+ Reply to Thread
Results 1 to 8 of 8

Thread: Questions about Tukey's HSD/Scheffe Test

  1. #1
    Points: 4,890, Level: 44
    Level completed: 70%, Points required for next Level: 60

    Posts
    25
    Thanks
    0
    Thanked 0 Times in 0 Posts

    Questions about Tukey's HSD/Scheffe Test




    I heard Tukey Test can be used for unequal sample sizes. My assumption is that you must use some kind of weighted average of n for all groups. Correct me if I am wrong.

    Anyone know the general formula to compute Scheffe Test? I've found a few resources on it, but they kind of had conflicting information.

    Also, is there a calculator of some sort to compute the q statistic for Tukey's HSD? The data I am trying to analyze has a k = 13 and the q tables on the web only have values up to k = 10.

    Thanks for your help.

  2. #2
    Super Moderator
    Points: 10,097, Level: 67
    Level completed: 12%, Points required for next Level: 353
    Dragan's Avatar
    Location
    Illinois, US
    Posts
    1,820
    Thanks
    0
    Thanked 153 Times in 136 Posts
    Correct me if I am wrong.

    Well, the computation of q is based on the harmonic mean of the samples (taken pairwise) i.e.

    q=(YBari-YBarj)/Sqrt[MSw/n*] where n*=2/(1/ni + 1/nj) and MSw is the Mean Square Within (or Error) from the overall ANOVA analysis. Note that YBari and YBarj are the means of group i and group j.

    This is as the Tukey-Kramer post hoc test.


    As to Scheffe test it is:

    Fcv=(k-1)F_alpa-rate(k-1,dfw), where the alpha-rate is, say .05, and dfw is the degrees of freedom within (error) from the overall ANOVA summary.

    Note that the Scheffe test is a conservative test and I would not use this.

    Last Part: If you give me your degrees of freedom within (dfw), I can provide you with the critical value.

    Note: If you use SPSS for k=13 groups then you should have any trouble doing this.

  3. #3
    Points: 4,890, Level: 44
    Level completed: 70%, Points required for next Level: 60

    Posts
    25
    Thanks
    0
    Thanked 0 Times in 0 Posts
    Quote Originally Posted by Dragan View Post
    As to Scheffe test it is:

    Fcv=(k-1)F_alpa-rate(k-1,dfw), where the alpha-rate is, say .05, and dfw is the degrees of freedom within (error) from the overall ANOVA summary.

    Note that the Scheffe test is a conservative test and I would not use this.

    Last Part: If you give me your degrees of freedom within (dfw), I can provide you with the critical value.

    Note: If you use SPSS for k=13 groups then you should have any trouble doing this.
    So, it is best to use Tukey's method for post-hoc pairwise comparisons? Do I have any other options? And, df(w) = 582.

  4. #4
    Super Moderator
    Points: 10,097, Level: 67
    Level completed: 12%, Points required for next Level: 353
    Dragan's Avatar
    Location
    Illinois, US
    Posts
    1,820
    Thanks
    0
    Thanked 153 Times in 136 Posts
    Quote Originally Posted by this_barb View Post
    So, it is best to use Tukey's method for post-hoc pairwise comparisons? Do I have any other options? And, df(w) = 582.
    The critical value of q = 4.68 for alpha=0.05.

    As to the first part, Tukey's test is often complained about as being too conservative as well (but Scheffe's test is even worse). There's also the Student-Newman-Kuels (S-N-K) but this is also complained about in the literature and texts as being too liberal.

    It appears your doing these calculations by hand calculator. Is that right?

    I would suggest that you get access to SPSS and enter your data and use the Ryan procedure (so-called REGWQ test). I think that would be your best way to go. This test is sort of a compramise between Tukey's conservative test and the S-N-K liberal test.

    If your stuck, and don't have access, I'd suggest you use Tukey's test.

