Pith. sign in

REVIEW 1 major objections 2 minor 11 references

Stacked Triple Differences

T0 review · 1 major / 2 minor · reviewed 2026-05-21 · grok-4.3

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read A regression with fully saturated fixed effects on stacked triple-difference data identifies a cell-size-weighted average of conditional average treatment effects.

desk verdict Stacked DDD adapts the stacked DiD framework to triple differences with a claimed identification result for a cell-size weighted average under saturated fixed effects, but the cross-stack fixed effects specification needs explicit verification to confirm the result holds. read the letter →

arxiv 2604.22982 v2 pith:26RXUZKT submitted 2026-04-24 econ.EM

Meng Hsuan Hsieh This is my paper
classification econ.EM
keywords stackedtripledifferencesstaggeredadoptionfixedeffectscausalinferencedifference-in-differenceseconometricmethods
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

The paper introduces stacked DDD to handle staggered adoption in triple differences designs. It builds self-contained stacks, each with four cells over an event window: treated and clean comparison cohorts, each split into treatment-eligible and ineligible units. Appending these stacks produces a unified dataset. A linear regression with fully saturated fixed effects applied to the stacked data identifies a strictly positive, cell-size-weighted average of the stack-level conditional average treatment effects. The method gives direct control over comparison groups and weights while maintaining regression transparency and complements GMM or imputation approaches.

What carries the argument

Self-contained stacks consisting of four cells over an event window—treated and clean comparison cohorts, each with treatment-eligible and treatment-ineligible units—appended into a unified dataset for estimation.

What would settle it

If the coefficient from the saturated regression on the stacked data does not match the cell-size-weighted average of the stack-level effects computed separately, the identification claim would be falsified.

Watch

Extended reading notes

Core claim

At each post-treatment event-time, a linear regression with fully saturated fixed-effects applied to the stacked dataset identifies a strictly positive, cell-size-weighted average of stack-level conditional average treatment effects, with stack weights proportional to stack-level cell sizes. Building on this characterization, alternative weighting schemes recover causal estimands with clear interpretations.

Load-bearing premise

Within each stack the clean comparison cohort satisfies the parallel changes-in-trends condition relative to the treated cohort, and appending stacks does not introduce cross-stack contamination in the fixed effects.

Editorial extensions

If this is right

  • Alternative weighting schemes can recover causal estimands with clear interpretations.
  • The approach relies on pairwise rather than global parallel changes-in-trends conditions.
  • Researchers obtain direct control over the comparison group for each treated unit and over the aggregation weights.
  • Stacked DDD trades some efficiency for regression-based transparency relative to GMM and imputation frameworks.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The stacking logic may extend to other multi-way difference designs with eligibility rules.
  • Monte Carlo studies could test how the method performs when stack-level parallel trends hold only approximately.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

1 major / 2 minor

Summary. The paper introduces stacked triple differences (stacked DDD) for staggered adoption settings. Self-contained stacks are formed, each consisting of four cells (treated and clean comparison cohorts, each split into treatment-eligible and ineligible units) over an event window. Appending stacks into one dataset, the central claim is that OLS with fully saturated fixed effects identifies a strictly positive, cell-size-weighted average of stack-level conditional average treatment effects (CATEs), with weights proportional to stack cell sizes. Alternative weighting schemes are outlined, and two empirical illustrations are presented where results differ from standard procedures.

Significance. If the identification result is correct, the approach supplies a regression-transparent estimator that grants explicit control over comparison cohorts and aggregation weights while relying only on pairwise parallel changes-in-trends within each stack. It complements GMM and imputation frameworks by emphasizing interpretability and direct weight selection over efficiency. The empirical examples illustrate that the choice of estimator can materially affect quantitative conclusions.

major comments (1)
  1. [Identification proof] Identification section (around the proof of the weighted-average result): the claim that fully saturated fixed effects recover the cell-size-weighted average of stack-level CATEs assumes stacks remain self-contained after appending. If any unit or time period appears in more than one stack, the global unit, time, or eligibility fixed effects are estimated from observations belonging to different stacks. This creates potential cross-stack leakage that mixes the within-stack contrasts, so the coefficient on the post-treatment interaction is no longer guaranteed to equal the claimed weighted average. The manuscript should clarify whether fixed effects are interacted with stack indicators or provide an explicit argument that the result survives without such interactions.
minor comments (2)
  1. [Abstract] The abstract states that two empirical illustrations are provided but gives no indication of the policy settings or the direction of the differences relative to existing estimators; a brief clause summarizing the applications would improve reader orientation.
  2. [Notation and setup] Notation for the event-time-specific estimands and the cell-size weights could be introduced with an explicit equation early in the identification section to make the subsequent weighting discussion easier to follow.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for the careful reading and constructive feedback on our manuscript introducing stacked triple differences. We address the major comment below.

read point-by-point responses

  1. Referee: [Identification proof] Identification section (around the proof of the weighted-average result): the claim that fully saturated fixed effects recover the cell-size-weighted average of stack-level CATEs assumes stacks remain self-contained after appending. If any unit or time period appears in more than one stack, the global unit, time, or eligibility fixed effects are estimated from observations belonging to different stacks. This creates potential cross-stack leakage that mixes the within-stack contrasts, so the coefficient on the post-treatment interaction is no longer guaranteed to equal the claimed weighted average. The manuscript should clarify whether fixed effects are interacted with stack indicators or provide an explicit argument that the result survives without such interactions.

