Pith. sign in

REVIEW 2 major objections 5 minor 37 references

Random Utility Model with Endogenously Assigned Menus

T0 review · 2 major / 5 minor · reviewed 2026-07-10 · grok-4.5

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

Pith's one-line read Any choice pattern can be produced by rational people when menus are assigned according to preferences; the paper recovers the causal shares that would arise if every menu were shown to everyone.

desk verdict Clean identification paper: endogenous menus kill RUM testability, and the author gives sharp closed-form bounds plus multi-market point ID under a clear steering restriction. read the letter →

arxiv 2607.08218 v1 pith:PSHOF4PR submitted 2026-07-09 econ.TH

Kai Wang This is my paper
classification econ.TH
keywords randomutilitymodelendogenousmenuscounterfactualchoiceprobabilitiespartialidentificationoutcome-basedsteeringtransportationfeasibilitystochastic
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

Stochastic choice theory has long treated menus as if they were randomly assigned to people. In practice, platforms, brokers, and retailers show menus that depend on what they think a person will like, so the people facing one menu are a selected group. The paper shows that once that selection is allowed, every observed choice pattern—no matter how anomalous—can be generated by a population of perfectly rational decision makers. The right object of study is therefore not the observed choice frequency but a counterfactual: the share that would choose an alternative if the menu were shown to the whole population. Using only the joint frequencies of menus and choices, the paper derives sharp bounds on those counterfactuals and on the assignment rule itself. When the same assignment rule operates across several markets whose preference mixes differ, both the counterfactuals and the rule become uniquely identified.

What carries the argument

The supermodular set function r that aggregates observed cell masses whose top cones lie inside a given set of rankings; by a classical core theorem its values at a top cone and its complement are exactly the sharp lower and upper bounds on the corresponding counterfactual share.

What would settle it

In multi-market data, check whether the market-by-alternative matrix of observed cell masses has full column rank equal to the number of alternatives on the menu; if the recovered steering probabilities fall outside (0,1] or fail to sum to one across menus for some ranking, the joint hypothesis is rejected.

Watch

Extended reading notes

Core claim

Under endogenous menu assignment any dataset of menu frequencies and conditional choice probabilities is rationalizable by some random-utility distribution, so classical axioms lose all testable content; the sharp identified set for the counterfactual choice probability ρ*(a|A) is nevertheless a closed interval whose endpoints are finite sums of the observed cell masses.

Load-bearing premise

For unique recovery the platform must assign each menu according only to the single alternative a person is predicted to choose from that menu, and must use the same rule in every market.

Editorial extensions

If this is right

  • Observed violations of Regularity or IIA can no longer be read as evidence against rationality when menus are endogenously assigned.
  • Analysts already record menu frequencies; those frequencies convert uninterpretable choice data into sharp bounds on the causal estimand.
  • The same bounds diagnose how strongly a platform steers people toward menus that match their predicted choice.
  • With enough markets and a common outcome-based rule, both the causal shares and the assignment probabilities are point-identified without covariates.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • Dynamic recommenders that update on past choices may generate the cross-period variation needed for identification without extra markets.
  • The same transportation argument could bound counterfactual consideration sets when both menus and attention are endogenous.
  • Insurance and retail scanner panels already contain the required menu and market variation; the bounds can be computed by summation alone.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

2 major / 5 minor

Summary. The paper studies the Random Utility Model when menus are assigned endogenously according to preferences rather than by an exogenous design. It shows that any observed pair of menu frequencies and conditional choice probabilities can be rationalized by some population of rational types and some assignment rule (Theorem 1). It then defines the causal estimand ρ*(a|A) as the population share that would choose a if menu A were assigned to everyone, and characterizes the sharp identified set for ρ* (and for steering probabilities) from observed (μ, ρ) via a transportation feasibility argument (Theorem 2), with closed-form endpoints obtained from a supermodular set function of cell masses (Theorem 3). Under common, outcome-based steering across markets with a full-rank market-by-alternative matrix, both the steering rule and market-level counterfactuals are point identified (Theorem 4). An appendix works the motivating decoy-style example through all three results.

