If the eta squared values in your ANOVA sum to 1.4, you do not have a calculation error: you have <strong>partial</strong> eta squared values labeled plain eta squared. They are different things, they are interpreted differently, and they are not compared against the same benchmarks.
Plain eta squared: η² = SSeffect / SStotal. Every effect shares the same denominator, so they divide up a pie of size 1 and cannot sum to more.
Partial eta squared: η²p = SSeffect / (SSeffect + SSerror). Each effect is compared only against itself and the error, ignoring the others. Since each has its own denominator, the sum is not bounded by 1.
In a two-factor design with an interaction, three partials of .35, .40 and .30 sum to 1.05 and are perfectly correct. The same effects as plain eta might be .18, .21 and .16.
The benchmarks are not the same. The classic "small, medium, large" cutoffs (.01, .06, .14) were proposed for plain eta. Applying them to a partial inflates the interpretation systematically, and that is what most papers do.
The partial depends on the design. Adding a factor that explains variance shrinks the error and RAISES the partial for every other effect without those effects changing at all. Which is why two studies with different partials can carry the same real effect.
And in meta-analysis they are not interchangeable. Pooling partials from different designs is one of the errors that most distorts a synthesis, because design heterogeneity slips in dressed as effect heterogeneity.
Use the symbol with its subscript, η²p or η²p, and say it in the table note as well: "η²p = partial eta squared". It costs six words and closes the ambiguity.
Where you can, give both: the partial for comparisons within the model, and the plain one so readers can see how much total variance each effect explains. Almost no package gives them together, but both come out of the same sums-of-squares table.
And check while you are there that each one matches its own F: that is done in effect size against F. How to report both is in how to report partial eta squared in APA 7.
It is not, if they are partial — and it is the clearest sign that they are. The error is the label: presenting them as plain eta squared and comparing them against plain-eta benchmarks.
No: they sum to at most 1, and what is left over is the error variance. If they sum to exactly 1 in a design with error, something is off.
Omega corrects eta’s optimistic bias and always comes out slightly smaller. It also has a partial version with the same denominator logic, and it can come out slightly negative when the effect is nil, which is correct even if it looks odd.
That is where partials run highest, because within-subject error tends to be small. Values of .40 or more are normal and do not mean a huge effect: they mean a design with little noise.