pith. sign in

arxiv: 2606.10681 · v1 · pith:HV4OOWWXnew · submitted 2026-06-09 · 💰 econ.TH

Limited belief propagation and contingent thinking

Andrew Ellis , Ran Spiegler This is my paper

Pith reviewed 2026-06-27 11:06 UTC · model grok-4.3

classification 💰 econ.TH
keywords belief updatingcontingent thinkingcorrelation neglectdirected acyclic graphlimited inferenceiterated expectationssocial learningpublic goods
0
0 comments X

The pith

An agent's updated beliefs after observing variables are represented by limited propagation of implications through their relation graph, with shorter inference chains on unobserved variables producing correlation neglect.

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

The paper models belief updating as a process that stops short of fully propagating an observation's consequences across all related variables. This is formalized by allowing the number of inference steps to differ between observed and unobserved variables in the graph of their relations. When fewer steps are taken on unobserved variables, the resulting beliefs neglect correlations and violate iterated expectations. The approach reinterprets existing experimental findings on contingent thinking and applies the same structure to public-good contribution and social learning.

Core claim

We provide a representation of updated beliefs that captures limited propagation of her observation's implications through the directed acyclic graph that represents the relations between all variables. Failure of contingent thinking occurs when she performs fewer inference steps from unobserved variables than observed ones, leading to correlation neglect and violations of iterated expectations.

What carries the argument

Limited propagation of an observation's implications through the directed acyclic graph of variable relations, parameterized by the relative number of inference steps taken on observed versus unobserved variables.

If this is right

  • The model nests standard Bayesian updating when inference steps are equal across observed and unobserved variables.
  • It produces correlation neglect as a direct consequence of unequal step counts.
  • It generates violations of iterated expectations without assuming non-Bayesian priors.
  • It yields distinct predictions for contribution games and social learning compared with full Bayesian benchmarks.

Where Pith is reading between the lines

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

  • The same limited-step rule could be tested in sequential decision problems where agents receive private signals about a common state.
  • Making unobserved variables more salient might increase the number of inference steps taken on them and reduce the documented biases.
  • The framework suggests a way to quantify how much additional information is needed to restore full propagation in a given graph.

Load-bearing premise

The relations between all variables can be represented by a directed acyclic graph, and the agent's updating process can be parameterized by the number of inference steps performed on observed versus unobserved variables.

What would settle it

Measure whether subjects' beliefs about an unobserved variable in a known three-variable chain equal the iterated expectation computed from an observed variable, or instead reflect fewer inference steps.

Figures

Figures reproduced from arXiv: 2606.10681 by Andrew Ellis, Ran Spiegler.

Figure 1
Figure 1. Figure 1: A Kakuro puzzle (c) 2007 by Octahedron80, https://commons.wikimedia.org/wiki/File:Kakuro_black_box.svg 3.1. New Experiments? Kakuro. In this subsection we propose a direction for new experiments on contingent thinking, using the recreational puzzle game Kakuro, also called Cross-Sum, as a template. The player’s objective in this game is to fill a crossword-puzzle-like grid with numbers ( [PITH_FULL_IMAGE:… view at source ↗
read the original abstract

An agent updates her beliefs over a set of variables after observing some of them. We provide a representation of updated beliefs that captures limited propagation of her observation's implications through the directed acyclic graph that represents the relations between all variables. Failure of contingent thinking occurs when she performs fewer inference steps from unobserved variables than observed ones, leading to correlation neglect and violations of iterated expectations. Our framework offers a new perspective on existing experiments about contingent thinking and suggests new directions. We characterize the model's relationship with familiar Bayesian and non-Bayesian benchmarks, and illustrate it with applications to public-good provision and social learning games.

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

2 major / 2 minor

Summary. The manuscript presents a representation of updated beliefs based on limited propagation of observations through a directed acyclic graph (DAG) of variable interdependencies. Failures of contingent thinking are formalized as agents conducting fewer inference steps from unobserved variables than from observed ones, which generates correlation neglect and violations of iterated expectations. The framework is characterized in relation to Bayesian updating and non-Bayesian models, and applied to public-good provision and social learning games.

