Estimating overdispersion in non-transformed scores of ANVUR’s department ranking procedure: a Beto-multinomial model

We introduce the Betoidal distribution and show how, in the absence of coarsening, maximum likelihood (ML) estimation for its characteristic parameter σ would match the well-established Normal framework. Then we provide details about how coarsening is handled and describe the ML estimation of the Beto-multinomial model. Such a model is then applied to the 2017 ANVUR ranking of academic department, for which results are presented. Some concluding remarks are given, and possible directions for adjusting the overall performance indicator ISPD as well as for future developments are sketched.