Mathematics
ANOVA (Analysis of Variance)
Quick fact
ANOVA was developed by Ronald Fisher in the 1920s and its name was originally applied to agricultural experiments where he analyzed the variance between different crop yields.
Why this is interesting
Imagine you're testing three different fertilizers on plant growth. If you just compare pairs with multiple t-tests, your chance of a false positive soars. ANOVA gives you a single, reliable test—how does it pull that off?
Read the full explanation
Understanding ANOVA (Analysis of Variance)
At its heart, ANOVA asks: Are the differences between group averages larger than the natural variation we'd see within each group? Suppose you have three classes taking different study methods. You collect exam scores. Even within one method, scores vary—some students score higher, some lower. That's called 'within-group variance.' If the teaching method truly matters, the scores from different methods should cluster apart, creating larger 'between-group variance.' ANOVA compares these two sources of variance. If the between-group variance is much larger than the within-group variance, then at least one method is likely different. The result is an F-statistic (named after Fisher) and a p-value that tells you if the difference is statistically significant.
A deeper explanation
ANOVA works by partitioning total variation in the data into two parts: sum of squares between groups (SSB) and sum of squares within groups (SSW). The total sum of squares (SST) equals SSB + SSW. Each sum of squares is divided by its degrees of freedom to get mean squares: MSB and MSW. The F-ratio = MSB/MSW follows an F-distribution under the null hypothesis that all group means are equal. A large F indicates that the between-group variation is not just random noise. This approach elegantly handles multiple groups by testing a single global hypothesis, avoiding the inflated error rate from multiple pairwise tests. If the global test is significant, post-hoc tests (like Tukey's HSD) can identify which specific groups differ. ANOVA is a special case of linear regression and forms the foundation for more complex designs like factorial ANOVA, repeated measures, and analysis of covariance (ANCOVA).