Kruskal-Wallis test Flashcards
What are key features of this test?
Non-parametric alternative to one-way anova for independent measures
Used to compare 3 or more groups
Data ranked before calculations performed
What do results of this test tell us?
How likely it is that the differences between ranks for each condition would occur under the null hypothesis (which claims that the ranks were drawn from identical populations i.e. intervention having no effect)
How is the test calculated?
Rank all scores, as if they belong to one single group
Then add the separate rank totals up for each group
If a real difference exists then ranks for one condition will be systematically higher or lower than those for another condition
Why is it not possible to have a one-tailed hypothesis with this test?
It only tells us WHETHER there is a significant difference between conditions, not what that difference is
How do you calculate degrees of freedom in this test?
Number of conditions - 1
What is the H statistic like?
Very similar to the chi-squared statistic so check significance in chi-square tables
Also means the KW test has to follow assumptions of chi-square e.g. no group can have fewer than 5 scores
What post-hoc tests can we do when we get a significant result?
Essentially 3 mann-whitney U tests (if we have three conditions)
What do we do if we want to calculate effect size?
We need the z-score and the total number of participants in the comparison
r=z/square root of N
How would you report the follow up post-hoc tests?
U statistic, followed by z-score, p value and finally effect size
How can we assess for normally distributed data?
Inspect histograms
Use Kolmogorov-Smirnov and Shapiro-Wilk tests (more objective, but affected by large samples so still confirm using visual inspection)
What is Levene’s test?
Assesses for homogeneity of variance - a non-significant result assumes rough homogeneity and we can use parametric tests
What does “adjusted for multiple comparisons” on SPSS mean?
Means a Bonferroni correction has been performed
