It is increasingly recognized that good reliability practice requires giving attention to the specification of joint distributions for failure times. If the marginal distributions are known then the problem of selecting a marginal distribution is equivalent to that of selecting a copula. In the competing risk situation, identification of the copula alone is important, for together with competing risk data it enables identification of the full joint distribution. In a number of papers with various co-authors, we have suggested using the minimally informative copula with given rank correlation. In this paper a method is described of assessing minimally informative copulae using observable quantities, which gives expert guidance on the ranges allowed for coherence. This is based on using a D1AD2 algorithm instead of the DAD algorithm used to determine copulae with given rank correlation. 1 Building subjective multivariate distributions There is a growing literature on building multivariate distributions, see for example (Joe, 1997). We are particularly interested in building models with fixed marginals.
[1]
C. Genest,et al.
The Joy of Copulas: Bivariate Distributions with Uniform Marginals
,
1986
.
[2]
Jonathan M. Borwein,et al.
Entropy minimization, DAD problems, and doubly stochastic kernels
,
1994
.
[3]
Adrianus M. H. Meeuwissen,et al.
Minimally informative distributions with given rank correlation for use in uncertainty analysis
,
1997
.
[4]
John Quigley,et al.
Eliciting engineering knowledge about reliability during design‐lessons learnt from implementation
,
2001
.
[5]
T. Bedford,et al.
Vines: A new graphical model for dependent random variables
,
2002
.
[6]
John Haigh,et al.
Probabilistic Risk Analysis: Foundations and Methods
,
2003
.