Open samuelsaari opened 3 years ago
I'm feeling pretty pessimistic about my ability to support this. mice
's pooled models do not offer any of the expected functions you'd expect from a "normal" model: no predict()
, no vcov()
, no coef()
, etc. Creating those from scratch could be rather labor-intensive and error-prone (for me). I'm not sure, for instance, whether it makes sense to think of an averaged variance-covariance matrix.
Because this would mostly require work in the jtools
package (its make_predictions()
function is essential to producing the plots in interactions
), I'm going to move it there.
As I check back in on this, I'll mention that the good folks over at mice
seem to be thinking about the problem that was holding me back when I last commented here. I'll keep an eye out for updates from them.
amices/mice#82
Solomon Kurz's blog post may be a useful reference down the line also.
Feature request: cat_plot & interact_plot do not support pooled regression objects (eg. mipo objects). It would be great if it did as there does not seem to be any other package available that would.
More info: https://stackoverflow.com/questions/67635243/plotting-an-interaction-term-in-moderated-regression-using-mice-imputation & https://gist.github.com/samuelsaari/a3be910f57469c7a001ceab9ffef6d6c