How to Run Correlation Analysis in SPSS (Pearson and Spearman)

Correlation analysis answers one of the most common questions in student research: as one variable goes up, what happens to the other? The coefficient it produces tells you two things at once, the strength of the relationship and its direction. Here is the complete SPSS procedure for both Pearson and Spearman correlation, from the menu clicks to the APA-formatted sentence you paste into your results section. The whole analysis takes about five minutes once you know where everything is.

Pearson or Spearman? Decide First

SPSS offers both coefficients in the same dialog box, so you need to know which one you want before you open it. Pearson's r is the right choice when both variables are continuous (interval or ratio scale), the relationship between them is linear, and both are roughly normally distributed. Exam scores against study hours, blood pressure against age, income against years of experience. That kind of data.

Spearman's r_s works with ranks instead of raw values. Use it when at least one variable is ordinal (single Likert items, satisfaction ratings, education level coded 1–5), when the relationship is monotonic but curved, or when your data fail the normality check. Spearman asks for less from your data, which makes it the fallback whenever Pearson's requirements are not met.

A short version of the decision: continuous, linear, normal → Pearson. Anything else → Spearman. For the full reasoning behind that rule, including a worked example where the two coefficients give different answers on the same data, see our Pearson vs Spearman comparison.

Setting Up Your Data

The Data View layout is simple: one row per participant, one column per variable. For a Pearson correlation between study hours and exam scores, you need two numeric columns, something like study_hours and exam_score, with each row holding one student's values. A student who studied 12 hours and scored 74 occupies one row: 12 in the first column, 74 in the second.

In Variable View, set the Measure column to Scale for both variables. This does not change the calculation, but it keeps your file honest and helps SPSS pick sensible defaults in chart dialogs. For a Spearman correlation, ordinal columns are fine too. A 5-point Likert rating column with values 1 through 5 works directly; set its Measure to Ordinal. There is no need to convert anything to ranks yourself. SPSS handles the ranking internally when you request Spearman.

Missing values are handled pairwise by default, meaning SPSS uses every case that has valid values on both variables in a given pair. If participant 23 skipped the satisfaction question, that participant still contributes to correlations not involving satisfaction.

Running a Pearson Correlation

With your data file open, follow these steps:

1. Click Analyze → Correlate → Bivariate.
2. Move both variables from the left panel into the Variables box. Select each one and click the arrow, or double-click it. You can add more than two variables; SPSS will correlate every pair.
3. Under Correlation Coefficients, make sure Pearson is checked. It is checked by default.
4. Leave Test of Significance on Two-tailed unless you have a directional hypothesis stated before data collection.
5. Check Flag significant correlations. This adds asterisks to significant results in the output, which makes large matrices much easier to scan.
6. Click OK.

The Output Viewer opens with a single table called Correlations. That table contains everything you need to report.

Running a Spearman Correlation

Same dialog, one different checkbox. Go to Analyze → Correlate → Bivariate, move your variables across, then under Correlation Coefficients uncheck Pearson and check Spearman. Click OK. The output table now shows Spearman's rho instead of Pearson's r, and the structure is identical.

You can also check both boxes at once. SPSS then prints two tables, one for each coefficient, which is useful when you want to see how much they diverge. A large gap between the two values usually points to outliers or a curved relationship, both of which argue for reporting Spearman. Just decide before you look at the p-values which coefficient you will report. Picking the one with the smaller p-value after the fact is a form of cherry-picking that a careful reviewer will catch.

Reading the Output Matrix

The Correlations table is a square matrix with your variables listed both across the top and down the side. Each cell where two variables meet contains three stacked numbers:

• Pearson Correlation (or Correlation Coefficient for Spearman): the top number. This is your r value, ranging from −1 to +1. For example, .42 means a positive relationship of moderate strength.
• Sig. (2-tailed): the middle number. This is the p-value. A value of .000 in SPSS means p < .001, not that the probability is literally zero. Report it as p < .001.
• N: the bottom number, the count of cases used for that specific pair. If N varies across cells, you have missing data.

Two structural features of the matrix trip students up. First, the diagonal always shows 1.000, because every variable correlates perfectly with itself. Ignore those cells. Second, the matrix is symmetric: the cell for study_hours × exam_score contains exactly the same numbers as the cell for exam_score × study_hours. With two variables, you only need to read one off-diagonal cell.

