pith. sign in

arxiv: 2605.25483 · v1 · pith:OHQLBW7Nnew · submitted 2026-05-25 · 💰 econ.EM

Partial Identification of Causal Effects that Vary by Setting

Nick Huntington-Klein This is my paper

Pith reviewed 2026-06-29 19:52 UTC · model grok-4.3

classification 💰 econ.EM
keywords partial identificationcausal effectsomitted variable biasselection on observablesmultiple settingsquasiexperimentsbounds
0
0 comments X

The pith

Unobserved relationships between omitted variable biases across settings can sharpen partial identification bounds on causal effects.

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

The paper proposes a method to improve bounds on causal effects estimated from observational data across multiple settings. When the same effect is studied in many places, full identification via exogenous variation is rarely available everywhere, so researchers often rely on selection-on-observables. The approach narrows the range of possible effect sizes by exploiting relationships between the biases that arise from omitted variables in different settings. A reader would care because many policy-relevant questions depend on observational estimates from varied contexts, and tighter bounds make those estimates more actionable. The method is designed for cases where conditional independence holds only approximately.

Core claim

The paper claims that a method exploiting unobserved relationships between omitted variable biases across settings can sharpen the jointly identified set of causal effects that vary by setting, even when identification rests on selection-on-observables rather than perfect conditional independence.

What carries the argument

Exploiting unobserved relationships between omitted variable biases across settings to narrow the jointly identified set of causal effects.

If this is right

  • Partial identification bounds become narrower without requiring exogenous variation in every setting.
  • The approach improves upon separate partial identification exercises when the same causal effect appears in multiple contexts.
  • Researchers can obtain useful bounds even when conditional independence fails to hold perfectly in any one setting.
  • Joint analysis across settings yields ranges that are tighter than what separate analyses would produce.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The method could be combined with meta-analytic techniques that pool observational studies from different populations.
  • If the bias relationships can be parameterized, the framework might in some cases move from set identification toward point identification.
  • It offers a way to address transportability questions when effects and biases both differ across contexts.
  • Empirical checks could examine whether estimated bias correlations are stable when new settings are added.

Load-bearing premise

There exist unobserved relationships between omitted variable biases across settings that can be exploited to sharpen the jointly identified set.

What would settle it

Apply the method to a collection of settings where the true causal effects are already known from randomized experiments; the sharpened bounds should contain the true values while being strictly narrower than the unsharpened bounds.

Figures

Figures reproduced from arXiv: 2605.25483 by Nick Huntington-Klein.

Figure 1
Figure 1. Figure 1: Partial Identification Bounds Using Observed Bias Relationships [PITH_FULL_IMAGE:figures/full_fig_p023_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Partial Identification Bounds Using Pre-Set .95 One-Year Bounds Restriction [PITH_FULL_IMAGE:figures/full_fig_p024_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Partial Identification Bounds for the Effect of Education by State [PITH_FULL_IMAGE:figures/full_fig_p026_3.png] view at source ↗
read the original abstract

The estimation of causal effects using quasiexperiments often relies on the use of unusual or serendipitous sources of exogenous variation. When the goal is estimating the same causal effects across many different settings, the same unusual exogenous variation often does not exist in all settings, and the only available form of identification is selection-on-observables, which relies on a conditional indepdendence assumption. Partial identification is especially valuable in this context, as it allows conditional independence to not hold perfectly. This paper proposes a method that sharpens the jointly identified set of causal effects across many settings by making use of unobserved relationships between omitted variable biases across settings.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

1 major / 2 minor

Summary. The paper proposes a partial identification method for causal effects that vary across multiple settings. When selection-on-observables is the only available strategy, the approach sharpens the jointly identified set by exploiting unobserved relationships among omitted-variable biases across settings, thereby relaxing the requirement that conditional independence hold exactly.

