How does multcompare function work?

Hey everyone,

I have done kruskal wallis test and then performed multcompare function (tukey test). Matrix X that I used in K-W test 2 columns each one corresponds to a disease, and for each disease that data is belongs to 70 patients so the matrix is 70x2. My question is that in the multcompare the comparison is done pairwise which means the first result I get corresponds to the comparison of the the first row? And regarding the example in matlab

For example, suppose one row contains the following entries.

2.0000 5.0000 1.9442 8.2206 14.4971

These numbers indicate that the mean of group 2 minus the mean of group 5 is estimated to be 8.2206, and a 95% confidence interval for the true mean is [1.9442, 14.4971].

here what is meant by group 2 and group 5; while we only have 2 columns in the matrix?!

Answer 1

[Tom Lane]

Can you be more specific about what code you ran? When I try this, I only get groups 1 and 2, comparing the two columns:

>> [~,~,s] = kruskalwallis(rand(30,2));
>> multcompare(s)
Note: Intervals can be used for testing but are not simultaneous confidence intervals.
ans =
    1.0000    2.0000  -10.1712   -1.3333    7.5045

Also, you wrote K-S test in one place; I assume this is a typo and you are not doing a Kolmogorov-Smirnov test.

[Yasmin Tamimi]

Oh! I meant the kruskal wallis test!! This is my code

[P,anovatab,stats] = kruskalwallis(X); where X is a 70x2

 c2 = multcompare(stats,'display','off')

by groups do you mean columns?! from what I understood that the multcompare compares between 2 groups(columns) only so how come in the matlab example there was a comparison between group 2 and 5!! so how can this happen?!

[Tom Lane]

If you mean the example on this page:

http://www.mathworks.com/help/toolbox/stats/multcompare.html

That would be something you would get after a comparison of at least five groups. For example, you might do a Kruskal-Wallis test on a matrix with five columns and see this in your output.

[Yasmin Tamimi]

Its because I thought that multcompare compares only between two columns and not more!! Thanx a lot Tom really appreciated!!

[Lingyun GUO]

Dear Tom & Yasmin::

I want to ask ：

I have 3 sets of data. It has been organized into two columns The first column is basedata; The second column is the grouping basis.

G1(:,1)=rand(30,1);

G1(:,2)=1;

G2(:,1)=rand(32,1);

G2(:,2)=2;

G3(:,1)= rand(30,1);

G3(:,2)=3;

X=[G1;G2;G3]; % size is (92*2).

[P,anovatab,stats] = kruskalwallis(X(:,1),X(:,2));

c2=multcompare(stats);

The answer is:

1 2 -18.21 -2.31 13.60 0.94

1 3 -11.96 4.20 20.36 0.82

2 3 -9.40 6.51 22.41 0.60

In order to verify that the data is accurate, I continued to carry out two separate KW test data.

A= [G1;G2];

[P,anovatab,stats] = kruskalwallis(A(:,1),A(:,2));

c2=multcompare(stats);

The answer is:

1 2 -10.34 -1.35 7.63 0.77

Got the p value of their difference. Why, Which p value can truly reflect the difference between G1 and G2?

I'm looking forward to your reply very much.