Significance. If the representation is formally established and the applications yield new insights, this model could serve as a useful tool for incorporating bounded rationality in belief formation into economic analysis. It provides a graph-theoretic approach to modeling limited contingent thinking that may explain experimental findings and guide future research in information economics and game theory.

major comments (2)
  1. Abstract: The central representation is described at a high level but the manuscript must supply the formal definition of the limited belief propagation operator and the step-count parameter to allow verification that the consequences (correlation neglect, iterated expectation violations) follow from the assumptions rather than being built in by construction.
  2. Abstract, paragraph 2: The assumption that variable relations are captured by a DAG and that updating is parameterized solely by the number of inference steps on observed vs. unobserved variables is load-bearing for the contingent thinking failure claim; the paper should discuss the robustness of results to alternative graph structures or step functions.
minor comments (2)
  1. Ensure that all notation for the DAG and inference steps is defined consistently throughout the paper.
  2. The applications section could include a brief comparison table of predictions under the new model versus standard Bayesian updating.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive comments. We respond to each major comment below.

read point-by-point responses

  1. Referee: Abstract: The central representation is described at a high level but the manuscript must supply the formal definition of the limited belief propagation operator and the step-count parameter to allow verification that the consequences (correlation neglect, iterated expectation violations) follow from the assumptions rather than being built in by construction.

    Authors: The formal definition of the limited belief propagation operator (denoted Φ_k) and the step-count parameter k appear in Definition 1 of Section 2. The operator is defined recursively along paths in the DAG, applying full Bayesian updating from observed variables but truncating propagation after k steps from unobserved variables. Theorem 1 derives correlation neglect and iterated-expectations violations as direct consequences of this truncation when k is finite and asymmetric; they are not imposed by construction. To address the abstract-level concern we will insert a concise statement of the operator and parameter into the abstract. revision: yes

  2. Referee: Abstract, paragraph 2: The assumption that variable relations are captured by a DAG and that updating is parameterized solely by the number of inference steps on observed vs. unobserved variables is load-bearing for the contingent thinking failure claim; the paper should discuss the robustness of results to alternative graph structures or step functions.

    Authors: Section 4.2 already examines robustness to alternative step functions (linear, exponential, and threshold) and shows that the core contingent-thinking failure persists whenever propagation depth is strictly smaller for unobserved variables. For graph structures we will add a new paragraph discussing extensions to graphs containing cycles (via local approximations) and to non-DAG representations that preserve the observed/unobserved asymmetry; the qualitative results remain intact under these alternatives. revision: partial

Circularity Check

0 steps flagged

No significant circularity detected

full rationale

The paper offers an axiomatic representation of belief updating via limited propagation steps on a DAG, with contingent-thinking failure defined directly in terms of asymmetric step counts on observed versus unobserved variables. This framework is presented as a modeling choice that captures correlation neglect and iterated-expectations violations by construction of the representation itself, rather than as a derived prediction from independent primitives. No load-bearing equations, fitted parameters renamed as predictions, or self-citation chains appear in the provided text; the contribution is explicitly positioned as a new representation that relates to Bayesian benchmarks without reducing to its own inputs.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

Only abstract available; ledger entries are inferred at the level of stated modeling choices rather than explicit equations.

axioms (2)
  • domain assumption Relations between variables are represented by a directed acyclic graph.
    Abstract states that the DAG represents the relations between all variables.
  • domain assumption Updated beliefs are represented by the number of inference steps performed on observed versus unobserved variables.
    Core modeling choice described in the abstract.

pith-pipeline@v0.9.1-grok · 5610 in / 1257 out tokens · 26984 ms · 2026-06-27T11:06:21.239124+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

49 extracted references · 17 canonical work pages

  1. [1]

    Mental models of causal structure in economics and psychology

    Sandro Ambuehl, Rahul Bhui, and Heidi C Thysen. Mental models of causal structure in economics and psychology. arXiv preprint arXiv:2603.29070, 2026

    arXiv 2026

  2. [2]

    Calford and Timothy N

    Evan M. Calford and Timothy N. Cason. Contingent reasoning and dynamic public goods provision. American Economic Journal: Microeconomics, 16 0 (2): 0 236--266, 2024. doi:10.1257/mic.20220111

    work page doi:10.1257/mic.20220111 2024

  3. [3]

