A properly reported repeated measures ANOVA in APA 7 carries one line that never shows up in a between-groups ANOVA: degrees of freedom with decimals. Something like F(1.56, 45.24) = 12.34, p < .001, η²p = .30. Those decimals are not a typo or an SPSS quirk. They are the fingerprint of a sphericity correction, and they are the first thing a reviewer who knows the design will look for.
This guide covers exactly what to report, in what order, and which of the two corrections applies to you. If what you need is the one-way between-groups ANOVA, that is in how to report an ANOVA in APA 7; and if you are still deciding whether your design calls for a repeated measures ANOVA or a mixed model, the comparison is in mixed models versus repeated measures ANOVA.
What a repeated measures ANOVA needs in APA 7
The full report has five pieces, in this order:
1. Mauchly's test. This comes first, because everything else depends on whether sphericity holds. It is written with its chi-square, its degrees of freedom and its p.
2. The correction applied, if one was needed, naming the correction and giving epsilon.
3. The F statistic with its degrees of freedom, already corrected if applicable.
4. The p value.
5. The effect size, usually partial eta squared (η²p) or generalized eta squared (η²G).
And if the effect is significant and the factor has more than two levels, add the pairwise comparisons with the adjustment method you used. None of this is optional in APA 7: the manual asks for an effect size for every contrast you report, and any journal with serious methodological review will send the manuscript back if Mauchly's result is missing.
Sphericity: what it is and why it forces you to change the degrees of freedom
Sphericity is the assumption that the variances of the differences between every pair of levels are equal. Watch the wording, because almost everyone states it wrong: it is not that the variances of the levels are equal, it is that the variances of the differences between levels are.
With two levels (pretest and posttest, nothing else) the assumption holds by definition: there is only one possible difference, so there is only one variance. That is why, in a two-measure design, SPSS hands you an empty Mauchly table and there is nothing to correct. From three levels up there are several differences and the assumption can fail.
When it fails, F becomes too optimistic: you reject the null more often than you should. The fix is not a different test, it is to penalize the degrees of freedom by multiplying them by a factor below 1, which is epsilon (ε). Lower degrees of freedom push the critical value of F up and make p more conservative. That is where the decimals come from.
Mauchly's test is not something to trust blindly
Mauchly's test has the same problem as every other assumption test: it depends on sample size. With 20 participants it misses violations that do matter; with 200 it flags trivial departures as significant. This is the same argument that applies to the Shapiro-Wilk normality test, and for the same reason.
Because of that, a good part of the methodological literature recommends simply correcting always, without looking at Mauchly. If epsilon equals 1 the correction changes nothing and you have lost nothing; if it is lower, you are already covered. That practice is defensible and increasingly common, but it has one practical consequence: if you decide to correct always, say so in the manuscript. A reviewer who sees decimal degrees of freedom and no Mauchly reported will assume you forgot the test, not that you made a decision.
Greenhouse-Geisser or Huynh-Feldt: which one is yours
There are two corrections and SPSS gives you both, which leaves you wondering which row to copy. The rule used in practice comes from Girden (1992) and depends on the Greenhouse-Geisser epsilon:
If Greenhouse-Geisser ε is below .75, use Greenhouse-Geisser. It is the more conservative correction and that is where it is needed.
If Greenhouse-Geisser ε is above .75, use Huynh-Feldt. Greenhouse-Geisser over-corrects in that range and you lose power for nothing.
There is a third option in the output, "lower bound", and you should not use it: it is the theoretical worst case and so conservative that it is almost never what you want.
What you cannot do is look at all three rows, find the one that gives p < .05, and copy that one. That is p-hacking, it shows, and it is exactly the kind of thing the Q1 paper reviewer flags when it reads a results section.
A full example, ready to copy
Picture an anxiety intervention study with 30 participants measured at three time points: before, after, and at three months. This is what would go in the Results section:
"Mauchly's test indicated that the assumption of sphericity had been violated, W = .72, χ²(2) = 9.34, p = .009, so the Greenhouse-Geisser correction was applied (ε = .78). The effect of time point on anxiety was significant, F(1.56, 45.24) = 12.34, p < .001, η²p = .30. Pairwise comparisons with Bonferroni adjustment showed a significant reduction between pretest (M = 24.30, SD = 5.12) and posttest (M = 18.70, SD = 4.88), p < .001, and between pretest and follow-up (M = 19.10, SD = 5.03), p = .002. The difference between posttest and follow-up was not significant, p = 1.000."
