5
$\begingroup$

I got the following concurvity values as a result of a GAM model.

enter image description here

Are they considered high or low? I don't know what is the acceptable value.

$\endgroup$

3 Answers 3

6
$\begingroup$

This is a late bump to a relatively old question, but may be of help to future visitors.

In Noam Ross' course on GAMs, GAMs in R, Chapter 2, section 10 "checking concurvity" says that observing a value over 0.8 for the worst case requires closer inspection of the model. So in the image you pasted there's only one variable over 0.8 (VV_PROM_S60), but there are many very close to that threshold, so you may want to check those too. The suggested way to have a closer look at the model is by using concurvity(model, full = FALSE) and carefully analyzing the pairwise concurvities.

$\endgroup$
3
$\begingroup$

I would like to get more clarity on this as well, but I did find one paper (Tree traits influence response to fire severity in the western Oregon Cascades, USA; 2019) that used the mgcv package where the authors used a cutoff of 0.3. It sounded like they applied that cutoff to all three measures of concurvity. Of course, even they said it was an arbitrary cutoff.

$\endgroup$
0
1
$\begingroup$

Another source, from a doctoral dissertation Multi-city time series analyses of air pollution and mortality data using generalized geoadditive mixed models by Lung-Chang Chien. Free online access.

So far, there is no strict criterion to identify the level of concurvity which can severely affect model fitting. Ramsay et al. (2003a) suggested using 0.5 to be the cutoff point. If the level of concurvity in nonparametric or semiparametric models is greater than 0.5, it is necessary to seek in order to eliminate or reduce this problem.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.