A1499
Title: Tree-based learning of structural changes in mixture models with uncertainty for rating data
Authors: Rosaria Simone - University of Naples Federico II (Italy)
Carmela Cappelli - University of Naples Federico II (Italy) [presenting]
Abstract: Qualitative data on individuals perceptions and subjective evaluations are collected through officialsurveys using rating scales, often administered repeatedly over multiple time waves. In many cases,only aggregated response distributions are made publicly available, with no access to individual-leveldata. In this context, we combine Atheoretical Regression Trees with CUB models to analyze ratingdata and identify structural changes over time in the main characteristics of the response distributions.The chosen modeling framework is well-suited to parameterize both the latent feeling and theuncertainty underlying the ordinal evaluation, with a time-varying structure. Then, the use ofAtheoretical Regression Trees involves introducing an artificial covariate that preserves temporalordering, enabling the segmentation of feeling and uncertainty measures into homogeneous timeintervals, which can then be interpreted in light of political and socio-economic events. We illustratethe proposed approach using data from the Consumer Confidence Survey issued monthly by theItalian National Institute of Statistics (ISTAT), assessing whether and to what extent individualschanged their judgments and expectations regarding price levels from 1994 to 2019. The proposedmethod demonstrates satisfactory performance when compared with the Bai and Perrons test forstructural change detection, which is used as the benchmark.
