Pith. sign in

REVIEW 5 minor 125 references

When treatment is assigned at the cluster level, the standard regression-based sensitivity analysis for omitted variable bias is systematically conservative: it treats within-cluster outcome variation as something a cluster-level confounder

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

A correction for Cinelli-Hazlett sensitivity analysis in clustered designs: scale the outcome-confounding partial R² by the between-group residual variance share (η²) to avoid counting irrelevant within-group variation.

T0 review reviewed 2026-08-02 challenge →

load-bearing objection A clean, correct correction to sensemakr for clustered treatment: unit-level sensitivity statistics are systematically conservative, and the η² fix works.

arxiv 2607.13334 v1 pith:R6NH5UHG submitted 2026-07-14 stat.ME

Omitted variable bias sensitivity analysis with clustered treatment assignment

Anton Strezhnev This is my paper
classification stat.ME MSC 62J05
keywords sensitivity analysisomitted variable biasclustered treatment assignmentpartial R-squaredpartial eta-squaredrobustness valuewithin-between regressioncluster-level confounding
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

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 studies a common setting: a policy or exposure is assigned to groups such as counties or schools, but the outcome is measured at the individual level and the regression is run on individuals. It argues that applying the standard sensitivity analysis for omitted variable bias to such a unit-level regression produces robustness statistics that are too pessimistic. The reason is that the outcome-confounder partial R² in the unit-level regression mixes between-cluster and within-cluster variation, while a group-level confounder can only act on the between-cluster part. The paper shows that dividing the unit-level robustness value and extreme-scenario threshold by the share of residual outcome variance that lies between clusters recovers exactly the statistics one would get from a properly aggregated cluster-level regression. Because many published analyses do not make this correction, the paper implies that clustered designs are often more robust to unmeasured confounding than their reported sensitivity output suggests.

Core claim

The paper establishes that in a cluster-only design with treatment assigned at the cluster level, applying the standard regression-based omitted-variable-bias sensitivity analysis to unit-level data yields systematically conservative robustness summaries. The unit-level outcome-confounder partial R² equals η² times the cluster-level partial R², where η² is the share of residual outcome variance explained by cluster membership. Consequently, the unit-level robustness value and extreme-scenario value are deflated by a factor that depends only on this ratio. Applying the inverse scaling to the unit-level f²—and interpreting the extreme scenario against the upper bound η² rather than 1—produces

What carries the argument

The central object is partial eta-squared (η²), also called the correlation ratio: the proportion of residual outcome variance explained by cluster indicators after conditioning on treatment and observed covariates, computed from an auxiliary regression of the unit-level residuals on cluster dummies. This one number carries the entire correction: it converts the unit-level outcome-confounder partial R² to the cluster-level partial R², rescales the unit-level f², and yields adjusted robustness values and extreme-scenario thresholds that match the cluster-level analysis exactly.

Load-bearing premise

The correction assumes a cluster-only design: all confounding enters through cluster-level covariates, and treatment assignment does not affect which cluster a unit belongs to; if within-cluster outcome variation can carry confounding, the corrected robustness statistics would be too optimistic.

What would settle it

Simulate data with cluster-level treatment, an omitted cluster-level confounder of known strength, and both between- and within-cluster outcome variation; fit a unit-level regression and compare the actual coefficient bias to the bias predicted by the corrected sensitivity analysis. If the corrected bound is violated when the confounder is truly cluster-level, the derivation fails. Separately, generate data where an omitted unit-level confounder varies within clusters and affects the outcome; the corrected bound should be exceeded, confirming the design assumption.

Watch this falsifier. Get emailed when new claim-graph text bears on it.

