Title: Test of mean difference in longitudinal data based on block resampling approaches
Authors: Hirohito Sakurai - National Center for University Entrance Examinations (Japan) [presenting]
Masaaki Taguri - National Center for University Entrance Examinations (Japan)
Abstract: The focus is on a two-sample problem, and propose two block resampling testing methods with permutation analogy for comparing the difference of two means in longitudinal data when the data of two groups are not paired. In order to detect mean difference of two samples, we consider the following four types of test statistics: (i) sum of absolute values of difference between two mean sequences, (ii) sum of squares of difference between two mean sequences, (iii) estimator of area-difference between two mean curves, and (iv) difference of kernel estimators based on two mean sequences. The considered block resampling techniques include circular block bootstrap and stationary bootstrap, and are used to approximate the null distributions of the above test statistics. Monte Carlo simulations are conducted to examine the size and power of the testing methods.