Pith. sign in

REVIEW 2 minor 1 cited by

The bridge score, a vector of treatment-specific mediator densities at a common value, balances baseline covariates and yields a sharp pointwise variance envelope on the unidentified mediator-outcome confounding function.

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 ↗

T0 review · grok-4.3

2026-06-30 18:10 UTC pith:V3ZBYLEK

load-bearing objection The bridge score gives a clean way to get sharp conditional bounds on mediator-outcome confounding, and the calibration plus Bayesian g-computation make it more usable than most sensitivity papers.

arxiv 2605.18724 v2 pith:V3ZBYLEK submitted 2026-05-18 stat.ME

Sensitivity analysis for causal mediation: bridge score, sharp sensitivity bounds, and calibration

Yuki Ohnishi , Fan Li This is my paper
classification stat.ME
keywords causal mediationsensitivity analysisbridge scorenatural direct effectnatural indirect effectconfounding functionbalancing scoreg-computation
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.

Causal mediation analysis separates total treatment effects into portions that act through a mediator and those that do not, but identification of natural direct and indirect effects depends on unverifiable assumptions about no mediator-outcome confounding. The paper introduces the bridge score to address this gap in sensitivity analysis. It shows that the score, built directly from the mediator densities under each treatment level, balances the covariates that matter for the mediator stage. Conditional on the score, the authors derive a sharp bound on how large the confounding function can be, expressed through latent outcome relevance and residual selection. They then supply a residual-budget calibration that turns the bound into a practical tool and combine it with Bayesian g-computation to propagate both data uncertainty and user-chosen sensitivity parameters.

Core claim

The central claim is that the bridge score, formed as a low-dimensional vector from the two treatment-specific mediator densities evaluated at a common mediator value, balances baseline covariates for the mediator stage relevant to natural effect identification. Conditional on the bridge score, a sharp pointwise variance envelope exists on the unidentified mediator-outcome confounding function, expressed in terms of latent outcome relevance and residual selection. This envelope supports operational sensitivity analysis through residual budget calibration based on local residual outcome variation, a complementary range bound for support restrictions, and a Bayesian g-computation algorithm tha

What carries the argument

The bridge score, a low-dimensional vector formed from the two treatment-specific mediator densities at a common mediator value, which balances baseline covariates and permits derivation of the sharp pointwise variance envelope on the confounding function.

Load-bearing premise

The bridge score balances baseline covariates for the mediator stage relevant to natural effect identification, allowing the sharp pointwise variance envelope on the confounding function to be derived conditional on this score.

What would settle it

A dataset or simulation in which the conditional variance of the mediator-outcome confounding function, given the bridge score, exceeds the envelope calculated from the specified latent outcome relevance and residual selection values.

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

If this is right

  • The sharp pointwise variance envelope supplies tight, operational bounds for sensitivity analysis of natural direct and indirect effects.
  • Residual budget calibration using local residual outcome variation converts the theoretical envelope into a usable sensitivity parameter.
  • A complementary range bound applies when support-based restrictions are imposed on the confounding function.
  • The pointwise bound reduces to a scalar functional that supports Bayesian g-computation inference combining data uncertainty with user-specified sensitivity corrections.

Where Pith is reading between the lines

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

  • The bridge-score construction could be generalized to settings with multiple mediators or time-varying mediators by extending the density-ratio vector.
  • Empirical checks could compare the calibrated bounds against fully observed confounding in randomized experiments with measured mediator-outcome confounders.
  • The scalar functional reduction might integrate directly into existing g-computation or weighting software for routine sensitivity reporting.

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 / 2 minor

Summary. The paper introduces the bridge score as a low-dimensional vector formed from the two treatment-specific mediator densities evaluated at a common mediator value m. It establishes that this score balances baseline covariates relevant to the mediator stage for natural direct and indirect effect identification. Conditional on the bridge score, the authors derive a sharp pointwise variance envelope for the unidentified mediator-outcome confounding function expressed in terms of latent outcome relevance and residual selection. They further propose a residual budget calibration method based on local residual outcome variation, record a complementary range bound under support restrictions, and outline a Bayesian g-computation procedure that combines observed-data posterior uncertainty with user-specified sensitivity parameters via a scalar functional reduction.

Significance. If the balancing property and sharpness of the derived envelope hold, the framework supplies a dimension-reduced, operational sensitivity tool for causal mediation that directly ties bounds to observable residual variation rather than fully nonparametric sensitivity parameters. The combination of the bridge score with the Bayesian g-computation algorithm offers a concrete route to inference that separates observed-data uncertainty from sensitivity uncertainty.

minor comments (2)
  1. The abstract states that the bridge score 'balances baseline covariates for the mediator stage relevant to natural effect identification,' but the precise statement of the balancing property (e.g., which covariates are balanced and under what conditional independence) should be stated as a formal proposition with proof sketch in the main text to allow readers to verify the subsequent envelope derivation.
  2. The description of the 'scalar functional reduction' used to operationalize the pointwise bound for the g-computation algorithm is not detailed in the abstract; a brief indication of how the functional is constructed (e.g., which moment or integral is taken) would clarify the computational steps.

Simulated Author's Rebuttal

0 responses · 0 unresolved

We thank the referee for the detailed summary of our manuscript and the positive assessment of its significance. The recommendation for minor revision is appreciated. No specific major comments were provided in the report.

