Title: Estimating dependence patterns in right-censored event time data using R-vine copula models
Authors: Nicole Barthel - Technische Universitaet Muenchen (Germany) [presenting]
Paul Janssen - Hasselt University (Belgium)
Candida Geerdens - Hasselt University (Belgium)
Claudia Czado - Technische Universitaet Muenchen (Germany)
Abstract: In many studies interest is in the time to a predefined event. Due to limited follow-up, instead of the true event times lower right-censoring times might be recorded for some sample units. The resulting lack of information has to be carefully taken into account by inference tools applied to right-censored data in order to arrive at a sound statistical analysis. If for the sample units multiple event times can be observed, the data might further exhibit complex association patterns, which claim elaborate dependence models. For this purpose, the flexible class of R-vine copulas was extended to right-censored event time data. We illustrate novel vine copula based estimation methods through several right-censored data examples: e.g. dependence between times until infection of the four udder quarters of cows is investigated. All four observation units are subject to right-censoring. To analyze data on recurrent asthma attacks in children, the subclass of D-vine copulas is used to capture the inherent temporal dependence. Additional challenges are unbalancedness of the data and dependent right-censoring induced by the serial data nature. Further, an outlook on ongoing research including R-vine copula based quantile prediction and quantile regression for right-censored data is given.