Does rstanarm::plot_nonlinear produce marginal or conditional plots?

Dear all,

I’m preparing a study using rstanarm::stan_gamm4 to model some continuous variables with splines and wanted to be sure about the output of the rstanarm::plot_nonlinear. My question stems from i.a. the fact that brms::conditional_smooths (invoked it seems by brms::marginal_smooths) yields the conditional effects of smoothers.

My understanding is that the marginal effects would be the posterior samples without conditioning on the other covariates whereas the conditional effects are of course conditioned on the covariates. Both are interesting but inherently different, so I only want to make sure (a) I didn’t get something wrong, and (b) that we share the results we intend.



I suppose these are of little use, but anyway my session info:

  • Operating System: Mac OS X Catalina
  • rstanarm Version: rstanarm_2.18.2

They are conditional on the observed values in the data of the other variables.

@bgoodri, thanks for your reply! Being conditional on the data, I suppose, is a direct—and desirable—consequence of their posterior nature. But they are marginals of the posterior then, right? At least, it seems they’re constantly referred to at such in i.a. text books on Bayesian modelling.