← Every exhibit
The staggered DiD estimand release gate
Release the estimand only after cohort, time, comparison, aggregation, and inference are named.
| Release field | Required input | Permitted statement | Stop signal |
|---|---|---|---|
| Cohort and time | First-treatment group, calendar period, event time | “This is the effect for cohort g at time t.” | The article says only “the treatment effect.” |
| Comparison | Never-treated or not-yet-treated set, with conditions | “This comparison supplies the stated contrast.” | Already-treated units silently serve as controls. |
| Outcome and horizon | Outcome unit, measurement window, post-treatment horizon | “The estimate concerns this outcome over this horizon.” | The outcome changes between sections. |
| Aggregation | Cohorts, periods, and weights used for the summary | “This overall result answers this weighted question.” | The summary is treated as a natural ATT. |
| Inference and sensitivity | Assignment level, uncertainty method, trend and heterogeneity checks | “Uncertainty and sensitivity were reviewed at this boundary.” | A pre-trend test is treated as proof. |
Swipe or scroll horizontally if the table is wider than your screen.
Reference & Evidence
Source: Author's release framework grounded in Callaway and Sant'Anna (2021), Goodman-Bacon (2021), Roth et al. (2023), and Wooldridge (2023). The prompts are synthetic and do not contain an effect estimate.
Each line is a claim from the register this journal publishes against, resolved from the register at build time.
- A The parameter, named: "the disaggregated causal parameter that we call the group time average treatment effect i e the average treatment effect for group g at time t where a group is defined by the time period when units are first treated" Callaway & Sant'Anna. (2021) ·
CS21-C1 - A Parallel trends is the identifying assumption: comparison groups "would have followed parallel paths over time which is the so called parallel trends assumption" Callaway & Sant'Anna. (2021) ·
CS21-C2 - A The aggregation step and the inference are separate contributions, both the authors' own: "We also propose different aggregation" schemes that highlight treatment-effect heterogeneity, and they prove the validity "of a computationally convenient bootstrap procedure to conduct asymptotically valid simultaneous (instead of pointwise) inference" Callaway & Sant'Anna. (2021) ·
CS21-C3 - A With variation in treatment timing, the two-way fixed-effects DD estimator is a weighted average of 2-by-2 comparisons, in the paper's own words: "This paper shows that the two-way fixed effects estimator equals a weighted average of all possible two-group/two-period DD estimators in the data" Goodman-Bacon. (2021) ·
GB21-C1 - A The comparisons include already-treated units as controls, and the weights can go negative: "Some compare units treated at two different times, using the later-treated group as a control before its treatment begins and then the earlier-treated group as a control after" its treatment begins, so "TWFEDD estimates a variance-weighted average of treatment effect parameters sometimes with" negative weights, and "Negative weights only arise when average treatment effects vary over time" Goodman-Bacon. (2021) ·
GB21-C2 - A A conventional two-way fixed-effects coefficient therefore need not equal a simple average treatment effect under staggered adoption: a causal reading of two-"way fixed effects DD estimates requires both a parallel trends assumption and treatment" "effects that are constant over time" Goodman-Bacon. (2021) ·
GB21-C3 - A A synthesis with advice attached: "this paper synthesizes recent advances in the econometrics of difference in differences did and provides concrete recommendations for practitioners" Roth, Sant'Anna, Bilinski & Poe. (2023) ·
RSP23-C1 - A TWFE is the baseline it complicates: the ATT "can be consistently estimated using a two way fixed effects twfe regression specification" under the canonical assumptions, and recent advances are classified as relaxing those Roth, Sant'Anna, Bilinski & Poe. (2023) ·
RSP23-C2 - A The recommendation is to choose deliberately: the paper begins "by articulating a simple set of canonical assumptions under which the econometrics of did are well understood" and then relaxes them one at a time Roth, Sant'Anna, Bilinski & Poe. (2023) ·
RSP23-C3 - A The paper develops simple nonlinear difference-in-differences strategies for panel data with staggered interventions and optional covariates: "I allow for general staggered interventions, with and without covariates" Wooldridge. (2023) ·
WOO23-C1 - A Under an index version of parallel trends, cohort-by-time effects are identified: "Under an index version of parallel trends, I show that average treatment effects on the treated (ATTs) are identified for each cohort and calendar time period in which a cohort was subjected to the intervention" Wooldridge. (2023) ·
WOO23-C2 - A The estimators extend familiar linear ideas but make the response specification a choice that has to be stated: "The pooled quasi-maximum likelihood estimators in the linear exponential family extend pooled ordinary least squares" Wooldridge. (2023) ·
WOO23-C3
Grades: A, verified against the printed page of the primary source · B, primary source, text layer only · C, authoritative secondary · D, reported.
Related exhibits
-
The synthetic parallel-trends screen
From the essay What are parallel trends? The assumption behind difference-in-differences
-
Why must causal inference name the cohort-time estimand before running regressions?
From the essay Staggered difference-in-differences needs a cohort-time estimand
-
The common-method-bias release sequence
From the essay Common method bias is not a checkbox