A. Lago, I. Van Keilegom, J. C. Pardo Fernández, J. de Uña Álvarez
Left truncation arises in many applied fields due to the way an experiment is designed or limitations in measurement instruments. It causes observational bias, which yields bias on the estimation. It is a frequent issue to determine whether the target variables from k independent populations follow the same distribution. From an adequate estimator of the density function, a test based on an integral distance between the estimator of the density function in each population and the one of the pooled sample is proposed. The asymptotic distribution is studied and, due to the difficulty of its application in practice, a bootstrap resampling plan is proposed to approximate the null distribution of the test statistic. The choice of the bandwidth will be addressed via Monte Carlo simulations and the proposed test will be compared to the Kolmogorov-Smirnov or the log-rank tests for left-truncated data. The performance of the test will be exemplified with real data regarding pregnancy times.
Keywords: Left truncation, bootstrap, density estimator, k-sample problems
Scheduled
GT09.NOPAR1 Invited Session. High dimension inference
November 7, 2023 3:30 PM
CC2: Conference Room