For many researchers, there seems to be confusion between ANOVA and Kruskal-Wallis tests. While both are used to determine whether there is a significant difference between the results of more than three groups, there are some differences between the two tests.
The ANOVA is used 90% of the time in research, whereas we use the Kruskal-Wallis test when our dependent variable is ordinal, and not a ratio or interval variable- otherwise we use the ANOVA test by default. However, there are actually three assumptions when carrying out an ANOVA test: normality, constant error variance (homoscedasticity), and independence. After an ANOVA test, we generate residual plot to see if these assumptions are violated. In the case that any of these assumptions are violated, we will have to use the Kruskal Wallis test.
Remember
- If your variables are interval or ratio, use the ANOVA test by default.
- After performing ANOVA testing, check for the assumptions using the residual plots.
- If the plots show that the assumptions are violated, perform the Kruskal-Wallis test instead of ANOVA.
Tanim Talino 