Significance tests of contrasts of effects of levels of a factor or of combinations of levels of two or more factors require the knowledge of the estimates of the variances of the estimators of these contrasts. In general, the expressions of these estimates for experiments with complex designs are not available in textbooks. Their algebraic derivation from the equation of the statistical model is tedious. A practical algorithm for obtaining the variance of the estimator of any contrast of parameters for balanced complete designs is presented. The algorithm is based on the expected values of mean squares.
treatment comparisons; balanced complete designs; inference from experiment
