How to Run a Paired-Samples t-Test in SPSS for Nursing Pre–Post Data

How to run a paired-samples t-test in SPSS for nursing pre–post data is a common question when the same participants are measured before and after an intervention. The procedure can show whether the mean within-person change differs from zero, but only when the design, variables and assumptions support it.

This guide explains how to structure nursing pre-test and post-test data, check the difference scores, run the test, interpret each SPSS table and report the result. All numerical results are illustrative; replace them with values from your verified dataset.

What Is a Paired-Samples t-Test?

A paired-samples t-test compares two related measurements. In a nursing pre–post study, each participant acts as their own reference: their post-intervention value is paired with their pre-intervention value. IBM’s paired-samples t-test documentation describes the procedure as comparing two variables for one group and calculating the difference within each valid pair.

The test converts every complete pair into a difference score. It then asks whether the population mean difference is plausibly zero. The test statistic can be expressed as:

t = mean difference ÷ standard error of the mean difference.

This focus on within-person differences distinguishes the method from an independent-samples t-test in SPSS, which compares two separate groups.

Is a Paired-Samples t-Test Suitable for Your Nursing Study?

Nursing research questions suited to paired analysis

The method may be suitable when a nursing dissertation asks whether:

• nurses’ sepsis-recognition scores change after simulation training;
• patients’ pain scores change after a nurse-led intervention;
• medication-adherence scores change following discharge education;
• student nurses’ clinical-confidence scores change after placement; or
• carers’ knowledge scores change after a structured teaching session.

The outcome should be continuous, or a composite scale score that the study can defensibly treat as approximately continuous. The two observations must be meaningfully linked. Merely having two columns does not establish pairing.

Paired-Samples Versus Independent-Samples t-Test

Decision point	Paired-samples test	Independent-samples test
Participants	The same participants measured twice, or valid matched pairs	Different participants in two groups
Nursing example	Pain before and after an intervention	Pain in intervention and control groups
Unit analysed	Within-pair difference	Difference between group means
Normality concern	Distribution of difference scores	Outcome within groups or model residuals
Variance equality	No Levene test is required	Variance assumptions affect the pooled test

If an intervention group and a separate control group are each measured at baseline and follow-up, a single paired t-test does not test whether the groups changed differently. That question usually needs an analysis designed for the group-by-time effect.

When Not to Use a Paired t-Test

Choose another method when:

• the outcome is nominal, binary or an individual ordinal item;
• the observations come from unrelated participants;
• there are three or more repeated time points;
• important clustering exists within wards, teams or hospitals;
• the difference scores are seriously non-normal and dominated by influential observations;
• the research question concerns prediction, association or time to an event; or
• substantial missing follow-up data require a model that can use incomplete repeated observations.

Use the SPSS statistical-test selection guide for nursing dissertations when the design does not fit a simple two-time-point comparison.

Prepare Nursing Pre–Post Data Correctly

Use one row per participant

For a basic paired analysis, each row normally represents one participant. Use separate columns for the two measurements, such as pain_pre and pain_post. A participant identifier should be unique.

participant_id	pain_pre	pain_post
001	7	4
002	6	5
003	8	4

Do not sort one time-point column independently. Doing so breaks the pairing and invalidates the analysis. Preserve an untouched original file and document corrections in a working copy. The nursing data-cleaning guide for SPSS provides a fuller checking workflow.

Define the direction of change

Decide whether the difference will be calculated as pre minus post or post minus pre. The choice changes the sign, not the two-sided significance. State the direction so readers can interpret negative and positive values correctly.

Check valid pairs

SPSS normally analyses a pair only when both values are present for that participant. Therefore, the paired sample size may be smaller than the number enrolled. Report recruitment, follow-up and analysis counts transparently. Do not substitute a group mean for missing outcomes simply to retain cases.

Assumptions of the Paired-Samples t-Test

Paired observations are correctly matched

Every pre-intervention score must belong to the same person as its post-intervention score. Reliable participant identifiers and careful data management establish this condition.

The outcome is suitable for mean-based analysis

The outcome should be measured on a continuous scale or be a defensible multi-item score. SPSS labelling a variable as “Scale” does not prove that the measurement is continuous or valid. See the guide to nursing Likert-scale data in SPSS before analysing questionnaire responses.

Pairs are independent of other pairs

One participant’s difference should not determine another participant’s difference. Data from nurses nested in a small number of wards, repeated admissions or linked family members may violate this condition and require a multilevel or otherwise clustered analysis.

