Kruskal-Wallis tests and many other analyses are covered in my book, SPSS Made Easy: www.amazon.co.uk/SPSS-Made-Easy-Statistical-Researchers/dp/B0DJGR4Z5K
Hi David, another fantastic video. How would you report this results if you had multiple DVs? Say for example how age group influences performance on three different subtests? Thank you!
Thanks Mary! You could just repeat something like the example I show at around 6:50 three times (i.e., once per subtest). Alternatively, you could think about putting the stats in tables (e.g., one for the Kruskal-Wallis test, one for the descriptive statistics). If you search, e.g., "Kruskal-Wallis APA table" on Google Images, you should be able to get some ideas.
Sir, I have 3 groups. My data is collected on 5 point Likert scale. Each of my construct is formative (not reflective) in nature, therefore there are different items which form a construct. I want to know significant differences between the three groups on each of the formative items of a construct. Shall i do KW test individually for all the items?
Hi Chef, if the items all measure the same constructs, it would be more common to complete the analysis on a variable that represents the sum or the mean of the items. Regarding your other question, yes, you could do a post-hoc test (e.g., Dunn's test) as the Kruskal-Wallis test just tells you whether there's a significant effect of the IV on the DV, not which specific groups differ significantly from each other.
@@DavidRobinsonPhD As i said, ,y items are formative in nature, it is not a reflective construct, so how can i combine them to a mean score for the analysis?
Hi , thanks for your question. The median is usually reported instead of the mean in the case of non-parametric tests that compare group differences. One reason is that these tests are often used with data that don't have a normal distribution, in which case the mean might not provide a useful description of the data.
Thank you so much for your explanation. Am doing my masters thesis and this test is the one I am using to test my hypothesis. You have really helped, I appreciate
Thank you for this great tutorial I have a problem that I listed here: I have three independent catagorial variables and two dependent continuous scale variables, so I want to compare means across these three independent catagorial variables. I think here the best choice would be three way manova, but I'm facing problem with my independent catagorial variables for example one of the independent variables was not normally distributed so I had to run kruskal wallis test, or I had another independent variable that violated the assumption of homogeneity of variance and so on. So I decided to run multiple separate anovas or kruskal wallis tests, because it's not possible to run three way manova. Is this the best approach for this problem?
I've had to complete a Kruskal Wallis and Mann Whitney U for my MSc dissertation as my data did not meet the parametric assumptions to complete an ANOVA. I was just wondering, are there any tables that would need to be included with either of these analyses? Or is there anything to add to the results section? Really useful and quick video, thank you!
Hi Tia, Thanks for your question. If you did multiple Kruskal-Wallis and/or Mann-Whitney U tests, it might be worth presenting the results in a table. If you search, e.g., "Kruskal-Wallis table APA" on Google Images, you should be able to find some examples. Hope that helps! David.
@@DavidRobinsonPhD thanks so much! I’ve have completed 3 Kruskal Wallis and 3 Mann Whitney U tests as I have 3 outcome variables. I will have a look into this thank you.
Hi David, thank you for such an amazing tutorial, it really helps in my thesis writing! However, i have some doubts and hopefully you could answer them: If the median are the same among 3 group (eg: group 1,2,3=1) but Kruskal Wallis test showed a significant result, what does it implies? once again, thanks for the great tutorial and have a nice day! :)
Thanks for this... How do you do it when you have 13 schools you evaluated and what to see the difference in performance of schools.. Would the schools be the 'grouping variable'? If the tool used has subscales, how do you get the pvalue for the whole subscales instead of per item? Thanks
Hi, Thanks for your question. Yes, it sounds like "school" would be the grouping variable (i.e., IV) in that case and performance would be the DV. If you haven't already, you'll probably need to calculate total or mean scores for the subscales that you used. You can then use these as your DVs instead of the individual items. I haven't yet made a video on how to calculate total or mean scores for subscales, but others have. E.g., ua-cam.com/video/GY8w4cTfA4o/v-deo.html Hope that helps! David.
@@DavidRobinsonPhD God bless you, very helpful. I prefer not to use the mean scores because of the argument around using it with ordinal data, I'll use the sum score. Thank you.
I have a doubt. Doing a post hoc for Kruskal Wallis test gives me a different p value for each pair compared to when performing individual man whittney tests for each pair. I am wondering whether I should report the p value obtained in Man whittney test or that obtained in post hoc test. Thank you for this video Sir
Hi Akashjyoti, thanks for your question. As there’s an increased risk of false positives when running multiple Mann-Whitney tests and as post-hoc tests usually account for this “multiple comparisons problem”, I would probably opt to report the post-hoc tests.
Hello David, thanks for this video! I´m looking for a way to report a significant post hoc test. I only find examples for how to report significant post hoc tests when the Mann-Whitney-U test is used (U = ..., p < ...). However, I used the Kruskal-Wallis 1 way ANOVA for three samples and pairwise comparisons. Do you know how to report these results? :)
Hi, Thanks for your question. There's an example here: www.sheffield.ac.uk/polopoly_fs/1.714567!/file/stcp-marshall-KruskalSPSS.pdf Hope that helps! David.
Hi David, It was wonderful to see your video. However I have one question and would be really grateful if you could answer that. In a study I did I had taken 3 independent groups where N. Was below 30 and it did not meet criteria for normality. So I did a Kruskal wallis as an. Omnibus test and then Mann Whitney U to determine where the significant difference is. However in a review, I was asked to quote from where I have derived I can use mann Whitney as post hoc since in some places it is still not used as post hoc. I have stated previous research papers where Mann Whitney U test has been used as post hoc , but do you have have any bol reference stating the same? Thanks in advance
@@sreetamachatterjee6945 Thanks for your question. Pallant's SPSS Survival Manual recommends Mann-Whitney U tests if you have a significant Kruskal-Wallis result. She also suggests that you apply Bonferroni corrections to avoid Type 1 errors. E.g., if you run three Mann-Whitney U tests, divide the normal alpha of .05 by three and then use .017 for determining whether a result is significant.
Oh okay. That is very helpful indeed 😊🙏. Thank you so much for replying. If I ever have any further questions, is it okay to ask it here or should emailing be better? Thanks once again 😊
Hi, Thanks for your question. It's basically the same. For example, if the result was significant, you could say something like, "A Kruskal-Wallis test revealed a statistically significant difference...", whereas if the result was not significant, you could say something like, "A Kruskal-Wallis test revealed that there was not a statistically significant difference..." In APA style, an equals symbol is used after "p" (unless the p value is below .001, in which case "
Hi Allan, It's uncommon to create a table for the Kruskal-Wallis, especially if you only ran one (as it's easy to just report the results in the text), but if you want to create, you could consider something like this: www.researchgate.net/figure/Kruskal-Wallis-test-results-of-the-mean-scores-obtained-in-first-grade-primary-school_tbl3_287805964 Best wishes, David.
Hi Anna, Thanks for your question. I'm not quite sure what you mean. Feel free to outline what you would like to do and I'll let you know what I think. Best, David.
@@DavidRobinsonPhD Thank you for your reply. So...Let's say you have also collected demographic data, such as gender, and you want to know whether the relation between time and the water sport was confounded or modified by other independent variables (i.e. gender). How would you do?
@@annacatharinaarmond8695 Hi Anna, Thanks for clarifying. That would ideally be handled at the design stage of the study, such as by randomly allocating participants to groups to ensure an equal distribution of gender across groups. However, if this wasn’t possible for some reason, you could consider a one-way analysis of covariance (ANCOVA), in which you would enter gender as a covariate. The ANCOVA is a parametric test so makes certain assumptions that the Kruskal-Wallis test doesn’t, but there unfortunately isn’t a non-parametric equivalent of the ANCOVA. If the data didn’t meet the assumptions of the ANCOVA, an alternative might be running a chi-squared test to assess whether or not gender varies significantly between groups; if it didn’t, you could have greater confidence that the Kruskal-Wallis results are meaningful (i.e., changes in the DV really do reflect changes in the IV). There are videos about ANCOVAs and chi-squared tests on my channel if you’re interested. Best, David.
Kruskal-Wallis tests and many other analyses are covered in my book, SPSS Made Easy:
www.amazon.co.uk/SPSS-Made-Easy-Statistical-Researchers/dp/B0DJGR4Z5K
Thank you, it was what I need it. Best wishes!
Thanks Florin, glad it helped!
Hi David, another fantastic video. How would you report this results if you had multiple DVs? Say for example how age group influences performance on three different subtests? Thank you!