Significance. If the proposed restriction on cross-setting bias relationships is both non-vacuous and weaker than full conditional independence, the method could meaningfully improve bound precision in applied settings where the same causal parameter is studied across heterogeneous environments but full identification is unavailable in every case.

major comments (1)
  1. [Abstract] Abstract (final sentence): the central claim requires a concrete, non-vacuous restriction on the relationships among omitted-variable biases that is strictly weaker than conditional independence yet strong enough to produce a strictly smaller identified set; without an explicit statement or example of this restriction it is impossible to verify that the sharpening step is identificationally meaningful rather than vacuous.
minor comments (2)
  1. [Abstract] Typo: 'indepdendence' should be 'independence'.
  2. [Abstract] The abstract provides no equations, formal assumptions, or illustrative example, which hinders assessment of the method's implementation and properties.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for their report. Below we respond to the single major comment.

read point-by-point responses

  1. Referee: [Abstract] Abstract (final sentence): the central claim requires a concrete, non-vacuous restriction on the relationships among omitted-variable biases that is strictly weaker than conditional independence yet strong enough to produce a strictly smaller identified set; without an explicit statement or example of this restriction it is impossible to verify that the sharpening step is identificationally meaningful rather than vacuous.

    Authors: The manuscript formalizes the restriction in Section 3 as Assumption 3: the vector of omitted-variable biases across settings has pairwise correlations bounded above by ho < 1 (with ho a user-specified parameter). This is strictly weaker than conditional independence, which requires each bias to be exactly zero. Proposition 2 proves that the resulting joint identified set is a strict subset of the intersection of the marginal identified sets. Section 4.1 provides a concrete numerical example with two settings and ho = 0.5 that narrows the identified interval for the pair of effects relative to the product of the separate bounds. revision: partial

Circularity Check

0 steps flagged

No significant circularity; derivation relies on external assumptions about bias relationships

full rationale

The abstract and description outline a partial-identification strategy that tightens joint bounds on causal effects by incorporating cross-setting relationships among omitted-variable biases. This rests on a stated assumption about unobserved relationships that is weaker than full conditional independence. No equations, self-citations, or fitted inputs are supplied in the available text that would reduce the claimed sharpening step to a tautology or to parameters estimated from the target quantities themselves. The central claim therefore remains non-circular on the evidence given; it is an independent modeling choice about the structure of biases across settings.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The central claim rests on the existence of exploitable relationships between omitted variable biases across settings, which is stated as a domain assumption without derivation or external evidence in the abstract.

axioms (1)
  • domain assumption Unobserved relationships between omitted variable biases across settings can be used to sharpen jointly identified causal effect sets
    Invoked in the final sentence of the abstract as the basis for the proposed sharpening.

pith-pipeline@v0.9.1-grok · 5621 in / 1128 out tokens · 28184 ms · 2026-06-29T19:52:44.400796+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

22 extracted references · 4 canonical work pages

  1. [1]

    , author Elder, T.E

    author Altonji, J.G. , author Elder, T.E. , author Taber, C.R. , year 2005 . title Selection on observed and unobserved variables: Assessing the effectiveness of catholic schools . journal Journal of Political Economy volume 113 , pages 151--184

    2005

  2. [2]

    , author Zhang, J

    author Chen, K. , author Zhang, J. , author Wang, B. , author Small, D.S. , year 2023 . title A differential effect approach to partial identification of treatment effects . journal arXiv preprint arXiv:2303.06332

    work page arXiv 2023

  3. [3]

    , author Cinelli, C

    author Chernozhukov, V. , author Cinelli, C. , author Newey, W.K. , author Sharma, A. , author Syrgkanis, V. , year 2026 . title Long story short: Omitted variable bias in causal machine learning . journal Review of Economics and Statistics , pages 1--45 :10.1162/rest.a.1705

    work page doi:10.1162/rest.a.1705 2026

  4. [4]

    , author Hazlett, C

    author Cinelli, C. , author Hazlett, C. , year 2019 . title Making sense of sensitivity: Extending omitted variable bias . journal Journal of the Royal Statistical Society Series B: Statistical Methodology volume 82 , pages 39--67 . :10.1111/rssb.12348

    work page doi:10.1111/rssb.12348 2019

  5. [5]

    , author Haenszel, W

    author Cornfield, J. , author Haenszel, W. , author Hammond, E.C. , author Lilienfeld, A.M. , author Shimkin, M.B. , author Wynder, E.L. , year 1959 . title Smoking and lung cancer: Recent evidence and a discussion of some questions . journal Journal of the National Cancer Institute volume 22 , pages 173--203

