Spearman's Rank: A Guide To
Spearman's Rank: A Guide To
Spearman's Rank: A Guide To
Spearman’s Rank
The Spearman’s Rank Correlation Coefficient is a statistical test that examines the degree to
which two data sets are correlated, if at all. While a scatter graph of the two data sets may give the
researcher a hint towards whether the two have a correlation, Spearman’s Rank gives the researcher
a numerical value on the degree of correlation, or indeed, the degree of non-correlation. It is a
relatively straight forward analysis for those researchers whom are not wholly confident in their
mathematical skills.
In order to use Spearman’s Rank the researcher must have paired sets of data that are in some way
related (such as the geographical site where they were collected in the field). It is a good idea for the
researcher to have at least ten pairs of data to use for the analysis: any fewer than this and the result
will be highly insignificant and more likely to be as a result of chance than of true correlation.
In the following example, the researcher is looking at whether for River X, the channel width
increases as the distance from the source increases. Theoretically, this should be true, but the
Spearman’s Rank analysis will tell the researcher whether it is true in this case that there is a
correlation and the strength of any such correlation.
1. The researcher should arrange the paired data in a table to allow for ease of analysis. This
can be done in a spreadsheet package or through hand written methods.
Distance
𝒅
Site from Rank (𝑹1) Width (m) Rank (𝑹2) 𝒅2
(𝑹1 – 𝑹2)
source (m)
1 150 0.40
2 300 0.80
3 450 1.00
4 600 0.95
5 750 1.20
6 900 1.10
7 1050 1.30
8 1200 1.40
9 1350 1.85
10 1500 2.40
11 1650 2.55
12 1800 3.20
13 1950 3.80
14 2100 3.60
15 2250 3.20
Total
This project was funded by the Nuffield Foundation, but the views expressed are those of the authors and not necessarily those of the Foundation.
2. Then researcher should rank each data, starting with 1 as the smallest figure and (in this
case) 15 as the largest. Where there might be two values that are equal, the researcher
should average the ranks and omit the ranking values they cover. For example, a set of
rankings may read: 1, 2, 3.5, 3.5, 5, 6, 7.25, 7.25, 7.25, 7.25, 11, 12, etc
Distance
𝒅
Site from Rank (𝑹1) Width (m) Rank (𝑹2) 𝒅2
(𝑹1 – 𝑹2)
source (m)
1 150 1 0.40 1
2 300 2 0.80 2
3 450 3 1.00 4
4 600 4 0.95 3
5 750 5 1.20 6
6 900 6 1.10 5
7 1050 7 1.30 7
8 1200 8 1.40 8
9 1350 9 1.85 9
10 1500 10 2.40 10
11 1650 11 2.55 11
12 1800 12 3.20 12.5
13 1950 13 3.80 15
14 2100 14 3.60 14
15 2250 15 3.20 12.5
Total
3. The difference ( 𝑑 ) between the two ranks should then be calculated by subtracting 𝑅1
from 𝑅2:
Distance
𝒅
Site from Rank (𝑹1) Width (m) Rank (𝑹2) 𝒅2
(𝑹1 – 𝑹2)
source (m)
1 150 1 0.40 1 0
2 300 2 0.80 2 0
3 450 3 1.00 4 -1
4 600 4 0.95 3 1
5 750 5 1.20 6 -1
6 900 6 1.10 5 1
7 1050 7 1.30 7 0
8 1200 8 1.40 8 0
9 1350 9 1.85 9 0
10 1500 10 2.40 10 0
11 1650 11 2.55 11 0
12 1800 12 3.20 12.5 -0.5
13 1950 13 3.80 15 -2
14 2100 14 3.60 14 0
15 2250 15 3.20 12.5 2.5
Total
2
4. 𝑑 should then be squared to remove any negative values. The total value of all the 𝑑2 can
also be calculated at this stage.
Distance
𝒅
Site from Rank (𝑹1) Width (m) Rank (𝑹2) 𝒅2
(𝑹1 – 𝑹2)
source (m)
1 150 1 0.40 1 0 0
2 300 2 0.80 2 0 0
3 450 3 1.00 4 -1 0
4 600 4 0.95 3 1 1
5 750 5 1.20 6 -1 0
6 900 6 1.10 5 1 1
7 1050 7 1.30 7 0 0
8 1200 8 1.40 8 0 0
9 1350 9 1.85 9 0 0
10 1500 10 2.40 10 0 0
11 1650 11 2.55 11 0 0
12 1800 12 3.20 12.5 -0.5 0.25
13 1950 13 3.80 15 -2 4
14 2100 14 3.60 14 0 0
15 2250 15 3.20 12.5 2.5 6.25
Total 12.5
5. One should then apply the Spearman’s Rank equation to calculate the coefficient value (𝑅)
(the value that tells the researcher the strength of the correlation).
6Ʃ𝑑 2
𝑅 = 1 −
(𝑛3 −𝑛)
where 𝑛 is the number of pairs of data collected and used (in this case 15). The sum of the 𝑑2 values
(Ʃ𝑑2) in this example is 12.5.
75 75
𝑅 = 1 − 𝑅 = 1 − 𝑅 = 1 − 0.022
(3375−15) 3360
𝑅 = 1 − 0.022 𝑅 = 0.978
The coefficient (𝑅) will be between a value of -1 and +1 where -1 indicates a perfect negative
correlation and +1 indicates a perfect positive correlation. A value of between -0.7 to +0.7 is
generally seen as being too weak to be thought of as a significant result.
3
Therefore, the data in this example shows a strong positive correlation between channel width and
the distance from the source.
6. To check whether the result is meaningful or is just down to chance, the value for 𝑅 can be
compared with the critical value for 𝑛 in the Spearman’s Rank significance table.
Below is the significance table for some values of 𝑛, but for analysis of larger sets of data,
extended significance tables can be found online.
Significance Level
𝑛 0.1 0.05 0.01
4 1.000 1.000 1.000
5 0.700 0.900 1.000
6 0.657 0.771 0.943
7 0.571 0.679 0.857
8 0.548 0.643 0.810
9 0.483 0.600 0.767
10 0.442 0.564 0.733
11 0.418 0.527 0.700
12 0.399 0.504 0.671
13 0.379 0.478 0.648
14 0.367 0.459 0.622
15 0.350 0.443 0.600
16 0.338 0.427 0.582
17 0.327 0.412 0.558
18 0.317 0.400 0.543
The critical value for this example, where there are 15 pairs of data ( 𝑛 = 15 ), is 0.443. As the value
of 𝑅 is greater than the critical value, we can say with 95% certainty that the results we have observed
have not occurred by chance. This means the results are highly significant and sound conclusions
can be drawn from them.