Strategies and statistical evaluation of Italy’s regional model for COVID-19 restrictions

Abstract This study presents a comprehensive assessment of the Italian risk model used during the COVID-19 pandemic to guide regional mobility restrictions through a colour-coded classification system. The research evaluates the variables selected by the Italian Ministry of Health and their effectiveness in supporting public health decision-making. The analysis adopts a statistical framework which combines data reduction and regression modelling techniques to enhance interpretability and predictive accuracy. Dimensionality reduction is applied to address multicollinearity and simplify complex variable structures, while an ordinal regression model is employed to investigate the relationship between the reduced set of variables and the colour regional classifications. Model performance is evaluated using classification error metrics, providing insights into the adequacy of the selected variables in explaining the decision-making process. Results reveal significant redundancy among the chosen variables, suggesting that excessive predictors may compromise information by reducing the efficiency and clarity of the model. To address this, the study proposes refined and robust predictive models for regional classification, and corresponding empirical distribution of the model parameters, offering a reliable tool of the proposed framework and to support public health decision-makers. By highlighting methodological improvements and offering a generalizable approach to multi-indicator risk classification, this study contributes to the ongoing development of quantitative tools for evidence-based policy-making. The findings provide insights that can inform regional health strategies and support more effective, data-driven decision-making during health crises. Although applied here to the Italian COVID-19 risk classification, our framework is a generalizable methodological approach that can be applied to any context where multiple indicators are aggregated into ordinal risk levels.