Analysis of Interval-Censored Data From Fractionated Experiments Using Covariance Adjustment

Abstract

Censored data are commonly observed in industrial experiments such as for life testing and reliability improvement. Analyzing censored data from highly fractionated experiments presents a challenging problem to experimenters because many traditional methods become inadequate. Motivated by the data from a fluorescent-lamp experiment, we consider in this article analyzing censored data from highly fractionated experiments using covariance adjustment based on multivariate multiple regression models, which make use of the joint distribution of multivariate response variables. The Bayesian approach is taken for the main statistical inference. The posterior distribution of the parameters is obtained using the data augmentation algorithm. We illustrate the methodology with the fluorescent-lamp experiment data. With the real example and a simulation study, we show that covariance adjustment can lead to both dramatic variance reduction and possible bias reduction.

KEY WORDS:

• Bayesian methods
• EM algorithm
• Fractional factorial design
• Interval censoring
• Invariant prior
• Markov chain Monte Carlo