Although experiments overwhelmingly show biodiversity increases ecosystem functioning, relationships in natural communities are more variable. This raises the question of whether and when theory would predict negative diversity effects. We argue that plant communities containing more stably coexisting core species should have higher biomass production. However, variation in numbers of transient species, whose abundances fluctuate strongly and which cannot stably coexist, may not consistently affect biomass. Distinguishing changes in core and transient species richness is critical. For instance, recent attempts to use novel causal modelling approaches have implied negative effects of biodiversity on biomass. However, we find these approaches also result in negative relationships when applied to experiments, where we know there is a causal, positive effect of diversity. We suggest that transient species contribute disproportionately to the variation in diversity isolated in these models. We highlight the need for improved approaches to analysing data from naturally assembled communities and call for increased attempts to compare results with experimental systems. Synthesis: understanding the functional consequences of biodiversity loss is critical but we need to be clear about what type of diversity change we are measuring and to focus on the loss of stably coexisting, core species.

The PaNDiv experiment was established in Autumn 2015. It contains factorial manipulations of plant diversity, functional composition, nitrogen, and foliar pathogen exclusion. The experiment contains 216 2m x 2m plots. Experimental plant communities vary in species richness (1, 4, 8, 20 species), functional composition and functional diversity, nitrogen enrichment (in the form of urea; 0, 100 kg.ha 1.y 1) and foliar fungicide treatment (Score Profi by Syngenta Agro AG, 24.8 % difenoconazole and Ortiva by Syngenta Agro GmbH, 22.8 % azoxystrobin). Plots are separated by 1m grass path,s and the whole experiment is mown twice a year, reflecting typical extensive grassland management in this area.

A pool of 20 common, perennial grassland species was used. To manipulate functional composition, in terms of resource use strategy, these were divided into 10 slow and 10 fast-growing species based on specific leaf area (SLA) and leaf Nitrogen. Species combinations were randomly selected from the respective species pool (i.e., fast, slow, or all), with the constraint that all mixtures contained grasses and non-leguminous forbs. Legumes were excluded as they cannot be easily assigned to fast or slow pools. The plots were arranged in four blocks, and all species compositions occurred once per block. Each composition received the four combinations of fungicide x nitrogen, randomly allocated per block. To maintain species compositions, the experiment is weeded three times per year.

We collected aboveground biomass twice per year, before the mowing from 2017 to 2023. The samples were taken by clipping plant material to 5cm above ground level, in two quadrats of 20 cm × 50 cm located in the centre of each plot. We weighed the biomass after 2 days of drying at 65°C. The plot target biomass production (i.e., of the sown species without weeds) was calculated by multiplying the percentage cover of weeds by the total biomass and subtracting this estimated weed biomass from the total weight. The percentage cover of all target species and weeds was measured on 1 m2 plot in the centre of each plot, at the same time as biomass was harvested.

For the analysis of this dataset used in the Allan et al. 2025 (doi 10.1111/1365-2745.70071) see https://github.com/noemiepichon/negative_diversity_effects_opinion