Difference scores are approximately normal

The normality assumption concerns the paired differences. Two raw score distributions can each look skewed while their differences are acceptable, or the reverse. Create a difference variable and examine its histogram, Q–Q plot, boxplot and Shapiro–Wilk result together.

A significant Shapiro–Wilk result is not an automatic command to abandon the test. Consider sample size, plot shape and influential observations. The site’s SPSS normality guide for nursing research explains this evidence-based assessment.

No materially influential errors or outliers

Investigate unusual differences against source records. Correct genuine data errors, but do not delete valid observations merely to obtain significance. Where a valid extreme case materially changes the conclusion, report a justified sensitivity analysis.

How to Calculate and Check Difference Scores in SPSS

1. Select Transform, then Compute Variable.
2. Name the new variable, for example pain_change.
3. Enter pain_post - pain_pre in Numeric Expression.
4. Select OK.
5. Choose Analyze, Descriptive Statistics, then Explore.
6. Move the change variable into the dependent list.
7. Under Plots, request a histogram and normality plots with tests.
8. Inspect the distribution and any labelled unusual cases.

A negative value means improvement only when lower scores represent a better outcome and the difference was defined as post minus pre. Always interpret direction using the scale’s meaning.

How to Run a Paired-Samples t-Test in SPSS for Nursing Pre–Post Data

1. Select Analyze.
2. Choose Compare Means and Proportions or Compare Means, depending on the SPSS version.
3. Select Paired-Samples T Test.
4. Place the pre-intervention variable first and the post-intervention variable second in one pair.
5. Select Options and retain a 95% confidence interval unless the approved plan specifies another level.
6. Select Continue, then OK.

According to IBM’s paired-test options guidance, SPSS uses cases with valid data for the tested pair. Therefore, check the displayed sample size before interpreting any result.

SPSS Syntax for a Reproducible Analysis

For variables named pain_pre and pain_post, the syntax is:

T-TEST PAIRS=pain_pre WITH pain_post (PAIRED)
  /CRITERIA=CI(.9500)
  /MISSING=ANALYSIS.

Save the syntax with the dataset and output. Syntax provides an auditable record, but it does not validate variable meaning, pairing or assumptions.

How to Interpret Paired-Samples t-Test Output in SPSS

Paired Samples Statistics

This table reports the mean, sample size, standard deviation and standard error for each time point. Confirm that the sample size is plausible and identify the direction of the raw change. These descriptive values do not, by themselves, test whether the mean paired difference is zero.

Paired Samples Correlations

This table shows the correlation between the two measurements. A high correlation indicates that participants’ relative positions are similar across time. It is not the intervention effect and should not replace the paired test. A weak correlation also does not automatically invalidate the analysis.

Paired Samples Test

This is the main inferential table. Read:

• the mean difference and its direction;
• the standard deviation and standard error of the differences;
• the lower and upper limits of the 95% confidence interval;
• the t statistic;
• degrees of freedom, normally complete pairs minus one; and
• the two-sided p-value.

If the interval excludes zero, the corresponding two-sided test will normally be statistically significant at the .05 level. More importantly, the interval shows the range of population mean changes reasonably compatible with the data.

Worked Nursing Example

Suppose 40 nurses complete a sepsis-recognition assessment before and after simulation training. The mean score rises from 68.4 (SD 9.7) to 75.1 (SD 8.9). If SPSS defines the pair as pre minus post, the mean difference is −6.7.

Assume the Paired Samples Test reports t(39) = −4.28, p < .001, with a 95% confidence interval from −9.87 to −3.53. This indicates evidence that the population mean score changed, and the interval’s negative direction reflects the selected subtraction order.

The result does not prove that training caused the change. Without an appropriate comparison group, alternative explanations can include repeated testing, concurrent learning, changes in staffing or other events.

Calculate and Report an Effect Size

Statistical significance is influenced by sample size. An effect size helps describe the standardised magnitude of the mean change. A common paired-design measure is Cohen’s d_z:

d_z = mean difference ÷ standard deviation of the difference scores.

For the illustrative example, if the standard deviation of the differences is 9.90, the absolute d_z is 6.70 ÷ 9.90 = 0.68. State the calculation used because several paired-design effect-size definitions exist and are not interchangeable.

Do not label an effect clinically important from a generic threshold alone. Compare the confidence interval and change magnitude with the instrument’s interpretation, prior evidence and any prespecified clinically meaningful difference.

APA-Style Reporting Example

“Sepsis-recognition scores increased from pre-training (M = 68.4, SD = 9.7) to post-training (M = 75.1, SD = 8.9). A paired-samples t-test indicated a mean pre-minus-post difference of −6.7 points, 95% CI [−9.87, −3.53], t(39) = −4.28, p < .001, d_z = 0.68.”

