pith. sign in

arxiv: 2605.20007 · v1 · pith:3UBLEXK2new · submitted 2026-05-19 · 📊 stat.ME

Identifying Interventional Joint Distributions via Extended Bridge Functions

Constantin Schott This is my paper

Pith reviewed 2026-05-20 03:36 UTC · model grok-4.3

classification 📊 stat.ME
keywords proximal causal inferencebridge functionsjoint interventional distributionsidentificationkernel methodscausal inferenceproxy variables
0
0 comments X

The pith

Extended bridge functions identify joint interventional distributions while retaining all proxy variables.

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

Standard proximal causal inference results identify only marginal interventional distributions and typically lose information from some of the proxy variables used to define the bridges. This paper introduces extended bridge functions that satisfy joint completeness and invertibility conditions, allowing recovery of the full joint interventional distribution of outcomes together with all proxies. If the new identification holds, applied researchers can target richer causal quantities such as conditional interventional laws without discarding proxy data that was already collected. The results are then plugged into existing proximal algorithms, where interventional kernels appear as intermediate objects and the whole procedure is recast as operations on kernels.

Core claim

By defining extended bridge functions, the paper derives identification formulas for joint interventional distributions that keep every relevant proxy variable; these formulas are then used to generalize proximal identification algorithms into a unified framework whose intermediate objects are interventional kernels operated on directly.

What carries the argument

Extended bridge functions, which extend ordinary outcome or treatment bridges so that they map the joint law of proxies and outcomes while preserving invertibility and completeness.

If this is right

  • Joint interventional distributions become identifiable targets rather than marginal ones.
  • Proximal algorithms can treat interventional kernels as first-class objects inside kernel operations.
  • All proxy variables used to construct the bridges can be retained in the final identified distribution.
  • A single kernel-based framework covers both standard marginal identification and the new joint case.

Where Pith is reading between the lines

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

  • The same construction may let researchers compute conditional interventional densities when the proxies carry high-dimensional information.
  • Links to kernel mean embeddings or reproducing-kernel Hilbert space methods in causal estimation become immediate.
  • Empirical checks could compare the recovered joint law against known marginals obtained from ordinary bridges.

Load-bearing premise

Extended bridge functions exist and satisfy the required completeness and invertibility conditions with respect to the joint distribution of the proxies and outcomes.

What would settle it

A simulation in which the true joint interventional distribution is known exactly and the extended-bridge estimator recovers it to arbitrary accuracy, or fails exactly when the completeness or invertibility conditions are violated.

Figures

Figures reproduced from arXiv: 2605.20007 by Constantin Schott.

Figure 1
Figure 1. Figure 1: (a) A causal graph depicting the causal relations between A (influenza vaccination), Y (respiratory-disease-related hospitalization) and X (age, gender) in the example medical trial. (b) The corresponding SWIG for the causal model in (a) after the intervention do(A = a). (c) Causal graph (a) with an added unmeasured confounder U (health awareness) influencing both treatment A and outcome Y . (d) Causal gra… view at source ↗
Figure 2
Figure 2. Figure 2: This figure gives an overview of different assumptions used in the PCI identification results in [PITH_FULL_IMAGE:figures/full_fig_p009_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: (a) A causal diagram representing the model used for proximal front-door identification with treatment bridge functions. We prove an identification formula for p(y ∥ a) in Theorem 5.1. (b) A SWIG representing the causal model in (a) after the intervention do(M = m). (c) A SWIG representing the causal model in (a) after the joint intervention do(A = a, M = m). Assumption A.20 (Exclusion restrictions). A(m) … view at source ↗
Figure 4
Figure 4. Figure 4: Comparison of different proximal front-door graphs: [PITH_FULL_IMAGE:figures/full_fig_p020_4.png] view at source ↗
read the original abstract

Existing identification results in proximal causal inference often focus on marginal interventional distributions using standard outcome or treatment bridge functions. These methods do not generally identify joint interventional distributions that contain all proxy variables that were used to define the corresponding bridge functions. In many applications, however, these joint interventional distributions are a natural target of interest. We introduce extended bridge functions and derive new identification results for joint interventional distributions that may retain all relevant proxy variables. We then apply these results to proximal identification algorithms, where interventional kernels naturally arise as intermediate objects, yielding a generalized framework based on kernel operations.

