You have two continuous variables and you want to measure the relationship between them. SPSS gives you two options on the Correlate menu: Pearson and Spearman. Most students pick Pearson by default, run the test, report the coefficient, and move on. In roughly 4 out of 10 projects we review, that default choice turns out to be wrong. The two coefficients measure different things, assume different data properties, and can produce very different results on the same dataset. Knowing when each one applies will save you a revision.
What Pearson's r Measures
Pearson's r quantifies the strength and direction of a linear relationship between two continuous variables. If you plot weekly exercise hours against resting heart rate for 100 participants and the dots fall roughly along a straight line sloping downward, Pearson captures that pattern. A value of r = −.62 tells you the relationship is negative (more exercise, lower heart rate), and by Cohen's benchmarks (.10 small, .30 medium, .50 large) the association is strong.
The coefficient comes with assumptions. Both variables must be measured on an interval or ratio scale. The data should be approximately bivariate normal, meaning each variable is roughly normally distributed and their joint distribution forms an elliptical cloud. There should be no extreme outliers, because Pearson's formula uses raw values, and a single participant earning ten times the sample mean can drag the coefficient substantially. You also need a normality check before reporting the result. Shapiro-Wilk tests on both variables, plus a quick scatterplot, take less than a minute and can prevent a misleading coefficient.
The linearity requirement is the part students overlook most often. Pearson's r only detects straight-line patterns. If the true relationship between two variables curves upward and then levels off, Pearson will underestimate the association because it tries to fit a line through a curve. Two variables can be perfectly related in a monotonic way and still produce a modest Pearson r.
What Spearman's ρ Measures
Spearman's rank correlation (rs or ρ) measures the strength and direction of a monotonic relationship. Instead of working with raw values, it first converts each variable to ranks and then computes Pearson's r on those ranks. This rank transformation is what gives Spearman its flexibility.
Because ranks compress the distance between data points, extreme values lose their pull. A participant earning €400,000 in a sample where everyone else earns between €20,000 and €80,000 gets the rank of 1 (or n, depending on direction), not a value five times larger than the next observation. That single change makes Spearman far more resistant to outliers than Pearson.
Spearman also handles ordinal data directly. If you measure patient satisfaction on a 5-point Likert scale and correlate it with waiting time in minutes, Spearman is the correct choice. The satisfaction variable has no true interval spacing (the difference between "satisfied" and "very satisfied" is not necessarily the same as the difference between "neutral" and "satisfied"), so Pearson's assumption of interval-level measurement is violated. In our experience, roughly half of all thesis correlation analyses involve at least one ordinal variable, and Spearman should be the default in those cases.
A Practical Example Where They Differ
Consider a study of 100 employees examining the relationship between annual income and job satisfaction (measured on a 7-point scale). Most employees earn between €25,000 and €65,000, but eight executives in the sample earn between €120,000 and €310,000. The scatterplot shows a clear monotonic trend: as income increases, satisfaction tends to increase. But the relationship is not linear. Satisfaction rises steeply across the €25,000–€65,000 range and then flattens out among the high earners, who all report satisfaction scores of 5, 6, or 7 regardless of whether they earn €120,000 or €310,000.
Pearson's r on this dataset returns .31. The eight extreme income values stretch the x-axis and pull the regression line flatter, weakening the apparent association. Spearman's rs returns .58. By converting incomes to ranks, Spearman removes the distortion caused by those extreme values and captures the consistent upward trend that is visible in the scatterplot. The difference between .31 and .58 could change how a supervisor evaluates the finding. One suggests a weak association; the other suggests a strong one. The Spearman value reflects what the data actually show.
How to Decide Between Them
Start with your scatterplot. Always. If the dots follow a straight-line pattern with no obvious curve and no extreme outliers, Pearson is appropriate. If the dots follow a monotonic curve (consistently going up or consistently going down, but bending along the way), Spearman captures that trend better. If you see a cluster of points with a few values far from the pack, those outliers will distort Pearson but barely affect Spearman.
