Statistical analyses
All statistical analyses were performed in the R statistical software package, version 3.3.2 (R Core Team, 2019). We used scatterplot to explore the bivariate relationship between four taxonomic diversity metrics (species richness, Shannon Index, Pielou Evenness and Simpson index) and AGB, and tested for their significant effects on AGB using separate simple linear models. To determine how multiple trait-functional diversity, single trait-functional diversity, functional dominance, and structural diversity influence AGB, we ran a multiple linear model incorporating all the 11 metrics. These metrics were functional richness (FRic), functional evenness (FEve), functional divergence (FDiv) and functional dispersion (FDis) for the multiple trait-functional diversity; functional divergence of wood density (FDvar WD) and of maximum plant height (FDvar Hmax) for single trait-functional diversity; community weight mean of wood density (CWM WD) and of maximum plant height (CWM Hmax) for functional dominance; and structural diversity [CV DBH: tree diameter variation; CV Ht: tree height variation; CV Nbp: variation of number of primary branches]. Due to this high number of independent variables and the likelihood of autocorrelation, a Multi-model Inference followed by a full averaging procedure was performed using the package “MuMIn” (Barton, 2018) to determine the optimal best model, which was selected based on the Akaike Information Criterion (AICc). Between two subset models that are equally supported (ΔAICc <2), the most parsimonious was the one with more independent variables. In addition to the significance fits of the predictors retained in the finally selected models, we computed for all predictors their relative importance. For easier interpretation of the results, bivariate relationships were also constructed.
Finally, we used Structural Equation Modelling (SEM) to assess the direct and indirect response of AGB to diversity metrics. To simplify the analytical framework of the SEM, we only considered the independent variables that were retained in the previous selected models. These were species richness, functional evenness, coefficient of variation of tree diameter and of the number of branches. We tested the a-priori model that AGB increased with increasing species richness, as result of positive mediation of functional evenness, and structural diversity.
To determine the individual mediation role of functional evenness, coefficient of variation of tree diameter and number of branches, we fitted three separate SEMs, and a fourth integrative SEM incorporating both direct and indirect paths between species richness and AGB via functional evenness, coefficient of variation of tree diameter and number of branches. The SEMs were fitted using “lavaan” package (Rosseel, 2012) in the R statistical software. The goodness of fit of the SEM was assessed using the Chi square statistic, comparative fit index (cfi), and root mean square error of approximation (rmsea) (Grace & Bollen, 2005; S. Mensah, du Toit, et al., 2018; S. Mensah, Veldtman, Assogbadjo, et al., 2016).