Circularity Check

0 steps flagged

No significant circularity

full rationale

The derivation begins by defining the bridge score explicitly as the low-dimensional vector of the two treatment-specific mediator densities evaluated at a shared mediator value, then proves its balancing property for the relevant covariates, and finally derives the sharp pointwise variance envelope on the confounding function conditional on that score. These steps are sequential mathematical consequences of the definition rather than reductions of the claimed results back to fitted parameters or self-citations by construction. The residual-budget calibration and Bayesian g-computation are presented as operationalizations that incorporate user-specified sensitivity parameters, not as predictions forced by the observed-data likelihood. No load-bearing self-citation chain, uniqueness theorem imported from the authors' prior work, or renaming of known empirical patterns appears in the abstract or described program.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 1 invented entities

The paper rests on the standard domain assumption that mediator-stage sequential ignorability cannot be verified empirically and introduces the new bridge score construct; no explicit free parameters or invented entities beyond the sensitivity parameters are described in the abstract.

free parameters (1)
  • sensitivity parameters for confounding function
    User-specified quantities that enter the bounds and calibration; exact form not detailed in abstract.
axioms (1)
  • domain assumption Mediator-stage sequential ignorability cannot be empirically verified
    Stated as the reason explicit sensitivity analysis is required.
invented entities (1)
  • bridge score no independent evidence
    purpose: Low-dimensional balancing score formed from treatment-specific mediator densities
    Newly formulated construct central to the sensitivity bounds.

pith-pipeline@v0.9.1-grok · 5724 in / 1459 out tokens · 37463 ms · 2026-06-30T18:10:22.316568+00:00 · methodology

0 comments
read the original abstract

Causal mediation analysis decomposes the total treatment effect into a portion operating through a hypothesized mediator and a residual direct portion. Identification of natural direct and indirect effects typically rests on the mediator stage of sequential ignorability, which cannot be empirically verified and requires explicit sensitivity analysis. We formulate the \emph{bridge score}, a mediator-stage balancing score, as a low-dimensional vector formed from the two treatment-specific mediator densities at a common mediator value, and show that it balances baseline covariates for the mediator stage relevant to natural effect identification. Conditional on the bridge score, we derive a sharp pointwise variance envelope on the unidentified mediator-outcome confounding function in terms of latent outcome relevance and residual selection. To make the bound operational for sensitivity analysis, we further introduce a residual budget calibration approach based on local residual outcome variation and record a complementary range bound for support-based restrictions. Finally, we show how the pointwise bound can be operationalized for inference through a scalar functional reduction and a Bayesian g-computation algorithm that combines observed-data posterior uncertainty with user-specified sensitivity uncertainty, rather than treating the unidentified sensitivity corrections as learned from the likelihood.

Figures

Figures reproduced from arXiv: 2605.18724 by Fan Li, Yuki Ohnishi.

Figure 1
Figure 1. Figure 1: Raw-scale and rank-based benchmark calibration for the immigration-framing [PITH_FULL_IMAGE:figures/full_fig_p017_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Sensitivity parameter sweeps for the immigration framing illustration. Left: [PITH_FULL_IMAGE:figures/full_fig_p018_2.png] view at source ↗

Review history (2 revisions) →

discussion (0)

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

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. Semiparametric Mediation Analysis with Separately Observed Mediator and Outcome under Unmeasured Confounding

    stat.ME 2026-06 unverdicted novelty 7.0

    A semiparametric data fusion framework restores identification of mediation effects from separately observed mediator and outcome data using shared IVs under unmeasured confounding and no-interaction plus latent align...

Reference graph

Works this paper leans on

4 extracted references · cited by 1 Pith paper

  1. [1]

    Brader, T., N. A. Valentino, and E. Suhay (2008). What triggers public opposition to immigration? anxiety, group cues, and immigration threat.American Journal of Political Science 52(4), 959–978. Cheng, C. and F. Li (2026, 02). Semiparametric causal mediation analysis of cluster- randomized trials for indirect and spillover effects.Biometrics 82(1), ujag0...

    2008

  2. [2]

    Imbens, G. W. (2003). Sensitivity to exogeneity assumptions in program evaluation.Amer- ican Economic Review 93(2), 126–132. le Cessie, S. (2016). Bias formulas for estimating direct and indirect effects when unmea- sured confounding is present.Epidemiology 27(1), 125–132. Li, F., P. Ding, and F. Mealli (2023). Bayesian causal inference: a critical review...

    2003

  3. [3]

    de Luna, and M

    Lindmark, A., X. de Luna, and M. Eriksson (2018). Sensitivity analysis for unobserved confounding of direct and indirect effects using uncertainty intervals.Statistics in Medicine 37(10), 1744–1762. McCandless, L. C., P. Gustafson, and A. Levy (2017). Bayesian sensitivity analysis for unmeasured confounding in causal mediation analysis.Statistics in Medic...

    2018

  4. [4]

    Tchetgen Tchetgen, E. J. (2013). Inverse odds ratio-weighted estimation for causal medi- ation analysis.Statistics in Medicine 32(26), 4567–4580. VanderWeele, T. J. (2010). Bias formulas for sensitivity analysis for direct and indirect effects.Epidemiology 21(4), 540–551. VanderWeele, T. J. and P. Ding (2017). Sensitivity analysis in observational researc...

    2013