A1337
Title: A general semiparametric model for zero-inflated fractional data
Authors: Yangzi Zheng - The Hong Kong Polytechnic University (China) [presenting]
Abstract: A frequent challenge encountered with production data is the interpretation andanalysis of the data having a high proportion of zeros. Much attention has been givento zero-inflated count data, whereas models for non-negative continuous data with anabundance of zero are much fewer. We consider zero-inflated fractional data and provide modeling to capture two types of zero in the context based on the regression model.We model zero due to missing by chance through independent binomial specificationand zero due to perfection using a general classification indicator. We solve the objective function by iterations that separate the classification part and regression part.We specify models hierarchically, applying the weighted least square for the regressionpart, and take advantage of the Nadaraya Watson estimator for the estimation of theclassification part. Our motivating dataset consists of a large amount of zero as wellas propositional data, which indicate the defective rate of some products in the clothesindustry. We nd that environmental features enable learning about both types of 0sas well as a positive percent rate. The proposed modeling enables the industry to extracta better understanding of factors of production due to perfection vs. missingness bychance which could help them to reduce the losses.