Analyses

Two-way ANOVA

Test two grouping factors, and whether they interact, on one numeric outcome.

Edit on GitHub

When to use it

Use two-way ANOVA when two categorical factors might both affect a numeric outcome, and you care whether they combine. For example: do exam scores depend on teaching method, on morning versus afternoon, and on the combination of the two?

A significant interaction means the teaching gap is not the same in the morning as in the afternoon. You then describe the pattern in each combination, not just the two averages.

Assumptions

Scores are numeric and independent. Cell sizes should be adequate, and the spread of scores should be similar across cells.

Tensr does not run Levene’s test or Shapiro–Wilk for this procedure. There is no effect-size option. The featured metric is R², the proportion of outcome variance the full model accounts for. If you need a homogeneity check, it is not on this page.

Running it in Tensr

Analyze → General Linear Model → Two-Way ANOVA. In chat: “Two-way ANOVA of score by teaching and time.”
Factor A and factor B are the two grouping columns. Value column is the numeric outcome.
Include interaction starts on. Post hoc starts at none, with the same choices as one-way ANOVA: tukey, bonferroni, scheffe, games_howell.

Options

Prop

Type

Reading the output

Teaching method and morning versus afternoon, on the same exam. Method was built to matter. Time of day was not, and the two were not built to interact.

SourceSum of squaresdfFp-value
method523.35124.018.021
time5.9510.091.763
method × time166.67621.28.283
Residual5,861.81190——
Group AGroup BMean differencep (adj)*CI lowCI highReject H₀
LectureOnline4.025.118-0.7718.822False
LectureWorkshop5.531.020*0.73510.328True
OnlineWorkshop1.506.736-3.296.303False

Method is associated with the exam, F = 4.018, p = .021. Time of day is not, F = 0.091, p = .763. The method × time interaction is not either, F = 1.28, p = .283. Tukey then compares teaching methods. Only some pairs clear .05, which is what a moderate gap looks like when the groups overlap.

Reporting (APA 7)

A two-way ANOVA on exam score (N = 96) found a method effect, F = 4.018, p = .021, no time-of-day effect, F = 0.091, p = .763, and no method × time interaction, F = 1.28, p = .283.

Coming from SPSS

Analyze → General Linear Model → Univariate.

SPSS Univariate can print partial η², estimated marginal means, and Levene’s test from the Options button. Tensr’s table has F and p, plus R² as a metric, and does not print partial η² or a homogeneity test. ANCOVA in Tensr is a separate menu item, not a covariate you add inside this dialog.

One factor only: one-way ANOVA. Three factors: three-way ANOVA. A numeric control variable: ANCOVA.