If this is right

  • For any regression of unit-level outcomes on cluster-level treatments, the standard reported robustness value is conservative; rescaling by 1/η gives the correct value, so the result is more robust than the unadjusted output suggests.
  • A cluster-level confounder can never explain more than η² of the residual outcome variation, so the extreme-scenario analysis should be evaluated against η² rather than 1.
  • Benchmarking against cluster-level covariates gives identical conclusions from unit-level and cluster-level regressions; benchmarking against raw unit-level covariates is misleading unless the cluster-level means of those covariates are added to the model.
  • Researchers can recover the cluster-level sensitivity results from an existing unit-level regression without refitting: compute η² from residuals and apply the inverse scaling to the reported f² and extreme-scenario statistics.
  • In the paper's reanalysis, the corrected robustness value rises from 7.4% to 12.7% and the extreme-scenario threshold from 0.59% to 1.83%, illustrating that published unit-level sensitivity analyses can materially understate robustness.

Where Pith is reading between the lines

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

  • Editorial inference: the same variance-compression mechanism should affect other regression-based sensitivity quantities, not just the robustness value and extreme scenario; an implementation that automatically reports η² alongside any unit-level sensitivity analysis would be a natural and low-cost addition.
  • Editorial inference: in unbalanced cluster designs, the correction is only exact when the cluster-level regression is weighted by cluster size; software and applied workflows that run unweighted cluster regressions would need to weight to reproduce the corrected unit-level result.
  • Editorial inference: if the cluster-only assumption is violated—for instance, if within-cluster outcome variation is associated with treatment through unit-level selection—the corrected statistics would overstate robustness. A check comparing within- and between-cluster associations of observed covariates with the outcome would help researchers judge whether the correction is safe.
  • Editorial inference: the paper's logic extends naturally to difference-in-differences and two-way fixed-effects settings, where repeated observations of the same unit play the role of within-cluster variation; deriving the analogous between-period variance ratio for those estimators would give applied researchers a similarly corrected sensitivity tool.
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

0 major / 5 minor

Summary. The paper studies Cinelli-Hazlett OVB sensitivity analysis in clustered observational designs with a cluster-level treatment. It shows that applying the framework to a unit-level regression rather than an appropriately weighted cluster-level regression yields a systematically conservative assessment of robustness: the unit-level outcome-confounder partial R² contains within-cluster residual variation that a cluster-level confounder cannot explain. The central result is that the unit-level partial R² equals η²_{Y|D,X} times the cluster-level partial R² (Eq. 12), and that calibrating the robustness value and extreme-scenario thresholds by this factor recovers equivalence (Eqs. 16–17). The paper also analyzes benchmarking against unit-level covariates, argues that cluster means are the appropriate benchmark, and proposes a Mundlak correction. A reanalysis of Reny et al. (2026) shows the reported robustness value rises from 7.4% to 12.7% (and the extreme-scenario threshold from 0.59% to 1.83%) after the cluster adjustment.

Significance. The result, if accepted, is practically important and non-obvious. It converts a common applied mistake into a conservative one, supplies simple formulas that can be computed from standard regression output plus η², and connects to the earlier clustered-treatment-assignment result of Hansen, Rosenbaum, and Small (2014). The derivation is transparent and the algebra in Eqs. (12)–(17) is checkable; I verified the key identities and the application's headline numbers. The paper ships a fork of sensemakr with the adjustments, which is a concrete strength for reproducibility. The scope is explicitly limited to the "cluster only design" in which cluster membership is unaffected by treatment and confounding is cluster-level; within that scope the claims are internally consistent. The TWFE discussion in the conclusion is more speculative, but it is not load-bearing for the main result.

