Method library WORKFLOW / 016

Research workflow/4 steps

Event-study DiD with Sun & Abraham (fixest)

@fixest ✓ P D · Jun 29, 2026 · 1.1k views

@misc{fixest,
  title        = {fixest},
  author       = {Laurent Bergé},
  howpublished = {\url{https://lrberge.github.io/fixest/}},
  note         = {Software / documentation}
}

In the margins · Peer notes@did_master · Nice complement to the did package — same idea (don't trust naive TWFE under staggered timing), FE flavour. 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

Assemble panel with cohort timing

Step notesField note & discussion1

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

What happens here

Panel id × period; cohort = the period each unit is first treated.

Reads fromNot specified Feeds into#2#3

Key code

data(base_stagg)

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

[fixest] Sun & Abraham event study — sunab()

Step notesField note & discussion1

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

What happens here

Fit feols with sunab(cohort, period) — interaction-weighted, robust to heterogeneous timing.

Formula

Y_{it}=\alpha_i+\lambda_t+\textstyle\sum_{e\ne-1}\beta_e\,\mathbb{1}[\,t-g_i=e\,]+\varepsilon_{it}

The estimator

Sun & Abraham event study — sunab() — Interaction-weighted event-study estimator robust to heterogeneous treatment timing, in a fast fixed-effects framework.

ATT Code @ HEAD ↗ Paper ↗

Reads from#1 Feeds into#4

Key code

m <- feols(y ~ sunab(year_treated, year) | id + year, base_stagg)

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

Naive TWFE comparison

Step notesField note & discussion1

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

What happens here

Estimate plain TWFE event-study to see the bias Sun-Abraham corrects.

Reads from#1 Feeds into#4

Key code

twfe <- feols(y ~ i(time_to_treat, ref = -1) | id + year, base_stagg)

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

iplot(): SA20 vs TWFE vs truth

Step notesField note & discussion1

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

What happens here

Overlay the event-study coefficients; Sun-Abraham tracks the true effect, TWFE drifts.

Reads from#2#3 Feeds intoNot specified

Key code

iplot(list(m, twfe))

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.

Data contract

Inputs & data

A panel with staggered treatment cohorts: unit id, period, cohort (first-treatment period), and an outcome.

Show data format & exampleHide example

Format — one row per (unit, period). cohort = first treated period.

 id  period  cohort   y
  1      1       3    2.0
  1      2       3    2.1
  2      1       5    1.8

Read the evidence

Results & interpretation

3 figures Workflow discussion 2

01FigureSun–Abraham event-study coefficients from sunab() — unbiased under heterogeneous timing.

02FigureNaive TWFE event study on the same data, showing the bias Sun–Abraham corrects.

03Figureiplot() overlay: Sun–Abraham tracks the true effect where TWFE drifts away.

Figures reproduced from fixest — Laurent Bergé — unofficial community showcase; all credit to the original authors.

Context & interpretation

From the fixest walkthrough. Build a staggered panel, fit the Sun-Abraham interaction-weighted event study, and compare it to naive TWFE. Unofficial summary.

Peer notes

Discuss the workflow (2)

2

@forest_fox · Jun 29, 2026

feols is stupid fast and sunab() makes the SA correction a one-liner. No reason to hand-roll event studies anymore.
5

@did_master · Jun 29, 2026

Nice complement to the did package — same idea (don't trust naive TWFE under staggered timing), FE flavour.

4

@splitpoint_sam · Jun 29, 2026

Exactly. iplot() overlaying SA20 vs TWFE vs truth is the clearest illustration of the bias I've seen.