ANOVA vs t-test: when to use which

A t-test compares the means of two groups, while ANOVA (analysis of variance) compares the means of three or more groups in a single test. Both are parametric methods that ask whether observed differences in means are larger than you would expect from chance. The rule for your dissertation is simple: count your groups first, because the number of groups, not the topic, decides which test you use.

Why you cannot just run several t-tests instead

It is tempting to compare three groups with three separate t-tests, but doing so inflates your Type I error rate. Each test carries a five percent risk of a false positive, and running several stacks that risk until a significant result is likely by accident alone. ANOVA solves this by testing all groups at once with a single omnibus test, holding the overall error rate at your chosen alpha. That is the core reason assessors expect ANOVA rather than a pile of pairwise comparisons, and it is part of the broader logic of matching the test to your design.

Matching the test to your design

The right choice depends on how your independent variable is structured, so it helps to be clear on which variable predicts which. A two-level grouping variable calls for a t-test; three or more levels call for one-way ANOVA. If you have two grouping variables at once, a two-way ANOVA lets you test each main effect plus their interaction. Both families assume a roughly normal distribution and homogeneity of variance, so confirm those before you run anything.

After a significant ANOVA: post hoc tests

A significant ANOVA tells you that at least one group mean differs, but not which groups differ from each other. To locate the difference you run post hoc tests, such as Tukey's HSD or Bonferroni-corrected comparisons, which control the error rate across all the pairwise contrasts. These are not the same as running raw t-tests, because the correction keeps your false positive risk in check. Report the omnibus result first, then the post hoc findings, each with an effect size.

One-way ANOVA as an extension of the t-test

The two tests are not separate inventions; a one-way ANOVA with exactly two groups gives the same verdict as an independent-samples t-test, and the relationship is exact: the ANOVA F statistic equals the square of the t value (F = t²), with the same p-value. Seen this way, ANOVA is the t-test generalised to handle any number of groups while keeping the overall error rate under control. That is why you do not need a special test for three groups and a different one for four; one framework scales across them. Understanding this link also clarifies why both rest on the same parametric footing, the topic of parametric versus rank-based tests, and why the same assumption checks apply to each.

When to use a t-test vs ANOVA in practice

The deciding factor is the number of levels in your independent variable, but a few practical cues sharpen the call. Reach for a t-test when you have a clean two-condition comparison, such as treatment versus control or before versus after, and reach for ANOVA the moment a third condition enters, because splitting that into multiple t-tests reintroduces the inflated false positive risk. If your design layers a second grouping factor, a two-way ANOVA earns its place by testing the interaction a t-test cannot see. When you are unsure how your variables map onto a design, work it through with the statistical test selector or the broader guide to reading your output so the choice is justified rather than habitual.

Reporting both tests cleanly

Whichever test you use, present it in a consistent format: the test statistic, degrees of freedom, the exact p-value, and an effect size such as Cohen's d for a t-test or eta squared for ANOVA. Lead with descriptive statistics so the reader sees the group means and spreads before the inferential verdict, and follow the conventions in reporting results in APA style. Done this way, your analysis chapter reads as a deliberate sequence rather than a scatter of outputs.