Proposal Description

Small Area Estimation (SAE) has become an indispensable set of tools for generating reliable statistics at finely disaggregated geographical or demographic levels. As the demand for localized, policy-relevant information grows—driven by needs in public health, poverty assessment, environmental monitoring, and social policy—the methodological development of SAE continues to expand. Producing accurate small-area indicators is crucial for identifying vulnerable populations, targeting social programs, designing public health interventions, and supporting evidence-based governance at regional and community scales. This invited paper session brings together four recent contributions that exemplify the innovation, diversity, and applied relevance of contemporary SAE research.
The session begins with the presentation “Small area estimation of categorical indicators using finite mixtures of multinomial logistic regression models” by Maria Francesca Marino (University of Florence, IT), co-authored with Maria Giovanna Ranalli. The contribution focuses on estimating categorical indicators—such as multidimensional poverty measures—by integrating finite mixture models into multinomial logistic regression frameworks. This approach allows the model to better reflect latent heterogeneity across areas, correcting for unobserved structures often present in social indicators and improving predictive accuracy.
The second presentation, delivered by María Bugallo (Miguel Hernández University of Elche, ES) and co-authored with Alexandro Aneiros-Batista and María José Lombardía, is titled “M-quantile Area-level Models for Robust Small Area Estimation without Reliance on Design-based Variances.” This work proposes an alternative area-level SAE framework that avoids dependence on design-based variance estimators. Such estimators are often unstable or unavailable in practical contexts, especially when using administrative data, non-probability samples, or complex survey designs. The M-quantile approach offers increased flexibility while maintaining interpretability, extending the applicability of SAE to data environments where traditional assumptions do not hold.
In the third talk, “A two-part small area model that accounts for heaping to map smoke habits,” Aldo Gardini (University of Bologna, IT) and Lorenzo Mori address a pervasive issue in self-reported health and behavioral data: heaping. Individuals often round or cluster their responses, distorting the distribution of key variables such as smoking intensity. The proposed two-part model explicitly accounts for both the underlying behavior and the reporting mechanism, enabling the production of more accurate and credible local-level indicators. This contribution demonstrates how carefully designed SAE models can significantly enhance public health surveillance and policy planning.
The final presentation, “Small area estimation in the context of non-probabilistic samples: applications and methodological considerations,” by Angelo Moretti (University of Utrecht, NL), tackles one of the most pressing contemporary challenges: producing valid small-area estimates from non-probability samples. As administrative data, web panels, and other big-data sources proliferate, statistical systems increasingly confront data collected without controlled sampling designs. This contribution outlines methodological strategies, practical issues, and applied examples that illustrate how SAE can extract meaningful information from such data sources.
Together, the four presentations highlight the vibrant evolution of SAE and its growing relevance in addressing the statistical needs of public health and social policy. The session showcases innovative methodologies that respond to real-world challenges, reinforcing the importance of SAE in delivering actionable, high-quality information where it is most needed.