Step 2: Variance component analysis
Variance component analysis was used to compare the relative importance
of each fixed effect included within the best performing models from
Step 1. We built a Bayesian hierarchical multilevel linear model for
each response variable considered in the mixed effects models,
implemented through R package ‘R2Jags’ (Su & Yajima 2021), following
Gelman & Hill (2007), Hector et al. (2011) and Catford et
al. (2014). These models were produced by taking the structure of the
best performing mixed-effects models and replicating this in JAGS (Just
Another Gibbs Sampler) multilevel structure, with variable intercepts
and slopes applied following Gelman & Hill (2007). Variance components
were then calculated from these multilevel models and presented on a
standard deviation scale to aid comparison between predictors (Gelman &
Hill 2007; Hector et al. 2011).