For interpreting the size of the coefficient, Cohen's (1988) benchmarks are the standard reference: .10 to .29 is weak, .30 to .49 is moderate, and .50 or above is strong. The same cutoffs apply to negative values; r = −.55 is a strong negative correlation. Note that significance and strength are separate questions. With n = 400, an r of .11 will be significant at p < .05 while telling you almost nothing practically useful.

Checking Assumptions for Pearson

Three checks, and none takes more than a minute.

Linearity. Build a scatterplot through Graphs → Legacy Dialogs → Scatter/Dot → Simple Scatter. Put one variable on each axis and look at the shape of the point cloud. A straight-line trend, in either direction, means Pearson is measuring the right thing. A curve that bends, flattens, or reverses means Pearson will underestimate the true association.

Normality of both variables. Run Analyze → Descriptive Statistics → Explore, move both variables into the Dependent List, click Plots, and check Normality plots with tests. Read the Shapiro-Wilk row for each variable. A significant result (p < .05) flags a departure from normality. Our normality testing guide covers how strict to be about this at different sample sizes.

Outliers. Go back to the scatterplot and look for points sitting far from the main cloud. Pearson's formula uses raw distances from the mean, so one extreme case can move the coefficient by .10 or more in a small sample. Check whether the value is a data-entry error before deciding what to do with it.

If linearity or normality fails, the fix is already in the same dialog: switch to Spearman. Ranks do not care about the shape of the distribution, and they follow curved monotonic trends without complaint.

Reporting in APA Format

APA 7 wants the statistic, the degrees of freedom (which is n − 2 for correlation), the coefficient, and the p-value, wrapped in a sentence that says what the relationship means. For Pearson:

There was a moderate positive correlation between study hours and exam score, r(98) = .42, p < .001.

The 98 in parentheses tells the reader you had 100 participants. For Spearman, the symbol changes to r_s and everything else stays the same:

Spearman's rank correlation showed a moderate positive association between class rank and satisfaction rating, r_s(78) = .38, p = .001.

Italicize r, r_s, and p. Drop the leading zero from the coefficient and the p-value because neither can exceed 1. Report exact p-values to two or three decimals when they are .001 or larger; use p < .001 below that. And say something about direction and strength in words. A bare string of numbers is not an interpretation.

Common Mistakes

Running Pearson on single Likert items. A 1-to-5 agreement rating is ordinal. The distance between "agree" and "strongly agree" is not guaranteed to equal the distance between "neutral" and "agree," which breaks Pearson's interval-scale requirement. Use Spearman for individual items. Averaged multi-item scale scores are a different matter and can usually be treated as approximately continuous.

Reading correlation as causation. An r of .42 between study hours and exam scores does not show that studying causes higher scores. Motivated students may both study more and perform better for reasons the correlation cannot see. Write "was associated with," not "led to" or "improved." Supervisors mark this specific wording error constantly.

Ignoring outliers that inflate r. One participant with extreme values on both variables can manufacture a correlation that does not exist in the rest of the sample. If your scatterplot shows a tight blob of 79 points and one point in the far corner, rerun the analysis without that case and compare. When the coefficient drops from .48 to .19, the outlier was doing the work, and your report needs to say so.

Frequently Asked Questions

Can I run a correlation with more than two variables in SPSS?

Yes. Move as many variables as you want into the Variables box in the Bivariate Correlations dialog. SPSS produces a matrix with every pairwise correlation. Keep in mind that each pair is tested separately, so with many variables the number of tests grows quickly and some will be significant by chance alone.

What does a negative correlation coefficient mean?

A negative coefficient means the two variables move in opposite directions: as one increases, the other tends to decrease. For example, r = −.45 between absences and exam score means students with more absences tend to score lower. The strength interpretation is the same as for positive values; only the direction differs.

Why does SPSS show a correlation of 1.000 on the diagonal?

The diagonal of the correlation matrix shows each variable correlated with itself, which is always a perfect 1.000. These cells carry no information about your research question. Read the off-diagonal cells, where each variable is paired with a different variable.