Jul 29, 2008

Confidence Intervals for glm and coxph models

Summary methods for linear and generalized linear models don't print nice confidence intervals. Use the intervals functions to get them easily. Following package nlme, intervals is generic and I have defined methods for glm and coxph. glm assumes a logistic regression model (family=binomial) an will show odds-ratios for covariates (not for the intercept).

fit<-glm(disease~gender, family=binomial)
intervals(fit)

           or 95% C.I. P-val
genderMale 0.68 ( 0.57 - 0.81 ) 0.0000

fit<-coxph(Surv(time,status)~gender)
intervals(fit)

           hr 95% C.I. P-val
genderMale 0.88 ( 0.55 - 1.42 ) 0.6061

No comments: