I am trying to compare continuous variables between four groups, let's say A, B, C, D.
While A and D groups are normally distributed, B and C aren't.
Which statistical method should I use to compare the four groups? I understand that for pairwise comparison (e.g. A and B or A and C), I should use non-parametric method, but what method should be used to compare all four groups simultaneously for significant statistical difference?
Thank you for your attention, and all comments regarding my question are welcome.
Have a nice day!
Jeong Uk Lim , Kruskal-Wallis test will work like a charm for your needs:
https://en.wikipedia.org/wiki/Kruskal%E2%80%93Wallis_one-way_analysis_of_variance
You can use the Kruskall-Walis test, it works similar to ANOVA. However, you should observe if your data, despite not having a normal distribution, can be compared using ANOVA, because although they do not have a normal distribution, if the variance shows little discrepancy between groups, if you have a considerable sample number (greater than 30, 50, 70) and the difference between the number of samples in each group is very small (3 or 5), it can force ANOVA assuming that the test has a greater detection power than its nonparametric equivalent, but it must make that clear in your research. Despite this suggestion from some authors, in any case I prefer to use Kruskall-Walis.
I agree with Daniel Medina Corrêa Santos that equality of variance is more important than normality in ANOVA. In general, ANOVA is relatively robust with regard to non-normality.
Rolando Gonzales Martinez Thank you! That's what I was looking for. I really appreciate your help :)
David L Morgan Daniel Medina Corrêa Santos I am pleased to learn that equality of variance was another important thing to consider when comparing continuous variables between multiple groups. Thank you!
What aspect of these groups do you want to compare? It seems many of the responders think it must be location/central tendency, but of course there are other things (like spread). Also, it is worth saying how you know two groups are Normally distributed. I assume these are known, and not just that the observe distributions are close. What is the n?
Jeong Uk Lim I hope that I have helped with your question!
Daniel Wright Thank you for the comment!. Variables are mostly clinical parameters: BMI, pulmonary function parameters such as forced expiratory volume per 1 second, symptom scores, etc. The four groups I mentioned above have smallest n of 33 and the biggest n of 130. The two non-normally distributed groups have n of 33 and 40. Following the previous recommendations, I underwent Kruskall-Wallis for evaluation. Would these aspects make difference? :)
Daniel Medina Corrêa Santos Of course it helped! Many thanks. I with you good luck in all that you do!
- Badges
- Science topic