Computation of Reliability Statistics for Success-Failure Experiments
Reliability is probability of success in a success-failure experiment. Confidence in reliability estimate improves with increasing number of samples. Assurance sets confidence level same as reliability to create one number for easier communication. Assuming binomial distribution for the samples, closed-form expression exists only for calculating confidence. The Wilson Score interval with continuity correction provides approximate closed-form expression for reliability. Brent's method was found to provide fast and accurate estimate for both reliability and assurance computations. Graphs and tables are provided for several number of samples. Two open-source python libraries are introduced for computing reliability, confidence, and assurance.
READ FULL TEXT