0

model raises a warning: glm.fit: algorithm did not converge. One of the suspected causes of this error in the model could be due to perfect separation (source: https://www.bookdown.org/rwnahhas/RMPH/blr-separation.html). I want to diagnose the variable causing the error caused by perfect separation by doing a boxplot as follows: I want to ask whether my method for diagnosing this is correct. Isn't it true that if perfect separation occurs, the resulting boxplot should be separate and not have quartiles where the boxplots intersect?

enter image description here

mutu
  • 1
  • 1
    Perfect separation is about the relationship between the linear predictor $x\beta$ and the target $y$, so it's not clear how this box plot will help you diagnose that. More information in the duplicate and [tag:perfect-separation] tag. – Sycorax Oct 04 '23 at 01:47
  • Is there another way to find the variable that causes perfect separation – mutu Oct 07 '23 at 13:07

0 Answers0