Significance. The paper identifies a genuine and under-appreciated identification problem in stochastic choice: cross-menu axioms and interpretations of ρ tacitly treat menu assignment as a hidden RCT. The 'anything goes' result (Theorem 1) cleanly shows that rationality loses all testable content once endogeneity is allowed, while the subsequent partial- and point-identification results restore a well-defined causal object. The use of Gale feasibility and the Shapley core of a supermodular set function is technically clean and yields closed-form bounds that are immediately computable from cell masses. The multi-market point-identification result is a natural and falsifiable extension. Strengths include complete classical proofs, an explicit worked example recovering both bounds and unique identification, and a clear separation between what requires no structure on assignment (sharp bounds) and what requires outcome-based common steering (point identification). If the results hold as stated, the paper should become a standard reference for applied work that uses scanner, platform, or insurance choice data where menus are selected.

major comments (2)
  1. [Section 5, Assumptions 1–2, Theorem 4] Section 5, Assumptions 1–2 and Theorem 4: Point identification rests on outcome-based common steering and rank(DA)=|A|. The paper correctly flags this as the price of uniqueness and notes that exogenous assignment is nested. For applied credibility it would help to state more explicitly what fails under mild violations—e.g., if assignment depends on the full ranking or on market-specific covariates correlated with preferences—and whether the sharp bounds of Theorems 2–3 remain the natural fallback. A short paragraph on this robustness hierarchy would strengthen the bridge from theory to the platform/retail settings motivating the paper.
  2. [Corollary 1, Theorem 3(ii)] Corollary 1 and Theorem 3(ii): The bounds apply to menus never observed and to alternatives never chosen, which is attractive. The closed form makes clear that the lower bound on ρ*(a|A) is positive only when some observed B⊇A records choice of a. When no such superset appears, the lower bound is zero and the upper bound may be loose. The paper should briefly discuss when the identified intervals are informative versus vacuous in typical datasets (e.g., nested vs non-nested observed menus), so readers can judge empirical usefulness beyond the tight motivating example.
minor comments (5)
  1. [Introduction] Introduction, near the end of the multi-market paragraph: 'I also how one can use this result' is missing the verb 'show'.
  2. [References / Literature Review] Several author names in the references and literature review have encoding artifacts (e.g., Barber´ a, Horta¸ csu, J¨ org). Clean these for production.
  3. [Introduction, Eq. (1)] Equation (1) and the surrounding text introduce ρ* before the formal framework in Section 3; a forward reference to the definition of T(a,A) would help readers who skip ahead from the introduction.
  4. [Appendix A] Appendix A.1–A.3 is excellent for verification. Consider adding one sentence in the main text pointing readers to the fact that the closed-form bounds and the three-market linear system recover the same numbers as the transportation inequalities, so the appendix is not merely illustrative but a consistency check.
  5. [Section 2] In the literature review, the contrast with Kono, Saito, and Sandroni (2025) is sharp and useful; a parallel one-sentence contrast with Manski (2007) already appears earlier—consider moving both into a single short 'what is missing vs what is contaminated' paragraph for clarity.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: pure transportation / supermodularity / linear-algebra identification with no fitted parameters or self-referential definitions.

full rationale

The paper's chain is self-contained classical revealed-preference analysis. Theorem 1 constructs an explicit rationalizing joint π by placing each observed cell mass d(A,a) on a single order that ranks a first; this is a pure existence proof, not a fit. Theorem 2 translates consistency of ν into a transportation problem whose feasibility conditions are exactly Gale's theorem; the identified set Θ(μ,ρ) is therefore the set of supplies that can meet the observed demands under the top-cone support restriction. Theorem 3 shows that the same set coincides with the core of the supermodular set function r that simply sums the cell masses whose cones lie inside a given set of orders; the closed-form bounds on ρ*(a|A) are then the standard core evaluations min = r(T(a,A)) and max = 1-r(L\T(a,A)). Theorem 4 imposes the economically motivated factorization dm(A,a)=γ(A,a)ρ*m(a|A) that follows from Assumptions 1–2 and recovers γ and the ρ*m by ordinary linear algebra once rank(DA)=|A|. The target estimand ρ* is defined independently of the observed conditional frequencies ρ (it is the population measure of the top cone under exogenous assignment). No parameter is fitted to data and then re-used as a prediction; there are no self-citations of uniqueness theorems; the classical results of Gale and Shapley are external and parameter-free. Consequently the derivation does not reduce to its inputs by construction.