Thanks Mary! You could just repeat something like the example I show at around 6:50 three times (i.e., once per subtest). Alternatively, you could think about putting the stats in tables (e.g., one for the Kruskal-Wallis test, one for the descriptive statistics). If you search, e.g., "Kruskal-Wallis APA table" on Google Images, you should be able to get some ideas.
@@DavidRobinsonPhD Thanks very much. Will do!
This was sooo helpful!! Thank you very much
Thanks Karla, glad it helped!
One more question. Dont we need to do Dunn's post hoc test?
Sir, I have 3 groups. My data is collected on 5 point Likert scale. Each of my construct is formative (not reflective) in nature, therefore there are different items which form a construct. I want to know significant differences between the three groups on each of the formative items of a construct. Shall i do KW test individually for all the items?
Hi Chef, if the items all measure the same constructs, it would be more common to complete the analysis on a variable that represents the sum or the mean of the items. Regarding your other question, yes, you could do a post-hoc test (e.g., Dunn's test) as the Kruskal-Wallis test just tells you whether there's a significant effect of the IV on the DV, not which specific groups differ significantly from each other.
@@DavidRobinsonPhD As i said, ,y items are formative in nature, it is not a reflective construct, so how can i combine them to a mean score for the analysis?
Hi, while explaining the results can I use mean score in my explanation instead of median?
Hi , thanks for your question. The median is usually reported instead of the mean in the case of non-parametric tests that compare group differences. One reason is that these tests are often used with data that don't have a normal distribution, in which case the mean might not provide a useful description of the data.
Thank you so much for your explanation. Am doing my masters thesis and this test is the one I am using to test my hypothesis. You have really helped, I appreciate
Thank you for this great tutorial
I have a problem that I listed here:
I have three independent catagorial variables and two dependent continuous scale variables, so I want to compare means across these three independent catagorial variables.
I think here the best choice would be three way manova, but I'm facing problem with my independent catagorial variables for example one of the independent variables was not normally distributed so I had to run kruskal wallis test, or I had another independent variable that violated the assumption of homogeneity of variance and so on.
So I decided to run multiple separate anovas or kruskal wallis tests, because it's not possible to run three way manova.
Is this the best approach for this problem?
I've had to complete a Kruskal Wallis and Mann Whitney U for my MSc dissertation as my data did not meet the parametric assumptions to complete an ANOVA. I was just wondering, are there any tables that would need to be included with either of these analyses? Or is there anything to add to the results section? Really useful and quick video, thank you!
Hi Tia,
Thanks for your question.
If you did multiple Kruskal-Wallis and/or Mann-Whitney U tests, it might be worth presenting the results in a table. If you search, e.g., "Kruskal-Wallis table APA" on Google Images, you should be able to find some examples.
Hope that helps!
David.
@@DavidRobinsonPhD thanks so much! I’ve have completed 3 Kruskal Wallis and 3 Mann Whitney U tests as I have 3 outcome variables. I will have a look into this thank you.
Hi David, thank you for such an amazing tutorial, it really helps in my thesis writing!
However, i have some doubts and hopefully you could answer them:
If the median are the same among 3 group (eg: group 1,2,3=1) but Kruskal Wallis test showed a significant result, what does it implies?
once again, thanks for the great tutorial and have a nice day! :)
Thanks for this... How do you do it when you have 13 schools you evaluated and what to see the difference in performance of schools.. Would the schools be the 'grouping variable'?
If the tool used has subscales, how do you get the pvalue for the whole subscales instead of per item? Thanks
Hi,
Thanks for your question.
Yes, it sounds like "school" would be the grouping variable (i.e., IV) in that case and performance would be the DV.
If you haven't already, you'll probably need to calculate total or mean scores for the subscales that you used. You can then use these as your DVs instead of the individual items. I haven't yet made a video on how to calculate total or mean scores for subscales, but others have. E.g., ua-cam.com/video/GY8w4cTfA4o/v-deo.html
Hope that helps!
David.
@@DavidRobinsonPhD God bless you, very helpful. I prefer not to use the mean scores because of the argument around using it with ordinal data, I'll use the sum score. Thank you.
I have a doubt. Doing a post hoc for Kruskal Wallis test gives me a different p value for each pair compared to when performing individual man whittney tests for each pair. I am wondering whether I should report the p value obtained in Man whittney test or that obtained in post hoc test. Thank you for this video Sir
Hi Akashjyoti, thanks for your question. As there’s an increased risk of false positives when running multiple Mann-Whitney tests and as post-hoc tests usually account for this “multiple comparisons problem”, I would probably opt to report the post-hoc tests.
