[Paper Review] Capturing positive network attributes during the estimation of recursive logit models: A prism-based approach

Yuki Oyama|arXiv (Cornell University)|Apr 4, 2022

Transportation Planning and Optimization4 citations

TL;DR

This paper proposes a prism-constrained recursive logit (Prism-RL) model that resolves numerical instability in recursive logit model estimation by restricting path choices via a state-extended network with a choice-stage constraint. The method enables reliable estimation even with positive network attributes (e.g., street greenery), outperforms standard RL models in fit and prediction, and successfully captures real-world pedestrian route choice behavior with detours and environmental preferences.

ABSTRACT

Although the recursive logit (RL) model has been recently popular and has led to many applications and extensions, an important numerical issue with respect to the computation of value functions remains unsolved. This issue is particularly significant for model estimation, during which the parameters are updated every iteration and may violate the feasibility condition of the value function. To solve this numerical issue of the value function in the model estimation, this study performs an extensive analysis of a prism-constrained RL (Prism-RL) model proposed by Oyama and Hato (2019), which has a path set constrained by the prism defined based upon a state-extended network representation. The numerical experiments have shown two important properties of the Prism-RL model for parameter estimation. First, the prism-based approach enables estimation regardless of the initial and true parameter values, even in cases where the original RL model cannot be estimated due to the numerical problem. We also successfully captured a positive effect of the presence of street green on pedestrian route choice in a real application. Second, the Prism-RL model achieved better fit and prediction performance than the RL model, by implicitly restricting paths with large detour or many loops. Defining the prism-based path set in a data-oriented manner, we demonstrated the possibility of the Prism-RL model describing more realistic route choice behavior. The capture of positive network attributes while retaining the diversity of path alternatives is important in many applications such as pedestrian route choice and sequential destination choice behavior, and thus the prism-based approach significantly extends the practical applicability of the RL model.

Motivation & Objective

• To address the numerical instability in recursive logit (RL) model estimation caused by infeasible value functions during parameter updates.
• To enable the inclusion of positive network attributes—such as street greenery—in route choice models, which are often excluded due to computational issues.
• To validate the prism-constrained RL model (Prism-RL) for practical estimation using real GPS data from Yokohama, Japan.
• To examine the impact of the choice-stage constraint (T) on estimation performance and model consistency.
• To demonstrate that Prism-RL improves model fit and prediction by implicitly limiting detours and loops without sacrificing path diversity.

Proposed method

• Introduces a prism-constrained path set using a state-extended network where each state tracks the number of choices made along a path.
• Imposes a choice-stage constraint T, limiting the maximum number of choices (e.g., intersections or turns) a path can include.
• Redefines the value function within the prism-constrained network to ensure numerical solvability regardless of utility parameter values.
• Applies the prism constraint in a data-driven way by setting T based on observed detour rates (e.g., 75% of paths have detour rate ≤ 1.33).
• Uses maximum likelihood estimation with iterative parameter updates, ensuring feasibility at every step via the prism constraint.
• Employs GPS-derived pedestrian route data from Yokohama to calibrate and validate the model against real-world behavior.

Experimental results

Research questions

• RQ1 Can the prism-constrained RL model overcome numerical instability in parameter estimation when positive network attributes are included?
• RQ2 How does the choice-stage constraint T affect estimation convergence and model performance?
• RQ3 Does the Prism-RL model better capture realistic pedestrian route choice behavior, including detours and environmental preferences?
• RQ4 Can the prism-based approach maintain or improve model fit and prediction accuracy compared to the standard RL model?
• RQ5 What is the computational cost of the Prism-RL model relative to the standard RL model in real-world applications?

Key findings

• The Prism-RL model successfully estimated parameters even when the initial and true parameters were infeasible for the standard RL model, due to value function non-convergence.
• The model captured a statistically significant positive effect of street greenery on pedestrian route choice, with a utility coefficient of β_green = 2.00 (p < 0.05), which the standard RL model failed to estimate reliably.
• Prism-RL achieved a better log-likelihood (-6729.71) than the standard RL model (-6730.22), indicating superior fit to real GPS data.
• Estimation time for Prism-RL increased linearly with T, but remained feasible—e.g., 31.22 seconds for T=100 with a complex parameter set, compared to 1.02 seconds for standard RL.
• The prism constraint effectively limited detour-prone and loop-heavy paths, improving realism without losing path diversity.
• The model maintained consistent estimation performance across different starting parameter values, including those far from the true values (e.g., β_init = (-4,3)), unlike the standard RL model.