    Authors: We appreciate the referee highlighting this important implementation detail. The manuscript defines stacks as self-contained to ensure that identification relies on pairwise within-stack changes in trends. To eliminate any possibility of cross-stack leakage when units or periods are shared across stacks, we will revise the identification section to specify that the saturated fixed effects (for units, time periods, and eligibility) are interacted with stack indicators. This guarantees that all contrasts remain strictly within each stack, so the OLS coefficient on the post-treatment interaction continues to recover the cell-size-weighted average of the stack-level CATEs as claimed. We will also add a brief discussion of this choice and its implications for the estimator. revision: yes

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity in the identification derivation

full rationale

The paper derives the identification result from the explicit construction of self-contained stacks (each with treated/clean cohorts and eligible/ineligible units) followed by appending and applying a linear regression with fully saturated fixed effects. This produces the cell-size-weighted average of stack-level CATEs as a consequence of the within-stack contrasts and the fixed-effects absorption, rather than by defining the target quantity in terms of the fitted coefficients themselves or by renaming a fitted input. No self-citation chain, ansatz smuggling, or uniqueness theorem imported from prior author work is invoked to force the central claim. The derivation is therefore self-contained against the stated assumptions about pairwise parallel trends within stacks.

Assumptions & free parameters 0 free parameters · 2 assumptions · 0 invented entities

The central claim rests on standard econometric assumptions for difference-in-differences and triple differences designs plus the novel stacking construction. No free parameters are introduced in the abstract. No new entities are postulated.

assumptions (2)
  • domain assumption Within each stack, the clean comparison cohort satisfies parallel changes-in-trends with the treated cohort for the eligible and ineligible subgroups.
    This is the key identifying assumption invoked to interpret the regression coefficients as causal effects.
  • domain assumption Appending independent stacks does not induce bias through the shared fixed effects structure.
    Invoked when moving from stack-level to pooled estimation.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Stacked Triple Differences." pith.science (2026). https://pith.science/paper/26RXUZKT

@misc{pith2026260422982,
  author       = {Pith},
  title        = {Pith review of: Stacked Triple Differences},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/26RXUZKT}},
  note         = {Machine review of arXiv:2604.22982}
}
read the original abstract

Triple differences (DDD) is a workhorse quasi-experimental design in applied economics. But, under staggered adoption, its conventional three-way fixed-effects (3WFE) implementation inherits the interpretation issues now well understood in the difference-in-differences literature. I introduce stacked DDD. I extend the stacked difference-in-differences approach to the DDD setting by creating self-contained stacks, each consisting of four cells over an event window: treated and clean comparison cohorts, each with treatment-eligible and treatment-ineligible units. Appending these stacks yields a unified dataset for estimating treatment effects. I prove that, at each post-treatment event-time, a linear regression with fully saturated fixed-effects applied to the stacked dataset identifies a strictly positive, cell-size-weighted average of stack-level conditional average treatment effects, with stack weights proportional to stack-level cell sizes. Building on this characterization, I outline alternative weighting schemes that recover causal estimands with clear interpretations. Stacked DDD complements recent GMM and imputation-based frameworks by trading efficiency for regression-based transparency, pairwise (rather than global) parallel changes-in-trends, and direct control over both the comparison group for each treated unit and the aggregation weights. I provide two empirical illustrations where stacked DDD yields substantially different quantitative conclusions compared to existing procedures.

Review history (2 revisions) →

Discussion (0). Sign in to comment.

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

11 extracted references · 11 canonical work pages

  1. [1]

    W., and Wooldridge, J

    Abadie, A., Athey, S., Imbens, G. W., and Wooldridge, J. M. (2023). When Should You Adjust Standard Errors for Clustering?Quarterly Journal of Economics, 138(1):1–35. (Cited on page 3.) Borusyak, K., Jaravel, X., and Spiess, J. (2024). Revisiting Event-Study Designs: Robust and Efficient Estima- tion.Review of Economic Studies, 91(6):3253–3285. (Cited on ...

    work page 2023

  2. [2]

