Li and Zhang (2010) considered the problem of construct
asymptotic confidence interval for the ratio of means of two two-parameter exponential distributions. Kharrati-Kopaei
et al. (2013) consider simultaneous fiducial generalized confidence intervals for differences of the location parameters of
several exponential distributions under heteroscedasticity. In the quality control study and the experimental design, a more
important parameter of interest is the mean lifespan of certain products. For example, it is known that the product quality
directly affects the competitive advantage of an enterprise in the market. The quality of the product and its lifespan are closely linked. Ifweassume that several component’s life of a mechanical system are all follow life distribution, it is necessary
for us to compare the mean life of these parts, timing to replacement and maintenance of these components to ensure the
reliability of the product; In experimental design we often consider comparing the life of one or more reference products
or one of more test products. Therefore, all pairwise differences of mean life of two or three products have become the
urgent problem to address. It is typically assumed that the product life follows a two-parameter exponential distribution
Exp(μ, θ), thus its mean life is δ = μ + θ, and the question of interest is to compare the differences δ’s from several such
distributions. Surprisingly, to the best of our knowledge, the literature seems scant in this area. In this paper, we will try to
fill this void by constructing simultaneous confidence intervals (SCIs) for differences of two-parameter exponential means
using a parametric bootstrap (PB) method. For more about the two-parameter exponential distribution family, see Lawless
(1982), Maurya et al. (2011) and the references therein.
The bootstrap approach is a computer method frequently used in applied statistics, which is a type of Monte Carlo method
applied on observed data, see Efron and Tibshirani (1993) for more information on this important topic. The bootstrap
method can be carried out in either parametric or nonparametric setting. However, the question addressed in this paper is
in a strict parametric setting, therefore a parametric bootstrap approach will be constructed accordingly.
The paper is organized as follows. The proposed parametric bootstrap method for constructing SCIs for differences of
means of several two-parameter exponential distributions is presented in Section 2, and a theorem on the asymptotic
correct coverage probability of the PB SCIs is given in Section 3. In Section 4 simulation results are present to evaluate
the empirical coverage probabilities and average volume of the proposed method in comparison to the fiducial generalized
simultaneous confidence intervals (FG SCIs) and a nonparametric bootstrap simultaneous confidence intervals (NPB SCIs)
procedures. Some concluding remarks are made in Section 5 and the FG SCIs and NPB SCIs methods are briefly described in
the Appendix.