@@DavidRobinsonPhD thank you so much Sir for your suggestion.
Hello David, thanks for this video! I´m looking for a way to report a significant post hoc test. I only find examples for how to report significant post hoc tests when the Mann-Whitney-U test is used (U = ..., p < ...). However, I used the Kruskal-Wallis 1 way ANOVA for three samples and pairwise comparisons. Do you know how to report these results? :)
Hi,
Thanks for your question.
There's an example here: www.sheffield.ac.uk/polopoly_fs/1.714567!/file/stcp-marshall-KruskalSPSS.pdf
Hope that helps!
David.
@@DavidRobinsonPhD Perfect, thank you!
Hi David,
It was wonderful to see your video. However I have one question and would be really grateful if you could answer that. In a study I did I had taken 3 independent groups where N. Was below 30 and it did not meet criteria for normality. So I did a Kruskal wallis as an. Omnibus test and then Mann Whitney U to determine where the significant difference is. However in a review, I was asked to quote from where I have derived I can use mann Whitney as post hoc since in some places it is still not used as post hoc. I have stated previous research papers where Mann Whitney U test has been used as post hoc , but do you have have any bol reference stating the same?
Thanks in advance
I meant to say book references. Sorry for the typographical error
@@sreetamachatterjee6945 Thanks for your question. Pallant's SPSS Survival Manual recommends Mann-Whitney U tests if you have a significant Kruskal-Wallis result. She also suggests that you apply Bonferroni corrections to avoid Type 1 errors. E.g., if you run three Mann-Whitney U tests, divide the normal alpha of .05 by three and then use .017 for determining whether a result is significant.
Oh okay. That is very helpful indeed 😊🙏. Thank you so much for replying.
If I ever have any further questions, is it okay to ask it here or should emailing be better?
Thanks once again 😊
@@sreetamachatterjee6945 No problem! Feel free to ask anything here.
Hello!
How do you report a non-significant result for Kruskal Wallis?
Hi,
Thanks for your question.
It's basically the same. For example, if the result was significant, you could say something like, "A Kruskal-Wallis test revealed a statistically significant difference...", whereas if the result was not significant, you could say something like, "A Kruskal-Wallis test revealed that there was not a statistically significant difference..." In APA style, an equals symbol is used after "p" (unless the p value is below .001, in which case "
Pls help me create an APA table format for Kruskal-Wallis Test
Hi Allan,
It's uncommon to create a table for the Kruskal-Wallis, especially if you only ran one (as it's easy to just report the results in the text), but if you want to create, you could consider something like this: www.researchgate.net/figure/Kruskal-Wallis-test-results-of-the-mean-scores-obtained-in-first-grade-primary-school_tbl3_287805964
Best wishes,
David.
Hi David, thank you for your video. How would you adjust this result, for example, for gender?
Hi Anna,
Thanks for your question.
I'm not quite sure what you mean. Feel free to outline what you would like to do and I'll let you know what I think.
Best,
David.
@@DavidRobinsonPhD Thank you for your reply. So...Let's say you have also collected demographic data, such as gender, and you want to know whether the relation between time and the water sport was confounded or modified by other independent variables (i.e. gender). How would you do?
@@annacatharinaarmond8695
Hi Anna,
Thanks for clarifying.
That would ideally be handled at the design stage of the study, such as by randomly allocating participants to groups to ensure an equal distribution of gender across groups. However, if this wasn’t possible for some reason, you could consider a one-way analysis of covariance (ANCOVA), in which you would enter gender as a covariate. The ANCOVA is a parametric test so makes certain assumptions that the Kruskal-Wallis test doesn’t, but there unfortunately isn’t a non-parametric equivalent of the ANCOVA. If the data didn’t meet the assumptions of the ANCOVA, an alternative might be running a chi-squared test to assess whether or not gender varies significantly between groups; if it didn’t, you could have greater confidence that the Kruskal-Wallis results are meaningful (i.e., changes in the DV really do reflect changes in the IV). There are videos about ANCOVAs and chi-squared tests on my channel if you’re interested.
Best,
David.
@@DavidRobinsonPhD Hi David, thank you very much for your answer. I will watch the videos and see what it's possible for me to do. Thank you again!