Comparison between constant and step‐stress tests for Weibull models

Compares optimum constant‐stress and simple step‐stress ALTs for the Weibull distribution. The purpose is to quantify the advantage of using step‐stress testing in comparison to constant‐stress testing when censoring is likely to occur at the lower levels of stress. Assumes that a log‐linear relationship exists between log scale parameter and stress and that the cumulative exposure model holds for the effect of changing stress in step‐stress test. The variance of the MLE of log scale parameter at design stress is used as the criterion for comparing ALTs. The efficiency of simple step‐stress tests relative to constant‐stress with Type I censoring is studied in terms of the ratio of these variances. Results are given for asymptotic variances and for finite sample sizes using simulation to estimate variances.