Estimation of the association parameters in hierarchically clustered survival data by nested Archimedean copula functions

Abstract

Statisticians are frequently confronted with highly complex data such as clustered data, missing data or censored data. In this manuscript, we consider hierarchically clustered survival data. This type of data arises when a sample consists of clusters, and each cluster has several, correlated sub-clusters containing various, dependent survival times. Two approaches are commonly used to analysis such data and estimate the association between the survival times within a cluster and/or sub-cluster. The first approach is by using random effects in a frailty model while a second approach is by using copula models. Hereby we assume that the joint survival function is described by a copula function evaluated in the marginal survival functions of the different individuals within a cluster. In this manuscript, we introduce a copula model based on a nested Archimedean copula function for hierarchical survival data, where both the clusters and sub-clusters are allowed to be moderate to large and varying in size. We investigate one-stage, two-stage and three-stage parametric estimation procedures for the association parameters in this model. In a simulation study we check the finite sample properties of these estimators. Furthermore we illustrate the methods on a real life data-set on Chronic Granulomatous Disease.