All Pairwise Comparisons among Means: Newman-Keuls Procedure (4 of 4)

previous


After the significant difference at span 3, differences not previously ruled out at span 2 are tested. The comparison of Means 3 and 4 has been ruled out by the failure of the comparison of Means 1 and 4 to be significant. Therefore, the only comparison left to be performed is between Means 4 and 5. The t is 2.86 which is less than the critical value for a span of 2 of 2.95, so the difference is not significant. The Newman-Keuls procedure has the advantage of being more powerful than the Tukey HSD. It is better at controlling the EER than the Fisher's LSD. However, there are patterns of population means that can lead to an inflated EER. For instance, if six population means were: 10, 10, 100,100, 1,000, and 1,000 then comparisons among sample means at span 2 would almost certainly be performed. The null hypothesis is true for three of these comparisons: Mean 1 versus Mean 2 Mean 3 versus Mean 4 Mean 5 versus Mean 6 Since the PCER for these comparisons is 0.05, the EER is above 0.05.


previous