I find it much more believable that I could estimate both nuisances consistently, but at slower rates, vs that I could pick 2 parametric models (without looking at data) & happen to get one exactly correct

Comments