minor comments (5)
  1. [Section 2] The balanced-cluster proof is given for Eq. (9), but the claimed extension to unbalanced clusters ('straightforward by weighting the cluster-level regression by N_g') is only sketched in Footnote 2. Since Eqs. (12)–(17) and the application depend on the unit-level/cluster-level equivalence, please spell out the weighted cluster-level regression and confirm that the partial-R² rescaling remains exact under imbalanced cluster sizes.
  2. [Section 6] The TWFE paragraph asserts that a 'similar discrepancy' applies to TWFE versus first-differences regressions, but no derivation is provided. TWFE includes unit fixed effects and time variation, so the variance decomposition in Eq. (12) does not directly apply. Either add a formal statement with appropriate assumptions or clearly mark the passage as an informal conjecture for future work.
  3. [Section 5, Tables 1–3] The corrected RV of 12.7% for the original specification is not the RV from the county-means regression in Table 3 (10.9%). Please state explicitly that the 12.7% is the cluster-adjusted RV for the original covariate set (including individual-level covariates), not the RV of a cluster-level regression with cluster means, and explain the difference in terms of the Mundlak discrepancy discussed in Section 4.
  4. [Section 3, Eq. (17)] The correction for the extreme-scenario statistic uses η², while Eq. (16) uses η. A reader may expect the same factor. Add one line showing that setting R_{Y_i∼Z|D,X}=η and converting the resulting f to R² yields Eq. (17), so the asymmetry is transparent.
  5. [Section 4, Eq. (28)] The decomposition of W_i into between and within components and the statement that 'the only variance in W that explains treatment is the between-cluster variance' requires X to contain only cluster-level covariates. Please state this assumption explicitly in Eqs. (28)–(31) or qualify the decomposition.

Circularity Check

0 steps flagged

No significant circularity: the correction is derived algebraically under an explicit cluster-only design; no fitted parameter or load-bearing self-citation is involved.

full rationale

The central rescaling result, Equation (12), is an algebraic consequence of the cluster-only design: since a cluster-level confounder Z_g is orthogonal to the within-cluster outcome residual, the unit-level outcome-confounder partial R^2 is exactly the between-cluster share eta^2 times the cluster-level partial R^2. The factor eta^2 is computed from the residuals of the unit-level regression as a descriptive statistic; it is not fitted to reproduce the cluster-level sensitivity summaries and is not defined in terms of them. Equations (13)-(17) then transform the unit-level bias factor, robustness value, and extreme scenario by this fixed scaling, so the claimed equivalence with the cluster-level sensitivity analysis follows from the algebra rather than from an equality imposed by construction. The paper contains no load-bearing self-citations: it relies on Cinelli and Hazlett, Hansen, Rosenbaum and Small, Huang et al., Mundlak, and other external work for background and assumptions, while the correction formulas themselves are proven in the text. The replication is against external data from Reny, Reeves, and Christenson (2026), and the adjusted statistics are reported as computations, not as predictions from fitted parameters. The paper also explicitly flags its own limitations, such as the homoskedasticity assumption in confidence-bound sensitivity analysis and the TWFE caveat, which are scope limits rather than evidence of circularity. Accordingly, the derivation is self-contained for the stated cluster-only design and no significant circularity is found.

Axiom & Free-Parameter Ledger

0 free parameters · 5 axioms · 0 invented entities

The method introduces no free parameters: η² is a statistic estimated from data in the application, not a parameter chosen to make the derivation work. The axiomatic burden consists of the standard linear OVB framework, the cluster-only design assumption, and the classical regression equivalence results.

axioms (5)
  • domain assumption Cluster-only confounding design: treatment assigned at cluster level; units' cluster memberships unaffected by treatment; confounding arises only from cluster-level covariates.
    Section 2 setup; if clusters are endogenous or unit-level confounders act directly, within-cluster variation can carry confounding and the correction overstates robustness.
  • standard math Equivalence of unit-level and cluster-level regressions for balanced clusters; extension to unbalanced by cluster-size weighting.
    Kloek (1981), Moulton (1986), Wooldridge (2003); used in Eq. 9 and throughout.
  • standard math Frisch-Waugh-Lovell theorem and Cinelli-Hazlett reparameterization of omitted variable bias in terms of partial R².
    Section 2, Eqs. 3–5; the entire framework rests on this.
  • standard math Variance decomposition var(Y) = var(bar Y) + var(tilde Y) for cluster means and within residuals.
    Eq. 12 and following; requires balanced clusters or weighting.
  • domain assumption Linear model sensitivity analysis assumes a single linear confounder with no interaction with treatment.
    Inherited from Cinelli and Hazlett (2020); the paper does not extend to nonlinear or heterogeneous-effect settings.

