Linking community-weighted trait mean values (CWM) and local environmental conditions have become a key framework to understand the processes underlying the structure of ecological communities. Two major statistical frameworks to make this link are (i) the correlation between CWM and environmental features and (ii) the fourth-corner correlation approach. The latter is a weighted correlation (sum of community abundances as weights) between CWM and environmental features; CWM and environment are both standardized by appropriate weights. Although these two methods are routinely applied by ecologists, individual variation within species and its effect on the estimation of CWM values and their links with environment have been largely ignored. Here we evaluate the effects of intraspecific trait variation on assessing trait-environment relationships and propose a mixed model implementation to account for intraspecific trait variation when information on individuals’ trait values is not available. These types of implementation are important given that lack of information regarding intraspecific variation is often the case, and that trait averages within species are used instead to calculate CWM values. Simulations generating varying levels of intraspecific trait variation were used to estimate its effect on community trait-environment correlations and the robustness of the mixed model version.
Results/Conclusions
We found that intraspecific trait variation can severely bias the estimation of trait-environment relationship by reducing the correlation, and thus reducing the statistical power to detect the relationship, particularly when environmental gradients are short. Considering intraspecific variation always improves the estimation, even when intraspecific variation does not greatly affect correlation values. However, including intraspecific variation in trait-environment relationship with the classical fourth-corner approach requires trait values for every individual, which is extremely data intensive. Our mixed model implementation of the fourth-corner correlation accounts for intraspecific variation, and lead to improved estimates of community trait-environment correlations without requiring any additional data. In conclusion, ignoring intraspecific trait variation can lead to severely biased estimates of community trait-environment relationships that can be attenuated by a mixed model implementation.