Part VI: NonParamteric Analogues
Rank Order Analogues to the SRD ANOVA
| The Kruskal-Wallis Test | Other Techniques |
 
 
This test is essentially a One-Way ANOVA with ordinal data. Another way to conceptualize this test is that is like doing a Mann-Whitney U with more than 2 groups.

This test is performed through the following steps:

  • As with the Mann-Whitney U, all of N scores are ranked, regardless of group membership.
  • Next, the ranks of the scores for each group are summed independently.
  • H, the test statistic, is derived using:

Picture (261x106, 2.7Kb)

  • When there are more than 5 people in each group, a chi square is used. Compare the value of H to a critical value of chi square (where df = number of groups -1) at the appropriate level of alpha. If the absolute value of H is greater than the critical, reject the null hypothesis.
 
 
Two other possibilities exist for the multiple group/ordinal dependent variable situation:
  • The Jonckheere Method: This is used for situations in which the independent variable is ordinal. (For Kruskall-Wallis, the IV is typically nominal, as it is in ANOVA). This method is, however, essentially the same as using Kendall's Tau (the measure of magnitude) corrected for tied scores.
  • The Ager-Brent Test: This test is useful in situations in which it is possible to partially order the groups that make up your independent variable (not really nominal, but not yet ordinal).

Neither of these tests is commonly used.