13.1 One-Way ANOVA

One-Way ANOVA

One-way ANOVA is a statistical method for comparing means across three or more groups. Instead of running multiple t-tests (which inflates your chance of a Type I error), ANOVA lets you test all the groups at once. This section covers the logic behind ANOVA, how to set up hypotheses, the key components of the analysis, and how to interpret results visually.

One-Way ANOVA

Purpose of One-Way ANOVA

The core question ANOVA answers: Are the differences between these group means larger than what we'd expect from random chance alone?

• Compares means of three or more groups (levels of an independent variable), such as age groups, treatment conditions, or dosage levels
• The independent variable is categorical with at least three levels (e.g., low, medium, high dosage)
• The dependent variable is continuous (e.g., weight, height, income)
• Why not just run several t-tests? With three groups you'd need three separate t-tests, and each one carries a risk of a false positive. ANOVA handles all groups in a single test, keeping the overall Type I error rate at your chosen $\alpha$ level.

Hypotheses for Multiple Group Comparisons

• Null hypothesis $H_0$: All group means are equal.
  • $\mu_1 = \mu_2 = \mu_3 = \ldots = \mu_k$, where $k$ is the number of groups
• Alternative hypothesis $H_a$: At least one group mean differs from the others.
  • At least one $\mu_i \neq \mu_j$, where $i \neq j$

Notice that the alternative hypothesis doesn't tell you which group is different. It only says that not all the means are equal. If ANOVA gives a significant result, you'll need follow-up tests (post-hoc tests) to pinpoint where the differences are.

ANOVA is inherently a two-sided (non-directional) test. It detects any difference among group means without specifying a direction. Unlike a t-test, you don't set up a one-tailed version of ANOVA.

Analysis of Variance (ANOVA) Components

• Grand mean: The overall mean of all observations combined, regardless of group. ANOVA compares how far each group mean falls from this grand mean.
• Omnibus test: ANOVA is called an omnibus ("overall") test because it evaluates whether any significant differences exist among the group means. It doesn't tell you which specific pairs differ.
• Effect size: Measures the magnitude of the differences between group means, not just whether they're statistically significant. A common effect size measure for ANOVA is eta-squared ($\eta^2$), which represents the proportion of total variance in the dependent variable explained by the group variable.
• Post-hoc (multiple comparison) tests: Conducted only after a significant ANOVA result. These are pairwise comparisons between specific groups to identify exactly where the differences lie. Common examples include Tukey's HSD and Bonferroni correction.

Interpreting ANOVA Results

Box Plots for ANOVA Visualization

Box plots display the distribution of the dependent variable for each group side by side, making them a natural companion to ANOVA.

Each box plot shows:

• The median (line inside the box)
• The interquartile range (IQR), which contains the middle 50% of the data (the box itself)
• Whiskers extending to the most extreme values within 1.5 times the IQR from the edges of the box
• Outliers plotted as individual points beyond the whiskers

When comparing box plots across groups:

• Non-overlapping boxes suggest the group medians (and likely means) differ substantially. For example, box plots of income levels across three countries with no overlap would hint at significant differences.
• Overlapping boxes don't automatically mean the differences are non-significant. The extent of overlap and the sample sizes both matter, so always check the actual ANOVA results.
• Similar box sizes across groups suggest homogeneity of variances, one of ANOVA's key assumptions. If one group's box is much wider or narrower than the others, that assumption may be violated.
• Outliers can pull group means and inflate variance, potentially affecting your ANOVA results. Investigate them before drawing conclusions.

Box plots complement the numerical output of ANOVA. They help you spot potential assumption violations (like unequal variances or non-normal distributions) and give you an intuitive sense of where group differences might be before you even look at the F statistic.