Proposal Description

Despite the rapid growth of large-scale digital data sources and the increasing capacity of AI systems to generate synthetic information, many applied domains continue to grapple with scarce, incomplete, or only partially labeled data. Missingness remains one of the most pervasive and consequential challenges in statistical modelling and machine learning. It manifests in many forms: unobserved or selectively observed covariates, structured or non-random missingness in causal inference, limited supervision in semi-supervised learning, incomplete features in predictive modelling, and small or highly imbalanced samples in small area estimation and official statistics. In each of these contexts, the quality of inference and prediction critically depends on how missingness and data scarcity are addressed.
Over the last decade, an impressive range of methodological developments has emerged to confront these issues. These include robust weighting and calibration techniques, flexible imputation and multiple imputation frameworks, data augmentation and generative modelling strategies, likelihood-based and Bayesian approaches, and hybrid methods that integrate statistical modelling with machine learning tools. Many of these techniques have been developed within distinct research communities—such as causal inference, survey sampling, semi-supervised learning, econometrics, and computational statistics—yet they share common conceptual foundations and face analogous theoretical and practical challenges.
This invited session aims to bring together leading researchers working on different facets of scarce and missing data to present recent methodological breakthroughs and discuss their implications for applied work. The session will highlight connections across domains that are not always in direct dialogue, emphasise the unifying principles underlying modern approaches, and explore the advantages and limitations of current strategies in real-world settings.
By fostering cross-disciplinary exchange, the session seeks to provide a more integrated perspective on missing and scarce data problems, promote collaboration across methodological traditions, and identify promising directions for future research — both in theory and in application.