The leaf economics spectrum (LES) describes global covariation in the traits of plant leaves. The LES is thought to arise from biophysical constraints and habitat filtering (ecological selection against unfit trait combinations along environmental gradients). However, the role of habitat filtering in generating the LES has not been tested experimentally. If the process of habitat filtering plays a role in generating the LES, the LES could weaken in communities that have yet to be filtered by the current environment, for example after abiotic environmental change. LES traits are commonly used to predict community and ecosystem processes, and if the LES weakens in unfiltered communities, LES-based models may no longer apply. In the glasshouse, we experimentally simulated three stages of habitat filtering in response to abiotic change: from unfiltered, to semi-filtered, to completely filtered communities. In each stage, we quantified the strength of the LES and assessed the accuracy of trait-based models of an important ecological process, pathogen infection. The strength of the LES increased with the completeness of habitat filtering, as did the accuracy of trait-based models of plant susceptibility to pathogen infection. Synthesis. Our results suggest that habitat filtering plays a fundamental role in strengthening the trait correlations of the LES and that trait-based models may be less accurate when communities have not been filtered by the current environment, for example, following rapid environmental change.