reviewed 2026-08-02 · how reviews work

0 comments
Cite this review

Pith. "Pith review of Omitted variable bias sensitivity analysis with clustered treatment assignment." pith.science (2026). https://pith.science/paper/R6NH5UHG

@misc{pith2026260713334,
  author       = {Pith},
  title        = {Pith review of: Omitted variable bias sensitivity analysis with clustered treatment assignment},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/R6NH5UHG}},
  note         = {Machine review of arXiv:2607.13334}
}
Share X Bluesky LinkedIn Reddit HN
abstract

Cinelli and Hazlett (2020) develops a sensitivity analysis method for the linear regression model that parameterizes omitted variable bias in terms of two partial $R^2$ parameters capturing the residual variation explained by an omitted confounder in the treatment and outcome respectively. This method is often applied to regressions fit to unit-level data when treatment is assigned at a higher level of aggregation -- as in clustered observational designs. This paper shows that despite the numerical equivalence of the unit-level regression and an appropriately weighted cluster-aggregated regression for estimating the treatment effect, the sensitivity analysis procedure yields different conclusions depending on the chosen level of analysis. The outcome-confounder partial $R^2$ reflects both between- and within- group variation but the latter is irrelevant to omitted variable bias as it by construction cannot be explained by a group-level confounder. Straightforward corrections to the robustness value and the extreme scenario analysis from the unit-level regression using Pearson's partial-$\eta$ recover equivalence between these two approaches. The paper concludes with a point of caution when benchmarking against unit-level covariates and recommends always including cluster-level averages of these covariates as regressors (Mundlak, 1978).

Figures

Figures reproduced from arXiv: 2607.13334 by Anton Strezhnev.

Figure 1
Figure 1. Figure 1: Sensitivity contour plot for main results of Reny, Reeves, and Christenson (2026) [PITH_FULL_IMAGE:figures/full_fig_p018_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Extreme scenario analysis for main results of Reny, Reeves, and Christenson (2026) [PITH_FULL_IMAGE:figures/full_fig_p019_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Sensitivity contour plots for the regressions with cluster-level covariates only [PITH_FULL_IMAGE:figures/full_fig_p021_3.png] view at source ↗

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

125 extracted references · 4 linked inside Pith

  1. [1]

    Journal of Human Resources , volume=

    An evaluation of instrumental variable strategies for estimating the effects of catholic schooling , author=. Journal of Human Resources , volume=

  2. [2]

    Journal of Machine Learning Research , volume=

    Copula-based sensitivity analysis for multi-treatment causal inference with unobserved confounding , author=. Journal of Machine Learning Research , volume=

  3. [3]

    Political science research and methods , volume=

    Explaining fixed effects: Random effects modeling of time-series cross-sectional and panel data , author=. Political science research and methods , volume=. 2015 , publisher=

    2015

  4. [4]

    The Annals of Mathematical Statistics , volume=

    Estimation of the parameters of a single equation in a complete system of stochastic equations , author=. The Annals of Mathematical Statistics , volume=

  5. [5]

    Journal of econometrics , volume=

    Random group effects and the precision of regression estimates , author=. Journal of econometrics , volume=. 1986 , publisher=

    1986

  6. [6]

    The American Statistician , volume=

    Heterogeneity and causality: Unit heterogeneity and design sensitivity in observational studies , author=. The American Statistician , volume=. 2005 , publisher=

    2005

  7. [7]

    Journal of the American Statistical Association , volume=

    A distributional approach for causal inference using propensity scores , author=. Journal of the American Statistical Association , volume=

  8. [8]

    Biometrika , volume=

    Variance-based sensitivity analysis for weighting estimators results in more informative bounds , author=. Biometrika , volume=

  9. [9]

    Political Analysis , volume=

    A selection bias approach to sensitivity analysis for causal effects , author=. Political Analysis , volume=

  10. [10]

    Statistics, Politics and Policy , volume=

    Unbiased estimation of the average treatment effect in cluster-randomized experiments , author=. Statistics, Politics and Policy , volume=. 2015 , publisher=

    2015

  11. [11]

    Epidemiology , volume=

    Sensitivity analysis without assumptions , author=. Epidemiology , volume=. 2016 , publisher=

