0

I am looking to understand the consequences with futur predictions using the predict(*) R function with a R glm object that didn't converged during modeling process. However, I am able to manually backfilled the coefficients and the other components needed to get a prediction. Should I be worried about weird prediction values or could it produce weird behavior during a prediction process?

Thank you,

John

EDIT: The fact that it doesn't converge is the exact behavior I wanted. I wanted to save timeprocess, because the coefficients are already known before modeling.

John E.
  • 137
  • 2
  • 10
  • 3
    You should not use a model fit that hasn't converged. Not for inference and not for predictions. Investigate why it doesn't converge and try to fix that. Maybe you need to collect more data. Maybe you can switch to a Bayesian approach. Maybe you just need to supply better starting values. Maybe your data does not support the model at all. – Roland Jul 28 '20 at 14:16
  • @Roland The fact that it doesn't converge is the exact behavior I wanted. I wanted to save timeprocess, because the coefficients are already known before modeling. – John E. Jul 28 '20 at 14:25
  • 2
    What does "the coefficients are already known before modeling" even mean? If you know the coefficients, you don't need a GLM object. – Roland Jul 28 '20 at 14:58

0 Answers0