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 / 1 minor

Summary. The manuscript introduces extended bridge functions to derive identification results for joint interventional distributions in proximal causal inference that retain all relevant proxy variables, then applies these results to proximal identification algorithms by treating interventional kernels as intermediate objects to obtain a generalized kernel-operation framework.

Significance. If the identification results hold under the stated completeness and invertibility conditions, the work would meaningfully extend proximal causal inference beyond marginal interventional distributions to joint distributions that include the defining proxies. This is relevant for applications where the full joint is the target quantity, and the kernel-based algorithmic framing could facilitate implementation.

major comments (1)
  1. [Section introducing extended bridge functions (and the subsequent identification theorem)] The identification of the joint interventional distribution rests on the existence and uniqueness of extended bridge functions satisfying completeness and invertibility with respect to the joint law of the retained proxies and outcomes. The manuscript asserts these properties but supplies neither the explicit integral equation defining the extended bridge functions nor a proof that solutions exist and are unique under standard proximal assumptions (e.g., conditional completeness of the proxy space). This is load-bearing for the central claim.
minor comments (1)
  1. [Abstract] The abstract refers to 'extended bridge functions' and 'interventional kernels' without a one-sentence characterization or pointer to the defining section, which would improve readability for readers familiar with standard bridge functions.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for the careful and constructive review. The major comment raises an important point about rigor in the foundational identification result, which we address below by committing to a targeted revision.

read point-by-point responses

  1. Referee: [Section introducing extended bridge functions (and the subsequent identification theorem)] The identification of the joint interventional distribution rests on the existence and uniqueness of extended bridge functions satisfying completeness and invertibility with respect to the joint law of the retained proxies and outcomes. The manuscript asserts these properties but supplies neither the explicit integral equation defining the extended bridge functions nor a proof that solutions exist and are unique under standard proximal assumptions (e.g., conditional completeness of the proxy space). This is load-bearing for the central claim.

    Authors: We agree that an explicit integral equation and a self-contained proof of existence and uniqueness are necessary to fully substantiate the central identification theorem. While the manuscript defines the extended bridge functions through their characterizing operator equations under the joint law of the retained proxies and outcomes, and invokes standard completeness and invertibility conditions from the proximal causal inference literature, we acknowledge that these elements could be stated more explicitly. In the revised manuscript we will (i) write out the precise integral equation that the extended bridge function must satisfy and (ii) supply a short proof of existence and uniqueness that directly applies the conditional completeness assumption on the proxy space together with the invertibility condition already stated in the paper. This revision will make the load-bearing step fully rigorous without altering the overall identification strategy. revision: yes

Circularity Check

0 steps flagged

No circularity: extended bridge functions introduced as new objects with independent identification derivations

full rationale

The paper introduces extended bridge functions as a novel extension of standard outcome/treatment bridge functions from proximal causal inference and derives identification results for joint interventional distributions that retain proxies. The abstract and reader's summary indicate these functions are defined to satisfy completeness and invertibility conditions with respect to the joint proxy-outcome distribution, after which the identification theorems follow. No quoted equations or steps reduce the claimed results to a self-definition, a fitted parameter renamed as a prediction, or a load-bearing self-citation chain whose prior work itself assumes the target result. The contribution is framed as building on external proximal literature rather than re-deriving quantities from its own fitted inputs, making the derivation self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 1 invented entities

The central addition is the definition of extended bridge functions whose existence and properties are taken as the basis for the new identification results.