    2016

  12. [12]

    Statistics & probability letters , volume=

    Bias of the regression estimator for experiments using clustered random assignment , author=. Statistics & probability letters , volume=. 2008 , publisher=

    2008

  13. [13]

    Statistics in Medicine , volume=

    Sensitivity analyses for unmeasured confounding assuming a marginal structural model for repeated measures , author=. Statistics in Medicine , volume=

  14. [14]

    Journal of the Royal Statistical Society Series A: Statistics in Society , volume=

    Interpretable sensitivity analysis for balancing weights , author=. Journal of the Royal Statistical Society Series A: Statistics in Society , volume=

  15. [15]

    arXiv preprint arXiv:2301.12396 , year=

    Sensitivity analysis of causal treatment effect estimation for clustered observational data with unmeasured confounding , author=. arXiv preprint arXiv:2301.12396 , year=

    Pith/arXiv arXiv

  16. [16]

    Journal of the Royal Statistical Society Series B , volume=

    Model-assisted analyses of cluster-randomized experiments , author=. Journal of the Royal Statistical Society Series B , volume=

  17. [17]

    arXiv preprint arXiv:2105.14705 , year=

    The equivalence of the delta method and the cluster-robust variance estimator for the analysis of clustered randomized experiments , author=. arXiv preprint arXiv:2105.14705 , year=

    Pith/arXiv arXiv

  18. [18]

    Oxford Bulletin of Economics and Statistics , volume=

    Computing robust standard errors for within-groups estimators , author=. Oxford Bulletin of Economics and Statistics , volume=

  19. [19]

    American Economic Review , volume=

    Estimating group effects using averages of observables to control for sorting on unobservables: School and neighborhood effects , author=. American Economic Review , volume=

  20. [20]

    Journal of the Royal Statistical Society Series B: Statistical Methodology , volume=

    Sensitivity analysis for inverse probability weighting estimators via the percentile bootstrap , author=. Journal of the Royal Statistical Society Series B: Statistical Methodology , volume=

  21. [21]

    Statistical models in epidemiology, the environment, and clinical trials , pages=

    Sensitivity analysis for selection bias and unmeasured confounding in missing data and causal inference models , author=. Statistical models in epidemiology, the environment, and clinical trials , pages=. 2000 , publisher=

    2000

  22. [22]

    Biometrika , volume=

    Sensitivity analysis for certain permutation inferences in matched observational studies , author=. Biometrika , volume=. 1987 , publisher=

    1987

  23. [23]

    Econometrica: journal of the Econometric Society , pages=

    On the pooling of time series and cross section data , author=. Econometrica: journal of the Econometric Society , pages=. 1978 , publisher=

    1978

  24. [24]

    biometrika , pages=

    Longitudinal data analysis using generalized linear models , author=. biometrika , pages=. 1986 , publisher=

    1986

  25. [25]

    The review of Economics and Statistics , pages=

    An illustration of a pitfall in estimating the effects of aggregate variables on micro units , author=. The review of Economics and Statistics , pages=. 1990 , publisher=

    1990

  26. [26]

    Educational and Psychological measurement , volume=

    The eta coefficient in complex ANOVA designs , author=. Educational and Psychological measurement , volume=. 1970 , publisher=

    1970

  27. [27]

    arXiv preprint arXiv:2406.15288 , year=

    Difference-in-Differences when parallel trends holds conditional on covariates , author=. arXiv preprint arXiv:2406.15288 , year=

    Pith/arXiv arXiv

  28. [28]

    Educational and psychological measurement , volume=

    Eta-squared and partial eta-squared in fixed factor ANOVA designs , author=. Educational and psychological measurement , volume=. 1973 , publisher=

    1973

  29. [29]

    The Quarterly Journal of Economics , volume=

    When should you adjust standard errors for clustering? , author=. The Quarterly Journal of Economics , volume=. 2023 , publisher=

    2023

  30. [30]

    2019 , institution=

    Design and analysis of cluster-randomized field experiments in panel data settings , author=. 2019 , institution=

