Main Content

kruskalwallis

Kruskal-Wallis test

Description

example

p = kruskalwallis(x) returns the p-value for the null hypothesis that the data in each column of the matrix x comes from the same distribution, using the Kruskal-Wallis test. The alternative hypothesis is that not all samples come from the same distribution. The Kruskal-Wallis test provides a nonparametric alternative to a one-way ANOVA. For more information, see Kruskal-Wallis Test.

example

p = kruskalwallis(x,group) returns the p-value for a test of the null hypothesis that the data in each categorical group, as specified by the grouping variable group comes from the same distribution. The alternative hypothesis is that not all groups come from the same distribution.

example

p = kruskalwallis(x,group,displayopt) returns the p-value of the test and lets you display or suppress the ANOVA table and box plot.

example

[p,tbl,stats] = kruskalwallis(___) also returns the ANOVA table as the cell array tbl and the structure stats containing information about the test statistics.

Examples

collapse all

Create two different normal probability distribution objects. The first distribution has mu = 0 and sigma = 1, and the second distribution has mu = 2 and sigma = 1.

pd1 = makedist('Normal');
pd2 = makedist('Normal','mu',2,'sigma',1);

Create a matrix of sample data by generating random numbers from these two distributions.

rng('default'); % for reproducibility
x = [random(pd1,20,2),random(pd2,20,1)];

The first two columns of x contain data generated from the first distribution, while the third column contains data generated from the second distribution.

Test the null hypothesis that the sample data from each column in x comes from the same distribution.

p = kruskalwallis(x)

Figure Kruskal-Wallis One-way ANOVA contains objects of type uicontrol.

Figure contains an axes object. The axes object contains 21 objects of type line. One or more of the lines displays its values using only markers

p = 3.6896e-06

The returned value of p indicates that kruskalwallis rejects the null hypothesis that all three data samples come from the same distribution at a 1% significance level. The ANOVA table provides additional test results, and the box plot visually presents the summary statistics for each column in x.

Create two different normal probability distribution objects. The first distribution has mu = 0 and sigma = 1. The second distribution has mu = 2 and sigma = 1.

pd1 = makedist('Normal');
pd2 = makedist('Normal','mu',2,'sigma',1);

Create a matrix of sample data by generating random numbers from these two distributions.

rng('default'); % for reproducibility
x = [random(pd1,20,2),random(pd2,20,1)];

The first two columns of x contain data generated from the first distribution, while the third column contains data generated from the second distribution.

Test the null hypothesis that the sample data from each column in x comes from the same distribution. Suppress the output displays, and generate the structure stats to use in further testing.

[p,tbl,stats] = kruskalwallis(x,[],'off')
p = 3.6896e-06
tbl=4×6 cell array
    {'Source' }    {'SS'        }    {'df'}    {'MS'        }    {'Chi-sq'  }    {'Prob>Chi-sq'}
    {'Columns'}    {[7.6311e+03]}    {[ 2]}    {[3.8155e+03]}    {[ 25.0200]}    {[ 3.6896e-06]}
    {'Error'  }    {[1.0364e+04]}    {[57]}    {[  181.8228]}    {0x0 double}    {0x0 double   }
    {'Total'  }    {[     17995]}    {[59]}    {0x0 double  }    {0x0 double}    {0x0 double   }

stats = struct with fields:
       gnames: [3x1 char]
            n: [20 20 20]
       source: 'kruskalwallis'
    meanranks: [26.7500 18.9500 45.8000]
         sumt: 0

The returned value of p indicates that the test rejects the null hypothesis at the 1% significance level. You can use the structure stats to perform additional followup testing. The cell array tbl contains the same data as the graphical ANOVA table, including column and row labels.

Conduct a followup test to identify which data sample comes from a different distribution.

c = multcompare(stats);
Note: Intervals can be used for testing but are not simultaneous confidence intervals.

