|1.||How can a single experiment produce multiple false alarms? Multiple false alarms and multiple misses?|
|2.||In the numerical example of Student–Newman range test, verify that the final outcome is the separation of the means into the following three subsets, each of which contains means that are not statsig different.|
|3.||You test four experimental conditions but the overall Anova falls somewhat short of statsig. However, your research assistant points out that your four experimental conditions form a clear a priori rank order and suggests that a linear trend test would be most effective. What do you do?|
|4.||a. Show that the one-for-two rule of Section 17.3.2 yields α3 =.074 for a = 3 conditions. Compare with Student–Newman procedure.|
|5.||In the familywise test of inverted-U shape (Section 17.4.2):|
|6.||With one-way design, a statsig range always demonstrates a two-mean difference, whereas a statsig F says only that not all true means are equal with little information about which means differ from which. Hence a range procedure such as Student–Newman might seem preferable to the overall F. Although the range procedures may have a little less power in most situations, they go beyond the overall F to say which means differ from which.|
So: Why not make range procedures standard and forget about overall Anova? As a bonus, much material of previous chapters could be omitted; learning statistics would be much easier. What reason can you see for retaining the overall F for one-way designs?