No. Our data is not normally distributed that's why we used a non parametric test such as Kruskal-Wallis. You don't necessary have to have more than one p value, but you have to make sure that you satisfy the following assumptions: 1. Your dependent variable should be measured at the ordinal or continuous level. 2. Your independent variable should consist of two or more categorical, independent groups. 3. You should have independence of observations, which means that there is no relationship between the observations in each group or between the groups themselves. For more info, you may check this link: statistics.laerd.com/spss-tutorials/kruskal-wallis-h-test-using-spss-statistics.php Hope this helps!
Thanks for the upload. In your data, the grouping is based on one variable. What if the groups are based on more than one (say 6) variables? in that case what test are we going to do to see if the groups are independent or not?
hi, first of all, thanx for yr very informative video... im facing bit problem now.. after i performed the split data, aggregate and means different..then my one way anova analysis has significant value.. that reject null hypothesis.... then, i dont know what to.. in your video, your p-value is not significant... can you give me advise? thank u...
Possibly, yes. I'm assuming that you do not want to combine the three dependent variables into a 'supervariable' as per a MANOVA. So, if analysing each dependent variable in separate analyses, Kruskal-Wallis should work for you.
thanks for the uploads, question - is it essential that the groups must be the same size? in this case each group is 15 participents - should i randomly select the higest common denonominator of participants to do this test?
No, this is not an ANOVA, which is a parametric statistical analysis. K-W is non-parametric in nature. Plus, a three-way ANOVA would imply three independent variables, this analysis has only one independent variables.
Sir what if when you perform ANOVA after getting the absolute differences, and teh ANOVA gives a significant result, like less than .05, so it means the non-parametric Levenes test for homogeneity of variance is not normal. will u still proceed to Kruskal Wallis?
+how2stats you say that homogeneity is assumed here and that it can be verified through descriptive stats like skewedness and kurtosis, only then to point to your figures and say that they are clearly different. are you saying a K-W test is not appropriate for your example please? (i do not believe homogeneity is assumed in a K-W test. in fact, i'm also certain it isn't. when we do not have homogeneity of variance, a K-W test is considered a comparison of mean ranks, that's all.)
I was wondering: where in Keselman's paper does it say Bonferroni correction is not necessary when comparing 4 or less means? I cannot seem to find this when reading the paper.
No, this is not an ANOVA, which is a parametric statistical analysis. K-W is non-parametric in nature. Plus, a three-way ANOVA would imply three independent variables, this analysis has only one independent variables.
Parts 1-5 of this video helped us a lot in our thesis! Highly recommended. AWESOME TUTOR!
Is your data homogenous ? have you tried it ? is it a condition to have more than p-value to perform kruskal test ?
No. Our data is not normally distributed that's why we used a non parametric test such as Kruskal-Wallis. You don't necessary have to have more than one p value, but you have to make sure that you satisfy the following assumptions:
1. Your dependent variable should be measured at the ordinal or continuous level.
2. Your independent variable should consist of two or more categorical, independent groups.
3. You should have independence of observations, which means that there is no relationship between the observations in each group or between the groups themselves.
For more info, you may check this link:
statistics.laerd.com/spss-tutorials/kruskal-wallis-h-test-using-spss-statistics.php
Hope this helps!
This video saved my life! Very helpful
and kruskal-Wallis?
Thanks for the upload. In your data, the grouping is based on one variable. What if the groups are based on more than one (say 6) variables? in that case what test are we going to do to see if the groups are independent or not?
hi, first of all, thanx for yr very informative video... im facing bit problem now.. after i performed the split data, aggregate and means different..then my one way anova analysis has significant value.. that reject null hypothesis.... then, i dont know what to.. in your video, your p-value is not significant... can you give me advise? thank u...
thanks for the upload just wanted to know if the order of groups matter for example 1,2,3 or 3,1,2
Thanks again
No, it doesn't matter.
Possibly, yes. I'm assuming that you do not want to combine the three dependent variables into a 'supervariable' as per a MANOVA. So, if analysing each dependent variable in separate analyses, Kruskal-Wallis should work for you.
Would you suggest using this approach when analysing an experiment(1 independent, 3 levels x 3 depedent) that has ordinal values?
thanks for the uploads, question - is it essential that the groups must be the same size? in this case each group is 15 participents - should i randomly select the higest common denonominator of participants to do this test?
my test is coming out saying there are not enough valid cases. I have 4 values for each group, is it because this is not enough?
I wonder that the identical population shape assumption is reasonable assumption?
there are 4 groups and sample sizes are not even?so can i use this method?how to apply????
Is your data homogenous ? have you tried it ? is it a condition to have more than p-value to perform kruskal test ?
No, this is not an ANOVA, which is a parametric statistical analysis. K-W is non-parametric in nature. Plus, a three-way ANOVA would imply three independent variables, this analysis has only one independent variables.
Sir what if when you perform ANOVA after getting the absolute differences, and teh ANOVA gives a significant result, like less than .05, so it means the non-parametric Levenes test for homogeneity of variance is not normal. will u still proceed to Kruskal Wallis?
what non parametric test is the equivalent of 2 way anova.. thanks..
the kruskal Wallis - this one above
PERMANOVA is a non parametric MANOVA
No, you can have unequal sample sizes.
How do you define each group? (i.e 1 - Location One?)
+how2stats you say that homogeneity is assumed here and that it can be verified through descriptive stats like skewedness and kurtosis, only then to point to your figures and say that they are clearly different. are you saying a K-W test is not appropriate for your example please?
(i do not believe homogeneity is assumed in a K-W test. in fact, i'm also certain it isn't. when we do not have homogeneity of variance, a K-W test is considered a comparison of mean ranks, that's all.)
this is aka 3 way ANOVA?
You're not supposed to if the Levene's test is statistically significant. You should consider doing a Median Test. I have a video for that too.
I was wondering: where in Keselman's paper does it say Bonferroni correction is not necessary when comparing 4 or less means? I cannot seem to find this when reading the paper.
Dude, you the man.
Thank you so much.
Yeah, numbers with decimal points are fine.
thanks very helpful
drinkling
good
No, this is not an ANOVA, which is a parametric statistical analysis. K-W is non-parametric in nature. Plus, a three-way ANOVA would imply three independent variables, this analysis has only one independent variables.