ANOVA Identity

The ANOVA identity decomposes each total deviation into a treatment part and a within-treatment part:

$$ y_{ij} - \bar{y}_{\cdot\cdot}

\left(\bar{y}{i\cdot} - \bar{y}{\cdot\cdot}\right) + \left(y_{ij} - \bar{y}_{i\cdot}\right). $$

Squaring and summing over all observations leads to the variance decomposition

$$ SST = SSTreat + SSE. $$

This is the core identity behind one-way ANOVA.