(c) Model comparisons
To compare among models, the model with the lowest Akaike’s Information
Criterion (AIC) was considered the best or most parsimonious model
(Burnham & Anderson 2002). Marginal and conditional
R2 values were calculated with the performance package
v.0.10.4 (Lüdecke et al. 2021, Nakagawa & Schielzeth 2013). Within each
model, to identify which variables most influenced patterns of genetic
diversity, we plotted marginal effects with the R package sjPlot
(Lüdecke 2021) and additionally examined the p -values of variable
coefficients. Model fits and spatial autocorrelation in the residuals
were checked with DHARMa v.0.4.3 (Hartig 2021). Moran’s I was
near zero for all models, and no significant spatial autocorrelation
(defined as p < 0.5) was found (Table S1). To assess
sensitivity to missing and rare data, all models were bootstrapped 1000x
with the boot package v.1.3.28 (Canty & Ripley 2021). All analyses were
performed in R v.4.2.3 (R Core Team 2023).
Finally, to identify whether global patterns varied across taxa, we ran
all models on a subset of 10 families (Scombridae, Lutjanidae,
Serranidae, Pomacentridae, Sebastidae, Engraulidae, Gadidae,
Syngnathidae, Rajidae, and Carcharhinidae), 1 family at a time. These 10
families were chosen as they (1) had a large amount of data
(> ~30 observations/dataset) and (2)
represented a broad range of life history traits.