Figure Multiple comparison of mean ranks contains an axes object. The axes object with title Click on the group you want to test, xlabel The mean ranks of groups 1 and 3 are significantly different contains 7 objects of type line. One or more of the lines displays its values using only markers

Display the multiple comparison results in a table.

tbl = array2table(c,"VariableNames", ...
    ["Group A","Group B","Lower Limit","A-B","Upper Limit","P-value"])
tbl=3×6 table
    Group A    Group B    Lower Limit     A-B      Upper Limit     P-value  
    _______    _______    ___________    ______    ___________    __________

       1          2         -5.1435         7.8       20.744         0.33446
       1          3         -31.994      -19.05      -6.1065       0.0016282
       2          3         -39.794      -26.85      -13.906      3.4768e-06

The results indicate that there is a significant difference between groups 1 and 3, so the test rejects the null hypothesis that the data in these two groups comes from the same distribution. The same is true for groups 2 and 3. However, there is not a significant difference between groups 1 and 2, so the test does not reject the null hypothesis that these two groups come from the same distribution. Therefore, these results suggest that the data in groups 1 and 2 come from the same distribution, and the data in group 3 comes from a different distribution.

Create a vector, strength, containing measurements of the strength of metal beams. Create a second vector, alloy, indicating the type of metal alloy from which the corresponding beam is made.

strength = [82 86 79 83 84 85 86 87 74 82 ...
            78 75 76 77 79 79 77 78 82 79];

alloy = {'st','st','st','st','st','st','st','st',...
         'al1','al1','al1','al1','al1','al1',...
         'al2','al2','al2','al2','al2','al2'};

Test the null hypothesis that the beam strength measurements have the same distribution across all three alloys.

p = kruskalwallis(strength,alloy,'off')
p = 0.0018

The returned value of p indicates that the test rejects the null hypothesis at the 1% significance level.

Input Arguments

collapse all

Sample data for the hypothesis test, specified as a vector or an m-by-n matrix. If x is an m-by-n matrix, each of the n columns represents an independent sample containing m mutually independent observations.

Data Types: single | double

Grouping variable, specified as a numeric or logical vector, a character or string array, or a cell array of character vectors.

  • If x is a vector, then each element in group identifies the group to which the corresponding element in x belongs, and group must be a vector of the same length as x. If a row of group contains an empty value, that row and the corresponding observation in x are disregarded. NaN values in either x or group are similarly ignored.

  • If x is a matrix, then each column in x represents a different group, and you can use group to specify labels for these columns. The number of elements in group and the number of columns in x must be equal.

The labels contained in group also annotate the box plot.

Example: {'red','blue','green','blue','red','blue','green','green','red'}

Data Types: single | double | logical | char | string | cell

Display option, specified as 'on' or 'off'. If displayopt is 'on', kruskalwallis displays the following figures:

  • An ANOVA table containing the sums of squares, degrees of freedom, and other quantities calculated based on the ranks of the data in x.

  • A box plot of the data in each column of the data matrix x. The box plots are based on the actual data values, rather than on the ranks.

If displayopt is 'off', kruskalwallis does not display these figures.

If you specify a value for displayopt, you must also specify a value for group. If you do not have a grouping variable, specify group as [].

Example: 'off'

Output Arguments

collapse all

p-value of the test, returned as a scalar value in the range [0,1]. p is the probability of observing a test statistic as extreme as, or more extreme than, the observed value under the null hypothesis. Small values of p cast doubt on the validity of the null hypothesis.

ANOVA table of test results, returned as a cell array. tbl includes the sums of squares, degrees of freedom, and other quantities calculated based on the ranks of the data in x, as well as column and row labels.

Test data, returned as a structure. You can perform followup multiple comparison tests on pairs of sample medians by using multcompare, with stats as the input value.

More About

collapse all

Kruskal-Wallis Test

The Kruskal-Wallis test is a nonparametric version of classical one-way ANOVA, and an extension of the Wilcoxon rank sum test to more than two groups. The Kruskal-Wallis test is valid for data that has two or more groups. It compares the medians of the groups of data in x to determine if the samples come from the same population (or, equivalently, from different populations with the same distribution).

The Kruskal-Wallis test uses ranks of the data, rather than numeric values, to compute the test statistics. It finds ranks by ordering the data from smallest to largest across all groups, and taking the numeric index of this ordering. The rank for a tied observation is equal to the average rank of all observations tied with it. The F-statistic used in classical one-way ANOVA is replaced by a chi-square statistic, and the p-value measures the significance of the chi-square statistic.

The Kruskal-Wallis test assumes that all samples come from populations having the same continuous distribution, apart from possibly different locations due to group effects, and that all observations are mutually independent. By contrast, classical one-way ANOVA replaces the first assumption with the stronger assumption that the populations have normal distributions.

Version History

Introduced before R2006a