Abstract

Summary 1. Although there is ample support for positive species richness–productivity relationships in planted grassland experiments, a recent 48‐site study found no diversity–productivity relationship (DPR) in herbaceous communities. Thus, debate persists about diversity effects in natural versus planted systems. Additionally, current knowledge is weak regarding the influence of evenness on the DPRs, how DPRs are affected by the variation in life‐history traits among constituent species in polycultures and how DPRs differ among biomes. The impacts of these factors on DPRs in forest ecosystems are even more poorly understood. 2. We performed a meta‐analysis of 54 studies to reconcile DPRs in forest ecosystems. We quantified the net diversity effect as log effect size [ln(ES)], the log ratio of the productivity in polycultures to the average of those in monocultures within the same type of mixture, site condition and stand age of each study. The first use of a boosted regression tree model in meta‐analysis, a useful method to partition the effects of multiple predictors rather than relying on vote‐counting of individual studies, unveiled the relative influences of individual predictors. 3. Global average ln(ES) was 0.2128, indicating 23.7% higher productivity in polycultures than monocultures. The final model explained 21% of the variation in ln(ES). The predictors that substantially accounted for the explained variation included evenness (34%), heterogeneity of shade tolerance (29%), richness (13%) and stand age (15%). In contrast, heterogeneity of nitrogen fixation and growth habits, biome and stand origin (naturally established versus planted) contributed negligibly (each ≤ 4%). Log effect size strongly increased with evenness from 0.6 to 1 and with richness from 2 to 6. Furthermore, it was higher with heterogeneity of shade tolerance and generally increased with stand age. 4. Synthesis. Our analysis is, to our knowledge, the first to demonstrate the critical role of species evenness, richness and the importance of contrasting traits in defining net diversity effects in forest polycultures. While testing the specific mechanisms is beyond the scope of our analysis, our results should motivate future studies to link richness, evenness, contrasting traits and life‐history stage to the mechanisms that are expected to produce positive net biodiversity effects such as niche differentiation, facilitation and reduced Janzen–Connell effects.