Summary1. Ecological studies commonly assume that it is possible to extrapolate from a response shown by a fixed set of species to the response when the species composition is allowed to change. However, as described by Simpson's paradox, this is not necessarily a reasonable expectation. 2. The impact of Simpson's paradox on an ecological question was tested using a metaanalysis of data on plant responses to arbuscular mycorrhizas. Although species-level response commonly declines as phosphorus availability increases, we hypothesized that the community-level response could either decline or remain constant. 3. As expected, mycorrhizal response of individual species declined significantly as P supply increased. The response averaged across multiple species was negative but not robust, so we cannot distinguish clearly between the hypotheses. 4. It is impossible to assume that community-level responses to environmental gradients are the same as those found at species level. We recommend that experimental tests of hypotheses should allow species identity to change with the environment.