axioms (1)
  • domain assumption Existence of extended bridge functions satisfying completeness and invertibility conditions for joint interventional distributions.
    The identification results for joint distributions rest on these functions being well-defined and usable in the proximal setup.
invented entities (1)
  • Extended bridge functions no independent evidence
    purpose: To enable identification of joint interventional distributions that retain all proxy variables.
    New concept introduced to overcome the limitation of standard bridge functions.

pith-pipeline@v0.9.0 · 5603 in / 1164 out tokens · 54602 ms · 2026-05-20T03:36:51.038055+00:00 · methodology

discussion (0)

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

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

59 extracted references · 59 canonical work pages · 1 internal anchor

  1. [1]

    Half-trek criterion for generic identifiability of linear structural equation models , year =

    Foygel, Rina and Draisma, Jan and Drton, Mathias , journal =. Half-trek criterion for generic identifiability of linear structural equation models , year =. doi:10.1214/12-aos1012 , publisher =

    work page doi:10.1214/12-aos1012

  2. [2]

    arXiv preprint arXiv:2512.24413 , year=

    Ringlein, Grace V. and Nguyen, Trang Quynh and Zandi, Peter P. and Stuart, Elizabeth A. and Parikh, Harsh , title =. 2025 , month = dec, abstract =. doi:10.48550/ARXIV.2512.24413 , eprint =

    work page doi:10.48550/arxiv.2512.24413 2025

  3. [3]

    , journal =

    Shpitser, Ilya and Wood-Doughty, Zach and Tchetgen Tchetgen, Eric J. , journal =. The. 2023 , month = aug, pages =. doi:10.48550/ARXIV.2108.06818 , eprint =

    work page doi:10.48550/arxiv.2108.06818 2023

  4. [4]

    Genericity of the completeness condition with constrained instruments , year =

    Loh, Isaac , journal =. Genericity of the completeness condition with constrained instruments , year =. doi:10.1016/j.econlet.2023.110998 , publisher =

    work page doi:10.1016/j.econlet.2023.110998 2023

  5. [5]

    and Ying, Andrew and Cui, Yifan and Shi, Xu and Miao, Wang , journal =

    Tchetgen Tchetgen, Eric J. and Ying, Andrew and Cui, Yifan and Shi, Xu and Miao, Wang , journal =. An Introduction to Proximal Causal Inference , year =. doi:10.1214/23-sts911 , publisher =

    work page doi:10.1214/23-sts911

  6. [6]

    Identifying causal effects with proxy variables of an unmeasured confounder , year =

    Miao, Wang and Geng, Zhi and Tchetgen Tchetgen, Eric J , journal =. Identifying causal effects with proxy variables of an unmeasured confounder , year =. doi:10.1093/biomet/asy038 , publisher =

    work page doi:10.1093/biomet/asy038

  7. [7]

    Half-trek criterion for identifiability of latent variable models , year =

    Barber, Rina Foygel and Drton, Mathias and Sturma, Nils and Weihs, Luca , journal =. Half-trek criterion for identifiability of latent variable models , year =. doi:10.1214/22-aos2221 , publisher =

    work page doi:10.1214/22-aos2221

  8. [8]

    Trek separation for Gaussian graphical models , year =

    Sullivant, Seth and Talaska, Kelli and Draisma, Jan , journal =. Trek separation for Gaussian graphical models , year =. doi:10.1214/09-aos760 , publisher =

    work page doi:10.1214/09-aos760

  9. [9]

    Determinantal Generalizations of Instrumental Variables , year =

    Weihs, Luca and Robinson, Bill and Dufresne, Emilie and Kenkel, Jennifer and Kubjas Reginald McGee II, Kaie and Reginald, McGee II and Nguyen, Nhan and Robeva, Elina and Drton, Mathias , journal =. Determinantal Generalizations of Instrumental Variables , year =. doi:10.1515/jci-2017-0009 , publisher =

    work page doi:10.1515/jci-2017-0009 2017

  10. [10]

