**Residual error**:
All ANOVA models have residual variation defined by the variation amongst
sampling units within each sample. This is always given by the last mean square
in ANOVA tables, and denoted 'ε' (epsilon) in the descriptions of fully replicated
models where it represents the error variance for at least some of the
treatment effects. Models without full replication may have no degrees of
freedom for measuring residual variation (e.g., randomised block, split plot,
and repeated measures models).

*Error variance* is
the random variation in the response against which an effect is tested,
containing all of the same components of variation estimated in the population except
for the test effect. The validity of ANOVA depends on three assumptions about
the error variance: (i) that the random variation around fitted values is the
same for all sample means of a factor, or across the range of a covariate; (ii)
that the residuals contributing to this variation are free to vary
independently of each other; (iii) that the residual variation approximates to
a normal distribution.

Doncaster, C. P. & Davey, A. J. H. (2007) *Analysis of Variance and Covariance: How to
Choose and Construct Models for the Life Sciences*. Cambridge: Cambridge
University Press.

http://www.southampton.ac.uk/~cpd/anovas/datasets/