a, Similarity in the growth rates of individual common species as predicted by the hierarchical model at three different resin phosphate concentrations. Each point represents the growth rate of a single species as estimated from the model, assuming intermediate moisture and a tree of 100 mm dbh. The graphs show the predicted species responses at intermediate resin phosphate (x axis, predicted growth − midP; as shown in Fig. 2a) against the predicted responses at low resin phosphate (predicted growth – lowP; left) and high resin phosphate (predicted growth – highP; right) concentrations. Only species that are common in the dataset (growth data available for >20 individuals) are plotted. The relative estimated responses are virtually identical across the entire phosphorus gradient. b, Observed growth rates as a function of species phosphorus affinities, with growth rates of individual trees ≥ 100 mm dbh shown in black and species-level median growth for the 362 species with estimated phosphorus affinities in blue. The y axis (growth) is log-transformed. The blue line shows a standard linear regression between log-transformed growth and phosphorus affinity (effect size), using median species growth (n = 362), weighted by species abundance. The slope (−0.13) is significantly different from zero (P = 0.00014), demonstrating that growth rates were greater for species with low-phosphorus affinity. c, Standing above-ground biomass (AGB) (top) and annual relative AGB growth (that is, standardized by the total AGB) (bottom) as a function of resin phosphate concentration. Data are from 32 plots across the phosphorus gradient in Panama. The resin phosphate scale is logarithmic. The linear regression relating standing AGB to log(resin phosphate), dry-season intensity and successional state revealed a slight negative but non-significant effect of phosphorus on biomass (slope = −7.9, P = 0.37). The same regression for relative AGB growth was likewise negative but not significant (slope = −0.002, P = 0.09). Biomass was significantly and negatively related to dry-season intensity (that is, more biomass at wetter sites) (P = 0.003), but relative biomass growth was not correlated with dry season intensity (P = 0.07).