    Caron and Tim Traynor

    Richard J. Caron and Tim Traynor. The zero set of a polynomial. Technical Report WMSR 05-02, Department of Mathematics and Statistics, University of Windsor, 2005

    2005

  4. [4]

    The origin of the winner's curse: A laboratory study

    Gary Charness and Dan Levin. The origin of the winner's curse: A laboratory study. American Economic Journal: Microeconomics, 1 0 (1): 0 207--236, 2009. doi:10.1257/mic.1.1.207

    work page doi:10.1257/mic.1.1.207 2009

  5. [5]

    Sequential cursed equilibrium

    Shani Cohen and Shengwu Li. Sequential cursed equilibrium. American Economic Review, 116 0 (3): 0 934--976, 2026

    2026

  6. [6]

    Cowell, A

    Robert G. Cowell, A. Philip Dawid, Steffen L. Lauritzen, and David J. Spiegelhalter. Probabilistic Networks and Expert Systems. Springer, New York, 1999

    1999

  7. [7]

    Correlation misperception in choice

    Andrew Ellis and Michele Piccione. Correlation misperception in choice. American Economic Review, 107 0 (4): 0 1264--92, April 2017. doi:10.1257/aer.20160093

    work page doi:10.1257/aer.20160093 2017

  8. [8]

    Subjective causality in choice

    Andrew Ellis and Heidi Thysen. Subjective causality in choice. Working paper, 2025

    2025

  9. [9]

    Enke and F

    B. Enke and F. Zimmerman. Correlation neglect in belief formation. Review of Economic Studies, 86, 2019

    2019

  10. [10]

    Behavioral equilibrium in economies with adverse selection

    Ignacio Esponda. Behavioral equilibrium in economies with adverse selection. American Economic Review, 98 0 (4): 0 1269--1291, 2008. doi:10.1257/aer.98.4.1269

    work page doi:10.1257/aer.98.4.1269 2008

  11. [11]

    Hypothetical thinking and information extraction in the laboratory

    Ignacio Esponda and Emanuel Vespa. Hypothetical thinking and information extraction in the laboratory. American Economic Journal: Microeconomics, 6 0 (4): 0 180--202, 2014. doi:10.1257/mic.6.4.180

    work page doi:10.1257/mic.6.4.180 2014

  12. [12]

    Contingent thinking and the sure-thing principle: Revisiting classic anomalies in the laboratory

    Ignacio Esponda and Emanuel Vespa. Contingent thinking and the sure-thing principle: Revisiting classic anomalies in the laboratory. The Review of Economic Studies, 91 0 (5): 0 2806--2831, 10 2024

    2024

  13. [13]

    Cursed equilibrium

    Erik Eyster and Matthew Rabin. Cursed equilibrium. Econometrica, 73 0 (5): 0 1623--1672, 2005

    2005

  14. [14]

    Extensive imitation is irrational and harmful

    Erik Eyster and Matthew Rabin. Extensive imitation is irrational and harmful. The Quarterly Journal of Economics, 129 0 (4): 0 1861--1898, 2014

    2014

  15. [15]

    Correlation neglect in financial decision making

    Erik Eyster and Georg Weizs \"a cker. Correlation neglect in financial decision making. Mimeo, 2010

    2010

  16. [16]

    Cursed sequential equilibrium

    Meng-Jhang Fong, Po-Hsuan Lin, and Thomas R Palfrey. Cursed sequential equilibrium. American Economic Review, 115 0 (8): 0 2616--2658, 2025

    2025

  17. [17]

    Monty hall's three doors: Construction and deconstruction of a choice anomaly

    Daniel Friedman. Monty hall's three doors: Construction and deconstruction of a choice anomaly. American Economic Review, 88 0 (4): 0 933--946, 1998

    1998

  18. [18]

    What comes to mind

    Nicola Gennaioli and Andrei Shleifer. What comes to mind. Quarterly Journal of Economics, 125 0 (4): 0 1399--1433, 2010. doi:10.1162/qjec.2010.125.4.1399

    work page doi:10.1162/qjec.2010.125.4.1399 2010

  19. [19]