    2019

  31. [31]

    Journal of econometrics , volume=

    Doubly robust difference-in-differences estimators , author=. Journal of econometrics , volume=. 2020 , publisher=

    2020

  32. [32]

    Journal of econometrics , volume=

    Difference-in-differences with multiple time periods , author=. Journal of econometrics , volume=. 2021 , publisher=

    2021

  33. [33]

    Omitted Variable Bias in Difference-in-Differences Designs , author=

  34. [34]

    The Annals of Statistics , pages=

    Nonparametric estimation of global functionals and a measure of the explanatory power of covariates in regression , author=. The Annals of Statistics , pages=. 1995 , publisher=

    1995

  35. [35]

    Handbook of economic field experiments , volume=

    The econometrics of randomized experiments , author=. Handbook of economic field experiments , volume=. 2017 , publisher=

    2017

  36. [36]

    Proceedings of the Royal Society of London

    On the partial correlation ratio , author=. Proceedings of the Royal Society of London. Series A, Containing Papers of a Mathematical and Physical Character , volume=. 1915 , publisher=

    1915

  37. [37]

    1905 , publisher=

    On the general theory of skew correlation and non-linear regression , author=. 1905 , publisher=

    1905

  38. [38]

    Econometrica , pages=

    OLS estimation in a model where a microvariable is explained by aggregates and contemporaneous disturbances are equicorrelated , author=. Econometrica , pages=. 1981 , publisher=

    1981

  39. [39]

    American Economic Review , volume=

    Cluster-sample methods in applied econometrics , author=. American Economic Review , volume=. 2003 , publisher=

    2003

  40. [40]

    Journal of the American Statistical Association , volume=

    Clustered treatment assignments and sensitivity to unmeasured biases in observational studies , author=. Journal of the American Statistical Association , volume=. 2014 , publisher=

    2014

  41. [41]

    1988 , publisher =

    Cohen, Jacob , title =. 1988 , publisher =

    1988

  42. [42]

    Annual Review of Economics , volume=

    Weak instruments in instrumental variables regression: Theory and practice , author=. Annual Review of Economics , volume=

  43. [43]

    Journal of the American Statistical Association , volume=

    Identification of causal effects using instrumental variables , author=. Journal of the American Statistical Association , volume=

  44. [44]

    Environmental Politics , volume=

    Rising seas, rising concerns: how climate change vulnerability shapes opinions towards policy , author=. Environmental Politics , volume=. 2026 , publisher=

    2026

  45. [45]

    Journal of the Royal Statistical Society Series A: Statistics in Society , pages=

    Sensitivity analysis for clustered observational studies with an application to the effectiveness of magnet nursing hospitals , author=. Journal of the Royal Statistical Society Series A: Statistics in Society , pages=. 2026 , publisher=

    2026

  46. [46]

    Journal of Economic Perspectives , volume=

    Instrumental variables and the search for identification: From supply and demand to natural experiments , author=. Journal of Economic Perspectives , volume=

  47. [47]

    Mostly harmless econometrics: An empiricist's companion , author=

  48. [48]

    Undergraduate econometrics instruction: Through our classes, darkly , author=

  49. [49]

    The Review of Economic Studies , year=

    When is TSLS actually LATE? , author=. The Review of Economic Studies , year=

  50. [50]

    Journal of the American Statistical Association , volume=

    Problems with instrumental variables estimation when the correlation between the instruments and the endogenous explanatory variable is weak , author=. Journal of the American Statistical Association , volume=

  51. [51]

    Mendelian randomization: methods for using genetic variants in causal estimation , author=

  52. [52]

    treatsens: A package to assess sensitivity of causal analyses to unmeasured confounding , author=

  53. [53]

    Journal of Research on Educational Effectiveness , volume=

    Assessing sensitivity to unmeasured confounding using a simulated potential confounder , author=. Journal of Research on Educational Effectiveness , volume=

  54. [54]

    Using geographic variation in college proximity to estimate the return to schooling , author=

  55. [55]

