逻辑斯特回归其实是GLM的一种。下图可以很好的说明这个关系

1. 1.回归函数
   

logit 回归要combine两个回应作为应变量，不能只是一个单纯的比例或者分数。或者本身就是binary 变量

fit <- glm(F~x1+x2+x3,data=mydata,family=binomial())

summary(fit)                                                        # display results

confint(fit)                                                           # 95% CI for the coefficients

exp(coef(fit))                                                       # exponentiated coefficients

exp(confint(fit))                                                   # 95% CI for exponentiated coefficients

predict(fit, type="response")                              # predicted values, 要是data frame

residuals(fit, type="deviance")                           # residuals

2. 2.参数检验
   

参见我的另一篇文章Parameter tests中的GLM部分

3. 3.  模型选择
   

anova(model, test=”Chisq”)

drop1(model, test=”Chisq”)

4. 可视化

cdplot(F~x, data=mydata)