Evaluation Of Multiple Comparison Methods With Quantitative Data

Objective Six common procedures of Multiple Comparisons were used in independent sample treatment means of tests, in order to give references to select the best method for analysis of quantitative data and the disciplinarian of these Multiple Comparisons. Methods In order to carry out the above mentioned goals, this study, using Monte Carlo type simulation experiment, are based on four areas: the condition of different distribution, number of experimental units within each group(balanced, partial balanced, null balanced), configuration of means, and the variance of each population(equal, unequal). In this study used LSD, Duncan, SNK, Bonferroni, Scheffe, Sidak methods did hypothesis test, using the Type I error rate, Type II error rate, power of test as the index of evaluation. Results All these six multiple comparisons are sensitive to unequal population variance. The comparison error rate of LSD and Duncan is equal toα, but their experimental error rate is large toα. The comparison error rate of SNK, Scheffe, Bonferroni, Sidak are smeller toα, as well as experimental error rate. The power of test of SNK, Scheffe, Bonferroni, Sidak are smeller to LSD and Duncan's. When unbalance design with small sample size, the power of SNK, Scheffe, Bonferroni, Sidak are little small. Conclusions In different situations analysis methods of selecting statistical should be different in order to interpret data reasonably in practice. When the first aim is to explore the difference between the means of each groups, the LSD and Duncan methods are first selected. When the aim is control the error rate of hypothesis test, the SNK, Scheffe, Bonferroni methods are first selected.