R help - Dec 2012 - Odds Ratio and Logistic Regression

Dear All,
I am learning the ropes about logistic regression in R.
I found some interesting examples

http://bit.ly/Vq4GgX
http://bit.ly/W9fUTg
http://bit.ly/UfK73e

but I am a bit lost.
I have several questions.
1) For instance, what is the difference between

glm.out = glm(response ~ poverty + gender, family=binomial(logit),
   data=mydata)

and

glm.out = glm(response ~ poverty * gender, family=binomial(logit),
   data=mydata)
? Which begs the question when I should use the "*" or "+"
sign when doing
a logistic regression on several explanatory variables. I think that in  
the former case I am allowing for an interaction between poverty and  
gender, but I would like to be sure about it.

2) Consider the following snippet


glm.out = glm(response ~ poverty + gender, family=binomial(logit),
   data=mydata)

where "response" is a dichotomous variable, poverty assumes only two  
values (Above poverty line and Below poverty line) and gender assumes only  
the Male or Female values.
The command above leads to the following output
#######################################
print(summary(glm.out))
Call:
glm(formula = response ~ poverty + gender, family = binomial(logit),
     data = mydata)

Deviance Residuals:
     Min       1Q   Median       3Q      Max
-2.2094   0.4269   0.4269   0.8033   1.1911

Coefficients:
                           Estimate Std. Error z value Pr(>|z|)
(Intercept)                 0.9656     0.1477   6.538 6.25e-11 ***
povertyBelow poverty line  -0.9978     0.3246  -3.074  0.00211 **
genderFEMALE                1.3840     0.2549   5.429 5.68e-08 ***
---
Signif. codes:  0 ?***? 0.001 ?**? 0.01 ?*? 0.05 ?.? 0.1 ? ? 1

(Dispersion parameter for binomial family taken to be 1)

     Null deviance: 494.81  on 499  degrees of freedom
Residual deviance: 457.13  on 497  degrees of freedom
AIC: 463.13

Number of Fisher Scoring iterations: 4
##############################################

To calculate then odds ratios, I should do the following

exp(coef(glm.out))
               (Intercept) povertyBelow poverty line               
genderFEMALE
                 2.6263831                 0.3687033                  
3.9909627

but here I am lost about the interpretation. For instance, what are the  
odds of a positive response for those above versus below the poverty line  
in males? In females?

I think that everything is there, but I cannot extract/interpret the info  
provided by R correctly.
Any help is appreciated.
Cheers

Lorenzo

Lorenzo:

Are you "learning the ropes" or "on the ropes"?

(Heh, heh -- I couldn't resist that bad bit of English slang. I apologize
to all non-native English speakers -- and maybe to the native ones, too.
But sometimes my evil inner twin just gets the better of me).

However, to try to be helpful. It sounds to me like you have some
considerable confusion about the basics of GLM's. I don't think this is
the
right forum to resolve that confusion. I think you either need to do some
more studying on your own or, better yet, consult a local statistical
expert, who should be able to help you out fairly quickly.

Cheers,
Bert

On Sun, Dec 30, 2012 at 10:14 AM, Lorenzo Isella
<lorenzo.isella@gmail.com>wrote:
> Dear All,
> I am learning the ropes about logistic regression in R.
> I found some interesting examples
>
> http://bit.ly/Vq4GgX
> http://bit.ly/W9fUTg
> http://bit.ly/UfK73e
>
> but I am a bit lost.
> I have several questions.
> 1) For instance, what is the difference between
>
> glm.out = glm(response ~ poverty + gender, family=binomial(logit),
>   data=mydata)
>
> and
>
> glm.out = glm(response ~ poverty * gender, family=binomial(logit),
>   data=mydata)
> ? Which begs the question when I should use the "*" or
"+" sign when doing
> a logistic regression on several explanatory variables. I think that in the
> former case I am allowing for an interaction between poverty and gender,
> but I would like to be sure about it.
>
> 2) Consider the following snippet
>
>
> glm.out = glm(response ~ poverty + gender, family=binomial(logit),
>   data=mydata)
>
> where "response" is a dichotomous variable, poverty assumes only
two
> values (Above poverty line and Below poverty line) and gender assumes only
> the Male or Female values.
> The command above leads to the following output
> ##############################**#########
> print(summary(glm.out))
> Call:
> glm(formula = response ~ poverty + gender, family = binomial(logit),
>     data = mydata)
>
> Deviance Residuals:
>     Min       1Q   Median       3Q      Max
> -2.2094   0.4269   0.4269   0.8033   1.1911
>
> Coefficients:
>                           Estimate Std. Error z value Pr(>|z|)
> (Intercept)                 0.9656     0.1477   6.538 6.25e-11 ***
> povertyBelow poverty line  -0.9978     0.3246  -3.074 0.00211 **
> genderFEMALE                1.3840     0.2549   5.429 5.68e-08 ***
> ---
> Signif. codes:  0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1
>
> (Dispersion parameter for binomial family taken to be 1)
>
>     Null deviance: 494.81  on 499  degrees of freedom
> Residual deviance: 457.13  on 497  degrees of freedom
> AIC: 463.13
>
> Number of Fisher Scoring iterations: 4
> ##############################**################
>
> To calculate then odds ratios, I should do the following
>
> exp(coef(glm.out))
>               (Intercept) povertyBelow poverty line
>  genderFEMALE
>                 2.6263831                 0.3687033
> 3.9909627
>
> but here I am lost about the interpretation. For instance, what are the
> odds of a positive response for those above versus below the poverty line
> in males? In females?
>
> I think that everything is there, but I cannot extract/interpret the info
> provided by R correctly.
> Any help is appreciated.
> Cheers
>
> Lorenzo
>
> ______________________________**________________
> R-help@r-project.org mailing list
>
https://stat.ethz.ch/mailman/**listinfo/r-help<https://stat.ethz.ch/mailman/listinfo/r-help>
> PLEASE do read the posting guide http://www.R-project.org/**
> posting-guide.html <http://www.R-project.org/posting-guide.html>
> and provide commented, minimal, self-contained, reproducible code.
>


