pith. sign in

arxiv: 2309.13159 · v3 · submitted 2023-09-22 · 💰 econ.EM

Nonparametric mixed logit model with market-level parameters estimated from market share data

Xiyuan Ren , Joseph Y. J. Chow , Prateek Bansal This is my paper

Pith reviewed 2026-05-24 06:54 UTC · model grok-4.3

classification 💰 econ.EM
keywords nonparametric mixed logitmarket share datainverse utility maximizationtravel mode choiceNew YorkBLP modelchoice elasticity
0
0 comments X

The pith

The nonparametric mixed logit model estimates market-specific taste parameters from choice share data by solving a multiagent inverse utility maximization problem.

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

The paper proposes estimating a mixed logit model nonparametrically by treating each market as an agent and recovering its unique taste parameters from observed market shares via inverse optimization. This avoids parametric assumptions on the distribution of tastes across markets. On New York travel data with over 120,000 markets, it achieves higher predictive accuracy than standard models like BLP while running much faster. The approach also yields parameters that integrate easily with supply-side models for policy analysis such as congestion pricing.

Core claim

By solving the multiagent inverse utility maximization problem, the model recovers market-level parameters that represent taste heterogeneity without parametric restrictions, leading to improved out-of-sample predictive accuracy on large-scale choice data.

What carries the argument

The multiagent inverse utility maximization problem that recovers market-specific parameters from choice shares.

If this is right

  • The model predicts mode choices with 81.78% out-of-sample accuracy compared to 65.30% for benchmarks.
  • Estimation completes in less than one-tenth the time required for the BLP model.
  • Price elasticities and diversion ratios show similar substitution patterns to parametric models.
  • Market-level parameters enable direct integration into supply-side optimization for transportation design.
  • Compensating variation analysis shows a $9 congestion toll affects about 60% of travelers.

Where Pith is reading between the lines

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

  • The method could extend to other discrete choice settings where only aggregate shares are observed.
  • Recovered parameters might reveal spatial patterns in preferences across census block groups.
  • Integration with supply models could allow joint estimation of demand and network design in one framework.

Load-bearing premise

Market-level choice shares are generated exactly by utility maximization with market-specific taste parameters that can be uniquely recovered by solving the multiagent inverse utility maximization problem.

What would settle it

If the recovered market-specific parameters fail to predict individual-level choices or out-of-sample market shares better than parametric alternatives on the same data.

Figures

Figures reproduced from arXiv: 2309.13159 by Joseph Y. J. Chow, Prateek Bansal, Xiyuan Ren.

Figure 1
Figure 1. Figure 1: Overlapping histograms of market-level true and estimated parameters in the multimodal scenario. In (a)-(f), x-axis is the value of parameters, y-axis is the probability density [PITH_FULL_IMAGE:figures/full_fig_p017_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: further presents the distribution of taste parameters corresponding to the first attribute (𝜃1,𝑡 ) in each taste cluster. The results indicate that when 𝑀 = 1, the model approximates the overall distribution well, but it cannot identify the number of unimodal Gaussian distributions (𝛼) used for data generation. When the number of clusters (𝑀) matches 𝛼 , the model more accurately captures the multimodal st… view at source ↗
read the original abstract

We propose a nonparametric mixed logit model that is estimated using market-level choice share data. The model treats each market as an agent and represents taste heterogeneity through market-specific parameters by solving a multiagent inverse utility maximization problem, addressing the limitations of existing market-level choice models with parametric estimation. A simulation study is conducted to evaluate the performance of our model in terms of estimation time, estimation accuracy, and out-of-sample predictive accuracy. In a real data application, we estimate the travel mode choice of 53.55 million trips made by 19.53 million residents in New York State. These trips are aggregated based on population segments and census block group-level origin-destination (OD) pairs, resulting in 120,740 markets. We benchmark our model against multinomial logit (MNL), nested logit (NL), inverse product differentiation logit (IPDL), and the BLP models. The results show that the proposed model improves the out-of-sample accuracy from 65.30% to 81.78%, with a computation time less than one-tenth of that taken to estimate the BLP model. The price elasticities and diversion ratios retrieved from our model and benchmark models exhibit similar substitution patterns. Moreover, the market-level parameters estimated by our model provide additional insights and facilitate their seamless integration into supply-side optimization models for transportation design. By measuring the compensating variation for the driving mode, we found that a $9 congestion toll would impact roughly 60 % of the total travelers. As an application of supply-demand integration, we showed that a 50% discount of transit fare could bring a maximum ridership increase of 9402 trips per day under a budget of $50,000 per day.

