Identifiability of within-person covariance effects

Identify the within-person covariance structure of the human Rel-A.I. reliance decisions and determine how that dependence changes uncertainty and contrast resolution, given that the released aggregate human rates contain no participant-level records.

Background

The human reference consists only of aggregate domain-by-marker rates, without participant-level choices or covariance information. The paper therefore uses independent-binomial working intervals and explores exchangeable-response correlations as sensitivity analyses. Because dependence among decisions by the same participant can widen cell intervals while narrowing within-person contrasts, the effect of the unknown covariance structure on uncertainty and directional conclusions cannot be recovered from the published table.

References

Positive within-person correlation can widen cell intervals yet narrow within-person contrast intervals; its net effect cannot be identified from the table.

— Can LLM User-Proxies Stand In for Humans in Reliance Measurement? Replicating Rel-A.I. with Simulated Users  (2610.25788 - Wang et al., 19 Oct 2026) in Section 3.5, “Uncertainty and resource measures”; Section 6, “Limitations”