Next, check your measurement scales. Any variable measured on an ordinal scale (Likert items, ranked preferences, education levels coded as 1–5) should be analyzed with Spearman. Pearson requires both variables to be interval or ratio. This is not a suggestion. If your supervisor or methodology textbook says "Pearson requires continuous, interval-level data," they mean Spearman is the alternative when that condition is not met.
Then run normality tests. If either variable fails the Shapiro-Wilk test at p < .05 and the sample is under 300, Spearman is the safer option. Pearson can tolerate mild departures from normality in large samples (the Central Limit Theorem helps with n > 30 for the sampling distribution of r), but with heavy skew or kurtosis, Spearman remains more reliable regardless of sample size. For a broader decision framework covering other tests too, see our guide on choosing between correlation and regression.
Pearson vs Spearman: Side-by-Side Comparison
| Feature | Pearson's r | Spearman's rs |
|---|---|---|
| What it detects | Linear relationships | Monotonic relationships (linear or curved) |
| Data type required | Continuous (interval or ratio) | Ordinal, interval, or ratio |
| Distribution assumption | Bivariate normality | None |
| Sensitivity to outliers | High (uses raw values) | Low (uses ranks) |
| Works with Likert scales | No (violates interval assumption) | Yes |
| SPSS path | Analyze → Correlate → Bivariate → Pearson | Analyze → Correlate → Bivariate → Spearman |
| APA reporting | r(98) = .45, p < .001 | rs(98) = .52, p < .001 |
Reporting in APA Format
For Pearson, report the statistic letter, degrees of freedom (which equals n − 2), the coefficient value, and the exact p-value: r(98) = .45, p < .001. If the p-value is between .001 and .05, report it exactly: r(73) = .28, p = .014. Always italicize the r and the p.
For Spearman, the format is the same but the symbol changes to rs: rs(98) = .52, p < .001. Some style guides use the Greek letter ρ instead of rs, but APA 7th edition recommends rs. Whichever symbol you choose, use it consistently throughout the paper. If Academic Stats Agent runs your correlation, the output is already formatted in APA 7 with the correct symbol and degrees of freedom.
One reporting mistake we see repeatedly: students write "Spearman's correlation was used because the data were not normal" without showing the normality test results that led to that decision. Always report the Shapiro-Wilk statistic for each variable (for example, W = .91, p = .003 for income) so the reader can verify your reasoning.
The Most Common Mistake
Defaulting to Pearson without checking the scatterplot. We see this in the majority of drafts that come to us for revision. The student opens SPSS, clicks Analyze → Correlate → Bivariate, leaves Pearson checked (it is the default), and reports the result. No scatterplot. No normality test. No thought about whether the relationship is linear or just monotonic.
The fix takes 30 seconds. Before running any correlation, create a scatterplot in SPSS (Graphs → Chart Builder → Scatter/Dot). Look at the shape. If the dots curve, if outliers sit far from the cloud, or if one variable is ordinal, switch to Spearman. That single step prevents the most frequent correction we make on correlation analyses. In a recent batch of 40 thesis projects, 17 had used Pearson when Spearman was the better fit. All 17 required re-analysis.
Key takeaway: Pearson's r measures linear association between two continuous, normally distributed variables. Spearman's rs measures monotonic association using ranks, handles ordinal data, and resists outliers. Always check your scatterplot and normality results before choosing. If the relationship curves, if outliers are present, or if either variable is ordinal, Spearman is the right coefficient.
Frequently Asked Questions
Which correlation should I use for Likert scale data?
Use Spearman correlation for single Likert items, because they are ordinal. If you sum or average several items into a scale score, the result is often treated as approximately interval, and Pearson becomes acceptable when the scores are roughly normally distributed.
Can Spearman correlation be used for continuous data?
Yes. Spearman works with continuous data and is the better choice when the data are not normally distributed, contain outliers, or the relationship is monotonic but not linear. It simply converts the values to ranks before computing the correlation.
Is Spearman correlation always lower than Pearson?
No. Spearman can be higher or lower than Pearson on the same data. For a perfectly monotonic but curved relationship, Spearman equals 1.0 while Pearson is below 1.0; with strong outliers, Pearson can be artificially inflated relative to Spearman.