I have a gam model where I am trying to describe animal mass as a function of distance from a feature. A simplified version of this model is gam(Mass~s(Time)+s(Distance)+s(Site, bs="re"), data=data), where Time is a necessary control, Distance is the variable of interest, and Site is a random effect as I have multiple observations per site.
The problem is that there is a lot of concurvity in my data. Below is the output of concurvity(model)
para s(Time) s(Distance) s(Site)
worst 1 1.0000000 1 1.0000000
observed 1 0.7855252 1 0.7583261
estimate 1 0.8306823 1 0.9185499
I found slides by Simon Wood here: https://statistique.cuso.ch/fileadmin/statistique/part-3.pdf that say that this is a common problem when you have a spatial covariate as a random effect, and your other variables are also functions of space. In these slides, he says "model averaging over the sampling distribution of the smoothing parameters can help", but I don't understand what this means.
Are there any solutions that would allow me to model distance, while not just ignoring the random effect and heirarchical structure of my data?
There are similar, unanswered questions here High concurvity with random effect terms in generalized additive mixed effects models GAMM and here concurvity in presence of random smooths
