Pearson vs Spearman: Which Correlation Do You Need?

Two Correlations, One Decision

You have two variables and one question: are they related? In SPSS, that question usually leads to one of two tests — Pearson’s correlation or Spearman’s rank correlation. They look similar, they sit next to each other in the same dialog box, and they both produce a number between −1 and +1. But they answer subtly different questions, and choosing the wrong one can hide a real relationship or invent one that is not there.

This guide explains what each correlation actually measures, how to run both in SPSS, how to read the output, and — most importantly — how to decide which one your data needs.

What Pearson’s Correlation Measures

Pearson’s r measures the strength and direction of a linear relationship between two scale variables. Linear means the points on a scatterplot roughly follow a straight line: as one variable goes up, the other goes up (positive r) or down (negative r) at a roughly steady rate.

Pearson works with the actual values of your data. Because of that, it makes three assumptions you should check:

• Both variables are scale (continuous) — for example, test scores, reaction times, or income.
• The relationship is linear. Check this with a scatterplot before you run the test.
• No extreme outliers, and roughly normal distributions. Pearson is sensitive to both: a single extreme value can drag r up or down considerably.

When these hold, Pearson is the more powerful choice — it uses all the information in your data, and its result is the one most readers expect to see.

What Spearman’s Correlation Measures

Spearman’s rho (ρ, often written as r_s) measures the strength and direction of a monotonic relationship. Monotonic is a weaker condition than linear: as one variable goes up, the other consistently goes up (or consistently down) — but not necessarily at a steady rate. A curve that always rises still counts.

The trick is in how Spearman works: SPSS first converts each value to a rank (smallest = 1, next = 2, and so on) and then computes Pearson’s correlation on the ranks instead of the raw values. That one change has big consequences:

• It works with ordinal data — Likert-scale ratings, rankings, or any variable where the order matters but the gaps between values do not.
• It does not assume normality, so skewed data is fine.
• It is far less sensitive to outliers, because the largest value is only ever one rank above the next, no matter how far away it sits.
• It detects curved relationships that Pearson would miss or underestimate.

The price is a small loss of statistical power when the data are perfectly well-behaved, and a result that describes ranks rather than the original units.

Pearson vs Spearman at a Glance

	Pearson (r)	Spearman (rho)
Relationship measured	Linear	Monotonic (linear or curved)
Data level	Scale (interval/ratio)	Ordinal, or scale
Uses	Actual values	Ranks
Normality assumed?	Yes, approximately	No
Sensitive to outliers?	Yes	Much less
Best when	Two scale variables, linear pattern, no outliers	Ordinal data, skewed data, outliers, curved pattern

How to Run Both Correlations in SPSS

Both tests live in the same dialog, so you can run them together and compare. Here is the click-by-click route, using two variables — say, StudyHours and ExamScore — as the example.

1. Click Analyze > Correlate > Bivariate…
2. Move both variables into the Variables box.
3. Under Correlation Coefficients, tick Pearson and Spearman.
4. Under Test of Significance, leave Two-tailed selected unless you have a directional hypothesis stated in advance.
5. Tick Flag significant correlations if you want asterisks on significant results.
6. Click OK.

If you also want a scatterplot to check linearity, go to Graphs > Chart Builder…, choose Scatter/Dot, drag the Simple Scatter icon into the preview, put one variable on the X-axis and the other on the Y-axis, and click OK. Add a fit line by double-clicking the chart and choosing Elements > Fit Line at Total.

SPSS Syntax for Pearson and Spearman

The syntax below runs both correlations in one command and is easy to save and re-run. Paste it into a syntax window (File > New > Syntax) and press the green Run arrow.

*** Pearson and Spearman correlations for the same two variables.
CORRELATIONS
  /VARIABLES=StudyHours ExamScore
  /PRINT=TWOTAIL NOSIG
  /MISSING=PAIRWISE.

*** Spearman only (for ordinal or skewed variables).
NONPAR CORR
  /VARIABLES=StudyHours ExamScore
  /PRINT=SPEARMAN TWOTAIL NOSIG
  /MISSING=PAIRWISE.

*** Scatterplot to check linearity before choosing.
GRAPH
  /SCATTERPLOT(BIVAR)=StudyHours WITH ExamScore
  /MISSING=LISTWISE.

Note the two commands: CORRELATIONS gives Pearson, while NONPAR CORR with PRINT=SPEARMAN gives Spearman. Running them side by side on the same variables makes the comparison immediate.

How to Interpret the Output

SPSS produces one correlation matrix per test. Each cell where the two variables meet shows three things:

• The correlation coefficient (r or rho), from −1 to +1. The sign tells you the direction; the size tells you the strength. As a rough guide, .10 is small, .30 is medium, and .50 is large — but always interpret the number in the context of your field, not by a fixed cut-off alone.
• Sig. (2-tailed), the p-value. Below .05, the correlation is statistically significant — unlikely to be zero in the population.
• N, the number of cases used. Check this: if N differs between the two tests, missing values are being handled differently.

