Is there a rule of thumb for excessive concurvity? For example, I've heard that a VIF of $>10$ suggests that multicollinearity may be a problem (e.g., in this CV answer), but I cannot find a similar rough guide for concurvity. I realise that such rules are very woolly and must be taken with a grain of salt, but a ballpark number is often useful, too...
Asked
Active
Viewed 337 times
1 Answers
1
I realise this is a very late response, but in case anyone else is looking for an answer on this too:
Noam Ross' free course on GAMs advises 0.80 for the "worst concurvity" measure (as opposed to estimate and observed concurvity).
The course does not provide a justification for this "threshold" though nor an explanation on why to look at worst concurvity rather than estimate or observed concurvity, or what the difference is between these estimates of concurvity. Hopefully someone with a greater understanding of GAMs and mgcv can chip in too! I for one am looking for an answer to the latter issue.
Jade
- 351