The one-parameter Weibull distribution as a means to reduce sample size—A cautionary tale

Abstract

With the continued improvement in electronic product reliability the requirement to test larger and larger samples is becoming more and more difficult to manage. As a result methods by which these sample sizes might be reduced are extremely attractive. This paper investigates an approach in which the one-parameter Weibull distribution is assumed to explain the failure mode, and looks at the consequences if this assumption turns out not to be valid. Specifically, the effect on calculated MTTF is examined from the point of view of variation in the shape parameter, and it is shown that massive errors can result for only slight variations in this parameter.