Assumptions & free parameters 0 free parameters · 5 assumptions · 2 invented entities

The paper works inside classical finite RUM. No free parameters are estimated. The load-bearing modeling choices are the maintained maximization hypothesis, the definition of the causal estimand, and (for point ID) the two steering assumptions. No new physical or economic entities are postulated beyond the standard preference types and the assignment rule.

assumptions (5)
  • domain assumption Each decision maker is characterized by a strict linear order and chooses the maximal element of the assigned menu (standard RUM maximization).
    Maintained throughout; used to define top cones T(a,A) and to restrict the support of the transport plan f (Section 3 and Theorem 2).
  • domain assumption The analyst observes the joint distribution of menus and choices (μ,ρ) or equivalently the cell masses d(A,a).
    Data requirement stated in Section 3; without μ the bounds cannot be formed.
  • ad hoc to paper Assumption 1 (common steering): the conditional menu-assignment rule is identical across markets.
    Imposed in Section 5 to obtain the factorization dm(A,a)=γ(A,a)ρ*m(a|A).
  • ad hoc to paper Assumption 2 (outcome-based steering): π(A|≻) depends on ≻ only through max(A,≻).
    Key restriction that makes γ(A,a) constant on each top cone and yields the linear system DA x = 1M (Section 5).
  • standard math Gale’s Feasibility Theorem (1957) and Shapley’s theorem on cores of supermodular set functions (1971).
    Invoked as Lemmas 1 and 3 to convert transport feasibility into the inequality description of Θ(μ,ρ) and the closed-form bounds.
invented entities (2)
  • Counterfactual choice probability ρ*(a|A) = ν(T(a,A))
    purpose: Causal estimand that restores an exogenous-assignment interpretation under endogenous menus.
    Defined in equation (1) and Section 3; the entire identification analysis targets this object.
  • Steering probability π(A|T(a,A))
    purpose: Diagnostic for the degree of menu endogeneity; bounded by the same program that bounds ρ*.
    Introduced after Corollary 1; used in Corollary 2 and the multi-market section.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Random Utility Model with Endogenously Assigned Menus." pith.science (2026). https://pith.science/paper/PSHOF4PR

@misc{pith2026260708218,
  author       = {Pith},
  title        = {Pith review of: Random Utility Model with Endogenously Assigned Menus},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/PSHOF4PR}},
  note         = {Machine review of arXiv:2607.08218}
}
read the original abstract

There is a largely overlooked assumption underlying stochastic choice theory: menus are assigned exogenously, as if by a hidden randomized controlled trial. This assumption is not innocuous, because in many real-world settings menus are assigned endogenously according to the decision makers' preferences. This paper studies Random Utility Model under menu endogeneity and shows that any seemingly anomalous choice behavior can be generated by a population of rational decision makers with heterogeneous preferences facing endogenously assigned menus. To address this problem, I propose a new causal estimand: the probability that an alternative would be chosen from a given menu if that menu were presented to the entire population. I then characterize sharp bounds on this estimand using observed menu and choice frequencies. In addition, I show that when choices are observed across multiple markets with different preference distributions and a common menu-assignment rule, the causal estimands from those markets and the menu assignment rule are uniquely identified.

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

37 extracted references · 37 canonical work pages

  1. [1]

    Abaluck, Jason, and Abi Adams-Prassl. 2021. ``What Do Consumers Consider Before They Choose? Identification from Asymmetric Demand Responses.'' Quarterly Journal of Economics, 136(3): 1611--1663

    work page 2021

  2. [2]

    Agarwal, Nikhil, and Paulo Somaini. 2026. ``Demand Analysis under Latent Choice Constraints.'' Review of Economic Studies, 93(4), 2215--2249

    work page 2026

  3. [3]