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

3 major / 1 minor

Summary. The manuscript proposes a nonparametric mixed logit model estimated from market-level choice share data by treating each market as an agent and solving a multiagent inverse utility maximization problem to recover market-specific taste parameters. It evaluates the model through simulations assessing estimation time, accuracy, and out-of-sample prediction, and applies it to New York travel mode choice data with 120,740 markets, claiming improved out-of-sample accuracy over MNL, NL, IPDL, and BLP models, faster computation, and utility for supply-side optimization and policy analysis such as congestion tolls and transit fare discounts.

Significance. If the inverse optimization step is shown to uniquely identify the market-level parameters without additional restrictions, the model provides a flexible nonparametric approach to taste heterogeneity that scales to large datasets and facilitates integration with supply models. The large-scale empirical application and direct comparisons to established benchmarks are notable strengths, as is the demonstration of policy-relevant quantities like compensating variation.

major comments (3)
  1. [model estimation / inverse problem section] The section describing the multiagent inverse utility maximization problem: explicit conditions or verification for uniqueness of the recovered market-specific taste vectors are not provided. This is load-bearing for the nonparametric claim, since standard share inversion identifies mean utilities only up to normalization and extending to per-market heterogeneity risks non-uniqueness or flat directions absent shown structure such as strict concavity.
  2. [simulation study] Simulation study section: recovery results may not test identification if data are generated from the same inverse process; an independent check of uniqueness under the stated assumptions is needed to support the reported estimation accuracy.
  3. [empirical application] Real-data application section (NY travel data with 120,740 markets): the out-of-sample accuracy gain (65.30% to 81.78%) requires clarification on the hold-out procedure and whether predictions are formed using the recovered market-specific parameters in a way that avoids reducing to in-sample fitted values by construction.
minor comments (1)
  1. [abstract] Abstract: the reported trip and resident counts (53.55 million trips by 19.53 million residents) should indicate whether they are exact or rounded.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the detailed and constructive comments. We address each major point below and will revise the manuscript to strengthen the identification arguments, add verification steps, and clarify the empirical procedures.

read point-by-point responses

  1. Referee: [model estimation / inverse problem section] The section describing the multiagent inverse utility maximization problem: explicit conditions or verification for uniqueness of the recovered market-specific taste vectors are not provided. This is load-bearing for the nonparametric claim, since standard share inversion identifies mean utilities only up to normalization and extending to per-market heterogeneity risks non-uniqueness or flat directions absent shown structure such as strict concavity.

    Authors: We agree that explicit uniqueness conditions are necessary to support the nonparametric claim. In the revised manuscript we will add a new proposition in the model estimation section establishing uniqueness of the market-specific taste vectors. The proof relies on the strict monotonicity of the market-share mapping under the mixed logit probability function combined with the assumption that the utility is strictly concave in the taste parameters; this rules out flat directions and ensures the inverse problem has a unique solution for each market. We will also include a brief numerical verification using the simulation design. revision: yes

  2. Referee: [simulation study] Simulation study section: recovery results may not test identification if data are generated from the same inverse process; an independent check of uniqueness under the stated assumptions is needed to support the reported estimation accuracy.

    Authors: The referee correctly notes that generating data from the same inverse process primarily tests numerical recovery rather than independent identification. We will revise the simulation section to include an additional Monte Carlo exercise in which data are generated from a parametric mixed logit (with random coefficients drawn from a known distribution) and then recovered using the nonparametric inverse procedure. Recovery accuracy and uniqueness diagnostics under this independent data-generating process will be reported to directly address the concern. revision: yes

  3. Referee: [empirical application] Real-data application section (NY travel data with 120,740 markets): the out-of-sample accuracy gain (65.30% to 81.78%) requires clarification on the hold-out procedure and whether predictions are formed using the recovered market-specific parameters in a way that avoids reducing to in-sample fitted values by construction.