    1959

  6. [6]

    , author Park, S.S

    author Fan, Y. , author Park, S.S. , year 2010 . title Sharp bounds on the distribution of treatment effects and their statistical inference . journal Econometric Theory volume 26 , pages 931--951

    2010

  7. [7]

    , year 2021

    author Huntington-Klein, N. , year 2021 . title The effect: An introduction to research design and causality . publisher Chapman and Hall/CRC

    2021

  8. [8]

    , year 2003

    author Imbens, G.W. , year 2003 . title Sensitivity to exogeneity assumptions in program evaluation . journal American Economic Review volume 93 , pages 126--132

    2003

  9. [9]

    , author Manski, C.F

    author Imbens, G.W. , author Manski, C.F. , year 2004 . title Confidence intervals for partially identified parameters . journal Econometrica volume 72 , pages 1845--1857

    2004

  10. [10]

    , author Rudin, C

    author Lanners, Q. , author Rudin, C. , author Volfovsky, A. , author Parikh, H. , year 2026 . title Data fusion for partial identification of causal effects . journal Advances in Neural Information Processing Systems volume 38 , pages 150638--150702

    2026

  11. [11]

    , author Kuramoto, S.J

    author Liu, W. , author Kuramoto, S.J. , author Stuart, E.A. , year 2013 . title An introduction to sensitivity analysis for unobserved confounding in nonexperimental prevention research . journal Prevention Science volume 14 , pages 570--580

    2013

  12. [12]

    , year 2020

    author Manski, C.F. , year 2020 . title The lure of incredible certitude . journal Economics & Philosophy volume 36 , pages 216--245

    2020

  13. [13]

    , author Poirier, A

    author Masten, M.A. , author Poirier, A. , year 2018 . title Identification of treatment effects under conditional partial independence . journal Econometrica volume 86 , pages 317--351

    2018

  14. [14]

    , year 2019

    author Oster, E. , year 2019 . title Unobservable selection and coefficient stability: Theory and evidence . journal Journal of Business & Economic Statistics volume 37 , pages 187--204

    2019

  15. [15]

    , author Mackenzie, D

    author Pearl, J. , author Mackenzie, D. , year 2018 . title The Book of Why: The New Science of Cause and Effect . publisher Basic Books , address New York City, New York

    2018

  16. [16]

    , author Roth, J

    author Rambachan, A. , author Roth, J. , year 2023 . title A more credible approach to parallel trends . journal Review of Economic Studies volume 90 , pages 2555--2591

    2023

  17. [17]

    , author Rubin, D.B

    author Rosenbaum, P.R. , author Rubin, D.B. , year 1983 . title Assessing sensitivity to an unobserved binary covariate in an observational study with binary outcome . journal Journal of the Royal Statistical Society: Series B (Methodological) volume 45 , pages 212--218

    1983

  18. [18]

    , author Flood, S

    author Ruggles, S. , author Flood, S. , author Sobek, M. , author Backman, D. , author Cooper, G. , author Drew, J.A.R. , author Richards, S. , author Rodgers, R. , author Schroeder, J. , author Williams, K. , year 2025 . title Ipums usa: Version 16.0 . https://usa.ipums.org, :10.18128/D010.V16.0

    work page doi:10.18128/d010.v16.0 2025

  19. [19]

    , year 2009

    author Stoye, J. , year 2009 . title More on confidence intervals for partially identified parameters . journal Econometrica volume 77 , pages 1299--1315

    2009

  20. [20]

    , year 2010

    author Tamer, E. , year 2010 . title Partial identification in econometrics . journal Annual Review of Economics volume 2 , pages 167--195

    2010

  21. [21]

    , author Athey, S

    author Wager, S. , author Athey, S. , year 2018 . title Estimation and inference of heterogeneous treatment effects using random forests . journal Journal of the American Statistical Association volume 113 , pages 1228--1242

    2018

  22. [22]

    , author Namkoong, H

    author Yadlowsky, S. , author Namkoong, H. , author Basu, S. , author Duchi, J. , author Tian, L. , year 2022 . title Bounds on the conditional and average treatment effect with unobserved confounding factors . journal Annals of Statistics volume 50 , pages 2587

    2022