| Popis: |
With all the multiple comparison tests cited above, the logical question is, "Which procedure is best?" The answer to this question depends on whether an individual's decisions regarding differences should be conservative or nonconservative. The most powerful tests (LSD and Duncan's) are nonconservative (ie, more likely to detect a significant difference when it doesn't exist). Scheff's S method is so conservative that it is not useful when pairwise comparisons are desired and is more appropriate for "data dredging," that is, looking at contrasts suggested by the data. Waller and Duncan have proposed a k ratio rule to determine whether one needs to use a conservative rule (such as Tukey's) or a nonconservative rule (such as Fisher's LSD). In practice, the SNK procedure is commonly used because it is a good compromise between power and conservatism. |
