The Effect of Baseline Selection in Hotelling’s T Two Sample Test for Shape Data

Hotelling’s T two sample test based on Bookstein coordinates or Kendall coordinates has been applied to test the difference between two groups of shape configurations. While this method has several advantages, some empirical evidences have shown that it is sensitive to the variation and correlation within landmarks and the selection of baseline. It is typical to select the two landmarks that form the baseline as far apart as possible when the variation and correlation within landmarks are both very small. However, it is not clear how a baseline should be selected when the variation and correlation within landmarks are large. The purpose of this study is to investigate the performance of the T statistic for different baseline selections under a range of variation and correlation within landmarks. A simulation study is conducted to investigate the effect of the baseline selection to the T statistic under different conditions of variation and correlation within landmarks. Some existing shape data will be analyzed based on the simulation findings.