Why mean indicator values are not biased

Abstract

AbstractQuestionsZelený & Schaffers (Journal of Vegetation Science 2012, 23: 419–431)) blame mean Ellenberg indicator values for producing ‘highly biased results’ that are ‘too good to be true’, statements causing considerable confusion among plant ecologists. I ask if (1) the detrimental phenomena they observed can be reproduced in yet another real‐world data set using the recently revised indicator values of Landolt et al. (Flora indicativa. Ecological indicator values and biological attributes of the flora of Switzerland and the Alps, 2010, Haupt, Bern, CH), (2) the issue of supposed bias also applies to species‐based functional traits, and (3) the assumed ‘bias’ also concerns principal components analysis.MethodsIn a first step, I probe the effect of randomizing indicator values by verification of classifications through ANOVA and CCA using published data. Then, I repeat the same procedure replacing the matrix of indicator values with a species by traits matrix. In a third step I apply testing with ANOVA to an ordination of principal components. Finally, patterns of alternative vegetation descriptors (species, mean indicator values, traits frequency, site factors) are compared using Mantel correlation.ResultsThe results at first glance confirm the failings of tests as found by Zelený & Schaffers. However, when using traits data instead of mean indicator values, precisely the same happens, raising the possibility that mean species‐based traits are also ‘biased’. The effect prevails when applied to PCA scores. I conclude that Zelený & Schaffers overlooked that unrealistic significances in ANOVA emerge from statistical dependence; they are artefacts of a non‐valid null model.ConclusionsRestrictions in the use of mean indicator values and species‐based traits concern statistical tests only that assume statistical independence from species scores when there is none. I conclude that mean indicator values deserve rehabilitation as there is no bias involved in their use.