    Exogeneity and robustness , author=

  56. [56]

    Long story short: Omitted variable bias in causal machine learning , author=

  57. [57]

    sensemakr: Sensitivity Analysis Tools for OLS , author=

  58. [58]

    International Conference on Machine Learning , year=

    Sensitivity analysis of linear structural causal models , author=. International Conference on Machine Learning , year=

  59. [59]

    Journal of the Royal Statistical Society Series B: Statistical Methodology , volume=

    Making sense of sensitivity: Extending omitted variable bias , author=. Journal of the Royal Statistical Society Series B: Statistical Methodology , volume=. 2020 , publisher=

    2020

  60. [60]

    Biometrika , volume=

    An omitted variable bias framework for sensitivity analysis of instrumental variables , author=. Biometrika , volume=. 2025 , publisher=

    2025

  61. [61]

    Review of Economics and Statistics , volume=

    Plausibly exogenous , author=. Review of Economics and Statistics , volume=

  62. [62]

    Journal of the National Cancer Institute , number=

    Smoking and lung cancer: recent evidence and a discussion of some questions , author=. Journal of the National Cancer Institute , number=

  63. [63]

    Instruments of development: Randomization in the tropics, and the search for the elusive keys to economic development , author=

  64. [64]

    Journal of Causal Inference , volume=

    To adjust or not to adjust? Sensitivity analysis of M-bias and butterfly-bias , author=. Journal of Causal Inference , volume=

  65. [65]

    Sociological Methodology , volume=

    Assessing bias in the estimation of causal effects: Rosenbaum bounds on matching estimators and instrumental variables estimation with imperfect instruments , author=. Sociological Methodology , volume=

  66. [66]

    Statistics in Medicine , volume=

    A flexible, interpretable framework for assessing sensitivity to unmeasured confounding , author=. Statistics in Medicine , volume=

  67. [67]

    Natural experiments in the social sciences: a design-based approach , author=

  68. [68]

    Sociological Methods & Research , volume=

    Handle with care: A sociologist's guide to causal inference with instrumental variables , author=. Sociological Methods & Research , volume=

  69. [69]

    Journal of the Royal Statistical Society: Series B (Methodological) , volume=

    Some problems in interval estimation , author=. Journal of the Royal Statistical Society: Series B (Methodological) , volume=

  70. [70]

    Darfur: a new history of a long war , author=

  71. [71]

    Sociological Methods & Research , volume=

    Impact of a confounding variable on a regression coefficient , author=. Sociological Methods & Research , volume=

  72. [72]

    Sociological Methodology , volume=

    Indices of robustness for sample representation , author=. Sociological Methodology , volume=

  73. [73]

    Educational Evaluation and Policy Analysis , volume=

    Does nbpts certification affect the number of colleagues a teacher helps with instructional matters? , author=. Educational Evaluation and Policy Analysis , volume=

  74. [74]

    Educational Evaluation and Policy Analysis , volume=

    What would it take to change an inference? Using Rubin's causal model to interpret the robustness of causal inferences , author=. Educational Evaluation and Policy Analysis , volume=

  75. [75]

    Journal of the American Statistical Association , volume=

    Flexible sensitivity analysis for observational studies without observable implications , author=. Journal of the American Statistical Association , volume=

  76. [76]

    Econometrica: Journal of the Econometric Society , pages=

    Partial time regressions as compared with individual trends , author=. Econometrica: Journal of the Econometric Society , pages=

  77. [77]

    Broken instruments , author=

  78. [78]

    Biometrika , year=

    Non-testability of instrument validity under continuous treatments , author=. Biometrika , year=

  79. [79]

    Journal of Conflict Resolution , volume=

    Angry or weary? How violence impacts attitudes toward peace among Darfurian refugees , author=. Journal of Conflict Resolution , volume=

  80. [80]

    Journal of Econometrics , volume=

    Comparing IV with structural models: What simple IV can and cannot identify , author=. Journal of Econometrics , volume=

Showing first 80 references.

This paper was first reviewed by deepseek-v4-flash on August 2, 2026.