    Authors: We will expand the empirical application section to detail the hold-out procedure: the 120,740 markets are randomly partitioned into an 80 % training set and a 20 % test set. The nonparametric mixed logit is estimated solely on the training markets, yielding an empirical distribution of taste parameters. Out-of-sample predictions for test markets are formed by integrating choice probabilities over this training-derived distribution; market-specific parameters recovered from the test markets themselves are never used. This ensures the reported accuracy gain reflects genuine out-of-sample performance rather than in-sample fitting. revision: yes

Circularity Check

0 steps flagged

No significant circularity; derivation relies on external inverse problem solution and out-of-sample benchmarks

full rationale

The paper recovers market-specific parameters by solving a multiagent inverse utility maximization problem from observed shares, then reports out-of-sample predictive accuracy (65.30% to 81.78%) against MNL/NL/IPDL/BLP on held-out New York data. No quoted equation or step shows the out-of-sample metric or recovered parameters reducing to the input shares by construction. The inverse problem is treated as an independent recovery step whose uniqueness is assumed under stated conditions; simulation and real-data comparisons to external models supply falsifiable checks outside any self-referential fit. This is the normal non-circular case for an estimation paper whose central output is benchmarked externally.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

The model rests on standard discrete-choice utility maximization plus the novel assumption that market-specific parameters are recoverable nonparametrically from shares alone.

axioms (2)
  • domain assumption Observed market shares are generated by utility maximization with market-specific taste parameters.
    Core premise invoked to justify the inverse optimization step.
  • domain assumption The multiagent inverse utility maximization problem admits a unique solution for the market-specific parameters.
    Required for consistent recovery from market share data.

pith-pipeline@v0.9.0 · 5850 in / 1384 out tokens · 24173 ms · 2026-05-24T06:54:31.765628+00:00 · methodology

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. Welfare, sustainability, and equity evaluation of the New York City Interborough Express using spatially heterogeneous mode choice models

    cs.CY 2024-08 unverdicted novelty 4.0

    Using synthetic citywide trip data and spatially heterogeneous mode choice models, the study projects the IBX light rail would attract 272k daily riders, save 28.1 minutes per trip on average, shift 16k trips from car...

Reference graph

Works this paper leans on

6 extracted references · 6 canonical work pages · cited by 1 Pith paper

  1. [1]

    K., & Orlin, J

    Ahuja, R. K., & Orlin, J. B. (2001). Inverse Optimization. Operations Research, 49(5), 771–783. Angrist, J. D., & Krueger, A. B. (2001). Instrumental variables and the search for identification: From supply and demand to natural experiments. Journal of Economic perspectives, 15(4), 69-85. Berry, S., Levinsohn, J., & Pakes, A. (1995). Automobile Prices in ...

    work page 2001

  2. [2]

    Berry, S. T. (1994). Estimating Discrete-Choice Models of Product Differentiation. The RAND Journal of Economics, 25(2),

    work page 1994

  3. [3]

    Bierlaire, M., & Frejinger, E. (2008). Route choice modeling with network-free data. Transportation Research Part C: Emerging Technologies, 16(2), 187–198. Bills, T. S., Twumasi-Boakye, R., Broaddus, A., & Fishelson, J. (2022). Towards transit equity in Detroit: An assessment of microtransit and its impact on employment accessibility. Transportation Resea...

    work page 2008

  4. [4]

    Y., Zhou, J., Ma, Z., Chow, J

    He, B. Y., Zhou, J., Ma, Z., Chow, J. Y. J., & Ozbay, K. (2020). Evaluation of city-scale built environment policies in New York City with an emerging-mobility-accessible synthetic population. Transportation Research Part A: Policy and Practice, 141, 444–

    work page 2020

  5. [5]

    Hess, S. (2010). Conditional parameter estimates from Mixed Logit models: Distributional assumptions and a free software tool. Journal of Choice Modelling, 3(2), 134–152. Huo, J., Dua, R., & Bansal, P. (2024). Inverse product differentiation logit model: Holy grail or not? Energy Economics, 131, 107379. Krueger, R., Bierlaire, M., Daziano, R. A., Rashidi,...

    work page 2010

  6. [6]

    Yin, H., & Cherchi, E. (2024). Willingness to pay for automated taxis: A stated choice experiment to measure the impact of in-vehicle features and customer reviews. Transportation, 51(1), 51–72

    work page 2024