**General Linear Model (GLM)**: A linear model is a
statistical model with linear (additive) combinations of parameter constants
describing effect sizes and variance components. Linear models can describe
non-linear trends in covariates, for example by transformation of the data or
fitting a polynomial model. GLM is a generic term
for parametric analyses of variance that can accommodate combinations of
factors and covariates, and unbalanced and
non-orthogonal designs. GLMs generally use an unrestricted model for analysing
combinations of fixed and random factors.

Significant effects are tested with the *F*
statistic, which is constructed from sums of squared deviations of observations
from means, adjusted for any non-orthogonality. This statistic assumes random sampling of independent replicates, homogeneous
within-sample variances, a normal distribution of the residual error variation
around sample means, and a linear response to any covariate. Transformations may
be necessary to the response and/or covariate to meet these assumptions. A
further generalization of GLM, the Generalized Linear Model (GLIM), accommodates
non-normally distributed response variables, and partitions the components of
variation using maximum likelihood rather than sums of squares.

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/