code

Code Template Library (Stata + Python)

Adapted from the verified Economic-Research-Journal-Skills/resources/code library (Stata 18 MP, 2026-06); translated to English and made venue-neutral.

A reproducible code skeleton for empirical causal-inference research, venue-neutral and ready to copy-and-adapt. It serves both as "ready-to-run" templates for each identification method and as a concrete layout for a one-click reproduction workflow.

Verified-on-Stata-18 (per the source library): the core command chains were run end-to-end on synthetic data on Stata 18 MP and ran successfully (2026-06): reghdfe (TWFE), bacondecomp, csdid, did_imputation, the event-study + parallel-trends joint test, esttab three-line-table export, ivreg2 + KP rk F, weakivtest effective F, rdrobust, and rddensity. All command syntax has been checked. For version-sensitive points (see below), defer to the latest official documentation.

Stata do-file rule (verified): in do-file mode a loop body must NOT share a line with { — a newline is required after the opening brace. This library follows that rule throughout (see the forvalues / foreach blocks).

How to use

1. Organize the project as below, with the scripts in this directory under code/:

project/
  ├── data/raw/         raw data (read-only)
  ├── data/clean/       cleaned data
  ├── code/             <- scripts in this directory
  ├── output/tables/
  └── output/figures/

2. Open stata/00_master.do, change global root to your project root, and uncomment the dependency-install block on the first run.
3. Run do code/00_master.do for a one-click pass: clean -> describe -> identify -> mechanism -> robustness -> export tables/figures.

Throughout the scripts, illustrative names are kept generic: Y (outcome), D / treat (treatment), id (unit), year, gvar (first treated period), and a controls list (size lev roa age group).

Stata scripts at a glance

File	Purpose
00_master.do	One-click master: paths, dependencies, fixed random seed
01_clean.do	Raw -> analysis sample: merge, recorded filtering, variable construction, winsorize
02_descriptive.do	Descriptive-statistics table; treatment/control balance
03_did_modern.do	TWFE baseline -> Bacon decomposition -> CS/BJS/SA/dCDH -> event study
04_iv.do	IV: KP rk F + effective F (weakivtest) + AR-robust inference (weakiv)
05_rdd.do	RDD: rdrobust bias-corrected robust CI + rddensity manipulation test
06_dml.do	Double machine learning: ddml + pystacked multi-learner
07_mechanism.do	Mechanism: estimate D->M only; argue M->Y by theory
08_robustness.do	Robustness system organized by identification threat + placebo + wild bootstrap
09_tables.do	Three-line tables (esttab) and event-study figure export

Python scripts at a glance

File	Purpose
python/clean_panel.py	Large-scale micro-panel cleaning (pandas), outputs analysis.dta
python/dml_doubleml.py	DoubleML PLR, multi-learner robustness comparison
python/event_study_plot.py	Publication-grade event-study / parallel-trends plot (matplotlib, >= 300 dpi)

Modern estimators and command quick reference

Method	Paper	Stata	Python / R
Staggered DID	Callaway & Sant'Anna (2021)	csdid	R did::att_gt
Staggered DID	Sun & Abraham (2021)	eventstudyinteract	R fixest::sunab
Staggered DID	Borusyak, Jaravel & Spiess (2024)	did_imputation	R didimputation
Staggered DID	Gardner (2022) two-stage	did2s	R did2s
Staggered DID	de Chaisemartin & D'Haultfoeuille	did_multiplegt_dyn	R DIDmultiplegtDYN
Bias diagnosis	Goodman-Bacon (2021)	bacondecomp	—
IV effective F	Montiel Olea & Pflueger (2013)	weakivtest	—
Weak-IV-robust	Anderson-Rubin / CLR	weakiv	—
RDD	Calonico-Cattaneo-Titiunik (2014)	rdrobust rddensity	R/Python same-name packages
DML	Chernozhukov et al. (2018)	ddml pdslasso	Python DoubleML
Few-cluster inference	Roodman et al. (2019)	boottest	—
Selection-bias bound	Oster (2019)	psacalc	—

Install uniformly with ssc install <pkg>, replace.

Version-sensitive points (defer to latest official docs)

• de Chaisemartin & D'Haultfoeuille: did_multiplegt_dyn supersedes the older did_multiplegt. Use the _dyn command and confirm option names against current docs.
• Weak-IV inference: Stata 19 ships built-in weak-IV-robust inference; on Stata 18 and earlier this library uses the community-contributed weakivtest / weakiv. Prefer the built-in path where available.
• Manipulation test: rddensity (Cattaneo-Jansson-Ma) supersedes the older McCrary (2008) density test.
• DID estimator choice: report at least one heterogeneity-robust estimator (CS / SA / BJS / dCDH) as the main result rather than TWFE alone; package option names evolve, so verify against the latest documentation.