Comparison of Various Methods Used to Determine the Sample Size Requirements for Meeting a 90/90 Reliability Specification

摘要

A common rule of thumb used in binomial-outcome reliability studies is that 22 trials with zero failures are needed to conclude a 90% minimal success rate with 90% confidence. This report explores the origins of this rule and shows how it is overly conservative. Alternative methods for constructing lower confidence bounds for binomial proportions are described which allow the 90/90 reliability criteria to be achieved with as few as 13 trials.