Keynote Speakers

Isabel Molina Peralta

Department of Statistics and Operations Research
Faculty of Mathematical Sciences
Complutense University of Madrid

Keynote: Conciliation: the Key to Success in Small Area Estimation

With all my collaborators

Abstract

Conciliation of design-based and model-based approaches to statistical inference is key to success in Small Area Estimation, as it provides a means to “borrow strength” across areas by combining data from different areas and integrating different data sources. Procedures that incorporate the sampling design into model-based estimators are reviewed, both for area means and for more general indicators, including poverty and/or inequality measures. Approaches that reconcile the two frameworks to estimate the mean squared error of model-based small-area estimators are also outlined. Finally, other research directions that combine both approaches are discussed.

Gauri Datta

Department of Statistics, University of Georgia, Athens, GA, USA, and CSRM, U.S. Census Bureau, USA

Keynote: A Bayesian Framework for Multi-Goals Small Area Inference: Estimation, Ranking and Benchmarking

Abstract

In small area inference, estimation of subpopulation means is often the primary goal of national statistics offices. For example, information on poverty, income, and access to primary healthcare at disaggregated levels is required in welfare programs. However, inference on the overall ranking of entities, such as small areas, hospitals, school districts, and other institutions, is equally critical in planning, policy making, and advocacy related to such programs. Ranking draws attention to unusually high or low performing subpopulations and provides investigators and policy makers with useful tools to establish priorities.

In human development, where resources are limited, authorities need to identify the most underprivileged or impoverished subpopulations to deliver relief. Estimates of ranks constructed exclusively from point estimates of parameters lack uncertainty quantification and may lead to imbalances and inequities. This is especially true in small area statistics, where there may be only a limited amount of directly observed data from each area, and the point estimates are subject to large estimation error. To address this deficiency, Klein et al. (2020, JRSS C, 69, 589–606) developed frequentist confidence sets for the rank vector.

As an alternative to this and another recent frequentist solution, we propose a novel Bayesian approach. Our solutions are built on strengths of the Bayesian paradigm: they identify the likely ranks for entities along with a probability distribution on the identified plausible ranks. The proposed solutions significantly outperform the state-of-the-art frequentist alternatives and, unlike their frequentist counterparts, can borrow from covariates as well as from other small areas. We evaluate our proposed Bayesian algorithms in terms of accuracy and stability using a simulation study and two applications of interest to the U.S. Census Bureau.

In a related problem, either by policy necessity or to ensure against possible model failure, model-based estimates of small area means are often required to be modified so that certain well-defined aggregates of these modified values agree with corresponding more reliable and direct benchmark values. Many existing benchmarked solutions are obtained by modifying the regular Bayesian estimates of the small area means in order to comply with the aggregation requirement, but variances of the benchmarked estimates are still computed under the regular model.

Sugasawa et al. (2024, arxiv:2407.17848v1) argued that incorporating the benchmark constraints perturbs the regular posterior distribution of the small area means to a new posterior distribution that automatically satisfies the constraints, produces benchmarked estimates, and more accurately measures uncertainty. Using samples from this modified posterior distribution, we carry out point and interval estimation for the means and ranks of the small areas.

Cristina Boboc

Department of Statistics and Econometrics, Faculty of Cybernetics, Statistics and Economic Informatics, Bucharest University of Economic Studies, Romania

Keynote: Measuring Skill Mismatch in the Romanian Labour Market: Are Small Area Estimation Methods a Solution?

Abstract

Romania’s labour market has a problem that unemployment figures do not capture: the right number of workers, but in the wrong jobs, with the wrong qualifications. Half of workers in elementary occupations are overqualified. Over half of agricultural workers lack formal certification. And these imbalances have persisted for a decade. Yet when a local policymaker asks what is happening in their county, the data simply is not there. This keynote explores what we know about skill mismatch in Romania, where our measurement tools fall short, and whether small area estimation can finally give us the local picture we need.