9.3.2.1.

Independent Groups: Mann-Whitney-U Test

The Mann-Whitney U test is used to compare differences between two independently measured
groups. It works with continuous ordinal data. For example, data on gender (e.g. male 1 and
female 2), education level (e.g. SD 1, SMP 2, SMA/K 3, bachelor 4, master 3, PhD 4),
employment (e.g. unemployed 1, employed 2), and the like. It is an alternative to t-test when the
normality of distributions is failed to be fulfilled.

Before using this Mann-Whitney U test, you also have to assess whether your data can actually
fulfill the requirements for its use. Firstly, make sure that the data, or the dependent variables, are
ordinal continuous. Again, in ELT, such scores will not be relevant and, as in the exercise below,
nominal continuous data of students’ test scores have to be ranked due to abnormality in
distribution and for analytical purposes. Secondly, there should be two independent variables. As
in independent group design, the variables will come from two measures of students’ post-tests.
Thirdly, the two data sets result from independent observation, for example, from different
samples or participants. Finally, the distribution in the two data sets is symmetrical.

Imagine that you experimented on two independent groups in teaching academic writing:
independent work (IW) and collaborative work (CW). Before the treatments you did not have
chances to give them a pre-test. After the treatments, you administered independent posttests and
the results are shown in the table. You repeated each treatment three or four times and in the end
you test the students with the posttests.

Table 7: Test Results

IW CW
No X No Y
Sample Sample
1 H 2 1 JK 2
2 N 2 2 YG 3
3 O 3 3 KL 3
4 Q 3 4 KO 3
5 I 4 5 BG 3
6 P 4 6 LI 3
7 G 6 7 TU 5
8 A 7 8 AS 6
9 C 7 9 DS 6
10 F 7 10 GF 6
11 L 7 11 TY 7
12 T 7 12 JH 7
13 B 8 13 DE 7
14 D 8 14 GH 7
15 E 8 15 RF 7
16 J 8 16 HK 7
17 M 8 17 PO 7
 18 R 8 18 KP 7
19 S 8 19 FR 8
20 K 9 20 HB 8

Certainly, you would be interested in identifying whether there is a significant difference

between the two treatments. In the previous sections, we have identified this by using t-test. But,
the data above do not qualify for t-test as the number of data is too small (N=20). In addition,
neither of data distribution is normal. The mean, median and modus scores are different. Thus,
you need to use a non-parametric test for independent group design.

One of statistic tests for independent group is the Mann-Whitney U test. As it works with
continuous ordinal data, you need to transform the continuous nominal data above into ordinal
ones. You do this by ranking them from the lowest to the highest score. As you have 40, you will
have to rank them from 1 to 40, as shown below. But there are ties, and each score in these ties
deserve the same rank. You do this by adding all the ranks and divided them with the number of
data, as shown in the previous sections. Thus, the rank should be shared and this shared rank is
called signed rank.

Table 8: Rank and Signed Rank

Signed Signed
No Data Rank No Data Rank
Rank Rank
1 2 1 2 21 7 21 24
2 2 2 2 22 7 22 24
3 2 3 2 23 7 23 24
4 3 4 7 24 7 24 24
5 3 5 7 25 7 25 24
6 3 6 7 26 7 26 24
7 3 7 7 27 7 27 24
8 3 8 7 28 7 28 24
9 3 9 7 29 7 29 24
10 3 10 7 30 7 30 24
11 4 11 11.5 31 8 31 34.5
12 4 12 11.5 32 8 32 34.5
13 5 13 13 33 8 33 34.5
14 6 14 15.5 34 8 34 34.5
15 6 15 15.5 35 8 35 34.5
16 6 16 15.5 36 8 36 34.5
17 6 17 15.5 37 8 37 34.5
18 7 18 24 38 8 38 34.5
19 7 19 24 39 8 39 34.5
20 7 20 24 40 9 40 40

With these signed ranks, you need to go back to your working table and assign the signed ranks
to the data in each group. For example, data 2 is assigned rank 2, data 3 is 7, data 6 is 15.5, data
8 is 34.5 and data 9 is 40. When all the data have been assigned ranks, you sum them up to
obtain Rank Sum of each group. This is shown in the table below.
Table 8: Working Table

IW CW
No X Y Rx Ry
Sample Sample
1 H 2 JK 2 2 2
2 N 2 YG 3 2 7
3 O 3 KL 3 7 7
4 Q 3 KO 3 7 7
5 I 4 BG 3 11.5 7
6 P 4 LI 3 11.5 7
7 G 6 TU 5 15.5 13
8 A 7 AS 6 24 15.5
9 C 7 DS 6 24 15.5
10 F 7 GF 6 24 15.5
11 L 7 TY 7 24 24
12 T 7 JH 7 24 24
13 B 8 DE 7 34.5 24
14 D 8 GH 7 34.5 24
15 E 8 RF 7 34.5 24
16 J 8 HK 7 34.5 24
17 M 8 PO 7 34.5 24
18 R 8 KP 7 34.5 24
19 S 8 FR 8 34.5 34.5
20 K 9 HB 8 40 34.5
∑ 458 357.5

Your next job is to run the Mann-Whitney U test on each group of scores and you do this by
using the following formula.
𝑁(𝑁+1)
Utest = Rank Sum – 2

N is the number of sample in each group.

Thus, Utest for the IW group is as follow

𝑁(𝑁+1)
UIW = Rank Sum – 2

20(20+1)
= 458- ( )
2
= 458 – 210
= 248
 𝑁(𝑁+1)
UCW = Rank Sum – 2

20(20+1)
= 357.5-( )
2
= 357.5 – 210
= 147.5

Now, you have two U values. You choose the lowest one, that is 147.5, and use it as U-count to
consult U-table. Note that the N of IW group is 20 and the N of CW group is coincidentally also
20. Draw a line from N1 20 and N2 20 and at the cross section of the lines is the maximum value
of U to reject H0. The number from the table is 127. Thus, to reject H0 you need a smaller U-
count. But, your U-count is 147.5. It is greater than U-table at alpha .05. Thus, the null
hypothesis is failed to be rejected and it should not be changed in favour of the alternate
hypothesis: there is no significant difference in the ranks.

U-Table