Method library WORKFLOW / 020

Research workflow/4 steps

Two-stage difference-in-differences (did2s)

@did2s ✓ P D · Jun 29, 2026 · 992 views

@misc{did2s,
  title        = {did2s},
  author       = {Butts and Gardner},
  howpublished = {\url{https://kylebutts.github.io/did2s/}},
  note         = {Software / documentation}
}

In the margins · Peer notes@forest_fox · did2s + fixest under the hood = blazing. My default for big staggered panels now. 2

The method workspace

Follow the analysis.

Select a node to explore the method

Method mapInput → 4 steps → output

START WITHInput data

ARRIVE ATResults

Read step 01

Inside the methodReasoning, equations & implementation

Data prep

Staggered panel + relative event time

Step notesField note & discussion1

Data preparation — shapes the raw inputs into what the estimator expects.

What happens here

Unit × period; treatment indicator and event time relative to first treatment.

Reads fromNot specified Feeds into#2

Key code

# id · year · treat · rel_year · y

Discussion on this step (0)

StatsOtterDiscussion starter · Editorial note

Before using this step, make its assumptions and its output explicit. A useful peer note explains what you would check, what could go wrong, or what would change your interpretation.

Questions to explore

Which assumption matters most here, and what evidence could challenge it?
What would you check before passing this output to the next step?

Peer discussion

Bring your experience, a question, or a different perspective.

Estimation

[did2s] Two-stage difference-in-differences

Step notesField note & discussion1

The core estimate — where the causal quantity itself is computed.

What happens here

Two-stage estimation: FE on untreated, then residuals on treatment.

Formula

\tilde Y_{it}=Y_{it}-\hat\alpha_i-\hat\lambda_t,\quad \tilde Y_{it}=\textstyle\sum_e\beta_e D^e_{it}+\varepsilon_{it}

The estimator

Two-stage difference-in-differences — Gardner's two-stage estimator: remove unit/time fixed effects from untreated obs, then regress residuals on treatment — robust to heterogeneous timing.

ATT Code @ HEAD ↗ Paper ↗

Reads from#1 Feeds into#3#4

Key code

did2s(df, yname="y", first_stage=~0|id+year, second_stage=~i(rel_year), treatment="treat", cluster_var="id")

Discussion on this step (0)

StatsOtterDiscussion starter · Editorial note

Before using this step, make its assumptions and its output explicit. A useful peer note explains what you would check, what could go wrong, or what would change your interpretation.

Questions to explore

Which assumption matters most here, and what evidence could challenge it?
What would you check before passing this output to the next step?

Peer discussion

Bring your experience, a question, or a different perspective.

Robustness check

Compare to TWFE / CS

Step notesField note & discussion1

A robustness check — does the headline result survive a different lens?

What happens here

Overlay against naive TWFE and Callaway-Sant'Anna to confirm agreement.

Reads from#2 Feeds into#4

Key code

# plot did2s vs TWFE vs CS event-study

Reference / docs ↗

Discussion on this step (0)

StatsOtterDiscussion starter · Editorial note

Before using this step, make its assumptions and its output explicit. A useful peer note explains what you would check, what could go wrong, or what would change your interpretation.

Questions to explore

Which assumption matters most here, and what evidence could challenge it?
What would you check before passing this output to the next step?

Peer discussion

Bring your experience, a question, or a different perspective.

Reporting

Event-study plot

Step notesField note & discussion1

Reporting — turn the numbers into a figure or table a reader can act on.

What happens here

Plot the dynamic effects with confidence bands.

Reads from#2#3 Feeds intoNot specified

Key code

iplot(est)

Discussion on this step (0)

StatsOtterDiscussion starter · Editorial note

Before using this step, make its assumptions and its output explicit. A useful peer note explains what you would check, what could go wrong, or what would change your interpretation.

Questions to explore

Which assumption matters most here, and what evidence could challenge it?
What would you check before passing this output to the next step?

Peer discussion

Bring your experience, a question, or a different perspective.

Data contract

Inputs & data

A staggered panel: unit id, period, treatment indicator, relative event time.

Show data format & exampleHide example

Format — one row per (unit, period).

 id  year  treat  rel_year   y
  1  2003      0       -2    4.0
  1  2005      1        0    5.2
  2  2003      0       NA    3.9

Read the evidence

Results & interpretation

3 figures Workflow discussion 2

01Figuredid2s event-study estimates of the dynamic ATT.

02FigureTwo-stage estimates under heterogeneous treatment effects.

03Figuredid2s vs TWFE vs Callaway–Sant'Anna — the estimators side by side.

Figures reproduced from did2s — Butts & Gardner — unofficial community showcase; all credit to the original authors.

Context & interpretation

The did2s vignette. A lightweight alternative to Callaway-Sant'Anna for staggered designs. Unofficial summary.

Peer notes

Discuss the workflow (2)

2

@did_master · Jun 29, 2026

Gardner's 2-stage trick is elegant: residualize on the untreated, then the second stage is just OLS. Fast on huge panels.

2

@splitpoint_sam · Jun 29, 2026

And it lines up with Callaway-Sant'Anna in practice — nice to show both agree.
3

@forest_fox · Jun 29, 2026

did2s + fixest under the hood = blazing. My default for big staggered panels now.