I wish to compare two (for gender independent variable) and more than two (for other independent variable) groups in my study. The data is not normally distributed, and I have both nominal and ordinal scales. From my reading, I'm beginning to think I will need to do chi-square and/or Kruskal-Wallis Test. But I'm not sure which to choose. Or whether it's possible to use the two tests? I just need clarification/help!
@Jamie Kennedy @Jochen Wilhelm @Jarosław Piotrowski @Matthew Hogben @Muayyad Ahmad Joanna Polanska
Kingsley Oluchi Ugwuanyi - The K-W test statistic is distributed as a chi-square, so if you are comparing groups, you can use K-W. The output in your program will probably have the chi-square test statistic and p value (SPSS does). The grouping variable is gender (and whatever else you are using), and the outcome variable is a rank.
If you have two levels in the grouping variable, you can use the Mann-Whitney test, which has a z score associated with the test stat. But I would likely stick with K-W for ease of explanation.
Thank you Matthew Hogben . Invariably you're saying that K-W is not necessarily a different test from chi-square.
And yes, I agree it's better to just stick with K-W, since it is only an extension of M-W test.
What is the nature of your DV? You could use a regression framework, it is much more flexible and informative. A non-parametric regression would be a good option as it is not subject to misspecification error. However, I am not sure if IBM SPSS has such feature... Another option would be a linear regression with robust standard error. But if your data is count or ordinal, for example, there are different regression families available for this type of data.
- Badges
- Science topic
- Similar topics
- Cognitive Science
- Thinking