I am doing Kruskal-Wallis test and I need to report the effect size as well. Thus, I am wondering what will be the best to calculate effect size for Kruskal-Wallis test?
#Kruskal-Wallis Test
#Nonparametric test
#Effectsize
Yes, there are a few that are used.
There's one sometimes called epsilon-squared, as described here ( http://tss.awf.poznan.pl/files/3_Trends_Vol21_2014__no1_20.pdf ).
There's also Freeman's theta.
Another approach is to use an effect size for the two-sample case, and use either the maximum for pairwise comparisons, or simply list all the pairwise comparisons. Here I would recommend either Cliff's delta or Vargha and Delaney's A, which are used in the two-sample Wilcoxon-Mann-Whitney test.
Hi Sal Mangiafico, Thank you for your response. I need to analyse per say perception of 3 groups (A, B, C) on their work burden (on a 5 points likert scale) in two different years (2017 and 2018). A lot of the participants are same for 2017 and 2018 although there are some new participants. And, now I am thinking that if I consider the Likert scale as interval data, I might be able to do parametric test. Do you have any suggestion or comment about this?
The F value with dfs should be able to give you info for converting test-stat into effect sizes.
Hope this helps,
Matt
Likert scales --- that is, it's several items summed or averaged into a scale value --- are often treated as interval data. This is common. I would still be sure to assess the model for the assumptions of parametric tests.
Matthew Kerry , Hi Matthew- Thanks you so much for your reply. I am also thinking about doing One-way ANOVA. However, someone also suggested to do Z-test. However, i am not convinced whether Z-test will be accurate or appropriate for analyzing perception of 3 groups (A, B, C) on their workload (on a 5 points likert scale) in two different years (2017 and 2018).
Is there any useful article or book that you suggest so I can out of this dilemma?
Hi Sal Mangiafico , thanks again. If possible, can you suggest any article or book that clearly outlines what kind of sig testing to choose for social and/or educational research?
Thank you.
Hi, Israt Jahan . David Morgan suggests some books and articles here: https://www.researchgate.net/post/How_can_one_help_people_create_simple_scales_from_Likert-scored_items
A quick reading of responses on that thread will give you some sense that there's a lot of discussion about how to best handle Likert item and Likert scale data.
From your description, it sounds like you will want to use a more complicated model than kruskal-wallis, so that you can account for both Group and Year. A further complication is that you have some of the same participants in both years. You might ignore this, or you might take this into account. As you suggest, if you are treating your scale data as interval, you might use a traditional general linear model. If you want to keep with nonparametric tests, an aligned ranks approach might be appropriate. (http://depts.washington.edu/madlab/proj/art/).
Hi Sal Mangiafico , Thank you so much for the link. I will have a read and I would like to use something simple but it has to be accurate :)
This has also been discussed here:
https://www.researchgate.net/post/Anyone-know-how-to-calculate-eta-squared-for-a-Kruskal-Wallis-analysis
Best wishes, Philippe
- Badges
- Science topic