Re: [R] Binary response GLM Question

Hi Josh, Thank you for your reply. The reason I thank "Y" (0 and 1) here as p is because I think each observation is just a bernulli trial, so in this case the binomial n=1. And yet R still fits it (with the logit link) . I know the expression for the logit link, so I assumed I can take y here as

Re: [R] Binary response GLM Question

Hi, Y is not the same as P. P is the conditional probability given the data matrix. So theoretically, P can take on any value in [0, 1], which means the odds can be anywhere from [0, +infty], not just 0 or undefined. In logistic regression, the logit link is pretty standard, so I do not think y

[R] Binary response GLM Question

Hi all, I have a problem with binary response data in GLM fitting. The problem is that the "y" take only 1 or 0, and if I use logit link, it is the log of the odds ratio, which is p/(1-p). In my situation, think "y" is "p", so sometimes the odds is 0, sometimes it is "1/0", which is (should be) un