    Aguiar, Victor H., and Nail Kashaev. 2025. ``Identification and Estimation of Discrete Choice Models with Unobserved Choice Sets.'' Journal of Business & Economic Statistics, 43(1): 204--215

    work page 2025

  4. [4]

    Aguiar, Victor H., Maria Jose Boccardi, Nail Kashaev, and Jeongbin Kim. 2023. ``Random Utility and Limited Consideration.'' Quantitative Economics, 14(1): 71--116

    work page 2023

  5. [5]

    Ballester, and Jay Lu

    Apesteguia, Jose, Miguel A. Ballester, and Jay Lu. 2017. ``Single-Crossing Random Utility Models.'' Econometrica 85: 661--674

    work page 2017

  6. [6]

    Ballester

    Apesteguia, Jose, and Miguel A. Ballester. 2023. ``Random Utility Models with Ordered Types and Domains.'' Journal of Economic Theory 211: 105674

    work page 2023

  7. [7]

    Pattanaik

    Barber \`a , Salvador, and Prasanta K. Pattanaik. 1986. ``Falmagne and the Rationalizability of Stochastic Choices in Terms of Random Orderings.'' Econometrica 54: 707--715

    work page 1986

  8. [8]

    Teitelbaum

    Barseghyan, Levon, Maura Coughlin, Francesca Molinari, and Joshua C. Teitelbaum. 2021. ``Heterogeneous Choice Sets and Preferences.'' Econometrica, 89(5): 2015--2048

    work page 2021

Show all 37 references
  1. [9]

    D., and Jacob Marschak

    Block, H. D., and Jacob Marschak. 1960. ``Random Orderings and Stochastic Theories of Responses.'' In Contributions to Probability and Statistics: Essays in Honor of Harold Hotelling, edited by I. Olkin, S. G. Ghurye, W. Hoeffding, W. G. Madow, and H. B. Mann, 97--132. Stanfor...

    1960

  2. [10]

    Caplin, Andrew, and Mark Dean. 2015. ``Revealed Preference, Rational Inattention, and Costly Information Acquisition.'' American Economic Review, 105(7): 2183--2203

    2015

  3. [11]

    Caradonna, Peter P., and Christopher Turansick. 2026. ``Identification in Stochastic Choice.'' Working paper

    2026

  4. [12]

    Cattaneo, Matias D., Xinwei Ma, Yusufcan Masatlioglu, and Elchin Suleymanov. 2020. ``A Random Attention Model.'' Journal of Political Economy, 128(7): 2796--2836

    2020

  5. [13]

    Chambers, Christopher P., and Christopher Turansick. 2025. ``The Limits of Identification in Discrete Choice.'' Games and Economic Behavior 150: 537--551

    2025

  6. [14]

    Conlon, Christopher T., and Julie Holland Mortimer. 2013. ``Demand Estimation under Incomplete Product Availability.'' American Economic Journal: Microeconomics, 5(4): 1--30

    2013

  7. [15]

    Covington, Paul, Jay Adams, and Emre Sargin. 2016. ``Deep Neural Networks for YouTube Recommendations.'' In Proceedings of the 10th ACM Conference on Recommender Systems, 191--198. New York: Association for Computing Machinery

    2016

  8. [16]

    Crawford, Gregory S., Rachel Griffith, and Alessandro Iaria. 2021. ``A Survey of Preference Estimation with Unobserved Choice Set Heterogeneity.'' Journal of Econometrics, 222(1): 4--43

    2021

  9. [17]

    Dardanoni, Valentino, Paola Manzini, Marco Mariotti, and Christopher J. Tyson. 2020. ``Inferring Cognitive Heterogeneity from Aggregate Choices.'' Econometrica, 88(3): 1269--1296

    2020

  10. [18]

    Dardanoni, Valentino, Paola Manzini, Marco Mariotti, Henrik Petri, and Christopher J. Tyson. 2023. ``Mixture Choice Data: Revealing Preferences and Cognition.'' Journal of Political Economy, 131(3): 687--715

    2023

  11. [19]

    Falmagne, Jean-Claude. 1978. ``A Representation Theorem for Finite Random Scale Systems.'' Journal of Mathematical Psychology, 18(1): 52--72

