In one of the studies, we assumed the "volumetric", chip-allocation task with answers ranging 1-100. We tricked CBC HB by creating the dummy "none". Now we need to estimate chip value of individual levels of some attributes. How to back-translate the resp-level utilities into chips?
I noticed that while utilities vary from -1.6 to +2.0, immense majority lies between -1 and +1. Hence I was considering TANH (arcus tangent) function to trim down the "outliers" and enhance differences in the mainstream. The results seems to be quite satisfactory, but perhaps you can offer a smarter approach?