Difference-in-Differences with Staggered Treatment: Package Comparison

Author

DiD Methods Comparison

Published

January 14, 2026

1 Overview

This book provides a comprehensive comparison of difference-in-differences (DiD) estimators designed for staggered treatment adoption with heterogeneous treatment effects. We benchmark multiple R and Python packages using synthetic data following the simulation design from @callaway2021difference.

1.1 Background

Traditional two-way fixed effects (TWFE) estimators can produce biased estimates when:

Treatment is adopted at different times across units (staggered rollout)
Treatment effects vary over time or across cohorts (heterogeneous effects)

Several new estimators have been developed to address these issues, including methods by:

Callaway & Sant’Anna (2021): Group-time average treatment effects
de Chaisemartin & D’Haultfoeuille (2020, 2024): Robust DiD with multiple periods
Sun & Abraham (2021): Interaction-weighted estimator
Borusyak, Jaravel & Spiess (2024): Imputation-based approach

1.2 Packages Compared

1.2.1 R Packages

Package	Method	Authors	Available
`did`	Group-time ATT	Callaway & Sant’Anna	Yes
`DIDmultiplegt` / `DIDmultiplegtDYN`	Robust DiD	de Chaisemartin & D’Haultfoeuille	Yes
`fixest` (sunab)	Interaction-weighted	Sun & Abraham	Yes
`didimputation`	Imputation	Borusyak, Jaravel & Spiess	Yes

1.2.2 Python Packages

Package	Method	Authors	Available
`csdid`	Group-time ATT	Callaway & Sant’Anna (port)	Yes
`diff_diff`	Multiple methods	Community	Yes
`did_multiplegt_dyn`	Robust DiD	de Chaisemartin & D’Haultfoeuille	No native Python package
`pyfixest`	Sun & Abraham via feols	Community port	Yes (partial)

1.3 Simulation Design

We generate synthetic panel data with:

1,000,000 units observed over 10 time periods
Staggered treatment adoption: Units receive treatment at different times (2012, 2014, 2016, 2018) or never
Heterogeneous treatment effects: Effects vary by treatment cohort and time since treatment
Unit and time fixed effects

This design allows us to test whether each estimator correctly recovers the true treatment effects.

1.4 Metrics

For each package, we report:

Execution time: How long the estimation takes
Estimated ATT: Compared to the true effect
Event study estimates: Dynamic effects by time relative to treatment
Memory usage (where available)

1.5 References

Callaway, B., & Sant’Anna, P. H. (2021). Difference-in-differences with multiple time periods. Journal of Econometrics.
de Chaisemartin, C., & D’Haultfoeuille, X. (2020). Two-way fixed effects estimators with heterogeneous treatment effects. American Economic Review.
Sun, L., & Abraham, S. (2021). Estimating dynamic treatment effects in event studies with heterogeneous treatment effects. Journal of Econometrics.
Borusyak, K., Jaravel, X., & Spiess, J. (2024). Revisiting event study designs: Robust and efficient estimation. Review of Economic Studies.

--- title: "Introduction" --- # Overview This book provides a comprehensive comparison of difference-in-differences (DiD) estimators designed for **staggered treatment adoption** with **heterogeneous treatment effects**. We benchmark multiple R and Python packages using synthetic data following the simulation design from @callaway2021difference. ## Background Traditional two-way fixed effects (TWFE) estimators can produce biased estimates when: 1. Treatment is adopted at different times across units (staggered rollout) 2. Treatment effects vary over time or across cohorts (heterogeneous effects) Several new estimators have been developed to address these issues, including methods by: - **Callaway & Sant'Anna (2021)**: Group-time average treatment effects - **de Chaisemartin & D'Haultfoeuille (2020, 2024)**: Robust DiD with multiple periods - **Sun & Abraham (2021)**: Interaction-weighted estimator - **Borusyak, Jaravel & Spiess (2024)**: Imputation-based approach ## Packages Compared ### R Packages | Package | Method | Authors | Available | |---------|--------|---------|-----------| | `did` | Group-time ATT | Callaway & Sant'Anna | Yes | | `DIDmultiplegt` / `DIDmultiplegtDYN` | Robust DiD | de Chaisemartin & D'Haultfoeuille | Yes | | `fixest` (sunab) | Interaction-weighted | Sun & Abraham | Yes | | `didimputation` | Imputation | Borusyak, Jaravel & Spiess | Yes | ### Python Packages | Package | Method | Authors | Available | |---------|--------|---------|-----------| | `csdid` | Group-time ATT | Callaway & Sant'Anna (port) | Yes | | `diff_diff` | Multiple methods | Community | Yes | | `did_multiplegt_dyn` | Robust DiD | de Chaisemartin & D'Haultfoeuille | **No native Python package** | | `pyfixest` | Sun & Abraham via feols | Community port | Yes (partial) | ## Simulation Design We generate synthetic panel data with: - **1,000,000 units** observed over **10 time periods** - **Staggered treatment adoption**: Units receive treatment at different times (2012, 2014, 2016, 2018) or never - **Heterogeneous treatment effects**: Effects vary by treatment cohort and time since treatment - **Unit and time fixed effects** This design allows us to test whether each estimator correctly recovers the true treatment effects. ## Metrics For each package, we report: 1. **Execution time**: How long the estimation takes 2. **Estimated ATT**: Compared to the true effect 3. **Event study estimates**: Dynamic effects by time relative to treatment 4. **Memory usage** (where available) ## References - Callaway, B., & Sant'Anna, P. H. (2021). Difference-in-differences with multiple time periods. *Journal of Econometrics*. - de Chaisemartin, C., & D'Haultfoeuille, X. (2020). Two-way fixed effects estimators with heterogeneous treatment effects. *American Economic Review*. - Sun, L., & Abraham, S. (2021). Estimating dynamic treatment effects in event studies with heterogeneous treatment effects. *Journal of Econometrics*. - Borusyak, K., Jaravel, X., & Spiess, J. (2024). Revisiting event study designs: Robust and efficient estimation. *Review of Economic Studies*.