    Loopy belief propagation: convergence and effects of message errors

    Alexander T Ihler, John W Fisher III, Alan S Willsky, and David Maxwell Chickering. Loopy belief propagation: convergence and effects of message errors. Journal of Machine Learning Research, 6 0 (5), 2005

    2005

  20. [20]

    Jakobsen

    Alexander M. Jakobsen. A model of complex contracts. American Economic Review, 110 0 (5): 0 1243--1273, 2020. doi:10.1257/aer.20190283

    work page doi:10.1257/aer.20190283 2020

  21. [21]

    Limited foresight may force cooperation

    Philippe Jehiel. Limited foresight may force cooperation. The Review of Economic Studies, 68 0 (2): 0 369--391, 2001

    2001

  22. [22]

    Revisiting games of incomplete information with analogy-based expectations

    Philippe Jehiel and Fr \'e d \'e ric Koessler. Revisiting games of incomplete information with analogy-based expectations. Games and Economic Behavior, 62 0 (2): 0 533--557, 2008. doi:10.1016/j.geb.2007.06.006

    work page doi:10.1016/j.geb.2007.06.006 2008

  23. [23]

    Kagel and Dan Levin

    John H. Kagel and Dan Levin. The winner's curse and public information in common value auctions. American Economic Review, 76 0 (5): 0 894--920, 1986

    1986

  24. [24]

    Kagel, Ronald M

    John H. Kagel, Ronald M. Harstad, and Dan Levin. Information impact and allocation rules in auctions with affiliated private values: A laboratory study. Econometrica, 55 0 (6): 0 1275--1304, 1987

    1987

  25. [25]

    Boundedly rational backward induction

    Shaowei Ke. Boundedly rational backward induction. Theoretical Economics, 14 0 (1): 0 103--134, 2019

    2019

  26. [26]

    Probabilistic Graphical Models: Principles and Techniques

    Daphne Koller and Nir Friedman. Probabilistic Graphical Models: Principles and Techniques. MIT Press, Cambridge, MA, 2009

    2009

  27. [27]

    A general algorithm for approximate inference and its application to hybrid bayes nets

    Daphne Koller, Uri Lerner, and Dragomir Anguelov. A general algorithm for approximate inference and its application to hybrid bayes nets. arXiv preprint arXiv:1301.6709, 2013

    Pith/arXiv arXiv 2013

  28. [28]

    Kagel, and Jean-Francois Richard

    Dan Levin, John H. Kagel, and Jean-Francois Richard. Revenue effects and information processing in english common value auctions. The American Economic Review, 86 0 (3): 0 442--460, 1996

    1996

  29. [29]

    Does polarization of opinions lead to polarization of platforms? the case of correlation neglect

    Gilat Levy and Ronny Razin. Does polarization of opinions lead to polarization of platforms? the case of correlation neglect. Quarterly Journal of Political Science, 10 0 (3): 0 321--355, 2015

    2015

  30. [30]

    Obviously strategy-proof mechanisms

    Shengwu Li. Obviously strategy-proof mechanisms. American Economic Review, 107 0 (11): 0 3257--3287, 2017. doi:10.1257/aer.20160425

    work page doi:10.1257/aer.20160425 2017

  31. [31]

    Barton L. Lipman. Decision theory without logical omniscience: Toward an axiomatic framework for bounded rationality. Review of Economic Studies, 66 0 (2): 0 339--361, 1999. doi:10.1111/1467-937X.00090

    work page doi:10.1111/1467-937x.00090 1999

  32. [32]

    Failures in contingent reasoning: The role of uncertainty

    Alejandro Martinez-Marquina, Muriel Niederle, and Emanuel Vespa. Failures in contingent reasoning: The role of uncertainty. American Economic Review, 109 0 (10): 0 3437--3474, 2019. doi:10.1257/aer.20171764

    work page doi:10.1257/aer.20171764 2019

  33. [33]

    A measure of complexity for strategy-proof mechanisms

    Lea Nagel and Roberto Saitto. A measure of complexity for strategy-proof mechanisms. In EC, page 1017, 2023

    2023

  34. [34]

