A0706
Title: Comparing the power of test statistics suitable for two sample compositional difference tests
Authors: Eva Budinska - Masaryk University Brno (Czech Republic)
Jan Bohm - Masaryk University Brno (Czech Republic) [presenting]
Abstract: Common task in compositional data analysis is to test null hypothesis H_0: There is no compositional difference between two groups. In our research we use Zero-Inflated Generalized Dirichlet Multinomial Model (ZIGDM) and real world data to generate artificial data to estimate power of various tests of H_0. Extensive simulations show that PERMANOVA with suitable dissimilarity metric (Canberra) is superior to other methods, especially tests using CLR transformation.
