Cross-Level/Ecological Inference

This article reports the main methodological approaches to the statistical problem. It describes the fundamental indeterminacy of the problem. It also provides a framework that coherently binds the variety of approaches that have been proposed to address this problem. Then, an overview of these various approaches and their respective contributions are mentioned. The ecological inference problem within the literature of partial identification and the recent work generalizing the use of logical bounds on possible solutions as an identification region for the general r × c problem are explained. It finally covers some admonitions about this fascinating problem that has enthralled decades of scholars from varied disciplines. The analysis by Duncan and Davis made clear that aggregate data only partially reveal the structure of individual behaviour. However, their contribution has largely been viewed as limited and an appreciation for the idea of bounds or an identification region has yet to fully emerge.