Evaluating product and process design changes with warranty data

This paper suggests evaluating field performance by analyzing hazard plots of the random variable time to first warranty claim. This approach corrects biases that arise when a single member contributes many repairs to the repair population and allows comparing the empirical lifetime distribution with the parametric model assumed at product design (e.g., Exponential or Weibull). Using automobile warranty data we demonstrate how using too broad a scope can mitigate changes in field performance by aggregating improving and degrading components. When comparing warranty performance across model years, the analyst should adjust the hazard plots to reflect common field service time.