The folded exponential transformation for proportions

Abstract

Summary. Several transformations have been suggested for proportions x. Most of these require that 0 < x < 1 and therefore fail when x = 0 or x = 1, which is quite common in practice. The paper proposes a transformation which allows x = 0 and x = 1. It is based on an extension of the exponential transformation and works well for moderately skewed unimodal distributions, whereas J-shaped or very skewed unimodel distributions are less well accommodated. Use of the transformation is exemplified with three data sets.