Modelling a proportion response variable using generalised additive models for location scale and shape

Abstract

In this paper two alternative approaches are proposed to model a response variable Y measured on the interval from zero to one, including both zero and one. The first proposed model employs a flexible four parameter distribution for 0 < Y < 1, for example a logit skew exponential power distribution, inflated by including point probabilities at 0 and 1. The second proposed model is a generalised Tobit model, obtained from a flexible four parameter distribution on (-infinity;+infinity), for example the skew exponential power distribution, by censoring below 0 and above 1. The proposed models are applied to a real data set and compared with current popular models.