diff-diff: Difference-in-Differences in Python#

diff-diff is a Python library for Difference-in-Differences (DiD) causal inference analysis. It provides sklearn-like estimators with statsmodels-style output for econometric analysis.

from diff_diff import DifferenceInDifferences

# Fit a basic DiD model
did = DifferenceInDifferences()
results = did.fit(data, outcome='y', treatment='treated', time='post')
print(results.summary())

Key Features#

20+ Estimators: Basic DiD, TWFE, Event Study, Synthetic DiD/Control, modern staggered estimators (Callaway-Sant’Anna, Sun-Abraham, Imputation, Two-Stage, Stacked, LP-DiD), reversible and heterogeneous-adoption designs (dCDH, HAD), distributional methods (Changes-in-Changes), Regression Discontinuity, and Bacon Decomposition diagnostics
Modern Inference: Robust standard errors, cluster-robust SEs, wild cluster bootstrap, and multiplier bootstrap
Assumption Testing: Parallel trends tests, placebo tests, Bacon decomposition, and comprehensive diagnostics
Sensitivity Analysis: Honest DiD (Rambachan & Roth 2023) for robust inference under parallel trends violations
Built-in Datasets: Real-world datasets from published studies (Card & Krueger, Castle Doctrine, and more)
High Performance: Optional Rust backend for compute-intensive estimators like Synthetic DiD and TROP
Publication-Ready Output: Summary tables, event study plots, and sensitivity analysis figures

Installation#

pip install diff-diff

For development:

pip install diff-diff[dev]

Quick Links#

Getting Started: Measuring Campaign Impact - Measuring campaign impact? Start here
Which Analysis Method Fits Your Problem? - Which method fits your business problem?
Quickstart - Installation and your first DiD analysis
Choosing an Estimator - Which estimator should I use?
aggregate_survey() - Have BRFSS/ACS/CPS microdata? Bridge it to a geographic panel for DiD
Basic Difference-in-Differences with diff-diff - Hands-on basic tutorial
Troubleshooting - Common issues and solutions
R Comparison - Coming from R?
API Reference - Full API reference

Explore the Documentation#

Getting Started

Install, run your first DiD analysis, and pick the right estimator for your design.

Getting Started

Practitioner Guide

Measuring campaign impact? A business-first path through DiD, no econometrics background required.

Practitioner Guide

Tutorials

28 hands-on notebooks, from basic 2x2 DiD to survey-weighted and spillover-aware designs.

Tutorials

User Guide

References, R and Python comparisons, benchmarks, and the methodology registry.

User Guide

API Reference

Complete reference for all estimators, results classes, diagnostics, and utilities.

API Reference

What is Difference-in-Differences?#

Difference-in-Differences (DiD) is a quasi-experimental research design that estimates causal treatment effects by comparing outcome changes over time between treated and control groups. It is one of the most widely used methods in applied economics, public policy evaluation, and social science research.

Why diff-diff?#

Complete method coverage: 20+ estimators from basic 2x2 DiD to cutting-edge methods like Efficient DiD (Chen et al. 2025), TROP (Athey et al. 2025), and HAD (de Chaisemartin et al. 2026)
Familiar API: sklearn-like fit() interface — if you know scikit-learn, you know diff-diff
Modern staggered methods: Callaway-Sant’Anna, Sun-Abraham, Imputation DiD, Two-Stage DiD, and Stacked DiD handle heterogeneous treatment timing correctly
Robust inference: Heteroskedasticity-robust, cluster-robust, wild cluster bootstrap, and multiplier bootstrap
Sensitivity analysis: Honest DiD (Rambachan & Roth 2023) for robust inference under parallel trends violations
Validated against R: Benchmarked against did, synthdid, and fixest — see Benchmarks
No heavy dependencies: Only numpy, pandas, and scipy

Supported Estimators#

Estimator	Description
`DifferenceInDifferences`	Basic 2x2 DiD with robust/clustered standard errors
`TwoWayFixedEffects`	Panel data with unit and time fixed effects
`MultiPeriodDiD`	Event study with period-specific treatment effects
`CallawaySantAnna`	Callaway & Sant’Anna (2021) group-time ATT for staggered adoption
`ChaisemartinDHaultfoeuille`	de Chaisemartin & D’Haultfoeuille (2020/2022) for reversible (non-absorbing) treatments
`SunAbraham`	Sun & Abraham (2021) interaction-weighted estimator
`ImputationDiD`	Borusyak, Jaravel & Spiess (2024) imputation estimator
`TwoStageDiD`	Gardner (2022) two-stage residualized estimator
`SpilloverDiD`	Butts (2021) ring-indicator spillover-aware DiD
`SyntheticDiD`	Synthetic DiD combining DiD and synthetic control
`SyntheticControl`	Abadie, Diamond & Hainmueller (2010) classic synthetic control
`StackedDiD`	Wing, Freedman & Hollingsworth (2024) stacked DiD
`EfficientDiD`	Chen, Sant’Anna & Xie (2025) efficient DiD
`TripleDifference`	Triple difference (DDD) estimator
`StaggeredTripleDifference`	Ortiz-Villavicencio & Sant’Anna (2025) staggered DDD with group-time ATT
`ContinuousDiD`	Callaway, Goodman-Bacon & Sant’Anna (2024) continuous-treatment dose-response DiD
`HeterogeneousAdoptionDiD`	de Chaisemartin, Ciccia, D’Haultfoeuille & Knau (2026) for designs with no untreated units
`LPDiD`	Dube, Girardi, Jorda & Taylor (2025) local-projections DiD
`WooldridgeDiD`	Wooldridge (2023, 2025) extended TWFE (ETWFE) via saturated OLS or QMLE
`ChangesInChanges`	Athey & Imbens (2006) distributional DiD with quantile treatment effects
`QDiD`	Quantile DiD comparison estimator applying DiD quantile-by-quantile
`RegressionDiscontinuity`	Calonico, Cattaneo & Titiunik (2014) sharp/fuzzy RD with robust bias-corrected inference
`TROP`	Triply Robust Panel with factor model adjustment (Athey et al. 2025)
`BaconDecomposition`	Goodman-Bacon decomposition diagnostics