B1369
Title: Modelling unbalanced hierarchical survival data using HAC-copula functions
Authors: Roel Braekers - Hasselt University (Belgium) [presenting]
Abstract: A copula model is introduced for hierarchically nested clustered survival times in which the different clusters and sub-clusters are possibly unbalanced. Due to the right censoring we do not fully observe each outcome variable. This, together with the hierarchical structure of the data, makes it very difficult to set-up a full likelihood function for a general copula model. To circumvent this problem, we focus to on the class of hierarchical nested Archimedean copula functions and use the properties of this copula family to simplify the full likelihood function. For the marginal survival time, we introduce both a parametric regression model and a semi-parametric Cox's regression model. Since maximizing the likelihood function for all parameters is computational difficult, we consider a two-stage estimation procedure in which we first estimate the marginal parameters and afterwards, estimate the association parameters. As a result, we obtaine the asymptotic consistency and normality of the association parameters. Next we compare the finite sample behaviour of the different estimators through a simulation study. Furthermore we illustrate the estimators on a practical example on the insemination time of dairy cows.
Journal