Clustered data often contain subgroups whose sizes vary within clusters. This paper formalises that variation as "informative subgroup size" (ISS): the composition of subgroups inside a cluster can carry information about the marginal distribution of outcomes for the units in those subgroups.
The authors build an estimator for the marginal association between paired outcomes in such clustered settings. They demonstrate it on data from the National Health and Nutrition Examination Survey, analysing caries and periodontal disease together — two dental conditions that naturally occur in clusters of teeth within individuals.
This digest draws on a single arXiv preprint, so there are no conflicting sources to compare. As a preprint, the work has not yet undergone peer review, and the findings should be treated as preliminary.