+ Reply to Thread
Results 1 to 5 of 5

Thread: Kruskal-Wallis and multiple comparisons

  1. #1
    ceperras
    Guest

    Kruskal-Wallis and multiple comparisons




    Hello all,

    I've run a Kruskal-Wallis analysis (in SAS) and found there is a significant difference between the groups. Therefore, I would like to find out between which groups that difference actually occurs. How do I proceed with multiple comparisons (in SAS) after a Kruskal-Wallis analysis?

    Thanks!
    Christyn

  2. #2
    Super Moderator
    Points: 9,859, Level: 66
    Level completed: 53%, Points required for next Level: 191
    Dragan's Avatar
    Location
    Illinois, US
    Posts
    1,785
    Thanks
    0
    Thanked 142 Times in 128 Posts
    Quote Originally Posted by ceperras View Post
    Hello all,

    I've run a Kruskal-Wallis analysis (in SAS) and found there is a significant difference between the groups. Therefore, I would like to find out between which groups that difference actually occurs. How do I proceed with multiple comparisons (in SAS) after a Kruskal-Wallis analysis?

    Thanks!
    Christyn
    See below for a post-hoc test for the KW test. Note: Whether SAS will do this (or not) I am not sure.

    http://www.talkstats.com/showthread.php?t=4894

  3. #3
    Dark Knight
    Points: 2,607, Level: 31
    Level completed: 5%, Points required for next Level: 143
    vinux's Avatar
    Posts
    1,905
    Thanks
    46
    Thanked 198 Times in 165 Posts
    Quote Originally Posted by ceperras View Post
    Hello all,

    I've run a Kruskal-Wallis analysis (in SAS) and found there is a significant difference between the groups. Therefore, I would like to find out between which groups that difference actually occurs. How do I proceed with multiple comparisons (in SAS) after a Kruskal-Wallis analysis?

    Thanks!
    Christyn
    You can do pairwise mann whitney using proc npar1way itself..
    Select only two group in the class using the where clause ( or another data step)
    In the long run, we're all dead.

  4. #4
    ceperras
    Guest
    Thanks for your help!

    I read that I can do the pairwise comparisons with a Wilcoxon two-sample test but will need to adjust the p-values with a Bonferroni correction (bc of the multiple comparisons).

    I've heard this can be done one of two ways. First way: adjusted p-value=(alpha/# of tests). Second way: use Proc Multtest in SAS. I tried both and got radically different results.

    Have either of you done anything like this?

    Thanks!

  5. #5
    Super Moderator
    Points: 9,859, Level: 66
    Level completed: 53%, Points required for next Level: 191
    Dragan's Avatar
    Location
    Illinois, US
    Posts
    1,785
    Thanks
    0
    Thanked 142 Times in 128 Posts

    Quote Originally Posted by ceperras View Post
    Thanks for your help!

    I've heard this can be done one of two ways. First way: adjusted p-value=(alpha/# of tests). Second way: use Proc Multtest in SAS. I tried both and got radically different results.

    Have either of you done anything like this?

    Thanks!
    I would recommend that you attempt to use of the technique that I suggested in the link I provide (above). This technique appropriately controls for familywise error.

+ Reply to Thread

           




Similar Threads

  1. kruskal-wallis multiple comparisons
    By l_schnapp in forum SPSS
    Replies: 1
    Last Post: 01-13-2011, 12:43 AM
  2. One way ANOVA vs Kruskal-Wallis
    By stat_18 in forum Other Software
    Replies: 6
    Last Post: 08-22-2010, 02:21 PM
  3. Is Kruskal Wallis the right test to use?
    By August 15 in forum Statistics
    Replies: 2
    Last Post: 08-03-2010, 05:43 AM
  4. Should I use Kruskal-Wallis?
    By brian in forum Statistics
    Replies: 3
    Last Post: 04-28-2009, 03:02 AM
  5. The Kruskal-Wallis Test
    By Keikle08 in forum Statistics
    Replies: 1
    Last Post: 03-11-2008, 09:55 PM

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