A1143
Title: A hybrid model for zero-inflated proportion data
Authors: Binyan Jiang - The Hong Kong Polytechnic University (Hong Kong) [presenting]
Abstract: The aim is to propose a novel hybrid approach to model zero-inflated proportion data by capturing two distinct types of zeros based on a regression framework. The first type of zero is attributed to "missing by chance," arising from random sampling or measurement errors. This type of zero is modelled using a binomial sampling process, accounting for the probability of observing a zero value due to chance. The second type of zero is the so-called "true zero", which is attributed to "unsuitability", reflecting the inherent characteristics of the process or phenomenon under study. This type of zero is handled using a general classification indicator, which separates the observations into suitable and unsuitable groups. The resulting model is, therefore, hybrid, with the regression part utilizing weighted least squares to model the "missing by chance" zeros and the non-zero observations and the classification part to identify the true zeros. A two-stage estimation procedure is employed that involves separating the classification part from the regression part. In particular, the optimal classification rule is identified, and as an illustration, a decision rule based on the nonparametric Nadaraya-Watson estimator is constructed. The consistency of the proposed estimation has been established, and the promising performance of the model has been further demonstrated through simulation and real data analysis.