  5. #5
    Points: 4,890, Level: 44
    Level completed: 70%, Points required for next Level: 60

    Posts
    25
    Thanks
    0
    Thanked 0 Times in 0 Posts
    Quote Originally Posted by Dragan View Post
    The critical value of q = 4.68 for alpha=0.05.

    As to the first part, Tukey's test is often complained about as being too conservative as well (but Scheffe's test is even worse). There's also the Student-Newman-Kuels (S-N-K) but this is also complained about in the literature and texts as being too liberal.

    It appears your doing these calculations by hand calculator. Is that right?

    I would suggest that you get access to SPSS and enter your data and use the Ryan procedure (so-called REGWQ test). I think that would be your best way to go. This test is sort of a compramise between Tukey's conservative test and the S-N-K liberal test.

    If your stuck, and don't have access, I'd suggest you use Tukey's test.
    Yes, since I don't have access to any of the statistical programs, I am using Excel and a calculator. Thank you for your timely response.

  6. #6
    Points: 4,890, Level: 44
    Level completed: 70%, Points required for next Level: 60

    Posts
    25
    Thanks
    0
    Thanked 0 Times in 0 Posts
    I am able to use SPSS in the psych lab at my college. I tried using the REGWQ post hoc test, but I'm not sure how to interpret the results. More specifically, it states that it shows homogeneous subsets of means, but I'm not even sure what a subset is. For reference, here is the results table.



    How should I go about interpreting this?
    Last edited by this_barb; 05-12-2008 at 02:57 PM.

  7. #7
    Super Moderator
    Points: 10,097, Level: 67
    Level completed: 12%, Points required for next Level: 353
    Dragan's Avatar
    Location
    Illinois, US
    Posts
    1,820
    Thanks
    0
    Thanked 153 Times in 136 Posts
    How should I go about interpreting this?[I]

    Reject the following hypotheses:

    Mu1=Muj for j=3,4,5,...,14
    Mu2=Muj for j=3,4,5,...,14
    Mu3=Muj for j=8,10,11,...,14
    Mu4=Muj for j=8,10,11,...,14
    Mu5=Muj for j=8,10,11,...,14
    Mu6=Muj for j=8,10,11,...,14
    Mu7=Muj for j=10,11,12
    Mu8=Mu12
    Mu9=Muj for j=10,11,12

    and that should do it.

  8. #8
    Points: 4,890, Level: 44
    Level completed: 70%, Points required for next Level: 60

    Posts
    25
    Thanks
    0
    Thanked 0 Times in 0 Posts

    Quote Originally Posted by Dragan View Post
    How should I go about interpreting this?[I]

    Reject the following hypotheses:

    Mu1=Muj for j=3,4,5,...,14
    Mu2=Muj for j=3,4,5,...,14
    Mu3=Muj for j=8,10,11,...,14
    Mu4=Muj for j=8,10,11,...,14
    Mu5=Muj for j=8,10,11,...,14
    Mu6=Muj for j=8,10,11,...,14
    Mu7=Muj for j=10,11,12
    Mu8=Mu12
    Mu9=Muj for j=10,11,12

    and that should do it.
    Ah, thank you very much. Now it makes sense.

+ Reply to Thread

           




Similar Threads

  1. Tukey's additivity test
    By suobs in forum SPSS
    Replies: 0
    Last Post: 09-06-2010, 09:08 PM
  2. Scheffe/Tukey help! Really basic
    By RainyMonday in forum Statistics
    Replies: 1
    Last Post: 05-01-2010, 12:36 PM
  3. Two-way ANOVA and Tukey's HSD test
    By chaos in forum Statistics
    Replies: 1
    Last Post: 08-11-2009, 08:25 AM
  4. How do you pronounce Scheffe'?
    By zoltan in forum General Discussion
    Replies: 0
    Last Post: 05-05-2009, 08:06 PM
  5. Reliability analysis and Tukey's nonadditivity test
    By statishtics in forum Statistics
    Replies: 0
    Last Post: 09-02-2008, 01:33 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