After presenting the result, discuss the estimate’s clinical meaning, uncertainty, study design and limitations. The guide to discussing SPSS results in a nursing dissertation shows how to move beyond repeating output.

Common Paired t-Test Mistakes

• Using separate groups: pre–post labels do not create pairing when different people were measured.
• Testing raw-score normality only: the relevant assumption concerns difference scores.
• Reporting only p: include descriptives, mean change, confidence interval and effect size.
• Ignoring subtraction order: this leads to incorrect statements about improvement or deterioration.
• Claiming causation: change over time is not necessarily caused by the intervention.
• Deleting outliers automatically: verify records and justify every exclusion.
• Ignoring missing follow-up data: complete-pair analysis may create attrition bias.
• Running several paired tests: repeated testing increases false-positive risk and requires a prespecified strategy.

What If the Assumptions Are Not Met?

First, check data accuracy and whether the research question, scale and pairing are correct. A transformation may alter interpretation and should not be used merely to chase significance. For seriously non-normal differences or influential valid observations, a Wilcoxon signed-rank test may be considered.

However, Wilcoxon is not simply a “non-parametric paired t-test” with identical hypotheses. Its interpretation depends on distributional conditions and ranks. For more than two time points, incomplete repeated data or clustered observations, seek advice on an appropriate repeated-measures or mixed-effects model.

Frequently Asked Questions

Do I need Levene’s test for a paired-samples t-test?

No. Levene’s test concerns equality of variance between independent groups. A paired analysis focuses on the distribution of within-pair differences.

Should I test normality of pre-test and post-test scores separately?

The central normality check is for the difference scores. Raw time-point plots remain useful for understanding the data, but they do not replace assessment of the differences.

Can I use a paired t-test for Likert-scale data?

An individual Likert item is ordinal and should not automatically be treated as continuous. A validated multi-item score may be defensible, depending on its construction, measurement properties, distribution and approved analysis plan.

What happens when a participant has no post-test score?

That participant normally contributes no complete pair to the test. Report the loss, investigate the reason and consider whether missingness could bias the result.

Does a significant paired t-test prove the intervention worked?

No. It shows evidence of mean change within the observed design. Causal interpretation depends on the design, comparison condition, bias control and alternative explanations.

Paired t-Test Dissertation Reporting Checklist

Before submitting your nursing results chapter, confirm that it allows another reader to reconstruct what you tested and why.

• Name the pre-test and post-test variables and explain what higher scores mean.
• State that the same participants supplied both measurements.
• Give the number of complete pairs included in the analysis.
• Report each time point’s mean and standard deviation.
• Define the subtraction order used for the difference score.
• Describe the histogram, Q–Q plot, Shapiro–Wilk result and any unusual differences.
• Explain what you did when an assumption or data-quality concern arose.
• Report the mean difference, 95% confidence interval, t, degrees of freedom and exact p-value.
• Name the paired-design effect-size formula and report its value.
• Separate statistical significance, clinical importance and causal interpretation.

Place essential findings in the chapter rather than sending the reader to raw output. Selected supporting tables can appear in the appendices when your university permits them. See the guides to reporting a nursing dissertation Chapter 4 and organising nursing dissertation appendices.

Conclusion

A defensible paired-samples t-test in SPSS begins with correctly linked nursing measurements and a clearly defined direction of change. Check difference scores, verify complete pairs, interpret the confidence interval alongside the p-value and report an explicit effect size. Most importantly, distinguish statistical change from clinical importance and causal evidence.

References

• IBM (2026) Paired-Samples T Test. IBM SPSS Statistics 32 documentation. Available at: https://www.ibm.com/docs/en/spss-statistics/32.0.0?topic=tests-paired-samples-t-test (Accessed: 8 September 2026).
• IBM (2026) Paired-Samples T Test Options. IBM SPSS Statistics documentation. Available at: https://www.ibm.com/docs/en/spss-statistics/30.0.0?topic=test-paired-samples-t-options (Accessed: 8 September 2026).
• National Institute of Standards and Technology (n.d.) Analysis of paired observations. NIST/SEMATECH e-Handbook of Statistical Methods. Available at: https://www.itl.nist.gov/div898/handbook/prc/section3/prc311.htm (Accessed: 8 September 2026).
• Ranganathan, P. and Aggarwal, R. (2021) ‘An introduction to statistics: choosing the correct statistical test’, Indian Journal of Critical Care Medicine, 25(Suppl 2), pp. S184–S186.

Featured image: Vitaly Gariev via Unsplash.