Closed hansvancalster closed 5 years ago
marginal_effects
excludes autocorrelation parameters in the predictions. What happens if you compare it to fitted(..., incl_autocor = FALSE)
?
Aah yes, I can conform that setting incl_autocor = FALSE
reproduces the default marginal_effects
plot. Great! I didn't see the argument in the help of marginal_effects
but I should have looked further for the arguments of extract_draws
which can be passed to the former function.
I also understand now the difference between fitted(..., incl_autocor = TRUE)
and fitted(..., incl_autocor = TRUE, newdata = data.frame(time = 1:n)
in the above example. With incl_autocor = FALSE
this will result in the same plot, but with incl_autocor = TRUE
and newdata
the expectations of the posterior predictions are no longer conditioned on the observed responses, which results in wider interval.
Thanks for implementing
cor_arma
for non-gaussian models in PR #650I saw some differences when trying to reproduce the output of
marginal_effects(..., method = "fitted")
starting from thefitted
procedure. The code that illustrates the differences follows:Is this intended behaviour or a bug?