Hello, I use model.matrix () to create my contrast to do a canonical analysis with rdatest() and I would like your opinion on my problem. DATA: I work with a microarray experiment composed 12 chips, 2 factors ( mutation , treatement) , each has 2 levels (absence , presence) , 3 biological replicat. DESIGN: no-mut1 no-mut2 no-mut3 mut1 mut2 mut3 TREATED_no-mut1 TREATED_no-mut2 TREATED_no-mut3 TREATED_mut1 TREATED_mut2 TREATED_mut3 CODE IN R: facteur_mutation = gl(2,3, length = 12) facteur_oxidant = gl(2,6) helmert = model.matrix (~facteur_mutation*facteur_oxidant, contrast = list(facteur_mutation ="contr.helmert", facteur_oxidant= "contr.helmert")) MY RESULT ISSUE FROM CANONICAL ANALYSIS: 63 % of variance explained by mutation 17% of variance explained by Treatement. QUESTION: Given that 20 % of variance is no explained , how to integrage ( in R code ) the factor " replication " I have tried to add in my original code this part : the line factor_replication = c(1,2,3,1,2,3,1,2,3,1,2,3) helmert = model.matrix (~facteur_mutation*facteur_oxidant* facteur_replication , contrast = list(facteur_mutation ="contr.helmert", facteur_oxidant= "contr.helmert", facteur_replication= "contr.helmert")) but I don't think that it's OK, because ( if I have good comprehension of what is helmert is ), the scalar product is no 0 ) . Moreover , the result "said"  that the variance explained by replication is 2 %... It's very low , too low... What Does you think about integrate replication factor in my model.matrix ? Thanks Greg Centre de recherche du CHUM. Montreal [[alternative HTML version deleted]]