When to use
- Produce model results (coefficients, marginal effects, event-study/LP estimates).
- Produce group summaries (means, shares, counts, rates).
- Update an analysis script that currently writes only free-form text.
Rules
- Write a machine-readable table to disk (
.csv) and a publication table to disk (.tex).
- Prefer LaTeX output in R scripts (
fixest::etable, knitr::kable, or gt).
- Keep output paths config-driven (
p_out(cfg, ...)), never hardcoded.
- Include key metadata columns when relevant: sample, spec, outcome, reference group, FE/cluster notes.
- Keep column names stable and explicit (
estimate, std_error, p_value, n_obs).
R patterns
- Prefer
fixest::etable(..., tex = TRUE) for regression tables.
- Prefer a paired coefficients CSV extracted from
summary(model)$coeftable.
- Use
knitr::kable(..., format = "latex", booktabs = TRUE) or gt::as_latex() for non-regression tables.
r
1# Regression table (LaTeX)
2tex_lines <- capture.output(fixest::etable(list(m1, m2), se.below = TRUE, digits = 3, tex = TRUE))
3writeLines(tex_lines, con = p_out(cfg, "tables", "model_main.tex"))
4
5# Coefficient CSV
6ct <- data.table::as.data.table(summary(m1)$coeftable, keep.rownames = "term")
7data.table::setnames(ct, c("Estimate", "Std. Error", "Pr(>|t|)"), c("estimate", "std_error", "p_value"))
8data.table::fwrite(ct, p_out(cfg, "tables", "model_main_coefficients.csv"))
Terminal presentation
- Present key results as a compact Markdown table in terminal responses.
- Show at minimum: term/group, estimate, standard error, p-value, and sample/spec label.
- State reference category directly above or below the table.
- Round for readability (typically 3-4 decimals) while preserving sign and significance.
