# [R] logistic regression - what is being predicted when using

(Ted Harding) Ted.Harding at manchester.ac.uk
Thu Feb 18 20:38:57 CET 2010

```On 18-Feb-10 18:58:57, Dimitri Liakhovitski wrote:
> Dear gurus,
> I've analyzed a (fake) data set ("data") using logistic regression
> (glm):
>
> logreg1 <- glm(z ~ x1 + x2 + y, data=data, family=binomial("logit"),
> na.action=na.pass)
>
> Then, I created a data frame with 2 fixed levels (0 and 1) for each
> predictor:
>
> attach(data)
> x1<-c(0,1)
> x2<-c(0,1)
> y<-c(0,1)
> newdata1<-data.frame(expand.grid(x1,x2,y))
> names(newdata1)<-c("x1","x2","y")
>
> Finally, I calculated model-predicted probabilities for each
> combination of those fixed levels:
>
> newdata1\$predicted <-predict(logreg1,newdata=newdata1, type="response")
>
> I am pretty sure the results I get (see the table below) are actual
> probabilities. But just in case - could someone please confirm that
> these are probabilities rather than log odds or odds?
> Thanks a lot!
>
> x1 x2 y predicted
> 1  0  0 0   0.08700468
> 2  1  0 0   0.19262901
> 3  0  1 0   0.27108334
> 4  1  1 0   0.48216220
> 5  0  0 1   0.53686154
> 6  1  0 1   0.74373367
> 7  0  1 1   0.81896484
> 8  1  1 1   0.91887072
> --
> Dimitri Liakhovitski

Yes, they are predicted probabilities of response Z=1.
You specified this by setting 'type="response"'.

See ?predict.glm (the method for 'predict' which is used for GLMs).

Ted.

--------------------------------------------------------------------
E-Mail: (Ted Harding) <Ted.Harding at manchester.ac.uk>
Fax-to-email: +44 (0)870 094 0861
Date: 18-Feb-10                                       Time: 19:38:54
------------------------------ XFMail ------------------------------

```