marginally non-significantp < 0.001
In multi-model inference, the best model of NMDS1 contained both PC1 ( Figure 7 ) and PC2, and was negatively associated with both (although PC2 was marginally non-significant), explaining 94.65% of the variance in NMDS1 ( b (PC1) = 0.294, t 12 = −14.40, p < 0.001; b (PC2) = 0.0783, t 12 = −2.16, p = 0.051).