Human disturbance and the magnitude and predictability of phenotypic divergence

Supplementary material: a step-by-step reproducible analysis

0.1 About this supplement

This computational supplement accompanies the manuscript:

Human disturbance and the magnitude and predictability of phenotypic divergence.

Authors removed for double-anonymous review.

It documents every analytical step, from data import to final figures, in a reproducible Quarto book. All code is written in R and all models are fitted with brms using the cmdstanr backend.

0.1.1 Reproduce the analysis in order

Start with the prepared data, calculate the effect sizes, build the dependency objects, check the model grid, and create the figures. Each chapter states its input, action, and saved output.

Start with data preparation Set up the software

0.2 Tutorial map

Step	Chapter	Main output
1	Prepare the data	Clean analysis data
2	Calculate lnM	Effect sizes and sampling variances
3	Check the data	Descriptive tables and plots
4	Build dependency objects	Sampling and phylogenetic matrices
5	Run the model grid	Fitted models and summaries
6	Create final figures	Publication figures
7	Record the environment	Versions, files, and render checks

TipTwo ways to follow the tutorial

Quick route: use the saved files in Rdata/ and render the book.

Full route: rebuild each saved file in chapter order. Model fitting requires more time and memory than rendering the saved results.

0.3 The PROCEED database

The analyses use the PROCEED v6.2 database of rates of evolutionary and phenotypic change. Each row represents one contrast between two populations or time points, with the raw summary statistics (means, standard deviations, sample sizes) needed to compute \(\ln M\).

0.4 Purpose of lnM

The effect size used in this book is \(\ln M\), which records the magnitude of divergence between two groups without retaining direction. Its formula and SAFE calculation are given in Section 3.1.

0.5 Location-scale meta-regression

The central statistical framework is location-scale meta-regression implemented in brms. Each model has the conceptual form:

bf(
  yi_lnM_safe ~ moderator + random_effects,
  sigma ~ moderator
)

The location submodel contains coefficients for the expected value of \(\ln M\). The scale submodel contains coefficients for the residual standard deviation, \(\sigma_{\ln M}\).

One independent model per moderator. No multi-moderator model is fitted. Each model uses the rows with complete data for its moderator.

0.6 Random effects

Every main model includes:

• (1 | ref_id) - non-independence among contrasts from the same reference.
• (1 | gr(sp_ncbi, cov = A)) - phylogenetic relatedness among species.
• (1 | gr(es_id_model, cov = V)) - known sampling variance from SAFE estimates, using the brms known-variance trick with a constant(1) prior on the corresponding SD.

sys_id (study system) is used only in sensitivity analyses. It is highly associated with ref_id and sp_ncbi.

0.7 Saved files and reproducibility

Long computations are saved as .rds files in Rdata/. The book reads these files by default, so a normal render does not refit models. To rebuild files, follow the full route in the Reproducibility chapter.

• Effect sizes: set recompute_effect_sizes <- TRUE in 02_lnm_effect_sizes.qmd.
• Models: set refit_models <- TRUE in 05_location_scale_model_grid.qmd.

Package versions, the CmdStan version, and the git commit are recorded in Section 19.1.