non-parametric test

[#1]

Hello everyone,

I'm currently working on a scientific article where I'm investigating the influence of geographical origin on the fatty acid composition of olive oil. Specifically, I have collected 9 samples from each of 4 different geographical origins. My goal is to determine if there are significant differences in the fatty acid composition among these 4 origins.

In my previous research, I often used ANOVA followed by the Tukey (HSD) test to assess differences among groups. However, in my current study, I'm encountering situations where the assumptions of normality and homogeneity of variance are not met for my data.

To address this issue, I've been considering non-parametric tests. I've come across suggestions to use the Kruskal-Wallis test, but I'm unsure about which post-hoc test to use in conjunction with it.

Could anyone provide guidance?

[#2 · Accepted solution]

Hello ELH, 

In my company we often use the Kruskal-Wallis and (Mood's) Median tests to compare the central tendency of groups and we use the Levene's test to compare the variation among groups for non-parametric data.

> Nonparametric > Wilcoxon / Kruskal-Wallis Tests

> Nonparametric > Nonparametric Multiple Comparisons > Wilcoxon Each Pair

> Nonparametric > Median Test

The Kruskal-Wallis Test compares the average rank of each population.    

In the presence of outliers, the average rank of a population may be biased significantly, thereby resulting in a decision error.

The Mood’s Median Test compares the number of data points that are above/below the median for each population. 

Hence, this test is more robust against the presence of outliers.

Hope this helps.  The study sounds delicious!!

[#3]

Thank you  

[#4]

Thank you sir for answering, you made my day