Abstract
Correlations between community‐weighted mean ( CWM ) traits and environmental gradients are often assumed to quantify the adaptive value of traits. We tested this assumption by comparing these correlations with models of survival probability using 46 perennial species from long‐term permanent plots in pine forests of Arizona. Survival was modelled as a function of trait × environment interactions, plant size, climatic variation and neighbourhood competition. The effect of traits on survival depended on the environmental conditions, but the two statistical approaches were inconsistent. For example, CWM ‐specific leaf area ( SLA ) and soil fertility were uncorrelated. However, survival was highest for species with low SLA in infertile soil, a result which agreed with expectations derived from the physiological trade‐off underpinning leaf economic theory. CWM trait–environment relationships were unreliable estimates of how traits affected survival, and should only be used in predictive models when there is empirical support for an evolutionary trade‐off that affects vital rates.