What Is Spearman Rank Correlation?
Imagine you ask ten students how many hours they studied for an exam, then rank them from first to tenth by their exam score. The student who studied the most also scored the highest, the second-most-studied student came second, and so on. The relationship is not a perfect straight line — but it is perfectly consistent. That consistency is exactly what Spearman rank correlation measures.
Spearman rank correlation, written as Spearman’s rho (ρ or rs), measures how strongly two variables move together in the same direction — not necessarily in a straight line, but in a steady, monotonic pattern. If one variable goes up and the other tends to go up too (or tends to go down), Spearman’s rho captures that. It works by comparing the ranks of your data rather than the raw values, which makes it one of the most forgiving and practical correlation coefficients in a researcher’s toolkit.
It is named after Charles Spearman, the English psychologist who introduced it in 1904. More than a century later, it remains the go-to correlation whenever your data are ranked, ordinal, skewed, or simply not well-behaved enough for Pearson’s r.
Spearman vs Pearson: The Key Difference in One Sentence
Pearson correlation asks: “Do these two variables form a straight line?” Spearman correlation asks a gentler question: “Do these two variables rise and fall together, whatever the shape?”
That single difference decides which one you should use. If your data are roughly normally distributed, measured on an interval or ratio scale, and the relationship looks linear, Pearson is the right choice. If any of those conditions fail — your data are ordinal, skewed, contain outliers, or the relationship curves — Spearman is the safer and more honest option.
When to Use Spearman Instead of Pearson
Reach for Spearman’s rho in any of these situations:
- Your data are ordinal. Likert-scale responses, rankings, grades, or competition placements are ordinal — the numbers tell you the order, not the exact distance between values. Pearson treats those distances as real; Spearman only needs the order.
- The relationship is monotonic but not linear. Perhaps exam scores rise quickly with the first few hours of study and then level off. Pearson would underestimate that relationship; Spearman captures it fully because it only cares that scores keep rising, not that they rise at a constant rate.
- Your data are skewed or not normally distributed. Reaction times, income, and citation counts are classic skewed variables. Ranking the data washes out the skew.
- You have outliers. One extreme value can drag Pearson’s r up or down dramatically. Because Spearman works with ranks, the largest value is simply “rank 1” — no more influential than any other rank.
- Your sample is small. With small samples, normality is hard to verify, and Spearman’s rank-based approach makes fewer assumptions.
Put simply: when Pearson’s assumptions look shaky, Spearman is your plan B — and often the better plan A.
Spearman vs Pearson: A Side-by-Side Comparison
| Feature | Pearson (r) | Spearman (rho / rs) |
|---|---|---|
| What it measures | Strength of a linear relationship | Strength of a monotonic relationship |
| Data it uses | Raw values | Ranks of the values |
| Measurement level | Interval or ratio | Ordinal, interval, or ratio |
| Assumes normality? | Yes, for significance testing | No |
| Sensitive to outliers? | Yes — one extreme value can distort r | Much less — ranks limit any value’s influence |
| Handles curved relationships? | No — it will underestimate them | Yes, as long as the trend keeps one direction |
| Range | −1 to +1 | −1 to +1 |
| Typical use | Well-behaved continuous data | Ranked, ordinal, skewed, or outlier-prone data |
How Spearman’s Rho Is Calculated (With a Worked Example)
The classic formula for Spearman’s rho is:
rs = 1 − (6 × Σd²) ÷ (n × (n² − 1))
Here, d is the difference between the two ranks for each case, and n is the number of cases. (Statistical software uses a slightly different version when there are tied ranks, but the idea is identical.)
A worked example. Suppose you rank 5 students by hours studied and by exam score:
| Student | Rank: hours studied | Rank: exam score | d (difference) | d² |
|---|---|---|---|---|
| A | 1 | 2 | −1 | 1 |
| B | 2 | 1 | 1 | 1 |
| C | 3 | 3 | 0 | 0 |
| D | 4 | 5 | −1 | 1 |
| E | 5 | 4 | 1 | 1 |
Σd² = 4, and n = 5. Plugging in: rs = 1 − (6 × 4) ÷ (5 × 24) = 1 − 24 ÷ 120 = 1 − 0.20 = 0.80. That is a strong positive monotonic relationship: students who studied more tended to score higher, with only small shuffles in the ranking.
In practice, you will almost never calculate this by hand — SPSS, R, Python, and even Excel handle it in seconds. In SPSS, go to Analyze → Correlate → Bivariate, move your variables across, and tick “Spearman” instead of “Pearson”. The worked example matters because it shows you what the number actually means: how closely two rankings agree.
