regression-table

name	regression-table
description	Formats estimation output as a publication-quality regression table with stars, SEs, and fit statistics. Use when creating a results table.
argument-hint	<notebook> [description]
allowed-tools	Bash, Read, Write, Edit, Glob, Grep
version	1.1.0
workflow_stage	writing
tags	["regression","tables","LaTeX","booktabs"]

Format Regression Table

Create a publication-quality regression table from estimation output in a notebook.

Arguments

• $ARGUMENTS — a notebook reference and/or description of the table (e.g., "notebook-02 OLS results" or "main regression table with 3 specifications")

Steps

1. Identify the source notebook and the estimation output:

   • If a notebook name is provided, read that notebook
   • If no notebook is specified, ask the user which notebook contains the regression results
   • Look for cells with estimation commands (Python: pyfixest, statsmodels, linearmodels; R: fixest, lm, felm; Stata: reg, reghdfe, ivregress)

2. Ask the user for table specifications:

   • Which models/columns to include
   • Dependent variable name(s)
   • Which coefficients to display (or "all", excluding constant by default)
   • Fixed effects to report as Yes/No rows
   • Clustering level for standard errors
   • Any custom notes for the table footer

3. Build the table manually as pipe-delimited Markdown (do NOT use pf.etable(type="md"), etable(markdown=TRUE), or esttab md — their output doesn't render correctly in the Quarto manuscript). Extract coefficients, SEs, and p-values from model objects and construct the table row by row:

   • Header row: Column numbers with model names: | | (1) OLS | (2) FE | ... |
   • Coefficient rows: Point estimate with significance stars on top, standard error in parentheses below
   • Significance stars: * p<0.10, ** p<0.05, *** p<0.01
   • Separator row: Empty row between coefficients and metadata
   • Fixed effects rows: Yes/No indicators
   • Summary rows: Observations, R-squared

4. Export the table in three formats to ../tables/:

   • <label>.md — Markdown (for {{< include >}} in manuscript)
   • <label>.csv — CSV (for data reuse)
   • <label>.tex — LaTeX with booktabs, following AER-style formatting conventions:
     • Use \toprule, \midrule, \bottomrule only — never \hline
     • Use \cmidrule{2-4} for partial rules spanning column groups (e.g., Panel A vs Panel B)
     • Wrap in \begin{threeparttable} with notes in \begin{tablenotes}[flushleft]\footnotesize to keep notes at table width
     • Use S columns from siunitx for decimal-aligned numbers, or d{2.3} for fixed decimal positions
     • For wide tables (many columns): reduce \tabcolsep or use \small/\footnotesize
     • Multi-panel tables: Use \multicolumn{N}{l}{\textit{Panel A: Description}} as a spanning header, then \midrule before Panel B

5. To embed in index.qmd, use {{< include >}} (NOT {{< embed >}}):

   **Table N: Caption text.**
   
   {{< include tables/<label>.md >}}
   
   ::: {.table-notes}
   *Note:* Dependent variable: ... Standard errors clustered by ... in parentheses.
   Significance levels: * p<0.10, ** p<0.05, *** p<0.01. All regressions include a constant (not reported).
   :::
   

Error handling

• If the notebook has no estimation output, report this and ask the user to run the regressions first.
• If the estimation output format is not recognized, ask the user to provide the raw coefficients and standard errors.

Common Pitfalls

• Tables exceeding page margins: Reduce font size (\small), reduce \tabcolsep, or split into Panel A/Panel B before shrinking further
• Missing SE documentation: Always state the type of standard errors (robust, clustered by X, HC1, etc.) in the table notes — reviewers will ask
• Inconsistent significance stars: Use the same convention across all tables in the paper: * p<0.10, ** p<0.05, *** p<0.01
• Missing dependent variable: Each column header should name or clearly indicate the dependent variable
• Forgetting constant suppression note: If the constant is excluded from the display, state "All regressions include a constant (not reported)" in notes
• Using \hline in LaTeX: Always use booktabs commands — \hline produces thicker, less professional lines