In-house processing of marginaleffects objects

[DominiqueMakowski]

Currently we call the tidy() function,

parameters/R/methods_marginaleffects.R

Lines 12 to 15 in 804cfcb

	out <- insight::standardize_names(
	marginaleffects::tidy(model, conf_level = ci, ...),
	style = "easystats"
	)

which is defined here: https://github.com/vincentarelbundock/marginaleffects/blob/a6730a3560dbf4220d4bf82da141400fa235b996/sandbox/tidy_old.R#L92

But this strongly limits our capabilities (especially in Bayesian cases). We should probably not rely on tidy and reimplement it using bayestestR for the processing of draws.

easystats/modelbased#302

[strengejacke]

We should probably submit bayestestR (easystats/bayestestR#691) the next days, then we should have the basis for this PR.

[DominiqueMakowski]

By daniel here about modelbased:

No, we fully support Bayesian models. For Bayesian models, we now call describe_posterior(), too. We could do this for the parameters package as well, but that doesn't affect modelbased.

I'm all for outsourcing its postprocessing to parameters, this feels like the right place to have it. So I'd move the postprocessing from modelbased to here and have modelbased call parameters().

[strengejacke]

Agree. But I think we can wait for 1.0.0 release to do this (both parameters and marginaleffects)

[mattansb]

I suggest, wherever the method is, to use the get_datagrid() function - it's what is used in bayestestR and it works great there 😄

[strengejacke]

Yes, I would call tidy() for frequentist models, because this is maintained in marginaleffects. And for Bayesian models, we call describe_posterior() instead.