latex-tables

LaTeX Tables

Construct tables as compact empirical arguments. Selection and ordering of quantities should make the relevant comparison, estimand, sample, and uncertainty available for scrutiny.

Tables and interpretive record

The work should leave:

1. a verified .tex table tied to the actual statistical output;
2. reproducible table-generation code when the project supports it;
3. a compiled or HTML-rendered preview; and
4. a short Table Interpretation Note explaining the table's evidentiary job, important comparisons, sample or estimand changes, and interpretation limits.

Constructing the table

1. Establish the table's job

Determine whether the table describes the sample, validates a measure, assesses balance, presents a primary estimand, compares theoretically meaningful heterogeneity, investigates mechanisms, or examines a specific vulnerability. Prefer one coherent job over a kitchen-sink table.

2. Trace every displayed quantity

Read the code and model output. Verify coefficients or quantities, uncertainty, outcome scale, sample, weights, fixed effects, clustering, reference categories, omitted periods, transformations, and model fit. Do not transcribe numbers from screenshots or invent absent statistics.

3. Choose rows and columns substantively

Order specifications to show a reasoned comparison. Make changes in sample, estimand, treatment, outcome, controls, fixed effects, weights, or uncertainty visible. Do not imply that columns estimate the same quantity when they do not. Do not select models by significance or hide inconvenient estimates.

4. Design for reading

Use booktabs, clear grouping, meaningful labels, aligned numbers, whitespace, and concise notes. Put units in labels or headings. Prefer confidence intervals or standard errors appropriate to the field and design; significance stars may be included when expected, but they must not carry the argument.

Avoid vertical rules, decorative complexity, tiny type, unexplained abbreviations, and excessively wide tables. Split a table when different panels serve different questions.

5. Write an informative caption and notes

State what is estimated, for whom, and over what period. Notes should disclose uncertainty, clustering, weights, fixed effects, sample restrictions, transformations, reference categories, and multiple-testing adjustments when these matter. Do not use notes to bury a changed sample or estimand.

6. Render and verify

Compile the containing document when possible. The bundled preview can provide a fast first check:

python3 <skill-dir>/scripts/table_html.py <table>.tex

Inspect alignment, headers, notes, width, symbols, escaping, and legibility. Compare rendered values with the source output and, when requested, verify that the manuscript references the correct table.

7. Interpret and revise

Write the table interpretation note. If the layout reveals incomparable models, unstable samples, opaque estimands, selective reporting, or an argument that depends only on stars, revise the table or analysis rather than polishing around the problem.

Criteria for judgment

• Every value is traceable to verified output.
• Rows and columns express a substantive comparison.
• The estimand, sample, scale, and uncertainty are visible.
• Model changes are transparent.
• The table remains readable at manuscript size.
• Captions and notes support scrutiny without overclaiming.
• Rendering is verified, not assumed from raw LaTeX.

Follow the project's existing R, Python, Stata, and LaTeX practice. Discuss software only when it affects the reported quantities or the integrity of the table.