Assessing significance of smooth terms with permutation testing

[avaruuser]

Hello, previously people have investigated the use of permutations for hypothesis testing of the smooth terms in GAMs (e.g. here). I'm relatively new to GAMs and wondered if you see any issues with how I implemented such a permutation testing procedure using pyGAM:

1. Compute a GAM with the DV and IVs.
2. Store the deviance explained from the GAM output using gam.score(IVs, DV)
3. Randomly shuffle one of the IVs n_perms times (e.g. 10,000 times).
4. Compute the GAM with the randomly shuffled IV.
5. Store the deviance explained for each random permutation.
6. Get p-value from the proportion of permutations that exceed the original deviance explained.
7. Repeat this process for the other IVs

Does this make sense to you?