r-environment.md

R Environment

EvidenceForge can be used as a pure workflow skill repository, but the meta-analysis examples become more useful with a small R environment.

Minimal Requirement

For the current minimal meta-analysis script:

install.packages("metafor")

Or run:

Rscript skills/meta-analysis-forge/scripts/install_r_packages.R

Minimal Run

Rscript skills/meta-analysis-forge/scripts/run_meta_analysis.R \
  --input examples/golden/meta-analysis-minimal/input/coding-sheet.csv \
  --outdir examples/golden/meta-analysis-minimal/output \
  --metric lnROM

Expected outputs:

• meta-summary.txt
• model-results.csv
• forest-plot.pdf
• funnel-plot.pdf

Validation Before R

Run the coding-sheet validator first:

python skills/meta-analysis-forge/scripts/validate_coding_sheet.py \
  --input examples/golden/meta-analysis-minimal/input/coding-sheet.csv

On Windows, if python points to the Microsoft Store alias, use:

py -3 .\skills\meta-analysis-forge\scripts\validate_coding_sheet.py `
  --input .\examples\golden\meta-analysis-minimal\input\coding-sheet.csv

The R script assumes the effects are already comparable. It does not decide that meta-analysis is appropriate.

Effect-Size Helpers

For mechanical extraction support:

source("skills/meta-analysis-forge/scripts/effect_size_helpers.R")

Available helpers include:

• se_from_ci()
• log_ratio()
• log_ratio_se_from_ci()
• fisher_z()
• fisher_z_se()
• smd_se_approx()
• lnrom()

These functions preserve transparency; they do not decide that effects are comparable.

Future R Packages

Possible future deterministic scripts may use:

• meta
• clubSandwich
• robumeta
• dmetar
• PRISMA2020

These are not required for the current minimal example.