Title: Assessment of zero inflated mixture models for ordinal data
Authors: Maria Iannario - University of Naples Federico II (Italy) [presenting]
Rosaria Simone - University of Naples Federico II (Italy)
Abstract: Excess of zeros is a commonly encountered phenomenon that limits the use of traditional regression models for analysing ordinal data which exhibit zero inflation. These data concern contexts where respondents express a graduated perception on a specific item or experiments identify levels of increasing assessments including zero (subjects are not susceptible to express the response). Specifically, the zero counts could be simply absent in the rating of respondents for the absence of the symptom or activity (structural zeros) or present with low frequency not observed because of sampling variation (sampling zeros). The focus is on modelling the data by means of zero-inflated mixture models by taking into account both excess of zeros and heterogeneity. It has been designed so as to discriminate between structured and unstructured zeros by placing particular emphasis on the uncertainty concerning the evaluation process. The performance of the proposed model is assessed through simulation studies and survey data.