Oversimplified models underestimate the role of local environmental filtering

Aims Fitting community assembly via trait selection (CATS) models is a useful method of estimating the relative role environmental in assembly. In simplest version CATS models, only linear and additive trait–environment relationships are supposed. This paper explores consequences neglecting non-linearity interactions by analyzing simulated field data. Location Northern Hungary, Central Europe. Methods Starting with Gaussian response curves, we discussed how more complex species–environment can be modeled regression. Using data, calculated bias unsatisfying linearity additivity assumptions introduce estimated filtering Field data collected forest understorey vegetation were used for illustrating magnitude underestimation real set. Results A non-linear relationship appeared even simplest, but realistic simulation (i.e., curve). Neglecting both resulted considerable strength correlation. Moreover, this was stronger middle part gradients, leading to spurious pattern. Conclusion At least, second-order polynomial regression model (including among traits) should fitted avoid filtering.