Abstract:
To compare the efficiency of five testing procedures for difference between two population means based on multiple dependent variables : Bonferroni-Holm procedure (BON), James-Holm procedure (JAM), OLS Closed Test procedure (OLS), GLS Closed Test procedure (GLS) and Westfall-Young procedure (WFY) by considering their capacity of controlling probability of type I error and power of the test when both populations have a multivariate normal distribution with the same covariance matrix which equals correlation matrix under the cases of 3, 5 and 7 dependent variables ; 10, 30 and 50 equal sample sizes ; equal and unequal correlation matrix design with correlation coefficient equal to 0.0 to 0.9 at 0.01 and 0.05 significant level (alpha). The data is obtained through simulation using Monte Carlo technique and repeating 500 times for each case. The results of this research can be summarized as follows : 1. The capacity of controlling probability of type I error. The number of dependent variables and design of correlation matrix do not effect the capacity of controlling probability of type I error of all procedures (except that of BON and JAM which decreases very little in case of unequal correlation matrix design), but sample size and significant level effect that of all procedures which increases when sample size or significant level increases. In addition, the capacity of controlling probability of type I error of BON varies inversely with correlation coefficient in cases of low and medium sample size but the capacity of controlling probability of type I error of OLS and GLS varies according to correlation coefficient. 2. Power of the test Power of the test of all procedures varies according to sample size and significant level but varies inversely with correlation coefficient. In case of equal correlation matrix design, power of the test of all procedures does not depend on the number of dependent variables, but power of the test of all procedures varies according to the number of dependent variables in case of unequal correlation matrix design. In most case both OLS and GLS have highest power of the test except some cases of high P and unequal correlation matrix design (including some cases which both OLS and GLS can not control the probability of type I error), JAM have highest power of the test when alpha = 0.01, WFY have highest power of the test when alpha = 0.05. Besides, in case of n equal to 10 and almost all level of p that GLS can not control the probability of type I error, only OLS have highest power of the test.