At the end of Sawtooth's "Tip's & Tricks of Key Driver Analysis" video from a couple of months ago, someone asked if Sawtooth's Johnson's epsilon will allow for binary outcomes, and the answer was along the lines of "not immediately, but eventually".
Is it inappropriate to use Johnson's epsilon for binary outcome variables? Or binary predictors? Can the issue be solved by using a different correlation metric for the correlation matrix that is used to calculate Johnson's epsilon?