Method library WORKFLOW / 035

Research workflow/4 steps

Honest sensitivity bounds for parallel-trends violations (HonestDiD)

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

@misc{honestdid,
  title        = {HonestDiD},
  author       = {Rambachan and Roth},
  howpublished = {\url{https://github.com/asheshrambachan/HonestDiD}},
  note         = {Software / documentation}
}

In the margins · Peer notes@tianzhuqin · I report the relative-magnitudes set next to the point estimate by default now — it pre-empts the 'but parallel tr… 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

Start from event-study coefficients

Step notesField note & discussion1

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

What happens here

Take β̂ (pre- and post-treatment event-time coefficients) and its covariance Σ from whatever DiD estimator you trust.

Formula

\hat\beta\sim\mathcal N(\beta,\Sigma),\qquad \beta=(\beta_{\text{pre}},\,\beta_{\text{post}})

Reads fromNot specified Feeds intoNot specified

Key code

es <- fixest::feols(y ~ i(rel_time, ref = -1) | id + period, df)
betahat <- coef(es); sigma <- vcov(es)

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.

Diagnostic / pre-tests

Read the pre-trends, don't just test them

Step notesField note & discussion1

A pre-flight check — run this before trusting any estimate downstream.

What happens here

A failed pre-trends test is low-power and a passed one isn't a green light. Treat the post-period bias δ as something to bound, not assume away.

Formula

\beta_{\text{post}}=\underbrace{\tau}_{\text{effect}}+\underbrace{\delta_{\text{post}}}_{\text{trend bias}}

Reads fromNot specified Feeds intoNot specified

Key code

# post-period coefficient = causal effect + differential trend

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.

Robustness check

Bound the deviation: relative magnitudes / smoothness

Step notesField note & discussion1

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

What happens here

Restrict δ to a transparent class — no bigger than M̄× the largest pre-period wiggle (relative magnitudes), or boundedly nonlinear (smoothness).

Formula

\Delta^{RM}(\bar M)=\Big\{\delta:\ |\delta_t|\le \bar M\!\max_{s<0}|\delta_s|\Big\}

Reads fromNot specified Feeds intoNot specified

Key code

delta_rm <- createSensitivityResults_relativeMagnitudes(
    betahat, sigma, numPrePeriods, numPostPeriods, Mbarvec = c(0.5,1,1.5,2))

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.

Inference

Robust confidence set & breakdown value

Step notesField note & discussion1

Uncertainty quantification — standard errors, intervals, and aggregation.

What happens here

Report the union of confidence intervals over the allowed δ, and the breakdown M̄* — how much trend the conclusion can survive.

Formula

\mathcal C_{1-\alpha}=\bigcup_{\delta\in\Delta(\bar M)}\mathrm{CI}(\tau;\delta),\qquad \bar M^{*}=\sup\{\bar M: 0\notin\mathcal C_{1-\alpha}\}

Reads fromNot specified Feeds intoNot specified

Key code

robust <- createSensitivityResults_relativeMagnitudes(...)
createSensitivityPlot_relativeMagnitudes(robust, originalResults)

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

An event-study: the vector of pre- and post-treatment coefficients and their covariance matrix (from any DiD / event-study estimator).

Show data format & exampleHide example

Format — the event-study coefficient vector betahat and its covariance sigma (e.g. from fixest/did).

library(HonestDiD)
# betahat: pre + post event-time coefficients
# sigma:   their covariance matrix; numPrePeriods / numPostPeriods

Read the evidence

Results & interpretation

2 figures Workflow discussion 2

01FigureEvent-study estimates (Medicaid expansion on insurance coverage): flat pre-trends, then a post-2013 jump — the input HonestDiD stress-tests.

02FigureRobust confidence sets as the allowed post-period trend deviation M̄ grows: the effect stays significant until M̄ ≈ 2 — its breakdown value.

Figures reproduced from HonestDiD — Rambachan & Roth — unofficial community showcase; all credit to the original authors.

Context & interpretation

⚠️ Unofficial community showcase of honestdid. Not affiliated with the authors; all credit to them.

Stop betting everything on a pre-trends test. Allow the post-treatment trend to deviate within a transparent class, and report the confidence set — and the breakdown value where the effect would vanish.

Peer notes

Discuss the workflow (2)

2

@did_master · Jun 29, 2026

Replacing the pre-trends test with explicit bounds is the right move. The breakdown M̄* is exactly what referees keep asking me for.
0

@tianzhuqin · Jun 29, 2026

I report the relative-magnitudes set next to the point estimate by default now — it pre-empts the 'but parallel trends' question.

1

@avg_treatment · Jun 29, 2026

Same. Pairing it with the event-study plot makes the whole argument legible in one figure.