A constrained quadratic spline as a model for the cumulative hazard function

Abstract

Probability distributions traditionally used in reliability analysis (e.g. exponential, Weibull, lognormal, normal, gamma and inverse Gaussian) do not always provide enough flexibility to model real-world lifetime data. Considered in this paper is a quadratic spline with a single free knot as a model of the cumulative hazard function for increasing failure rate distributions. A parsimonious (essentially closed form) solution with a simple numerical search is proposed for estimating spline parameters. Automotive warranty data examples are discussed as an illustration.