    (Cited on page 5.) Sant’Anna, P. H. C. and Zhao, J. (2020). Doubly robust difference-in-differences estimators.Journal of Econo- metrics, 219(1):101–122. (Cited on page 3.) Shastry, G. K. and Tortorice, D. L. (2025). Effective health aid: Evidence from gavi’s vaccine program.Amer- ican Economic Journal: Economic Policy, 17(1):540–74. (Cited on pages 4, 31...

    work page arXiv 2020

  3. [3]

    Computing the FWL residual within each stack. By Step 1, the OLS estimator forτ e reduces to a bivariate FWL regression (B.12)bτ e = P g∈Gtrg(e) P i∈Sg eRe(i, g+e, g) g∆Y i,g+e,g P g∈Gtrg(e) P i∈Sg eRe(i, g+e, g) 2 , where the time summation collapses entirely tot=g+ebecause eRe(i, t, g) = 0fort̸=g+e. I now derive the FWL residual eRe and the resulting we...

    work page 2021

  4. [4]

    APPENDIXE. THREE-WAYFIXED-EFFECTS INEVENT-STUDYDESIGNS The conventional approach to DDD estimation is the three-way fixed effects (3WFE) regression (E.1)Y i,t =α i +γ t +δ Si,t +θD i,t +ϵ i,t , whereα i are unit fixed effects,γ t are time fixed effects,δ Si,t are group-by-time fixed effects, andD i,t = 1{t≥S i}Qi is the treatment indicator. Despite its si...

    work page 2021

  5. [5]

    The group-time mean Dg,t equals the fraction of group-g units that are eligible, Dg,t =n g,1/ng,· fort≥g(since all eligible group-gunits are treated) and Dg,t = 0for t < g

    The unit-level time mean is Di,· = (T−g+ 1)/T. The group-time mean Dg,t equals the fraction of group-g units that are eligible, Dg,t =n g,1/ng,· fort≥g(since all eligible group-gunits are treated) and Dg,t = 0for t < g. Thus Dg,· = ((T−g+ 1)/T)(n g,1/ng,·). STACKED TRIPLE DIFFERENCES 23 Now consider a second cohortg ′ > gand a not-yet-treated eligible uni...

    work page 2020

  6. [6]

    The stacked estimator avoids negative weights because each stack produces a singleATT(g, t)estimate, and the aggregation weightsω g are chosen by the researcher to be non-negative

    The stacked estimator avoids both pathologies. The stacked estimator avoids negative weights because each stack produces a singleATT(g, t)estimate, and the aggregation weightsω g are chosen by the researcher to be non-negative. It avoids forbidden comparisons because each stack restricts the comparison group to units withS i =g c > g+K, ensuring no compar...

    work page 2023

  7. [7]

    Property (i), own-period weights sum to one

    I now prove each property by summing (E.12) overg. Property (i), own-period weights sum to one. Fixℓ=eand sum (E.12) overg∈ G trg X g∈Gtrg ωe g,e =e ⊤ e  X t E[ ¨Ri,t ¨R⊤ i,t]   −1 E ¨Ri,g+e X g∈Gtrg Rg,e(i, g+e) . SinceP g Rg,e(i, t) =R e(i, t)for all(i, t)(summing over cohorts recovers the aggregate event-time indicator), this becomes the coefficien...

    work page 2021

  8. [8]

    The specifi- cation (E.6) recovers interpretable causal parameters only under the joint restrictions of DDD-PCT, treatment effect homogeneity across cohorts, and no anticipation

    Thereforeµ e = ATTe.■ Theorem E.5 clarifies the conditions under which the 3WFE event-study regression is valid. The specifi- cation (E.6) recovers interpretable causal parameters only under the joint restrictions of DDD-PCT, treatment effect homogeneity across cohorts, and no anticipation. In practice, these conditions are rarely satisfied simul- taneous...

    work page 2021

Show all 11 references
  1. [9]

    The event- time indicatorR 0(i,2015)lights up for cohort 2015 but not for cohort 2013; conversely,R 2(i,2015)lights up for cohort 2013 but not

    2015

  2. [10]

    forbidden comparisons

    After removing unit and time fixed effects, these indicators retain residual correlation because the relative-time composition of the sample changes across time. The group-by-time fixed effectsδ Si,t absorb some of this variation but do not eliminate it, since the within-group...

    2021

  3. [11]

    forbidden comparison

    The group mean for groupgis DSi=g,· = 1 ngT X j:Sj=g TX t=1 Dj,t =p Q=1|S=g T−g+ 1 T . Substituting into (E.19) for a unit withS i =g,Q i = 1: eDi,t =1{t≥g} − T−g+ 1 T −p Q=1|S=g 1{t≥g}+p Q=1|S=g T−g+ 1 T = (1−p Q=1|S=g) 1{t≥g} − T−g+ 1 T .(E.20) The factor(1−p Q=1|S=g)is the ...

    2021

Pith tools

Reviewed May 21, 2026 · model on record in the stance chip above.