Data transformation and analysis
For each species at each site, the mean value for each trait was used. Where traits were
reported separately for sun leaves and shade leaves, sun-leaf data were used. Similarly, if
data were presented separately for recently matured and old leaves, that of the recently
matured leaves were chosen. Leaf traits were approximately log-normally distributed
across the data set, as were rainfall and VPD. Accordingly, these variables were
log10-transformed before analysis. MAT, PETand solar radiation were left untransformed
because their distribution across sites was approximately normal.
Standardized major axis slopes24 with 95% confidence intervals were fitted to bivariate
trait relationships because our aim was to describe the best-fit lines or central axes of these
‘scaling’ relationships. Ordinary multiple regression was used for analyses exploring the
additional predictive power of climate variables on leaf trait relationships. Regressions
were first run including interaction terms between climate variables and leaf traits; where
the interaction was non-significant (P , 0.05), we re-ran models with main effects only.
In several cases, principal components analyses run on data subsets defined by growth
form or plant functional type had to be re-run with one ormore leaf traits removed, owing
to insufficient data. These cases are indicated in Table 2. Variance component analyses
were based on the decomposition of analysis of variance (ANOVA) type I sums of squares.
Principal components analyses, variance components and regression analyses were run in
SPSS for Windows version 11.01.
Data transformation and analysisFor each species at each site, the mean value for each trait was used. Where traits werereported separately for sun leaves and shade leaves, sun-leaf data were used. Similarly, ifdata were presented separately for recently matured and old leaves, that of the recentlymatured leaves were chosen. Leaf traits were approximately log-normally distributedacross the data set, as were rainfall and VPD. Accordingly, these variables werelog10-transformed before analysis. MAT, PETand solar radiation were left untransformedbecause their distribution across sites was approximately normal.Standardized major axis slopes24 with 95% confidence intervals were fitted to bivariatetrait relationships because our aim was to describe the best-fit lines or central axes of these‘scaling’ relationships. Ordinary multiple regression was used for analyses exploring theadditional predictive power of climate variables on leaf trait relationships. Regressionswere first run including interaction terms between climate variables and leaf traits; wherethe interaction was non-significant (P , 0.05), we re-ran models with main effects only.In several cases, principal components analyses run on data subsets defined by growthform or plant functional type had to be re-run with one ormore leaf traits removed, owingto insufficient data. These cases are indicated in Table 2. Variance component analyseswere based on the decomposition of analysis of variance (ANOVA) type I sums of squares.
Principal components analyses, variance components and regression analyses were run in
SPSS for Windows version 11.01.
การแปล กรุณารอสักครู่..
