Aim: Decades of experimental research have conclusively shown a positive relationship between species richness and ecosystem function. However, authoritative reviews find no consensus on how species loss affects function in natural communities. We analyse experimental and observational data in an identical way and test whether they produce similar results.

Location: North America and Europe (experimental communities); global (natural communities).

Time period: Experimental communities: 1998–2013; natural communities: 1982–2018.

Major taxa studied: 

Experimental communities: temperate grassland plants; natural communities: temperate grassland plants, tropical forest trees, kelp forest producers
and native bees.

Methods: We used an approach inspired by the Price equation to analyse 129 datasets from experimental and natural communities worldwide. We tested how the
effects of species loss on ecosystem function varied with dominance and the nonrandomness
of species loss and, in turn, how these two factors differed between
experiments and observations.

Results: Studies carried out in experimental and natural communities reached different conclusions regarding the effects of species loss. First, species loss had greater effects on ecosystem function in experiments than in nature. Second, the importance of species loss was negatively correlated with dominance in nature because as dominance increased, lost species were increasingly those contributing little to ecosystem function. Although experimental and natural communities exhibited similar levels of dominance, an analogous relationship was not possible in experiments because the order of species loss was randomized by design. Main conclusions: Species loss was sometimes, but not always, the major driver of loss of function in nature. Variation in the importance of species loss was not messy and context dependent; instead, it was predicted by functional dominance. Although results from experimental and natural communities were similar in several key ways, they differed in that species loss was a consistent predictor of ecosystem function in experiments and not in nature.

Data was collected from sources described in the original paper and processed into species (rows) by sites (columns) matrices where each matrix cell is the species' contribution to function at that site. Data is in an RDS format as a list with three entires. The first entry has one matrix for each of the 129 datasets. The second and third entry describe, for each matrix, the specific ecosystem function and whether it was natural vs. experiment, respectively.