a basic question, though i need some help because this
is somewhat away from my current issue, but nevertheless interesting
to me... Lets assume i have some estimated probabilities, say
estimated by a logit model. i know i can also state them as an odds
ratio.
Now i´d like to state these odds ratios as a reference to a specific
outcome of my investigated variable.
for example, if my covariate of interest is race and possible outcomes
are white, black and hispanic, whereas the latter are minorities in my
case - how can i state the odds ratio in such a way that white is the
reference (always 1) and other races' odds ratio are relative to the
reference. e.g. hispanics are 1.5 times more likely to ...
Is creating 3 binary dummies for race the right way ? And if so how
can i go on.
As i said, i know this is rather basic, i am thankful for any links /
references...
thanks in advance !
______________________________________________
R-help@r-project.org mailing list
https://stat.ethz.ch/mailman/listinfo/r-help
PLEASE do read the posting guide http://www.R-project.org/posting-guide.html
and provide commented, minimal, self-contained, reproducible code.
[[alternative HTML version deleted]]