    Kathleen Ngangou \'e and Georg Weizs \"a cker

    M. Kathleen Ngangou \'e and Georg Weizs \"a cker. Learning from unrealized versus realized prices. American Economic Journal: Microeconomics, 13 0 (2): 0 174--201, 2021. doi:10.1257/mic.20180289

    work page doi:10.1257/mic.20180289 2021

  35. [35]

    Cognitive limitations: Failures of contingent thinking

    Muriel Niederle and Emanuel Vespa. Cognitive limitations: Failures of contingent thinking. Annual Review of Economics, 15: 0 307--328, 2023. doi:10.1146/annurev-economics-091622-124733

    work page doi:10.1146/annurev-economics-091622-124733 2023

  36. [36]

    Modeling the change of paradigm: Non- B ayesian reactions to unexpected news

    Pietro Ortoleva. Modeling the change of paradigm: Non- B ayesian reactions to unexpected news. American Economic Review, 102 0 (6): 0 2410--2436, 2012. doi:10.1257/aer.102.6.2410

    work page doi:10.1257/aer.102.6.2410 2012

  37. [37]

    Alternatives to B ayesian updating

    Pietro Ortoleva. Alternatives to B ayesian updating. Annual Review of Economics, 16: 0 545--570, 2024. doi:10.1146/annurev-economics-100223-050352

    work page doi:10.1146/annurev-economics-100223-050352 2024

  38. [38]

    Overconfidence in political behavior

    Pietro Ortoleva and Erik Snowberg. Overconfidence in political behavior. American Economic Review, 105 0 (2): 0 504--35, 2015

    2015

  39. [39]

    Martin J. Osborne. An Introduction to Game Theory. Oxford University Press, New York, 2004

    2004

  40. [40]

    Causality: Models, Reasoning, and Inference

    Judea Pearl. Causality: Models, Reasoning, and Inference. Cambridge University Press, Cambridge, 2nd edition, 2009

    2009

  41. [41]

    The Book of Why: The New Science of Cause and Effect

    Judea Pearl and Dana Mackenzie. The Book of Why: The New Science of Cause and Effect. Basic Books, New York, 2018

    2018

  42. [42]

    Failures of contingent thinking

    Evan Piermont and Peio Zuazo-Garin. Failures of contingent thinking. arXiv:2007.07703 [econ.TH], revised January 2026, 2026. URL https://arxiv.org/abs/2007.07703

    arXiv 2007

  43. [43]

    Modeling bounded rationality

    Ariel Rubinstein. Modeling bounded rationality. MIT press, 1998

    1998

  44. [44]

    `` I sn't everyone like me'': On the presence of self-similarity in strategic interactions

    Ariel Rubinstein and Yuval Salant. `` I sn't everyone like me'': On the presence of self-similarity in strategic interactions. Judgment and Decision Making, 2016

    2016

  45. [45]

    Samuelson and Max H

    William F. Samuelson and Max H. Bazerman. The winner's curse in bilateral negotiations. In Vernon L. Smith, editor, Research in Experimental Economics, volume 3, pages 105--137. JAI Press, Greenwich, CT, 1985

    1985

  46. [46]

    On the M onty H all problem

    Steve Selvin. On the M onty H all problem. The American Statistician, 29 0 (3): 0 134, 1975. Letter to the Editor

    1975

  47. [47]

    B ayesian networks and boundedly rational expectations

    Ran Spiegler. B ayesian networks and boundedly rational expectations. Quarterly Journal of Economics, 131 0 (3): 0 1243--1290, 2016. doi:10.1093/qje/qjw011

    work page doi:10.1093/qje/qjw011 2016

  48. [48]

    Can agents with causal misperceptions be systematically fooled? Journal of the European Economic Association, 18 0 (2): 0 583--617, 2020

    Ran Spiegler. Can agents with causal misperceptions be systematically fooled? Journal of the European Economic Association, 18 0 (2): 0 583--617, 2020

    2020

  49. [49]

    Pseudo- B ayesian updating

    Chen Zhao. Pseudo- B ayesian updating. Theoretical Economics, 17 0 (1): 0 253--289, 2022. doi:10.3982/TE4535

    work page doi:10.3982/te4535 2022