A0665
Title: Statistical modelling of variables on the unit interval (the closed interval between zero and one)
Authors: Jonas Andersson - Norwegian School of Economics (Norway) [presenting]
Abstract: Variables representing fractions sometimes have values that are exactly zero or exactly one. This prohibits the direct use of some probability distribution, e.g., the beta distribution, as an appropriate representation. I will discuss the modelling of such variables. This is exemplified by using the experimental data collected by a prior study. Some aspects of the drivers of the declaration rate are shown by using a zero-one inflated beta regression model. It turns out that some of the drivers of the declaration rate affect the three parts of the declaration rate distribution, the zero declarers, the full declarers and the intermediate declarers, differently. It is found that the effect of taxpayers' monitoring, i.e., their knowledge about other taxpayers' evasion, increases the probability of declaring zero. Among the individuals declaring a part of their income, the effect is significantly positive; they declare more. Another result is that, for the average experiment participant, the probability of fully declaring or declaring nothing of income increases as the experiment progresses. Furthermore, time dependence in models on [0,1] is discussed.
