If you have measured the same people more than once, you probably need a repeated-measures ANOVA. It is the test you reach for when each participant contributes several scores — for example, before, during, and after an intervention — and you want to know whether the average score genuinely changed across those occasions, rather than just wobbling by chance.
Unlike a one-way ANOVA, which compares different groups of people, a repeated-measures ANOVA compares the same people with themselves. That is a big advantage: because each person acts as their own control, individual differences in overall level (some people simply score higher than others) are stripped out of the comparison. The test is therefore often more sensitive to real change over time or across conditions than a between-groups design with the same number of participants.
This tutorial walks you through the whole process in SPSS: setting up your data, checking the assumptions, running the analysis click by click, reading the output, and reporting the result in APA style (7th edition). No advanced statistics background is assumed.
When to Use a Repeated-Measures ANOVA
Use a repeated-measures ANOVA when all of the following are true:
- You have one group of participants measured on the same continuous outcome at three or more time points, or under three or more conditions.
- The measurements are related: the same person provides every score, so the scores are not independent of one another.
- You want to test whether the mean of the outcome differs across those time points or conditions.
Classic examples include test scores measured at the start, middle, and end of a course; anxiety ratings before, immediately after, and one month after a therapy programme; and reaction times under three different task conditions completed by every participant.
If you have only two time points or conditions, a paired-samples t-test is the simpler and correct choice — it is, in fact, mathematically equivalent to a repeated-measures ANOVA with two levels. If you have different groups of people in each condition, use a one-way (between-groups) ANOVA instead. And if you have both a grouping factor and repeated measurements — say, a treatment group and a control group each measured three times — you need a mixed ANOVA, which is a different analysis covered in a separate guide.
Setting Up Your Data in SPSS
Repeated-measures ANOVA in SPSS uses wide-format data: one row per participant, and one column per measurement occasion. For a study with three time points, your Data View should have three columns, for example pre_test, post_test, and follow_up, plus one row for each participant. Do not stack all the scores into a single column with a separate grouping column — that is the long format used by some other procedures, and the repeated-measures dialog will not accept it.
Before you run anything, give each variable a clear label (for example, “Score before intervention”) and check for missing values and obvious data-entry errors, such as a score far outside the possible range of your scale. A single mistyped value can distort the whole analysis, so a quick scan of descriptives first is time well spent.
Assumptions to Check Before You Run the Test
A repeated-measures ANOVA rests on a few assumptions. The most important ones are:
- Normality: the differences between each pair of conditions should be approximately normally distributed. With reasonably sized samples, the test is fairly robust to modest departures, but severe skew or extreme outliers deserve attention. You can inspect this with histograms of the difference scores, or with a Shapiro-Wilk test on them.
- Sphericity: the variances of the differences between all pairs of conditions should be roughly equal. This is the repeated-measures counterpart of the equal-variances assumption, and SPSS tests it for you with Mauchly’s test, covered below.
- No extreme outliers: a single wild score from one participant can pull a condition mean and inflate the error term. Boxplots of each occasion are a quick visual check.
- Interval or ratio outcome: the dependent variable should be measured on a continuous scale, not a simple yes/no or category.
If sphericity is violated, you do not abandon the analysis — you simply read a corrected row of the output, as explained in the interpretation section. If normality is badly violated, consider a nonparametric alternative such as the Friedman test, or transforming the outcome, and say so transparently in your write-up.
Running the Test in SPSS: Click by Click
Follow these steps exactly:
- Click Analyze in the top menu, then General Linear Model, then Repeated Measures.
- In the Repeated Measures Define Factor(s) dialog, SPSS asks you to name your within-subjects factor. Type a sensible name in the Within-Subject Factor Name box, for example “time”, and type the number of levels (for three measurement occasions, type 3). Click Add, then click Define.
- In the next dialog, you will see your measurement variables on the left and a Within-Subjects Variables box on the right. Move your three occasion variables into that box in chronological order: first occasion at the top, last occasion at the bottom. The order matters for the plots and pairwise comparisons, so get it right.
- Click Plots if you would like a profile plot (a line graph of the means across occasions). Move your factor into the Horizontal Axis box and click Add, then Continue. This plot is not required for the test, but it makes the pattern of change instantly visible.
- Click Options. Tick Descriptive statistics and Estimates of effect size. Under Display Means for, move your factor over and tick Compare main effects, leaving the confidence interval adjustment at Bonferroni (the default). This gives you pairwise comparisons between occasions. Click Continue.
- Click OK to run the analysis.
That is the whole procedure. The output will appear in the SPSS Viewer window.
SPSS Syntax for Repeated-Measures ANOVA
If you prefer syntax, or you want a reproducible record of exactly what you ran, paste the following into a syntax window and adapt the variable names to your own data. The variable names here (pre_test, post_test, follow_up) are placeholders for your three occasions:
GLM pre_test post_test follow_up /WSFACTOR=time 3 Polynomial /METHOD=SSTYPE(3) /PLOT=PROFILE(time) /EMMEANS=TABLES(time) COMPARE ADJ(BONFERRONI) /PRINT=DESCRIPTIVE ETASQ /CRITERIA=ALPHA(.05) /WSDESIGN=time.
Each line does one job: the WSFACTOR line defines the within-subjects factor and its three levels; the PLOT line requests the profile plot; the EMMEANS line requests the pairwise comparisons with a Bonferroni adjustment; and the PRINT line asks for descriptives and partial eta-squared as the effect size. Running syntax rather than menus also makes it easy to re-run the identical analysis after a data correction.
How to Interpret the SPSS Output
Work through the output tables in this order:
Mauchly’s Test of Sphericity
Find Mauchly’s Test of Sphericity first. If its significance value (p) is greater than .05, sphericity can be assumed, and you read the row labelled “Sphericity Assumed” in the next table. If p is .05 or smaller, sphericity is violated, and you should read one of the corrected rows instead — Greenhouse-Geisser is the usual default choice, and Huynh-Feldt is a slightly less conservative alternative. Whichever row you read, say so when you report the result, and use the corrected degrees of freedom shown in that row.
Tests of Within-Subjects Effects
This is the main result. Look at the row for your factor (for example, “time”) in the column set you selected above, and read across to the F value and its significance (p) value. If p is less than your alpha level (conventionally .05), the means differ significantly across occasions: scores did not stay flat. If p is .05 or greater, you do not have evidence of a change, and you stop there — there is no justification for hunting through pairwise comparisons after a non-significant overall test.
Next, check the effect size. In the same table, partial eta-squared tells you what proportion of the variance in scores (after removing individual differences) is associated with the factor. As a rough guide, values around .01 are small, around .06 are medium, and around .14 or above are large, but treat these benchmarks as a starting point, not a verdict — the practical meaning of an effect always depends on your field and your measure.
Pairwise Comparisons
If the overall test is significant, the Pairwise Comparisons table (from the EMMEANS request) tells you where the differences lie: which occasions differ from which. Each row compares two occasions, gives the mean difference, and gives a Bonferroni-adjusted p value. The Bonferroni adjustment already corrects for the fact that you are making several comparisons, so you can read these p values directly against .05. Report only the comparisons that matter for your research question, and note the direction of each difference (for example, scores at follow-up were higher than scores before the intervention).
Descriptives and the Profile Plot
Finally, use the Descriptive Statistics table and the profile plot to describe the pattern in plain language. State the mean and standard deviation at each occasion, and describe the shape: did scores rise steadily, rise then fall back, or change only between two specific occasions? A significant test with a tiny mean difference may be statistically real but practically unimportant, so always pair the p value with the actual means.
What If Sphericity Is Violated?
Do not panic, and do not switch to a different test. Sphericity violations are common with real data, especially when you have four or more occasions. Simply report the Greenhouse-Geisser (or Huynh-Feldt) corrected result, using its adjusted degrees of freedom, and note in your write-up that the correction was applied because Mauchly’s test was significant. If you also want a result that does not assume sphericity at all, the multivariate tests table (Pillai’s trace and Wilks’ lambda) in the same output provides one — but for most theses and journal articles, the corrected univariate result is the standard report.
Reporting in APA Style (7th ed.)
Report a repeated-measures ANOVA with the factor name, the F statistic with its degrees of freedom, the p value, and the effect size. Use the corrected degrees of freedom if sphericity was violated, and state that the correction was applied. A template you can adapt:
“A one-way repeated-measures ANOVA was conducted to compare scores across the three time points. Mauchly’s test indicated that the assumption of sphericity had been violated, so Greenhouse-Geisser corrected results are reported. There was a significant effect of time on scores, F(df1, df2) = x.xx, p = .xxx, partial eta-squared = .xx. Bonferroni-adjusted pairwise comparisons showed that [describe which occasions differed and in which direction].”
Replace the placeholders with your own values from the Tests of Within-Subjects Effects table, and give the means and standard deviations for each occasion in the text or in a table. If sphericity was not violated, simply delete the sentence about the correction and report the “Sphericity Assumed” row. Always report exact p values as SPSS gives them (for example, p = .013), except that SPSS prints .000 for very small values — report those as p < .001, never as p = .000.
Key Takeaways
- A repeated-measures ANOVA compares the same participants across three or more occasions or conditions, using each person as their own control.
- Your data must be in wide format: one row per participant, one column per occasion.
- Check Mauchly’s test first. If sphericity is violated, report the Greenhouse-Geisser corrected row.
- A significant overall F only tells you that the means differ somewhere — use the Bonferroni-adjusted pairwise comparisons to find out where.
- Always report the F statistic, degrees of freedom, p value, and partial eta-squared, alongside the means and standard deviations.
Frequently Asked Questions
How is a repeated-measures ANOVA different from a one-way ANOVA?
A one-way ANOVA compares different groups of people, while a repeated-measures ANOVA compares the same people measured several times. Because the same participants appear in every condition, individual differences are removed from the error term, which usually makes the repeated-measures test more powerful.
Can I use a repeated-measures ANOVA with only two time points?
You can, but a paired-samples t-test is simpler and gives the same answer. Reserve the repeated-measures ANOVA for three or more occasions or conditions.
What should I do if Mauchly’s test is significant?
Use the Greenhouse-Geisser corrected row of the Tests of Within-Subjects Effects table and report the corrected degrees of freedom. You do not need a different test.
What is the nonparametric alternative if my data are badly non-normal?
The Friedman test is the usual nonparametric counterpart for one group measured on three or more occasions. Follow it with Wilcoxon signed-rank tests (with a correction for multiple comparisons) if it is significant.
Do I need equal group sizes for a repeated-measures ANOVA?
There are no groups to balance — every participant contributes a score at every occasion. What matters instead is complete data: participants with a missing occasion are typically dropped from the analysis by default, so check how much data would be lost before you run the test.
Need this done on your data? Get a free quote → /services/