The abstract is read at the publisher; the summary is JClub's.
A 95% empirical confidence interval for the ensemble mean of GCM impacts does not reliably cover the mean: coverage can approach 1 with <40 GCMs or 0 with >40.
- Why it matters: Health-impact projections often report ensemble means with eCIs, but the interval’s nominal level may not match its true coverage for the intended estimand, risking misleading uncertainty.
- What they did: The authors define estimands/predictands and analyze eCI coverage as observed-data sample size grows with fixed GCM count and infinitely many coefficient draws, including simulations and a London reanalysis.
- The result: For the mean over considered GCMs, coverage converges to 1 with fewer than 40 GCMs unless impacts are all equal, and can converge to 0 with more than 40; in simulated heat-related mortality, 95% eCI coverage was ≥0.99 in 22 of 26 settings.
The findingWhy it mattersWhat they didThe result
AI summary of the abstract · may contain errors · not medical advice · read the paper.