coefficient of logistic regression

in logistic regression,
I want to know that it is possible to get probability values of each 
predictors by
using following formula for each predictor one by one (keeping constant 
the others)
<<< exp(coef)/(1+exp(coef)) >>>

At 11:54 AM 6/3/2003 +0300, orkun wrote:

in logistic regression,
I want to know that it is possible to get probability values of each 
predictors by
using following formula for each predictor one by one (keeping 
constant the others)
<<< exp(coef)/(1+exp(coef)) >>>

Dear Ahmet,

This will almost surely give you nonsense, since it produces a fitted 
probability ignoring the constant in the model (assuming that there is 
one), setting other predictors to 0 and the predictor in question to 
1. What is it that you want to do?

I hope that this helps,
 John

-----------------------------------------------------
John Fox
Department of Sociology
McMaster University
Hamilton, Ontario, Canada L8S 4M4
email: jfox at mcmaster.ca
phone: 905-525-9140x23604
web: www.socsci.mcmaster.ca/jfox
-----------------------------------------------------

Hello

in logistic regression,
I want to know that it is possible to get probability values of each
predictors by
using following formula for each predictor one by one (keeping constant
the others)
 <<< exp(coef)/(1+exp(coef)) >>>

thanks in advance
Ahmet Temiz

John Fox wrote:

At 11:54 AM 6/3/2003 +0300, orkun wrote:

in logistic regression,
I want to know that it is possible to get probability values of each 
predictors by
using following formula for each predictor one by one (keeping constant 
the others)
<<< exp(coef)/(1+exp(coef)) >>>

Dear Ahmet,

This will almost surely give you nonsense, since it produces a fitted 
probability ignoring the constant in the model (assuming that there is 
one), setting other predictors to 0 and the predictor in question to 1. 
What is it that you want to do?

I hope that this helps,
 John

thank you

Say, I just want to find each predictor's particular effect on dependent 
variables.
Actual model is to prepare landslide susceptibility map on GIS. So  I want 
to know
what the effect as probability value comes from each predictor. For 
instane what is the effect
of  slope on landslide susceptibility. Should I keep others constant ?

kind regards

Dear Ahmet,

Sorry for the slow response, but I've been busy all today, 
coincidentally teaching a workshop on logistic regression.

Tom Blackwell sent you a useful suggestion for interpreting 
coefficients on the odds scale. If you want to trace out the partial 
relationship of the fitted probability of response to a particular 
predictor holding others constant, you can set the other predictors to 
typical values and let the predictor in question vary over its range, 
transforming the fitted log-odds to the probability scale.

You may be interested in my effects package (on CRAN or at 
<http://socserv.socsci.mcmaster.ca/jfox/Misc/effects/index.html>), 
which makes these kinds of displays for linear and generalized-linear 
models, including those with interactions.

Regards,
 John

At 03:06 PM 6/3/2003 +0300, orkun wrote:

John Fox wrote:

At 11:54 AM 6/3/2003 +0300, orkun wrote:

in logistic regression,
I want to know that it is possible to get probability values of 
each predictors by
using following formula for each predictor one by one (keeping 
constant the others)
<<< exp(coef)/(1+exp(coef)) >>>

Dear Ahmet,

This will almost surely give you nonsense, since it produces a 
fitted probability ignoring the constant in the model (assuming that 
there is one), setting other predictors to 0 and the predictor in 
question to 1. What is it that you want to do?

I hope that this helps,
 John

thank you

Say, I just want to find each predictor's particular effect on 
dependent variables.
Actual model is to prepare landslide susceptibility map on GIS. So  I 
want to know
what the effect as probability value comes from each predictor. For 
instane what is the effect
of  slope on landslide susceptibility. Should I keep others constant ?

kind regards

-----------------------------------------------------
John Fox
Department of Sociology
McMaster University
Hamilton, Ontario, Canada L8S 4M4
email: jfox at mcmaster.ca
phone: 905-525-9140x23604
web: www.socsci.mcmaster.ca/jfox
-----------------------------------------------------

Ahmet  -

In a logistic regression model, fitted probabilities make
sense for individual cases (rows in the data set), as well
as for future cases (predictions) for which no outcome
(success or failure) has been observed yet.  Fitted
probabilities are calculated from the matrix formula:

 Pr[success]  =  exp( X %*% beta) / (1 + exp( X %*% beta)

where  X  is an [n x (p+1)] matrix, containing all p predictor
variables as columns, preceded by a column of 1s for the
intercept, and  beta  is the [(p+1) x 1] vector of logistic
regression coefficients.

One can interpret the sign and the magnitude of an individual
regression coeffient by saying that an increase of 1 unit in
predictor variable [i] will increase or decrease the odds of
success by a multiplier of  exp(beta[i]).  When  beta[i] > 0
the odds increase, because  exp(beta[i]) > 1,  and when
beta[i] < 0  the odds decrease, because  exp(beta[i]) < 1.

I hope this explanation helps.

-  tom blackwell  -  u michigan medical school  -  ann arbor  -

On Tue, 3 Jun 2003, orkun wrote:

 

Hello

in logistic regression,
I want to know that it is possible to get probability values of each
predictors by
using following formula for each predictor one by one (keeping constant
the others)
<<< exp(coef)/(1+exp(coef)) >>>

thanks in advance
Ahmet Temiz
   

Dear Mr. Fox

thank you very much all.

Because of related to your answer. I ask you directly if you don't mind
I studied several ways after my email.
I wonder whether pgeo<-predict.glm(glm.ob,type="terms")
gives same result with probability value I asked before.
I tried on it. But it gives "Error in rep(1/n,n) %*% model.matrix(object): 
non conformable
arguments" .

By the way , your teaching notes is available on the internet ?