Answers are grounded in this paper and cite their sources.
For fixed-effect models one can always, according to the Gauss-Markov Theorem, uniquely determine independent variables called source identifiers, each corresponding to a source of variation. When linearly combined, source identifiers can generate all possible expected values for the response variable. The co-effect method uses regression of the response variable on source identifiers. Corresponding regression coefficients are, by definition, unbiased estimates of co-effects, and satisfy the same restrictions as those imposed on main effects and interaction effects in standard analysis of variance. with balanced data, co-effect analysis gives results identical to those of the standard method; with unbalanced data, however, results can be significantly different . An example is given where predicted genetic interaction can be easily observed using the co-effect method (Р≈10-14) while Yates' weighted-squares-of-means method does not detect any interaction effects (P>O.l) .
No takes yet. Share an insight, caveat, or question.
André Plante (1992) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: