1

I'm new here - apologies if there are any problems.

I have two GAMs that I've fit (using mgcv::gam()), and the two models are comparable in AICc values. With GLMs, I've read that one option when you have competing models can be to average the models, to present one model where variables are weighted based on their AICc values. Is this appropriate or possible to do with a GAM, where the model is composed of many basis functions?

There is a somewhat related question here about model selection for GAMs, and the answer says for model selection you should use select=TRUE to penalize unnecessary terms out of your model - but I am wondering if model averaging could be an alternative if I'm simply wanting to explain how my predictors affect my response.

sheep
  • 11
  • 2
  • 1
    Welcome to CV, sheep! – Alexis Jan 09 '22 at 07:28
  • 1
    What is different about the two models? How were they specified? And make sure that no other models were entertained or screened out, which would invalidate many aspects of inference. – Frank Harrell Jan 09 '22 at 13:19
  • What is the aim of the modelling? Prediction or something else? And what are you planning on averaging: the model terms/smooths or the model predictions/estimates of the response? – Gavin Simpson Jan 10 '22 at 09:24

0 Answers0