A Methodological Note on Empirical Confidence Intervals for the GCM-Ensemble Mean in Projecting Climate Change Impacts on Health

The study addresses the methodology of constructing confidence intervals for climate change health impact projections using multiple general circulation models (GCM). Researchers found that commonly used 95% empirical confidence intervals (eCI) constructed from the 2.5th and 97.5th percentiles of individual model impacts do not provide the expected coverage level for the GCM-ensemble mean. A simulation study demonstrated that the coverage of empirical intervals deviates from the nominal level in both directions, whereas a new aligned interval achieves coverage near 95% across all examined settings. Analytically derived exact coverages under a location-shift model agree with simulation results. In a reanalysis of heat-related mortality projection in London, empirical intervals were consistently wider. The study recommends distinguishing between empirical intervals and confidence intervals for the GCM-ensemble mean.