Graphical analysis and guidelines for step-stress testing

Two graphical techniques for analyzing step-stress Arrhenius exponential data are presented and compared to maximum likelihood estimation using simulated data. Analysis shows that the graphical technique using least squares estimation is unbiased. Levene's test for equality of variance also shows that the least squares estimator is as efficient as the maximum likelihood estimator. An analysis of outliers provides further evidence that the graphical technique is a valid alternative to maximum likelihood estimation. Indeed, both the maximum likelihood estimator and the least squares estimator are similarly affected by stress-steps that contain no failures. These results, in conjunction with design of experiment analysis are used to formulate guidelines for conducting step-stress testing. A novel algorithm is also proposed that identifies when to switch to the next higher stress-step in order to avoid too many failures at the lower stress-steps. Finally a worked example is included to illustrate the graphical technique.