Answers are grounded in this paper and cite their sources.
Ergodicity concerns whether statistical relationships observed in cross-sectional data generalize to intraindividual processes over time. Recent work suggests human behaviour is largely non-ergodic, threatening the validity of translating group-level theories and interventions to individual prediction. In the present article, I propose an epistemological intervention. Specifically, I argue that linear models may be poorly suited to evaluating ergodicity owing to important hidden assumptions. For estimates to be consistent across levels, linear models require not just that the focal relationship X→Y be consistent, but that all variances, covariances and resulting statistical controls be consistent across levels. When any single covariance differs between group and individual levels, the entire coefficient structure changes. I argue that a more viable path to locating ergodic phenomena is to forgo standard linear models in favour of structural relationships. I propose a framework for structural generalizability that utilizes a formal structural operator (Φ) to map raw data onto relational forms that remain invariant to metric properties. I present formal definitions and preliminary empirical examples for three such structures: network topology, partitions and clusters and logical relationships. The proposed framework shifts the ergodicity debate from an ontological question about the nature of groups and individuals to an epistemological question about appropriate levels of analysis. This article is part of the theme issue ‘Ergodicity and ergodicity breaking in the sciences’.
No takes yet. Share an insight, caveat, or question.
Aaron Jason Fisher (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: