Causal Inference Methods in Economics

Author: Gavin Qu

Applied causal inference techniques from an Economics Masters course (EC963), implemented in Python. Each lab replicates a classic empirical study using modern econometric methods.

Labs

Lab	Paper	Methods
Lab 1: Randomized Experiments	Chattopadhyay & Duflo (2004); Gerber, Green & Larimer (2008)	Difference in means, t-tests, OLS, balance checks, subgroup analysis
Lab 2: Matching	Eggers & Hainmueller (2009); Dehejia & Wahba (1999)	Exact matching, nearest-neighbor matching, Mahalanobis distance, ATT estimation
Lab 3: Difference-in-Differences	Dinas et al. (2019); Card & Krueger (1994)	Manual DiD, interaction regression, parallel trends, two-way fixed effects
Lab 4: Instrumental Variables	Sesame Street experiment; Acemoglu, Johnson & Robinson (2001)	ITT, LATE, Wald estimator, first-stage F-test, 2SLS
Lab 5: Regression Discontinuity	Meyersson (2014)	Sharp RDD, bandwidth selection, local linear regression, LOESS
Midterm: COVID-19 Curfew Evaluation	France's 2021 reinforced curfew policy	DiD, cross-sectional comparison, before-after analysis, RDD discussion

Each lab is available in two formats:

• Jupyter notebooks in notebooks/ — interactive .ipynb files, good for step-by-step exploration
• Quarto documents in quarto/ — plain-text .qmd files that render to a navigable HTML website with table of contents, cross-references, and cleaner diffs in version control

Getting Started

Prerequisites

• Python 3.11+
• uv — fast Python package manager (replaces pip/venv)
• Quarto — only needed if you want to render the .qmd files

Install dependencies

uv sync

This creates a .venv/ virtual environment and installs all required packages (numpy, pandas, statsmodels, linearmodels, etc.). No need to manually create a venv or run pip install.

Run the Jupyter notebooks

uv run jupyter notebook

Then open any notebook in notebooks/ from the browser.

Render the Quarto site

cd quarto
uv run quarto preview   # live preview with hot-reload
uv run quarto render     # build static HTML to quarto/_output/

The rendered site includes a navbar, table of contents, and all code output and figures.

Project Structure

├── pyproject.toml       # Python dependencies (managed by uv)
├── data/                # Datasets (.csv, .dta) organized by lab
├── notebooks/           # Jupyter notebooks (.ipynb)
└── quarto/              # Quarto documents (.qmd)
    ├── _quarto.yml      # Site configuration (theme, navigation)
    ├── index.qmd        # Landing page
    └── *.qmd            # One file per lab

Methods Overview

• Randomized Experiments: The gold standard for causal inference. Random assignment ensures treatment and control groups are comparable in expectation.
• Matching: Constructs a comparison group by pairing treated units with similar untreated units based on observed covariates, under the Conditional Independence Assumption.
• Difference-in-Differences: Compares changes over time between treated and control groups, identifying causal effects under the parallel trends assumption.
• Instrumental Variables: Addresses endogeneity by using an exogenous source of variation (instrument) that affects the outcome only through the treatment.
• Regression Discontinuity: Exploits a known threshold in a continuous assignment variable, comparing units just above and below the cutoff.

License

MIT