Collective wisdom: Methods of confidence interval aggregation

We report the results of a meta-analysis study of the relative accuracies for a range of methods for aggregating confidence interval estimates of unknown quantities. We found that a simple “trim-and-average” method—that is, remove outliers and then average—produced the most accurate estimates. Our results show that more complicated methods of confidence interval aggregation, which factor in confidence levels and estimate imprecisions, do not produce estimates more accurate than those produced by the simple trim-and-average method.