Prompt file imported from Zsijiang/My-personal-repository (.github/prompts/review-r.prompt.md). Copyright stays with the author.

Review R Code

Run the comprehensive R code review protocol. Produces a report without editing any files.

Steps

1. Identify scripts to review:

   • Specific .R filename → review that file only
   • LectureN → review all R scripts for that lecture
   • all → review all scripts in scripts/R/

2. For each script, evaluate:

Code Quality

• Packages loaded with library() at the top (never require())
• set.seed() present if any randomness
• No setwd() (use relative paths)
• No rm(list = ls()) (use script isolation instead)
• All functions have meaningful names (verb_noun pattern)
• No magic numbers — use named constants

Reproducibility

• Script runs from top to bottom without errors
• All input files exist at stated paths
• saveRDS() called for all computed objects
• Output directories created with dir.create(..., recursive = TRUE)
• Session info saved: writeLines(capture.output(sessionInfo()), "output/session_info.txt")

Statistical Correctness

• Appropriate standard errors (clustered, robust, or default — and justified)
• No p-hacking indicators (no specification search without pre-registration)
• Effect sizes reported alongside p-values
• Sample restrictions documented and justified
• Multicollinearity checked for regression models

R/Econometrics-Specific

• fixest used for panel models (not lfe or felm)
• modelsummary used for tables (preferred over stargazer)
• ggplot2 used for all figures
• bg = "transparent" in all ggsave() calls for Beamer compatibility
• Axis labels in sentence case with units
• Figure dimensions explicitly set in ggsave()

Performance

• No redundant data copies (use pipes, not intermediate objects when possible)
• Large joins checked for cartesian products
• complete.cases() or explicit na.action for regressions

3. Produce a detailed report saved to quality_reports/[script_name]_r_review.md:

   • Severity: Critical / High / Medium / Low
   • Location: line number
   • Issue description
   • Suggested fix

4. Present summary:

   • Total issues per script
   • Breakdown by severity
   • Top 3 most critical issues

Severity Definitions