Difference between method = fitted vs. predict in marginal_effects function

Hi guys,

This is, hopefully, a relatively simple question: what is the difference between “method = "fitted"” and “method = "predict"” in the marginal_effects function? That is, can someone explain the difference between “marginal predictions of the regression curve” and “plot marginal predictions of the responses”?

I believe I have answered this question several times on threads in the Stan fourms, the old brms google group, and / or in (now closed) github issues. Would you mind taking another look if you can find an answer in these sources?

1 Like

Of course. If I find one I’ll post it here for posterity.

Edit:method = "predict"” accounts for the residual (observation-level) variance, the uncertainty in the fixed coefficients, and the uncertainty in the variance parameters for the grouping factors. “method = "fitted"” accounts only for the uncertainty in the fixed coefficients and the uncertainty in the variance parameters for the grouping factors.

https://github.com/paul-buerkner/brms/issues/82

6 Likes