(oral conference presentation at EVS meeting in Ljubljana, 2014)
Ellenberg indicator values represent estimated optima of plant species along main ecological gradients. Values assembled by Ellenberg are valid mostly in central Europe, but similar systems were developed also in other parts of Europe (e.g. Landolt’s values for Switzerland, Borhidi’s for Hungary, Jurko’s for Slovakia, Zarzycki’s for Poland, Hill’s for Great Britain, Pignatti’s for Italy) and also outside Europe (Klinka’s values for British Columbia). Means of Ellenberg indicator values for species occurring in given vegetation plot (sometimes weighted by their abundances) are commonly used to infer habitat conditions from species composition. To evaluate how accurate is such inference, mean Ellenberg values are routinely correlated or regressed with measured environmental variables, and a good fit is considered as a proof that Ellenberg values are good estimates of real environmental conditions.
The problem is that one can get strong and significant relationship between mean Ellenberg indicator values and measured environmental variable even in case that species Ellenberg values are replaced by random numbers, and mean of such random numbers therefore have no ecological meaning. This paradox situation occurs in cases of environmental variables which have a strong effect on species composition. The reason is that weighted mean of species indicator values and environmental variable are not independent from each other. They are linked by changes in species composition: mean Ellenberg values are derived from species composition (by calculation), while environmental variables influence species composition. When relating mean Ellenberg values to environmental variables, it is important to separate the relationship between environmental variable and species composition (which is not of our interest in this case) from relationship between environmental variable and indicator values (which is what we are testing).
In a previous study, Zelený & Schaffers (2012) pointed out similar issue with testing the significance of relationship between mean Ellenberg values and scores from ordination diagrams or results of cluster analysis, which was explained by circularity of reasoning. Here, I will show that this issue is more general and includes also testing the relationship between mean Ellenberg values and environmental variables or experimental treatments. I will propose a way how to deal with this problem, illustrate it on examples of real vegetation data and demonstrate a simple software (https://davidzeleny.net/wiki/doku.php/eiv:start), which calculates the correct significance values for regression or correlation between mean Ellenberg values and other variables (environmental variables, ordination scores, cluster assignment etc.).
Zelený D. & Schaffers A.P. (2012): Too good to be true: pitfalls of using mean Ellenberg indicator values in vegetation analyses. Journal of Vegetation Science, 23: 419-431.