Updated stats guidance favors Bonett-Price for small-sample AI evals

New guidance on small-sample AI evaluation compares 95% confidence intervals, paired p-values and FWER correction, finding the Bonett-Price method outperforms previous recommendations for paired binary data.

2026-08-27 ~ 2026-08-27 · 2 related posts

• Better Statistical Methods for Small-Sample AI Evaluations — IanArawjo · 2026-08-27
• Guide to Statistical Methods for Small-Sample AI Evals — IanArawjo · 2026-08-27