More detailed histograms when only a small amount of data is outliers
This is a suggestion from Rob Levy.
Where, there are big differences at a few points, but 99% of points are within ±0.01. This is where signed(exponent(logged(differences)))) would be extremely helpful (your logit scale?)
X = Max(Absolute difference) A = Min(absolute difference) that is not zero Range = log(X) – log(A) Divvy up range into 25 (or fewer is probably okay) intervals of equal log Final plot will have 25 intervals on left (<0.0), 25 intervals on right (>0.0), plus one with exactly 0.0. Now you can have histogram!