How to Interpret Spearman’s Rho
Like Pearson’s r, Spearman’s rho ranges from −1 to +1:
- +1: a perfect positive monotonic relationship — the rankings match exactly.
- 0: no monotonic relationship at all.
- −1: a perfect negative monotonic relationship — as one ranking rises, the other falls in exact mirror image.
For judging strength, researchers commonly use Cohen’s (1988) guidelines: around .10 is small, .30 is medium, and .50 is large. But treat these as rough signposts, not laws. In some fields a rho of .30 is a major finding; in others it is barely worth reporting. Always interpret the number in the context of your field, your sample size, and your research question — and always report the p-value alongside it.
One warning that applies to every correlation: correlation is not causation. A rho of .80 between study hours and exam scores does not prove that studying caused the higher scores. Motivation, prior knowledge, or sleep could be driving both. Spearman tells you that two rankings move together — nothing more, and nothing less.
Common Mistakes When Using Spearman Correlation
1. Using Spearman when Pearson is the better fit. Spearman is not automatically “safer” for every dataset. If your data meet Pearson’s assumptions, Pearson uses more information (the actual values, not just their ranks) and is more powerful. Use Spearman when its assumptions fit your data better, not out of habit.
2. Forgetting that Spearman only detects monotonic relationships. If the relationship is U-shaped — say, both very low and very high anxiety hurt performance, while moderate anxiety helps — Spearman will report a rho near zero. It cannot see curves that change direction. Always plot your data first.
3. Reporting rho without the sample size and p-value. A rho of .60 from 8 participants is far less convincing than the same value from 200. Report all three: rho, n (or degrees of freedom), and p.
4. Ignoring tied ranks. When two cases share the same value, they receive the average of the ranks they would have occupied. Most software handles this automatically, but the simple hand-calculation formula above is slightly off when there are many ties — another reason to use software for real analyses.
5. Interpreting rho as a percentage. A rho of .50 does not mean “50% related”. If you want a percentage-style interpretation, square Pearson’s r — but that logic does not transfer neatly to ranks, so report rho as a strength of association and leave percentages out of it.
How to Report Spearman Correlation in APA Style (7th Edition)
APA 7th edition reports Spearman’s rho as rs, italicized, with the degrees of freedom in parentheses — degrees of freedom are n − 2, just as for Pearson. Include the value to two decimal places, the p-value, and the sample size if it is not clear from the context.
Example sentence: “There was a strong positive association between hours studied and exam score, rs(28) = .62, p = .001.”
If the result is not significant, say so plainly: “The association between sleep quality ranking and stress ranking was not significant, rs(45) = .18, p = .23.”
When you report several correlations, present them in a table with the rho values, sample sizes, and significance levels — and state in the table note that Spearman’s rho was used because the data were ordinal (or skewed, or contained outliers). Reviewers want to know why you chose Spearman, not just that you did.
Frequently Asked Questions
Can I use Spearman correlation with Likert scale data?
Yes — this is one of its most common uses. Likert responses are ordinal, so ranking-based Spearman is a natural fit. It is the standard choice when correlating two Likert-scale variables, or a Likert variable with another ranked variable.
What sample size do I need for Spearman correlation?
There is no hard minimum, but with fewer than about 10 cases the result is very unstable and significance tests have little power. For a thesis, aim for at least 20–30 cases, and remember that the p-value depends heavily on n: the same rho can be significant with 100 cases and non-significant with 15.
Is Spearman correlation the same as Pearson correlation on ranked data?
Essentially, yes. When there are no tied ranks, Spearman’s rho is mathematically equal to Pearson’s r computed on the ranks instead of the raw values. That is exactly why it is robust: ranking is what tames outliers and skew.
When should I choose Kendall’s tau instead of Spearman?
Kendall’s tau is another rank correlation, and it is often preferred when your sample is small or when there are many tied ranks — common with short Likert scales. Spearman is more widely known and easier to interpret; Kendall’s tau is slightly more robust with ties. For most student research, either is defensible — just justify your choice.
Can Spearman correlation be negative?
Yes. A negative rho means that as one variable’s ranking rises, the other’s tends to fall — for example, higher stress rankings paired with lower sleep-quality rankings. A rho of −.70 is just as strong as +.70; only the direction differs.
The Bottom Line
Choose Pearson when your data are continuous, roughly normal, and the relationship is linear. Choose Spearman when your data are ranked or ordinal, skewed, outlier-prone, or the relationship curves without changing direction. Spearman’s rho is not a weaker version of Pearson — it answers a different, often more realistic question: do these two things rise and fall together? For much of the data researchers actually collect, that is exactly the question worth asking.