-- 

Bert Gunter
Genentech Nonclinical Biostatistics

Internal Contact Info:
Phone: 467-7374
Website:
http://pharmadevelopment.roche.com/index/pdb/pdb-functional-groups/pdb-biostatistics/pdb-ncb-home.htm

	[[alternative HTML version deleted]]

Took some googling, but it was worth it! :)

Best,
Simone

Sent from my iPhone. Please excuse brevity and odd typos

Il giorno 30/dic/2012, alle ore 19:44, Bert Gunter <gunter.berton at
gene.com> ha scritto:
> Are you "learning the ropes" or "on the ropes"?

At 18:14 30/12/2012, Lorenzo Isella wrote:
>Dear All,
>I am learning the ropes about logistic regression in R.
>I found some interesting examples
>
>http://bit.ly/Vq4GgX
>http://bit.ly/W9fUTg
>http://bit.ly/UfK73e
>
>but I am a bit lost.
>I have several questions.
>1) For instance, what is the difference between
>
>glm.out = glm(response ~ poverty + gender, family=binomial(logit),
>   data=mydata)
>
>and
>
>glm.out = glm(response ~ poverty * gender, family=binomial(logit),
>   data=mydata)
>? Which begs the question when I should use the "*" or
"+" sign when doing
>a logistic regression on several explanatory variables. I think that in
>the former case I am allowing for an interaction between poverty and
>gender, but I would like to be sure about it.
I think you need to (re)-read any introductory 
text on R, in particular about the use of 
formulae. The asterisk implies an interaction. 
This also answers your second question I think.

>2) Consider the following snippet
>
>
>glm.out = glm(response ~ poverty + gender, family=binomial(logit),
>   data=mydata)
>
>where "response" is a dichotomous variable, poverty assumes only
two
>values (Above poverty line and Below poverty line) and gender assumes only
>the Male or Female values.
>The command above leads to the following output
>#######################################
>print(summary(glm.out))
>Call:
>glm(formula = response ~ poverty + gender, family = binomial(logit),
>     data = mydata)
>
>Deviance Residuals:
>     Min       1Q   Median       3Q      Max
>-2.2094   0.4269   0.4269   0.8033   1.1911
>
>Coefficients:
>                           Estimate Std. Error z value Pr(>|z|)
>(Intercept)                 0.9656     0.1477   6.538 6.25e-11 ***
>povertyBelow poverty line  -0.9978     0.3246  -3.074  0.00211 **
>genderFEMALE                1.3840     0.2549   5.429 5.68e-08 ***
>---
>Signif. codes:  0 ???***??? 0.001 ???**??? 0.01 
>???*??? 0.05 ???.??? 0.1 ??? ??? 1
>
>(Dispersion parameter for binomial family taken to be 1)
>
>     Null deviance: 494.81  on 499  degrees of freedom
>Residual deviance: 457.13  on 497  degrees of freedom
>AIC: 463.13
>
>Number of Fisher Scoring iterations: 4
>##############################################
>
>To calculate then odds ratios, I should do the following
>
>exp(coef(glm.out))
>               (Intercept) povertyBelow poverty line
>genderFEMALE
>                 2.6263831                 0.3687033
>3.9909627
>
>but here I am lost about the interpretation. For instance, what are the
>odds of a positive response for those above versus below the poverty line
>in males? In females?
>
>I think that everything is there, but I cannot extract/interpret the info
>provided by R correctly.
>Any help is appreciated.
>Cheers
>
>Lorenzo
>
>
Michael Dewey
info at aghmed.fsnet.co.uk
http://www.aghmed.fsnet.co.uk/home.html