Abstract:
The purposes of this research were to compare type I error rates and power of the test in completely randomized design for 14 statistics in Post Hoc comparison produres, namely, LSD, Tukey's HSD, Bonferroni Tukey's b, Sidak, Duncan, Scheffe's, Hochberg's GT2, R-E-G-WF, Gabriel, R-E-G-WQ, Waller-Duncan, S-N-K, and Dunnett When each population had equal variances and normally distributed using [alpha]=0.05. Sample sizes of concern were to be equal and unequal. Each category, data were simulated into 3 to 8 groups with small, medium, and large sample sizes. For each statistic under experimental situation, the Monte Carlo experiment was repeated 10,000 times. The finding were summarized as follows: 1. When the sample sizes were equal, LSD and Duncan can control the Type I error rates in every cases. There were seven methods, Sidak, Dunnett, Tukey's Waller-Duncan, S-N-K, Gabriel and R-E-G-WF, can control the Type I error rates in some cases. Furthermore, there were five methods, Bonferroni, Tukey's HSD, Hochberg's GT2, Scheffe's and R-E-G-WQ, cannot control Type I error rates. Moreover, when the sample sizes were unequal, there were three methods, LSD, Waller-Duncan, can control the Type I error rates in every cases. There were five methods, Dunnett, Tukey's b, S-N-K, Gabriel and R-E-G-WF, can control the Type I error rates in some case. In addition, there were six methods, Bonferroni, Sidak, Tukey' HSD, Hochberg's GT2, Scheffe's and R-E-G-WQ, cannot control Type I error rates. 2. In every procedures that leads to find power of the test, power of the test would be increased according to the amount of groups and sizes of sample. LSD and Waller-Duncan methods gave the highest power of the test in every cases when having 3-4 groups. Gabriel method gave the highest power of the test in every cases when having 5-8 groups.