Additional file 4: Table S4.

Title of data: Generalized linear mixed models fitted to variance in host shoot dry weight. Description of data:+ estimated G-matrix not positive definite. Host = test host species, block = spatial block, strain = Bradyrhizobium genotype, sp_orig = host species from which Bradyrhizobium genotypes were isolated, ge_orig = host genus from which Bradyrhizobium genotypes were isolated. All models assumed a gamma distribution with a logit link function. Model 16 (bold face) provided the best fit (i.e., minimized AICc). Adding the block by host species interaction improved the fit (e(AICmin-AICi)/2 = 5.21 x 10–13) but the proportion of variance explained by this component (4.91% + 2.42%, Additional file 7: Table S7) was only marginally significant. The next best fitting model (model 1) included the fixed effect of host and the random effects of block and Bradyrhizobium genotype, with interactions of host with strain and with block. Residuals plots suggested that this model did not perform much worse than the best model, but nesting Bradyrhizobium genotype into host species from which it was isolated greatly improved the relative likelihood that the model fit the data (model 16 versus model 1; e(AICmin-AICi)/2 = 1.28 × 10–12)). Examination of Q-Q plots revealed that all other models provided much poorer fits to the data.

Format: CSV Size: 1KB Download file

Ehinger et al. BMC Ecology 2014 14:8   doi:10.1186/1472-6785-14-8