A0961
Title: Tests for left-truncated and right-censored data
Authors: Juan-Carlos Pardo-Fernandez - Universidade de Vigo (Spain)
Jacobo de Una-Alvarez - Universidade de Vigo (Spain)
Adrian Lago - Universidade de Vigo (Spain) [presenting]
Abstract: The comparison of distributions is not only a classical task in statistics but also a useful tool employed in many applied fields. A great variety of tests based on different quantities or functions related to a random variable have been developed for complete data. Regarding failure times, it is also common not to know precisely the time when an individual experiments with the event of interest. Such an individual is said to be censored. In addition, it may also happen that the individuals cannot be included in the study since the event of interest occurs before the observation time. This phenomenon is called left truncation. Both censoring and truncation yield biased estimators, which implies that usual inferential methods are no longer adequate. In particular, not taking into account truncation and censoring causes inconsistent tests. On top of that, the literature on tests for left-truncated and right-censored data is vaguely developed, with the well-known rank-based tests the only option to tackle such a task. The aim is to propose a test to address the comparison of populations with left-truncated and right-censored data. Its asymptotic null distribution will be studied. As an alternative, a bootstrap resampling plan will be proposed to approximate the null distribution of the test statistic. The proposed method will be studied via Monte Carlo simulations. Finally, the test will be compared to the classical log-rank test.
