Bias adjustment

[strengejacke]

See easystats/modelbased#57
Would be cool to have such a feature (if it's no yet implemented?)

[vincentarelbundock]

That's interesting. In marginal effects, the 4 core functions and quantities are: marginaleffects, comparisons, predictions, and marginalmeans.

In general, we can't really do back-transformations on contrasts or marginal effects: #390

In predictions() the back transformations are delegated to insight::get_predicted, so we would have to implement it there.

I guess that means that the bias correction could only potentially be implemented in the marginalmeans() function. In that context, I think it would be very easy to pass a function to the transform_post argument: https://vincentarelbundock.github.io/marginaleffects/reference/marginalmeans.html

Any function supplied to this argument will be applied to the estimates and confidence intervals at the very end of the computation process, which I guess is what we need here. We could also possibly include a string shortcut like transform_post="whatever" to make things easier for the user.

Frankly, I'm not sure I'm expert enough in this bias correction to implement it correctly. But I'm definitely open to the idea.

[strengejacke]

Frankly, I'm not sure I'm expert enough in this bias correction to implement it correctly

Damn, that was the main reason I opened an issue here... 😉

[vincentarelbundock]

Will be happy to continue the conversation if someone has an idea for how to implement this in marginaleffects, but since I don't have the knowledge, I'm unlikely to do this myself. So going to close the issue to keep things tidy in the repo, but feel free to re-open or keep the conversation going if you have thoughts.