Relative Efficiency
The efficiency of the Page test for ordered alternatives when compared with its normal distribution alternative (the t test) is the same as that of the Friedman two-way analysis of variance by ranks test (see Sec.7.2.6). However, compared to the Friedman test, the Page test is more powerful in its ability to detect ordered alternatives.
7.3.5 References
The Page test was proposed by Page (1963). It is related to the Spearman rank-order correlation coefficient (see Chap. 9.). The power of the test has been considered by Hollander (1967).
7.4 DISCUSSION
Three nonparametric statistical tests for testing H0 in the case of k related samples or matched groups were presented in this chapter. The first, the Cochran Q test, is useful when the measurement of the variable under study is categorical (in a nominal scale with two levels or dichotomized ordinal scale). This test enables the researcher to determine whether it is likely that the k related samples could have come from the same population with respect to the proportion or frequency of "successes" in the various samples or conditions. That is, it is an overall test of whether the k samples exhibit significantly different frequencies of "successes" than would be expected by chance.
The second nonparametric statistical test presented, the Friedman two-way analysis of variance by ranks, Fr, is appropriate when the measurements of the variables are at least ordinal. It tests the probability that the k related samples could nave come from the same population with respect to the mean rankings. That is, it is an overall test of whether the values of the data vary as a function of the conditions under which they were observed.
The Friedman test should be used in preference to the Cochran Q test whenever the data are appropriate for its use (i.e., whenever the variables are at least ordered). The Friedman test has the advantage of having tables of exact probabilities for very small samples, whereas the Cochran Q test should not be used when N (the number of rows or sets of observations) is very small.
If, when using the Friedman test, the hypothesis H0 that there is no difference among medians is rejected, multiple comparisons may be done to determine which conditions are different from each other. If the researcher has a more precise hypothesis concerning the difference between one condition (say, a control condition)