The negative log-gamma prior distribution for Bayesian assessment of system reliability
This paper presents the negative log-gamma distribution as a prior distribution useful for Bayesian assessment of system reliability. When the scale parameter is held fixed, the negative log-gamma distribution is closed under products, making it convenient for specifying priors for series systems. In particular, for series systems, negative log-gamma component priors can be specified to give an exact desired system prior and vice versa. We consider pass/fail data at the system and component levels for both static and time-varying data collection schemes and propose two new prior distributions for analyzing time-varying reliability. Finally, we consider an application of the negative log-gamma to a missile reliability problem and illustrate diagnostics useful for developing the priors.