A Note On Experiment Design For Accelerated Life Tests.

A statistician would be considered extremely incautious if, on designing an experiment to determine the regression of y on x, she/he sampled only from two values of x and ignored the possibility of nonlinearity. The point of this note is to point out that even within a limited class of physical models, the same is true of a statistician who designs an accelerated life test in which devices (e.g., capacitors, lasers) are distributed among several levels of accelerating stresses but no single device ever sees more than one level of stress (a constant stress accelerated life test). For that limited class of physical models, a class of sufficient designs (sufficient in the sense of making all potentially identifiable parameters identifiable) is characterized.