+ Reply to Thread
Results 1 to 6 of 6

Thread: Maximum number of groups in Kruskal Wallis one-way analysis of variance

  1. #1
    Points: 1,584, Level: 22
    Level completed: 84%, Points required for next Level: 16

    Posts
    12
    Thanks
    1
    Thanked 0 Times in 0 Posts

    Maximum number of groups in Kruskal Wallis one-way analysis of variance




    Is there a maximum number of groups in Kruskal Wallis' one-way ANOVA? I am looking at a data set with 42 separate groups and I am unsure if there would be some sort of issue in running that many groups under Kruskal Wallis.

    I remember reading once that ANOVA should not be ran with more than 26 groups although I am not sure what the logic behind that number exactly is

  2. #2
    TS Contributor
    Points: 17,749, Level: 84
    Level completed: 80%, Points required for next Level: 101
    Karabiner's Avatar
    Location
    FC Schalke 04, Germany
    Posts
    2,540
    Thanks
    56
    Thanked 640 Times in 602 Posts

    Re: Maximum number of groups in Kruskal Wallis one-way analysis of variance

    Why Kruskal-Wallis? And why 42 groups? I guess that technically it is possible, but what would you do with a "statistical significant" result (if such a result is possible at all in your study)?

    With kind regards

    K.

  3. #3
    Points: 1,584, Level: 22
    Level completed: 84%, Points required for next Level: 16

    Posts
    12
    Thanks
    1
    Thanked 0 Times in 0 Posts

    Re: Maximum number of groups in Kruskal Wallis one-way analysis of variance

    Quote Originally Posted by Karabiner View Post
    Why Kruskal-Wallis? And why 42 groups? I guess that technically it is possible, but what would you do with a "statistical significant" result (if such a result is possible at all in your study)?

    With kind regards

    K.
    I thought KW test would be the most appropriate given that the distributions are not normal. There are 42 groups because there are 42 very distinct classifications in a particular categorical variable. A statistical significance would mean that one or more of the groups was significantly different from rest of the groups in terms of the median or the shape of distribution was significantly different depending on whether or not the distributions differ significantly. Are there other things that I should be considering?

  4. #4
    TS Contributor
    Points: 14,811, Level: 78
    Level completed: 91%, Points required for next Level: 39
    Miner's Avatar
    Location
    Greater Milwaukee area
    Posts
    1,171
    Thanks
    34
    Thanked 405 Times in 363 Posts

    Re: Maximum number of groups in Kruskal Wallis one-way analysis of variance

    Quote Originally Posted by feignfeign View Post
    I thought KW test would be the most appropriate given that the distributions are not normal. There are 42 groups because there are 42 very distinct classifications in a particular categorical variable.
    Are the shapes similar? That is an assumption of the KW test.

    Quote Originally Posted by feignfeign View Post
    A statistical significance would mean that one or more of the groups was significantly different from rest of the groups in terms of the median or the shape of distribution was significantly different depending on whether or not the distributions differ significantly. Are there other things that I should be considering?
    Not quite. It would mean that one or more of the groups are different from each other, not from the rest of the groups.

    By comparison: The null hypothesis for a 1-way ANOVA is that all means are equal, while the null hypothesis for an ANOM is that all means are equal to the group mean. These null hypotheses are distinctly different. In ANOVA, two means significantly different from each other, regardless of the other means would provide a low p-value. In ANOM, 1 mean significantly different from the group mean would provide a low p-value. Your wording seemed to imply a KW null interpretation similar to the ANOM.

  5. #5
    TS Contributor
    Points: 17,749, Level: 84
    Level completed: 80%, Points required for next Level: 101
    Karabiner's Avatar
    Location
    FC Schalke 04, Germany
    Posts
    2,540
    Thanks
    56
    Thanked 640 Times in 602 Posts

    Re: Maximum number of groups in Kruskal Wallis one-way analysis of variance

    Quote Originally Posted by feignfeign View Post
    I thought KW test would be the most appropriate given that the distributions are not normal.
    If your dependent variable is interval scaled and your sample size is large enough
    (which must be the case here), then why bother about normality? You can use a
    oneway anova then.
    A statistical significance would mean that one or more of the groups was significantly different from rest of the groups in terms of the median or the shape of distribution was significantly different depending on whether or not the distributions differ significantly.
    Yes. But how helful would this information be for you?
    Usually, one wants to identify which particular groups differ from
    other groups, but you'd have to perform 861 pairwise
    comparisons in that case.

    With kind regards

    K.

  6. #6
    Points: 1,584, Level: 22
    Level completed: 84%, Points required for next Level: 16

    Posts
    12
    Thanks
    1
    Thanked 0 Times in 0 Posts

    Re: Maximum number of groups in Kruskal Wallis one-way analysis of variance


    Quote Originally Posted by Karabiner View Post
    If your dependent variable is interval scaled and your sample size is large enough
    (which must be the case here), then why bother about normality? You can use a
    oneway anova then.

    Yes. But how helful would this information be for you?
    Usually, one wants to identify which particular groups differ from
    other groups, but you'd have to perform 861 pairwise
    comparisons in that case.

    With kind regards

    K.


    You're right. I thought about and decided to change the categorization schemes to create 18 groups to make the posthoc tests more manageable


    Quote Originally Posted by Miner View Post
    Are the shapes similar? That is an assumption of the KW test.

    Not quite. It would mean that one or more of the groups are different from each other, not from the rest of the groups.

    By comparison: The null hypothesis for a 1-way ANOVA is that all means are equal, while the null hypothesis for an ANOM is that all means are equal to the group mean. These null hypotheses are distinctly different. In ANOVA, two means significantly different from each other, regardless of the other means would provide a low p-value. In ANOM, 1 mean significantly different from the group mean would provide a low p-value. Your wording seemed to imply a KW null interpretation similar to the ANOM.
    my understanding is that the shapes do not have to be similar in order perform a kb test but the interpretation of the results would be different in that a significant result indicates a difference in the shape of distribution between the groups at a population level. Perhaps not a very helpful inferential insight, but something, i suppose

+ Reply to Thread

           




Posting Permissions

  • You may not post new threads
  • You may not post replies
  • You may not post attachments
  • You may not edit your posts






Advertise on Talk Stats