Methodology: One-Way Analysis of Variance (ANOVA)
The One-Way Analysis of Variance (ANOVA), pioneered by Sir Ronald A. Fisher, is the fundamental parametric test for comparing the means of three or more independent experimental conditions. It determines whether at least one group mean is statistically significantly different from the others without inflating the family-wise Type I error rate.
1. The Omnibus Partitioning of Variance
ANOVA partitions the total sum of squares ($SS_{\text{Total}}$) into two orthogonal components: the variation between sample means ($SS_{\text{Between}}$) and the intrinsic random error within groups ($SS_{\text{Within}}$):
$$ SS_{\text{Total}} = SS_{\text{Between}} + SS_{\text{Within}} $$ $$ SS_{\text{Between}} = \sum_{k=1}^{K} n_k (\bar{x}_k - \bar{x}_{\text{grand}})^2, \quad df_{\text{Between}} = K - 1 $$ $$ SS_{\text{Within}} = \sum_{k=1}^{K} \sum_{i=1}^{n_k} (x_{ik} - \bar{x}_k)^2 = \sum_{k=1}^{K} (n_k - 1)s_k^2, \quad df_{\text{Within}} = N - K $$2. The Fisher F-Statistic & p-Value Computation
Under the null hypothesis ($H_0: \mu_1 = \mu_2 = \dots = \mu_K$), the ratio of the Mean Square Between to the Mean Square Within follows an $F$-distribution:
$$ MS_{\text{Between}} = \frac{SS_{\text{Between}}}{K - 1}, \quad MS_{\text{Within}} = \frac{SS_{\text{Within}}}{N - K} $$ $$ F = \frac{MS_{\text{Between}}}{MS_{\text{Within}}} \sim F(df_{\text{Between}}, df_{\text{Within}}) $$The exact upper-tail $p$-value $P(F \ge f)$ is computed via the regularized incomplete beta function:
$$ p = I_{\frac{df_2}{df_1 \cdot f + df_2}}\left(\frac{df_2}{2}, \frac{df_1}{2}\right) $$3. Effect Size Estimation: Eta-Squared (η²) and Omega-Squared (ω²)
A statistically significant $p$-value does not convey the magnitude of the experimental effect. Standard guidelines (APA 7th edition, ICMJE) require reporting standardized effect sizes:
$$ \eta^2 = \frac{SS_{\text{Between}}}{SS_{\text{Total}}} $$ $$ \omega^2 = \frac{SS_{\text{Between}} - (K - 1)MS_{\text{Within}}}{SS_{\text{Total}} + MS_{\text{Within}}} $$Interpretation (Cohen, 1988): $\eta^2 \approx 0.01$ indicates a small effect; $\eta^2 \approx 0.06$ indicates a medium effect; $\eta^2 \ge 0.14$ indicates a large effect. Because $\eta^2$ exhibits positive sample bias, $\omega^2$ is strongly recommended for smaller sample sizes.
4. Core Parametric Assumptions
- Independence of Observations: Each participant or unit must appear in exactly one group.
- Normality: Residuals within each treatment condition should be approximately normally distributed (robust to moderate departures when $n \ge 20$).
- Homoscedasticity: Homogeneity of group variances ($\sigma_1^2 = \sigma_2^2 = \dots = \sigma_K^2$). If variances are severely unequal and sample sizes are unbalanced, Welch's adjusted ANOVA is indicated.
5. References
- Cohen, J. (1988). Statistical Power Analysis for the Behavioral Sciences (2nd ed.). Lawrence Erlbaum Associates.
- Fisher, R. A. (1925). Statistical Methods for Research Workers. Oliver & Boyd.
- Keppel, G., & Wickens, T. D. (2004). Design and Analysis: A Researcher's Handbook (4th ed.). Pearson Prentice Hall.