Now compare the two coefficients. Three patterns are common:

• Pearson and Spearman are similar (for example, r = .62 and rho = .60). The relationship is roughly linear, the data are well-behaved, and either coefficient is defensible. Report Pearson.
• Spearman is noticeably larger than Pearson (for example, r = .35 and rho = .68). The relationship is monotonic but curved — the points rise steadily but not in a straight line. Spearman is the honest summary; a scatterplot will show the curve.
• Pearson is larger, or the two disagree wildly. Look for outliers. A single extreme case can inflate Pearson, and Spearman’s rank-based calculation will usually be closer to the pattern in the bulk of the data. Inspect the scatterplot and the boxplots before deciding.

Never report one coefficient while hiding the other if you ran both. If they disagree, the disagreement itself is a finding — say so, and explain what the scatterplot shows.

How to Decide Which One You Need

Work through these questions in order:

1. Is either variable ordinal? If yes — Likert items, rankings, ordered categories — use Spearman. Pearson is not valid for ordinal data.
2. Is the relationship linear? Look at the scatterplot. If the points follow a straight line, Pearson is fine. If they follow a curve that always rises or always falls, use Spearman.
3. Are the distributions roughly normal, without extreme outliers? Check with a histogram or boxplot, or a Shapiro-Wilk test. If the data are badly skewed or contain extreme values you cannot justify removing, use Spearman.
4. If the answers above all point to Pearson, use Pearson. It is the standard, most powerful choice for well-behaved scale data.
5. When in doubt, run both and compare. If they agree, report Pearson and mention that Spearman gave the same picture. If they disagree, report Spearman and explain why — the scatterplot is your evidence.

A common mistake is choosing Spearman “to be safe” on perfectly normal data. It is not wrong, but it throws away information and makes your result harder to compare with the rest of the literature, where Pearson is the default. Use Spearman when you have a reason, not as a reflex.

Reporting in APA Style (7th ed.)

Report the correlation coefficient, the degrees of freedom, the sample size, and the p-value, in text or in a table. For Pearson, the degrees of freedom are N − 2; for Spearman, report N instead.

Example wording for a Pearson result:

Study hours were positively associated with exam scores, r(58) = .62, p < .001.

Example wording for a Spearman result:

There was a strong positive monotonic relationship between study hours and exam scores, r_s(60) = .60, p < .001.

If you are reporting both because they disagreed, say so plainly:

Pearson’s r was .35, p = .007; however, the scatterplot showed a curved, monotonic pattern, so Spearman’s rho is the more appropriate summary, r_s(60) = .68, p < .001.

Report exact p-values (for example, p = .007) except when p is smaller than .001, which you write as p < .001. Do not write p = .000. Add the direction and a plain-language interpretation — a bare coefficient without a sentence explaining it is not a result.

Key Takeaways

• Pearson measures linear relationships using actual values; Spearman measures monotonic relationships using ranks.
• Use Pearson for two scale variables with a linear pattern, roughly normal distributions, and no extreme outliers.
• Use Spearman for ordinal data, skewed distributions, outliers, or curved but consistently rising or falling relationships.
• Both tests run from the same SPSS dialog (Analyze > Correlate > Bivariate) — tick both boxes and compare.
• When Pearson and Spearman disagree, check the scatterplot: a curve means Spearman, an outlier means inspect the case before trusting either.
• Report the coefficient, degrees of freedom or N, and the exact p-value, with a plain-language interpretation.

Frequently Asked Questions

Can I use Spearman with two scale variables?

Yes. Spearman works on any two variables that can be ranked, including scale variables. It is the right choice when scale data are skewed, contain outliers, or show a curved monotonic pattern. It is not limited to ordinal data.

What does it mean if Pearson is significant but Spearman is not (or the other way round)?

It usually means the relationship is linear but not monotonic in the ranks, or the other way round — or that outliers are distorting one of the two. Inspect the scatterplot first. If a few extreme cases are driving Pearson, Spearman is more trustworthy; if the pattern is linear but the ranks are noisy, Pearson may be the better summary.

Should I transform my data instead of switching to Spearman?

Sometimes. If a skew is the only problem, a log or square-root transformation can make Pearson valid and keeps the result in the original measurement framework. But a transformation does not fix a genuinely curved relationship or ordinal data — in those cases, Spearman is the simpler and more honest choice.

How do I report a Spearman correlation in APA style?

Write it as r_s(N) = .xx, p = .xxx, where N is the sample size (Spearman has no degrees of freedom in the same way Pearson does). Add a sentence describing the direction and strength in plain words.

Is Spearman just a weaker version of Pearson?

No. Spearman answers a different question — whether the relationship is consistently directional, not whether it is straight. On curved monotonic data it is more accurate than Pearson, not weaker. It only loses power when the data are perfectly linear and normal, which is exactly when Pearson is the right tool anyway.

Need this done on your data? Get a free quote → /services/