SPSS TUTORIALS BASICS DATA ANALYSIS T-TEST ANOVA CHI-SQUARE TEST

# Spearman Rank Correlations – Simple Tutorial

A Spearman rank correlation is a number between -1 and +1 that indicates to what extent 2 variables are monotonously related.

## Spearman Correlation - Example

A sample of 1,000 companies were asked about their number of employees and their revenue over 2018. For making these questions easier, they were offered answer categories. After completing the data collection, the contingency table below shows the results.

The question we'd like to answer is is company size related to revenue? A good look at our contingency table shows the obvious: companies having more employees typically make more revenue. But note that this relation is not perfect: there's 60 companies with 1 employee making $50,000 -$99,999 while there's 89 companies with 2-5 employees making $0 -$49,999. This relation becomes clear if we visualize our results in the chart below.

The chart shows an undisputable positive monotonous relation between size and revenue: larger companies tend to make more revenue than smaller companies. Next question. How strong is the relation? The first option that comes to mind is computing the Pearson correlation between company size and revenue. However, that's not going to work because we don't have company size or revenue in our data. We only have size and revenue categories. Company size and revenue are ordinal variables in our data: we know that 2-5 employees is larger than 1 employee but we don't know how much larger.
So which numbers can we use to calculate how strongly ordinal variables are related? Well, we can assign ranks to our categories as shown below.

As a last step, we simply compute the Pearson correlation between the size and revenue ranks. This results in a Spearman rank correlation (Rs) = 0.81. This tells us that our variables are strongly monotonously related. But in contrast to a normal Pearson correlation, we do not know if the relation is linear to any extent.

## Spearman Rank Correlation - Basic Properties

Like we just saw, a Spearman correlation is simply a Pearson correlation computed on ranks instead of data values or categories. This results in the following basic properties:

• Spearman correlations are always between -1 and +1;
• Spearman correlations are suitable for all but nominal variables. However, when both variables are either metric or dichotomous, Pearson correlations are usually the better choice;
• Spearman correlations indicate monotonous -rather than linear- relations;
• Spearman correlations are hardly affected by outliers. However, outliers should be excluded from analyses instead of determine whether Spearman or Pearson correlations are preferable;
• Spearman correlations serve the exact same purposes as Kendall’s tau.

## Spearman Rank Correlation - Assumptions

• The Spearman correlation itself only assumes that both variables are at least ordinal variables. This excludes all but nominal variables.
• The statistical significance test for a Spearman correlation assumes independent observations or -precisely- independent and identically distributed variables.

## Spearman Correlation - Example II

A company needs to determine the expiration date for milk. They therefore take a tiny drop each hour and analyze the number of bacteria it contains. The results are shown below.

For bacteria versus time,

• the Pearson correlation is 0.58 but
• the Spearman correlation is 1.00.

There is a perfect monotonous relation between time and bacteria: with each hour passed, the number of bacteria grows. However, the relation is very non linear as shown by the Pearson correlation.
This example nicely illustrates the difference between these correlations. However, I'd argue against reporting a Spearman correlation here. Instead, model this curvilinear relation with a (probably exponential) function. This'll probably predict the number of bacteria with pinpoint precision.

## Spearman Correlation - Formulas and Calculation

First off, an example calculation, exact significance levels and critical values are given in this Googlesheet (shown below).

Right. Now, computing Spearman’s rank correlation always starts off with replacing scores by their ranks (use mean ranks for ties). Spearman’s correlation is now computed as the Pearson correlation over the (mean) ranks.

Alternatively, compute Spearman correlations with $$R_s = 1 - \frac{6\cdot \Sigma \;D^2}{n^3 - n}$$
where $$D$$ denotes the difference between the 2 ranks for each observation.

For reasonable sample sizes of N ≥ 30, the (approximate) statistical significance uses the t distribution. In this case, the test statistic $$T = \frac{R_s \cdot \sqrt{N - 2}}{\sqrt{1 - R^2_s}}$$
follows a t-distribution with $$Df = N - 2$$
degrees of freedom.

This approximation is inaccurate for smaller sample sizes of N < 30. In this case, look up the (exact) significance level from the table given in this Googlesheet. These exact p-values are based on a permutation test that we may discuss some other time. Or not.

## Spearman Rank Correlation - Software

Spearman correlations can be computed in Googlesheets or Excel but statistical software is a much easier option. JASP -which is freely downloadable- comes up with the correct Spearman correlation and its significance level as shown below.

SPSS also comes up with the correct correlation. However, its significance level is based on the t-distribution: $$t = \frac{0.77\cdot\sqrt{4}}{\sqrt{(1 - 0.77^2)}} = 2.42$$
and $$t(4) = 2.42,\;p = 0.072$$
Again, this approximation is only accurate for larger sample sizes of N ≥ 30. For N = 6, it is wildly off as shown below.

# Tell us what you think!

*Required field. Your comment will show up after approval from a moderator.

# THIS TUTORIAL HAS 7 COMMENTS:

• ### By Jon K Peck on July 16th, 2019

It is true that the t approximation deteriorates with very small sample sizes. I have discussed this with the SPSS statisticians, and they may make an improvement, but I can't speak for product management. I should point out, however, that a bootstrapped alternative is available here if the bootstrap option is licensed.
Also, I should point out that the alternative formula given above is not valid if there are ties in the data. I don't know whether ties affect the small-sample alternative methods, but the R cor.test function reverts to the t approximation if they are present.

• ### By Jon K Peck on July 16th, 2019

p.s. I have now confirmed that the exact method for small samples is not correct in the presence of ties.

WELL DONE

• ### By Ruben Geert van den Berg on July 18th, 2019

Hi Jon!

First off, ties can't occur if you have respondents rank a number of options as in the example I presented. For this example, the JASP p-value corresponds to the p-value I found in text books and it's pretty different than the SPSS p-value.

I'm not sure about the best method if ties are present in some other dataset. The best formula for the significance of Spearman correlations in small samples is way beyond the scope of my simple introduction.

A comparison among the JASP p-values, the SPSS p-values and bootstrapped p-values could provide some answers but I don't find it interesting enough to spend any time on.

• ### By Jon K Peck on July 18th, 2019

My point was not that the significance approximation should not be improved for very small samples but that the exact method and the quick formula posted are not a general solution.

So, I wondered how the accuracy varies with sample size. I generated two random datasets with no ties. One is normal data and the other is binomial with some fuzz to eliminate ties. I compared the exact and t-based results for sample sizes of 5, 10, ..., 30 and calculated the percentage error in the t significance approximation with the exact result.
With N=5, the errors are 8.7% and 19.4%, respectively. I.e., for normal, the exact value was ..6833 vs .6238 and .233 vs .188 for t.
But by N=10, the percentage differences were .61% and .59% (.0033 and .0030 in the sig levels), and the differences continued to shrink rapidly.
So, even by an N of 10, the error in the t approximation was negligible.
Of course, I only looked at these two random samples, but this suggests that the approximation error does not matter except with extremely small data.