Testing for Independence in Multivariate Duration Models

Modelling multivariate failure times in a competing risks setting is often performed by assuming independence between risks. However, by wrongly assuming independence, seriously biased parameter estimates may result. The aim of this paper is to evaluate a test for independence previously proposed in the literature. The test is an information matrix test which is evaluated by Monte Carlo methods. The basic duration model used in the simulations is of the mixed proportional hazards type with Weibull distributed latent failure times. <p> We find the sampling distribution of the test statistic to deviate from the anticipated asymptotic one. However, by applying bootstrap methods it seems that proper critical values can be obtained. Further, the power of the test is found to be highly asymmetrical with respect to the sign of the correlation between risks.