MODELLING THE INCIDENCE OF POVERTY IN ITALY: A COMPARISON OF THREE STATISTICAL METHODS

This paper is devoted to the problem of “explaining ” poverty in Italy, in the sense of studying its multivariate relationship with individual and background characteristics, such as: age, education, employment, gender, residential zone, etc. The identification of the individual and background characteristics that are associated with its incidence is very important because it can be used for anti-poverty policies. In the literature, alternative measures exist for the definition of the concept of poverty. The two main approaches are: i) the traditional approach that dichotomises the population into poor and non poor people by means of the so called poverty line; ii) the totally fuzzy and relative approach that allows us to analyse poverty in a multidimensional perspective avowing the use of an arbitrary threshold value. In the context of the traditional approach, generally, the statistical method used for studying the relationship of individual poverty status with personal characteristics is a logit or probit analysis for a dummy dependent variable that is a poverty indicator. Pudney (1999) has shown that this approach entails an inconsistency because monotonicity of the poverty rate as a function of the poverty line may be violate