Generalised Additive Mixed Model

[TomKealy]

Is your feature request related to a problem? Please describe

I want to reproduce the GAMM portion of this post:

https://petolau.github.io/Analyzing-double-seasonal-time-series-with-GAM-in-R/

as the R function gamm currently has no python equivalent. Since this is a broad class of models which currently has no equivalent and since you already have GLMGam functionality, this could be a good addition to statsmodels.

Describe the solution you'd like

I volunteer to produce this!

We could extend the Linear Mixed Model class like the GLM class has been extended for statsmodels.gam.GLMGam.

Describe alternatives you have considered

Since gamm has no python coverage, there is no alternative solution

Additional context

I will volunteer to produce this feature, but I will need some guidance.

[josef-pkt]

Do you have a single time series or panel/longitudinal data?

AFAIC, our mixed model does not handle AR error correlation.
It won't be the best model to handle time series data.

I don't know how difficult it would be to add penalized splines to Mixed model, i.e. how to get to a general GAMM version.
I can help with any parts related to penalized splines.

[kshedden]

I looked at this closely back in May. I made some updates in #5898 specifically to begin this process. But I didn't have time to work on it after than point. It will require deep familiarity with this very complicated code to add this feature.

We have totally separate codebases for mixed linear and mixed GLM. The latter uses a non-standard variational Bayes fitting approach. I have not thought at all about adding GAM features to our mixed GLM.