# Style guide — tables, figures and files (Dufresne lab, v2.3, 2026-09)

These conventions make every student's output machine-readable and comparable across replications. Follow them exactly.

## Files and folders

- All deliverables go to `out/`. Tables in `out/tables/`, figures in `out/figures/`, code in `out/code/`.
- Tables are **CSV** files (UTF-8, comma separator, header row, `.` as decimal mark, no thousands separators, no currency symbols).
- Figures are **PNG**, at least 1200 px wide, white background, with axis titles and the reference period marked.
- Code: every script actually used to produce the deliverables, numbered in execution order (`01_clean.py`, `02_estimate.py`, …).
  Any language is accepted (Python, R, Stata, Julia) as long as the scripts are in `out/code/`.

## Table 1 — descriptive statistics (`table1_descriptives.csv`)

Long format, one row per (variable, group, period). Columns, in this order:

`variable,group,period,n_obs,n_firms,mean,sd,min,max`

- `variable` ∈ {`employment`, `log_employment`, `revenue_kcad`}
- `group` ∈ {`all`, `treated`, `control`}
- `period` ∈ {`pre`, `post`, `all`} (`pre` = years before the program start, `post` = program years)
- At minimum, provide `employment` for every group × {`pre`, `post`} combination (6 rows). Add the other variables if useful.
- Statistics are computed on the **cleaned analysis sample**, with `mean` and `sd` rounded to 2 decimals.

## Table 2 — regression results (`table2_did.csv`)

One row per estimated coefficient of interest. Columns, in this order:

`model,term,coefficient,std_error,t_stat,p_value,n_obs,n_firms,n_clusters,firm_fe,year_fe,cluster_var`

- The main specification uses `model = did_twfe` and `term = treated_post`.
- `firm_fe` and `year_fe` are `yes`/`no`. `cluster_var` is the name of the clustering variable (e.g. `region`).
- Coefficients and standard errors with 4 decimals; `n_obs` is the number of firm-year observations used by the estimator.
- Event-time coefficients may be added as extra rows (`model = event_time`, `term = treated_x_<year>`).

## Figure 1 — event-time plot (`fig1_event_time.png`)

- x-axis: calendar year (2019–2024); y-axis: coefficient on treated × year (log points), reference year 2021 fixed at 0.
- Point estimates with 95 % confidence intervals (clustered standard errors), a horizontal line at 0 and a vertical line
  between 2021 and 2022 marking the program start.
- Title: "Event-time estimates: effect of the wage subsidy on log employment".

## Report (`report.docx`)

- Word document, at least 600 words of body text, sections in this order: 1. Research question; 2. Data; 3. Data cleaning
  (with the count of records affected at each step); 4. Method; 5. Results (quote the ATT, its clustered standard error and
  N exactly as in Table 2); 6. Reference check (cite each problematic entry by its BibTeX key); 7. Limitations.
- Numbers quoted in the text must match the tables to the displayed precision.
- Name the author "Research assistant (PSE-2026)" and date the document.