Dear R members, I understand the main principles why R-Vegan does not provide p-values for the biplot scores and/or canonical coefficients (see also post on stackoverflow). (i) We can obtain linear regression statistics and refit an ordination result as multiple response linear model (lm, see as.mlm.cca). This regression ignores residual unconstrained variation in the data. However, constrained ordination is based on iteration with regression. My question is now, how does ordination considers this unconstrained variation? By the unimodal distribution of the data (cca). By the selected distance matrix (Chi, Euclidian)? Or is the difference based on the fact, that ordination is a multivariate analyses? (ii) I think question (i) is the reason why I get difference between biplot scores (integral of rda) and scores() (equivalent of regression coefficients) scores(PFcompUZL_h_rda, choices = 1:4, display = "bp", scaling = 0) scores(PFcompUZL_h_rda) Many thanks for your answer Sibylle [[alternative HTML version deleted]]