Notice four details that get missed constantly:
Degrees of freedom carry two decimals and are not rounded to integers. If you write F(2, 58) after correcting, you are reporting the uncorrected ANOVA.
Epsilon takes two decimals and no leading zero, because it cannot exceed 1.
A Bonferroni p can be exactly 1.000 and is written that way. The adjustment multiplies p by the number of comparisons and truncates at 1; it is not an output error.
Means and standard deviations go inside the sentence when there are few of them. If there are many, they go in a table and only the contrast stays in the text.
Partial eta squared in repeated measures: the warning almost nobody gives
The η²p that SPSS returns in a repeated measures design uses the sum of squares of the effect plus its specific error term as the denominator, not total variability. In within-subjects designs that error term is small, because between-person variability has already been partialled out. The result: the η²p of a repeated measures ANOVA comes out systematically higher than that of an equivalent between-groups design, and not because the effect is bigger.
That has two consequences that actually affect you. First, you cannot compare your η²p with one from a between-groups study as if they measured the same thing. Second, if someone puts your study into a meta-analysis, that η²p will give them trouble.
The alternative is generalized eta squared (η²G), proposed by Olejnik and Algina (2003) precisely so that effect sizes are comparable across designs. More and more psychology journals ask for it in within-subjects designs. If your target journal says nothing, the most defensible move is to report both, or to report η²p and add one sentence saying it is not comparable with between-groups designs.
If you already have the result written up and need to pull the effect size without redoing the maths, the effect size calculator converts it, and effect size from text extracts it straight from the paragraph.
The table, when you need one
With a single within-subjects factor and three levels the table is unnecessary: everything fits in the text. A table starts earning its place when you have a mixed design (one within-subjects factor and one between-groups), several within-subjects factors, or more than one dependent variable.
In that case the APA 7 table takes a top rule, a rule under the header and a bottom rule, with no vertical lines. The columns are: source of variation, df, F, p and the effect size. Corrected degrees of freedom go in the df column with their decimals, and the correction applied is explained in the table note, not in a column.
If you would rather not wrestle with rules in Word, the APA 7 table generator hands it back formatted and ready to paste.
Mistakes that come up again and again
Reporting Mauchly and then using the uncorrected degrees of freedom. The most common of them all. You report that sphericity failed and then copy the "Sphericity assumed" row. Either you correct or you do not, but what the text says and what the numbers say have to match.
Not naming the correction. Printing decimal degrees of freedom without naming Greenhouse-Geisser or Huynh-Feldt leaves the reader unable to tell which of the two they are looking at, and they are different.
Reporting Mauchly in a two-level design. There is nothing to report: with two measures sphericity always holds. If your Mauchly table comes out empty, that is not an output failure.
Omitting the adjustment method for pairwise comparisons. "Post hoc comparisons were significant" says nothing. Bonferroni, Holm and Tukey give different answers, and you have to say which.
Using a repeated measures ANOVA when there are missing data. Repeated measures ANOVA drops the entire participant if a single measure is missing. In a longitudinal study with attrition, that can take half your sample away without you noticing. There, a mixed model is not a statistical preference: it is the only option that uses all the data.
This same repeated measures ANOVA, with Greenhouse-Geisser applied to real data rather than a textbook example, is part of chapter 4 of the book "Report Your Results in APA 7": ten chapters running through one single study, with the dataset and the scripts to reproduce every number.
Before you submit
The Results section of a within-subjects design is where technical objections pile up most, because there is a run of decisions and each one leaves a trace. If you want to see what will be raised before an editor raises it, upload the manuscript to the Q1 paper reviewer: it is free, needs no sign-up, and tells you exactly what is missing in each sentence.
References
Girden, E. R. (1992). ANOVA: Repeated measures. Sage.
Greenhouse, S. W., & Geisser, S. (1959). On methods in the analysis of profile data. Psychometrika, 24(2), 95-112.
Huynh, H., & Feldt, L. S. (1976). Estimation of the Box correction for degrees of freedom from sample data in randomized block and split-plot designs. Journal of Educational Statistics, 1(1), 69-82.
Mauchly, J. W. (1940). Significance test for sphericity of a normal n-variate distribution. The Annals of Mathematical Statistics, 11(2), 204-209.
Olejnik, S., & Algina, J. (2003). Generalized eta and omega squared statistics: Measures of effect size for some common research designs. Psychological Methods, 8(4), 434-447.