[R] Interpreting Logistic Regression

[Bernardo Rangel Tura]

Em Qua, 2008-08-20 às 23:54 -0700, Madhavi Bhave escreveu:

Hi Madhavi,


> Hi !
> 
> This is Madhavi from Mumbai, India. Incidently this is my first post.

You are wellcome! 
> 
> I am working on Credit Scoring Model and using R, I have run the logistic regression. I have received following Output.
> 
> I have two questions
> 
> (a) What is the significance of "family = binomial(link = logit)". Why do I have to mention Binomial? Is it because my dependent variable assumes only two values 0 and 1? Can I write name of some other Statistical distribution (say Poisson or Negative Binomial) in place of Binomial? How will it affect my results?

Well the logistitc regression is a generalized linear model. Your
specification is just "binomial" function with "logit" link function so
"summary" mention this especification ...

The binomial model is choose because your data have a thorical binomial
distribution (two type of outcome wtih fix probabilty of outcome and
independent observations)

For ohert distribuitions existing other links functions see more details
in:  ?family 
> 
> (b) How do I interpret the "R" result as given below? I know all the variables are significant. How do I get Log Likelihood ratio, Odds ratio etc.?

well odds ratio:

model<- glm (formula,family=binomial)
exp(coef(model))

Log Likehood

model$deviance
-- 
Bernardo Rangel Tura, M.D,MPH,Ph.D
National Institute of Cardiology
Brazil