    1978

  12. [20]

    Fiorini, Samuel. 2004. ``A Short Proof of a Theorem of Falmagne.'' Journal of Mathematical Psychology, 48(1): 80--82

    2004

  13. [21]

    Gale, David. 1957. ``A Theorem on Flows in Networks.'' Pacific Journal of Mathematics, 7(2): 1073--1082

    1957

  14. [22]

    Gaynor, Martin, Carol Propper, and Stephan Seiler. 2016. ``Free to Choose? Reform, Choice, and Consideration Sets in the English National Health Service.'' American Economic Review, 106(11): 3521--3557

    2016

  15. [23]

    Gibbard, Peter. 2021. ``Disentangling Preferences and Limited Attention: Random-Utility Models with Consideration Sets.'' Journal of Mathematical Economics, 94: 102468

    2021

  16. [24]

    Gul, Faruk, and Wolfgang Pesendorfer. 2006. ``Random Expected Utility.'' Econometrica, 74(1): 121--146

    2006

  17. [25]

    Honka, Elisabeth, Ali Horta c su, and Maria Ana Vitorino. 2017. ``Advertising, Consumer Awareness, and Choice: Evidence from the U.S. Banking Industry.'' RAND Journal of Economics, 48(3): 611--646

    2017

  18. [26]

    Kitamura, Yuichi, and J \"o rg Stoye. 2018. ``Nonparametric Analysis of Random Utility Models.'' Econometrica 86: 1883--1909

    2018

  19. [27]

    Knott, Aaron, Andrew Hayes, and Scott A. Neslin. 2002. ``Next-Product-to-Buy Models for Cross-Selling Applications.'' Journal of Interactive Marketing, 16(3): 59--75

    2002

  20. [28]

    Kono, Haruki, Kota Saito, and Alec Sandroni. 2025. ``Random Utility with Unobservable Alternatives.'' American Economic Review, forthcoming

    2025

  21. [29]

    Liu, Hui, and Yao Luo. 2025. ``Demand Analysis under Price Rigidity and Endogenous Assortment: An Application to China's Tobacco Industry.'' Working paper, arXiv:2501.17251

    2025 arXiv

  22. [30]

    Lops, Pasquale, Marco de Gemmis, and Giovanni Semeraro. 2011. ``Content-Based Recommender Systems: State of the Art and Trends.'' In Recommender Systems Handbook, edited by Francesco Ricci, Lior Rokach, Bracha Shapira, and Paul B. Kantor, 73--105. Boston: Springer

    2011

  23. [31]

    Manski, Charles F. 2007. ``Partial Identification of Counterfactual Choice Probabilities.'' International Economic Review, 48(4): 1393--1410

    2007

  24. [32]

    Manzini, Paola, and Marco Mariotti. 2014. ``Stochastic Choice and Consideration Sets.'' Econometrica, 82(3): 1153--1176

    2014

  25. [33]

    Masatlioglu, Yusufcan, Daisuke Nakajima, and Erkut Y. Ozbay. 2012. ``Revealed Attention.'' American Economic Review, 102(5): 2183--2205

    2012

  26. [34]

    McFadden, Daniel, and Marcel K. Richter. 1991. ``Stochastic Rationality and Revealed Stochastic Preference.'' In Preferences, Uncertainty and Rationality,, edited by J. S. Chipman, D. McFadden, and M. K. Richter, 161--186. Boulder: Westview Press

    1991

  27. [35]

    Shapley, Lloyd S. 1971. ``Cores of Convex Games.'' International Journal of Game Theory, 1(1): 11--26

    1971

  28. [36]

    Tomlinson, Kiran, Johan Ugander, and Austin R. Benson. 2021. ``Choice Set Confounding in Discrete Choice.'' In Proceedings of the 27th ACM SIGKDD Conference on Knowledge Discovery and Data Mining. New York: Association for Computing Machinery, 1571--1581

    2021

  29. [37]

    Turansick, Christopher. 2022. ``Identification in the Random Utility Model.'' Journal of Economic Theory 203: 105489

    2022

Pith tools

Reviewed July 10, 2026 · model on record in the stance chip above.