    A General Identification Condition for Causal Effects , year =

    Tian, Jin and Pearl, Judea , booktitle =. A General Identification Condition for Causal Effects , year =

    work page

  11. [11]

    doi:10.1080/01621459.2023

    Cui, Yifan and Pu, Hongming and Shi, Xu and Miao, Wang and Tchetgen Tchetgen, Eric , journal =. Semiparametric. 2023 , issn =. doi:10.1080/01621459.2023.2191817 , publisher =

    work page doi:10.1080/01621459.2023.2191817 2023

  12. [12]

    Proximal mediation analysis , year =

    Dukes, Oliver and Shpitser, Ilya and Tchetgen Tchetgen, Eric J , journal =. Proximal mediation analysis , year =. doi:10.1093/biomet/asad015 , publisher =

    work page doi:10.1093/biomet/asad015

  13. [13]

    Causal inference with hidden mediators , year =

    Ghassami, Amiremad and Yang, Alan and Shpitser, Ilya and Tchetgen Tchetgen, Eric , journal =. Causal inference with hidden mediators , year =. doi:10.1093/biomet/asae037 , publisher =

    work page doi:10.1093/biomet/asae037

  14. [14]

    , publisher =

    Adams, Ernest W. , publisher =. The Logic of Conditionals: An Application of Probability to Deductive Logic , year =. doi:10.1007/978-94-015-7622-2 , ranking =

    work page doi:10.1007/978-94-015-7622-2

  15. [15]

    Probabilistic evaluation of counterfactual queries , year =

    Alexander Balke and Judea Pearl , journal =. Probabilistic evaluation of counterfactual queries , year =

    work page

  16. [16]

    Spatial Interaction and the Statistical Analysis of Lattice Systems , year =

    Julian Besag , journal =. Spatial Interaction and the Statistical Analysis of Lattice Systems , year =

    work page

  17. [17]

    Probability and Measure , year =

    Billingsley, Patrick , publisher =. Probability and Measure , year =

    work page

  18. [18]

    and Hill, Christopher S

    Carlstrom, Ian F. and Hill, Christopher S. , journal =. Ernest W. Adams The Logic of Conditionals. Dordrecht: D. Reidel, 1975. xiii + 155 pp. 22.00. , year =

    work page 1975

  19. [19]

    2023 , note =

    Cheridito, Patrick and Eckstein, Stephan , title =. 2023 , note =. doi:10.48550/ARXIV.2303.14085 , eprint =

    work page doi:10.48550/arxiv.2303.14085 2023

  20. [20]

    , title =

    Clifford, Peter and Hammersley, John M. , title =. 1971 , publisher =

    work page 1971

  21. [21]

    Probability: Theory and Examples , year =

    Durrett, Richard , publisher =. Probability: Theory and Examples , year =

    work page

  22. [22]

    Maß- und Integrationstheorie , year =

    Elstrodt, Jürgen , publisher =. Maß- und Integrationstheorie , year =. doi:10.1007/978-3-662-57939-8 , ranking =

    work page doi:10.1007/978-3-662-57939-8

  23. [23]

    Logical and algorithmic properties of independence and their application to

    Geiger, Dan and Pearl, Judea , journal =. Logical and algorithmic properties of independence and their application to. 1990 , issn =. doi:10.1007/bf01531004 , publisher =

    work page doi:10.1007/bf01531004 1990

  24. [24]

    , journal =

    Holland, Paul W. , journal =. Statistics and Causal Inference , year =. doi:10.1080/01621459.1986.10478354 , publisher =

    work page doi:10.1080/01621459.1986.10478354 1986

  25. [25]

    Pearl's Calculus of Intervention Is Complete

    Huang, Yimin and Valtorta, Marco , title =. 2012 , note =. doi:10.48550/ARXIV.1206.6831 , keywords =

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.1206.6831 2012

  26. [26]

    Foundations of Modern Probability , year =

    Kallenberg, Olav , publisher =. Foundations of Modern Probability , year =. doi:10.1007/978-3-030-61871-1 , issn =

    work page doi:10.1007/978-3-030-61871-1

  27. [27]

