Run and Interpret Pearson Correlation in IBM SPSS Statistics

Reader outcome

This guide explains how to prepare two numeric variables, inspect the relationship, run Pearson correlation through IBM SPSS Statistics, preserve a reproducible syntax record, interpret the correlation table and transfer the result into a thesis. It does not decide whether Pearson correlation is substantively appropriate; that decision belongs in the linked method guide.

Version and scope

This guide is scoped to IBM SPSS Statistics 31 documentation verified on 27 July 2026. IBM documents both the Bivariate Correlations interface and the CORRELATIONS command. Menu wording and available options can change between releases, so compare the steps with the installed version.

Prepare the data

Each row should represent one observational unit and each column one variable. Confirm that:

• both variables are numeric and use meaningful value definitions;
• special missing codes have been declared as missing rather than analysed as real scores;
• every pair belongs to the same case;
• duplicate rows, impossible values and unit inconsistencies have been investigated;
• the number of valid paired observations is known.

Keep a data dictionary. Variable labels help readers, but short, stable variable names make syntax more reproducible.

Inspect linearity and influential observations

Create a scatterplot before running the coefficient. The official IBM guidance identifies linearity and outlier screening as important data considerations for Pearson correlation. A curved pattern, separate clusters or a single influential point can make the coefficient misleading.

Record the inspection decision in the analysis log. Do not delete a case simply because it changes significance. Verify whether it is an error, a legitimate extreme value or evidence that the proposed model is incomplete.

Run the Bivariate Correlations procedure

In the graphical interface, open the Bivariate Correlations procedure, move the two variables into the analysis list, select Pearson correlation and choose a one- or two-sided significance test according to the pre-specified hypothesis. A two-sided test is the safer default when the direction was not justified in advance.

Choose the missing-data rule deliberately. Pairwise exclusion can use different cases for different correlations. Listwise exclusion uses a common complete set for all variables in the procedure. Neither option is universally correct; the choice should follow the analysis plan and be reported.

Where available and required, request descriptive statistics and confidence intervals. Before clicking the final run command, use the interface’s paste or syntax-generation function so the procedure is preserved as syntax.

Reproducible syntax

A minimal pattern is:

CORRELATIONS
  /VARIABLES=study_hours exam_score
  /PRINT=TWOTAIL
  /MISSING=LISTWISE.

A documented confidence-interval pattern in version 31 is:

CORRELATIONS
  /VARIABLES=study_hours exam_score
  /PRINT=TWOTAIL
  /STATISTICS DESCRIPTIVES
  /CI METHOD=FISHER CILEVEL=95
  /MISSING=LISTWISE.

The exact subcommands should be checked in the installed version. Keep the syntax file with a date, data version and short comment explaining why the missing-data rule and tail choice were selected.

Read the output table

The correlation matrix normally presents the Pearson coefficient, significance value and number of observations for each pair. Read them as a set:

• Pearson Correlation gives direction and linear association strength.
• Sig. gives the p-value for the selected one- or two-sided test.
• N gives the effective number of cases for that pair.
• A requested confidence interval gives a range of population values compatible with the sample under the model.

Do not interpret the diagonal values of 1.000 as findings; they are each variable correlated with itself. In a larger matrix, confirm whether N changes across pairs, especially under pairwise deletion.

Worked interpretation

Assume a synthetic dataset of 30 students produces r = .46, two-sided p = .010 and a 95% confidence interval [.12, .70]. A defensible interpretation is that the sample shows a positive linear association between weekly independent-study hours and examination score. The uncertainty interval is broad enough to discourage false precision. The analysis does not demonstrate that additional study hours caused the higher scores.

Thesis reporting template

A scatterplot indicated an approximately linear positive association between weekly independent-study hours and examination score, with no obviously dominant observation. Pearson’s correlation, calculated in IBM SPSS Statistics 31 using listwise exclusion, showed a positive association, r(28) = .46, 95% CI [.12, .70], p = .010, N = 30. Because the data were observational, the result is interpreted as association rather than causation.

Replace every value with the actual output. Describe transformations, weights, filters or split-file settings that affected the analysis.

Quality checks before export

• Save the dataset, syntax and output with stable names.
• Re-run the syntax from a clean session or a documented copy of the data.
• Verify that filters, weights and split-file settings are not accidentally active.
• Confirm that the table’s N matches the intended missing-data rule.
• Compare the coefficient with the scatterplot and descriptive statistics.
• Export a clean table rather than a full unedited output viewer file.

Common errors

• Reporting the significance marker without the coefficient or confidence interval.
• Selecting a one-tailed test after seeing the direction of the result.
• Treating user-missing codes as valid numbers.
• Using pairwise deletion without noticing that sample sizes differ.
• Failing to preserve syntax, making the analysis difficult to reproduce.
• Interpreting Pearson correlation as evidence of causation.
• Copying an SPSS table into a thesis without adapting labels, notes and formatting to the required style.

Source and verification notes

The procedure and syntax discussion is based on IBM’s version 31 CORRELATIONS reference and Bivariate Correlations documentation. The worked example and reporting language are original and synthetic. Any screenshots created from the installed version must avoid exposing confidential or identifiable data.