Nonparametric bayesian estimation of a survival curve with dependent censoring mechanism

Summary Let (X ₁,Y ₁),..., (X _N ,Y _N ) be a random sample from a bivariate distribution functionF. Based on observing only (δ _i ,Z _i ) whereδ _i =1 ifX _i ≦X _i and =0 otherwise andZ _i =min{X _i ,Y _i } fori=1,...,n, we obtain the Bayes estimator ofF whenF is a Dirichlet process under the usual integrated squared error loss function. It should be pointed out here thatX _i andY _i neednot be independent which is the usual assumption in survival analysis models. The effect of this dependence can be seen clearly in the estimators obtained and also in the given example which illustrates the estimator when Freunds' bivariate exponential distribution is taken as the parameter of the Dirichlet process. The research of this author is supported in part by an NSF Grant MCS80-03244. The research of this author is supported by the NIH Grant NO: 1 R01 GM 28405.