    Lectures on Graphical Models , year =

    Lauritzen,. Lectures on Graphical Models , year =

    work page

  28. [28]

    Lewis , publisher =

    David K. Lewis , publisher =. Counterfactuals , year =

    work page

  29. [29]

    Lewis , journal =

    David K. Lewis , journal =. Causation , year =. doi:10.2307/2025310 , publisher =

    work page doi:10.2307/2025310

  30. [30]

    Counterfactual Theories of Causation , year =

    Menzies, Peter and Beebee, Helen , booktitle =. Counterfactual Theories of Causation , year =

    work page

  31. [31]

    Monge, Gaspard , publisher =. M. 1781 , added-at =

    work page

  32. [32]

    Gibbs and

    Moussouris, John , journal =. Gibbs and. 1974 , issn =. doi:10.1007/bf01011714 , publisher =

    work page doi:10.1007/bf01011714 1974

  33. [33]

    , publisher =

    Munkres, James R. , publisher =. Topology , year =

    work page

  34. [34]

    Introduction to Causal Inference from a Machine Learning Perspective: Course Lecture Notes , year =

    Brady Neal , publisher =. Introduction to Causal Inference from a Machine Learning Perspective: Course Lecture Notes , year =

    work page

  35. [35]

    A Measure-Theoretic Axiomatisation of Causality , year =

    Junhyung Park and Simon Buchholz and Bernhard Schölkopf and Krikamol Muandet , note =. A Measure-Theoretic Axiomatisation of Causality , year =. doi:10.48550/ARXIV.2305.17139 , eprint =

    work page doi:10.48550/arxiv.2305.17139

  36. [36]

    Probabilistic Reasoning in Intelligent Systems , year =

    Pearl, Judea , publisher =. Probabilistic Reasoning in Intelligent Systems , year =

    work page

  37. [37]

    Causal diagrams for empirical research , year =

    Pearl, Judea , journal =. Causal diagrams for empirical research , year =. doi:10.1093/biomet/82.4.669 , publisher =

    work page doi:10.1093/biomet/82.4.669

  38. [38]

    The Book of Why , year =

    Pearl, Judea and Mackenzie,Dana , editor =. The Book of Why , year =

    work page

  39. [39]

    Richardson and James M

    Thomas S. Richardson and James M. Robins , title =. 2013 , organization =

    work page 2013

  40. [40]

    Optimal Transport for Applied Mathematicians: Calculus of Variations, PDEs, and Modeling , year =

    Santambrogio, Filippo , publisher =. Optimal Transport for Applied Mathematicians: Calculus of Variations, PDEs, and Modeling , year =. Progress in Nonlinear Differential Equations and Their Applications , ranking =. doi:10.1007/978-3-319-20828-2 , issn =

    work page doi:10.1007/978-3-319-20828-2

  41. [41]

    A Course in the Calculus of Variations: Optimization, Regularity, and Modeling , year =

    Santambrogio, Filippo , publisher =. A Course in the Calculus of Variations: Optimization, Regularity, and Modeling , year =. doi:10.1007/978-3-031-45036-5 , issn =

    work page doi:10.1007/978-3-031-45036-5

  42. [42]

    Villani , Optimal Transport, vol

    Villani, Cédric , publisher =. Optimal Transport , year =. doi:10.1007/978-3-540-71050-9 , issn =

    work page doi:10.1007/978-3-540-71050-9

  43. [43]

    Funktionalanalysis , year =

    Werner, Dirk , publisher =. Funktionalanalysis , year =

    work page

  44. [44]

    Lecture Notes on Causal Inference , year =

    Qingyuan Zhao , publisher =. Lecture Notes on Causal Inference , year =

    work page

  45. [45]

    Identification of joint interventional distributions in recursive semi-Markovian causal models , year =

    Shpitser, Ilya and Pearl, Judea , booktitle =. Identification of joint interventional distributions in recursive semi-Markovian causal models , year =

    work page

  46. [46]

