Analyses

Three-way ANOVA

Test three grouping factors and their interactions on one numeric outcome.

Edit on GitHub

When to use it

Use three-way ANOVA when three categorical factors might affect one numeric outcome. For example: do practice scores depend on sleep, caffeine, and practice schedule, including the ways those factors combine?

Read the three-way interaction first. If it is significant, the two-way patterns are not the same at every level of the third factor, and the main effects are a blunt summary.

Assumptions

Each factor needs at least two levels, and Tensr asks for at least eight complete rows. A balanced design (the same number of scores in every combination) uses Type II sums of squares. An unbalanced or incomplete design uses Type III.

Tensr does not run Levene’s test or Shapiro–Wilk here. η² in the table is each term’s sum of squares divided by the total sum of squares, not partial η². There is no switch for that.

If a sum of squares comes out negative, the report warns that the model may be overspecified. Tensr then drops the three-way term and refits the two-way interactions. If squares are still negative, the run stops.

Running it in Tensr

Analyze → Compare Means → Three-Way ANOVA. In chat: “Three-way ANOVA of score by sleep, caffeine, and practice.”
Factor A, factor B, and factor C are the three grouping columns. Value column is the numeric outcome.
Include interactions starts on. Confidence level starts at 0.95. There is no post-hoc control on this procedure.

Options

Prop

Type

Reading the output

Method, time of day, and a north/south campus label on the same exam. Only method was built to move the scores.

SourceSSdfFpη²
method523.35123.858.0250.08
time5.9510.088.7680.0009073
campus1.42610.021.8850.0002174
method × time166.67621.229.2980.025
method × campus88.88320.655.5220.014
time × campus46.34310.683.4110.007
method × time × campus28.41620.209.8110.004
Residual569784———
FactorLevelMean
methodLecture68.209
methodOnline72.234
methodWorkshop73.741
timeAfternoon71.644
timeMorning71.146
campusNorth71.273
campusSouth71.517

Three-way ANOVA for score, n = 96. Three-way ANOVA for score, n = 96. Metrics: R² = 0.131.

Reporting (APA 7)

Three-way ANOVA for score, n = 96. Report the estimate in the table. This procedure is not summarised by one p-value.

Coming from SPSS

Analyze → General Linear Model → Univariate, with three fixed factors. Tensr’s stored path is Analyze → General Linear Model → Three-Way ANOVA. SPSS can print partial η² and post-hoc tests from that dialog. Tensr prints η² as SS / SS total and has no post-hoc option here.

Two factors: two-way ANOVA. One factor: one-way ANOVA.