Minitab Tutorial for Nonparametric Statistics: Rank Tests 1

Measurements of height and weight of 22 Truman students were taken, resulting in the following data:
Height (in) 69 70 65 72 76 70 70 66 68 73
Male
Weight (lb) 192 148 140 190 248 197 170 137 160 185

Height (in) 65 61 67 65 70 62 63 60 66 66 65 64
Female
Weight (lb) 110 105 136 135 187 125 147 118 128 175 147 120

When you enter it in Minitab, it should look like this:

Say we want to determine if the average height for females

at Truman is 63 inches. When running a t-test, the two-
sided hypothesis would be:

H o :  F  63 inches & H A :  F  63 inches

Since the non-parametric tests look at median, we need to rewrite these as:

H o :  F  63 inches & H A :  F  63 inches

Wilcoxon Signed-Rank Test (One-Sample):

To test the height of females according to our hypothesis go to Stat: Nonparametric: 1-Sample
Wilcoxon. We enter ‘FemaleHeight’ as our variable and choose the ‘test median’ option. Enter 63 in
the ‘test median:’ box. In this case we have a two-sided hypothesis, so we choose ‘not equal,’ but
notice the ‘less than’ and ‘greater than’ options in the drop down menu for the alternative
hypothesis. These will be used for one-sided hypotheses.

The output looks like this:

The hypothesis is given, so make sure it
matched what you had intended to test.

Our p-value is 0.100, so at the 95% level, we fail to reject the null hypothesis and conclude that it is
possible that the median height of women could be 63 inches.
Minitab Tutorial for Nonparametric Statistics: Rank Tests 2

Sign Test (One-Sample):

The sign test is very similar and we will use the same hypothesis. This time go to Stat:
Nonparametric: 1-Sample Sign.

Our output looks like this:

The p-value this time is 0.2266, so once again we fail to reject
the null hypothesis.

Sign and Wilcoxon Tests (Matched Pairs):

We can also run the Wilcoxon and Sign Tests for matched-ordinal data. The process is very similar
to what we just did, except this time we must calculate the difference of our two samples.Then we
set the ‘test median’ to 0. Notice how this will affect our hypothesis. If there is no difference in our
pairs, then the differences should be equal to 0, but if a difference is found then the median
difference will not be equal to 0.
H0: RT-RB = 0 HA: RT-RB 0

The p-value for the Wilcoxon test is 0.036, so we reject the null hypothesis and conclude that the
difference is not equal to 0. This means that we have evidence to conclude that the two samples are
different.

The same process is used for the sign test. The p-value
is 0.0313, so again we reject the null hypothesis.
Minitab Tutorial for Nonparametric Statistics: Rank Tests 3

Mann-Whitney Test (Two Independent Samples):

This test can be used to compare two samples that are independent of each other, such as height of
males verse females. Our hypothesis is:

H 0 :F   M & H A :F   M

This time we go to Stat: Nonparametric: Mann-Whitney. Enter ‘MaleHeight’ as the first sample,
and ‘FemaleHeight’ as the second sample. The output looks like this:

We see that the p-value is 0.0017, and so we reject

the null hypothesis and conclude that we have
statistically significant evidence that the heights of
males verse females are different.

Kruskal-Wallis Test(3+ Samples):

Next we will look at a test that will allow us to compare more than two samples. The following
example with essentially look at four samples, and will compare year in school to the number of
credit hours that students are taking. Consider the following data set:
3 2 3 4 1 2 2 1 3 2 1 2 4 3 1 3 4 1 2 1 1 4 3 2 1 4

15 14 12 15 15 15 17 15 12 15 17 18 9 15 12 12 15 18 15 15 15 9 12 14 13 15

Our null hypothesis would be that year in school has no effect on credit hours taken. The alternate
hypothesis is that there is a difference between the four groups. After entering the data, we go to
Stat: Nonparametric: Kruskal-Wallis. We enter year in school as the ‘factor’ and credit hours as
the ‘response.’ Here is the output:

We see that the p-value is 0.161, so we cannot reject the

null hypothesis. We have found no difference in number of
credit hours taken based on year in school.

Friedman Test:
For this test we will look at scores of Judge Gymnast Score
gymnasts from three different judges based on 1 1 9
what gym the attend. Here is the data set:
2 1 8
Minitab Tutorial for Nonparametric Statistics: Rank Tests 4

3 1 9
1 2 7
2 2 9
3 2 6
1 3 8
2 3 7
3 3 8
1 4 8
2 4 7
3 4 9
1 5 6
2 5 7
3 5 4 Since score is what is being measured, this is the
1 6 8 response variable. The treatment is the judge, and
the gymnast is the block variable. Here is our
2 6 9
output:
3 6 9
1 7 9
2 7 6
3 7 10
1 8 7
2 8 9
3 8 7
1 9 7
2 9 8 Since the p-value is 0.697, we fail to reject the null
hypothesis, and conclude that no one judge scores
3 9 10 differently than all of the other judges.

We want to determine if scoring of any one judge

differs from the other judges. Our null hypothesis
is that there is no difference. Our alternate
hypothesis is that one judge scores differently than
the others.

We go to Stat: Nonparametric: Friedman.