The very concise explanation of GLMs in my course describes the link function without saying much about the error distribution. From what I read in other places such distribution seems much more important than it would seem in my course.
For GLMs in one the most useful answers on this site I read
"The link function is the cumulative probability function that the error terms follow"
I would be extremely grateful if someone provides an intuitive explanation of the above (for GLM), and of the actual importance/role of error distributions in general, beyond punctual and terse statements like those of my course (that just goes "to apply xzy we assume that the error is/has distribution ..." (no explanation of the why)
I notice only now a question that seems the same as mine (please correct if wrong), at What difference (if any) exists between the Response Distribution and Error Distribution in GLMs?
Asked
Active
Viewed 487 times
2
-
2See https://stats.stackexchange.com/questions/48594/purpose-of-the-link-function-in-generalized-linear-model – kjetil b halvorsen Feb 12 '18 at 22:25
-
In the link you posted I am unable to see the answer to my question, it speaks about link function modeling parameters of the distribution, every course says so no problem with that. My questions are: 1) intuitive explanation of why "The link function is the cumulative probability function that the error terms follow" 2) importance of error distribution in general (if such a general question makes sense and can be answered it in a Q&A) Maybe the thread you posted answers that but I am unable to see it because of problems with 2) – Feb 13 '18 at 14:47