Automotive component reliability: should it be measured in time or mileage?

This paper is a sequel to the work published in 2004 RAMS Transactions on the nonparametric estimation of the marginal in time and mileage failure distributions. Depending on the failure mode, accumulated time in service (hereafter just time) and mileage could be either a survival variable (i.e. the one "driving" failures and representing failure distribution) or a cumulative use variable. The question is "Which is which?" Proper resolution of this dilemma with subsequent data analysis helps classify the underlying failure mode in terms of assembly-related failures, wear-related failures, etc., which, in turn, is critical to developing the appropriate corrective actions? We review a few available statistical procedures devised to identify a "good" scale to model the failure distribution. We propose, however, that these procedures must be used with caution in the automotive industry applications, since the variability of mileage accumulation of nominally identical vehicles in a population can mislead the choice of the appropriate survival variable. We conclude that most important criterion for selecting the appropriate survival variable is engineering relevance of the failure mode