    Negative Controls: A Tool for Detecting Confounding and Bias in Observational Studies , year =

    Lipsitch, Marc and Tchetgen Tchetgen, Eric and Cohen, Ted , journal =. Negative Controls: A Tool for Detecting Confounding and Bias in Observational Studies , year =. doi:10.1097/ede.0b013e3181d61eeb , publisher =

    work page doi:10.1097/ede.0b013e3181d61eeb

  47. [47]

    A new approach to causal inference in mortality studies with a sustained exposure period—application to control of the healthy worker survivor effect , journal =

    Robins, James , journal =. A new approach to causal inference in mortality studies with a sustained exposure period—application to control of the healthy worker survivor effect , year =. doi:10.1016/0270-0255(86)90088-6 , publisher =

    work page doi:10.1016/0270-0255(86)90088-6

  48. [48]

    Causality: Models, Reasoning, and Inference , year =

    Pearl, Judea , publisher =. Causality: Models, Reasoning, and Inference , year =

    work page

  49. [49]

    Chapter 77 Linear Inverse Problems in Structural Econometrics Estimation Based on Spectral Decomposition and Regularization , year =

    Carrasco, Marine and Florens, Jean-Pierre and Renault, Eric , pages =. Chapter 77 Linear Inverse Problems in Structural Econometrics Estimation Based on Spectral Decomposition and Regularization , year =. Handbook of Econometrics , doi =

    work page

  50. [50]

    Causal Inference Under Unmeasured Confounding With Negative Controls: A Minimax Learning Approach , year =

    Nathan Kallus and Xiaojie Mao and Masatoshi Uehara , journal =. Causal Inference Under Unmeasured Confounding With Negative Controls: A Minimax Learning Approach , year =

    work page

  51. [51]

    Inference on strongly identified functionals of weakly identified functions , year =

    Bennett, Andrew and Kallus, Nathan and Mao, Xiaojie and Newey, Whitney K and Syrgkanis, Vasilis and Uehara, Masatoshi , journal =. Inference on strongly identified functionals of weakly identified functions , year =. doi:10.1093/jrsssb/qkaf075 , eprint =

    work page doi:10.1093/jrsssb/qkaf075

  52. [52]

    and Zhu, Y

    Mastouri, A. and Zhu, Y. and Gultchin, L. and Korba, A. and Silva, R. and Kusner, M. and Gretton, A. and Muandet, K. , booktitle =. Proximal Causal Learning with Kernels: Two-Stage Estimation and Moment Restriction , year =

    work page

  53. [53]

    Equivalence and synthesis of causal models , year =

    Verma, Thomas and Pearl, Judea , booktitle =. Equivalence and synthesis of causal models , year =

    work page

  54. [54]

    Identifiability in Causal Bayesian Networks:

    Yimin Huang and Marco Valtorta , booktitle =. Identifiability in Causal Bayesian Networks:. 2006 , pages =

    work page 2006

  55. [55]

    and Evans, Robin J

    Richardson, Thomas S. and Evans, Robin J. and Robins, James M. and Shpitser, Ilya , journal =. Nested Markov properties for acyclic directed mixed graphs , year =

    work page

  56. [56]

    Identification Is Easy, Factorization Is Hard , year =

    Ilya Shpitser , note =. Identification Is Easy, Factorization Is Hard , year =

    work page

  57. [57]

    and Ashworth, Jason J

    Sproston, Nicola R. and Ashworth, Jason J. , journal =. Role of C-Reactive Protein at Sites of Inflammation and Infection , year =. doi:10.3389/fimmu.2018.00754 , publisher =

    work page doi:10.3389/fimmu.2018.00754 2018

  58. [58]

    and Robins, J.M

    Hernan, M.A. and Robins, J.M. , publisher =. Causal Inference: What If , year =

    work page

  59. [59]

    Optimal treatment regimes for proximal causal learning , year =

    Shen, Tao and Cui, Yifan , booktitle =. Optimal treatment regimes for proximal causal learning , year =

    work page