Incentive-driven transition to high ride-sharing adoption

The study addresses when and under which conditions users of app-based ride-hailing services choose shared rides over single-occupancy trips. Although ride-sharing can increase vehicle utilization, reduce required fleet size, mitigate congestion, and lower environmental impacts, overall adoption remains low, and varies widely across urban areas even where demand is high. Prior work established that shareability and operational efficiency rise with demand density, but it is unclear whether user willingness to share rides follows the same trend. The authors propose a game-theoretic framework that captures user incentives and interactions to explain observed adoption patterns and to identify conditions leading to high versus low ride-sharing adoption. The central hypothesis is that the balance between financial discounts and perceived disutilities (detours and inconvenience) drives a transition between distinct adoption regimes, potentially discontinuous in the high-demand limit. Empirical analyses of large-scale trip records from New York City and Chicago are used to assess whether predicted regimes appear in real-world data.

Past research has focused on algorithmic and operational aspects of ride-sharing, including dynamic trip-vehicle assignment and shareability networks, demonstrating substantial efficiency gains and improved service quality in dense demand conditions. Studies suggest ride-sharing is most efficient in densely populated areas due to easier matching with minimal detours. Behavioral studies and qualitative analyses of user experiences identified three primary determinants of willingness to share: financial savings (discounts), expected detours and uncertainty about trip duration, and the inconvenience of sharing a vehicle with strangers (privacy, comfort). Evidence also indicates heterogeneity in preferences across socio-economic groups and urban contexts. However, the literature lacks a mechanistic model linking these incentives and user interactions to aggregate adoption patterns across varying demand levels, and it does not explain why adoption can decrease with increasing demand despite higher operational shareability.

Modeling framework: The authors develop a game-theoretic model where users decide between a single ride and a shared ride based on the expected utility difference relative to a single ride. Utility for a shared ride is U_share = U_single + ε d_single − ξ d_det − ζ d_inc, where ε is the per-distance financial discount, ξ and ζ are user-specific weights for detour (d_det) and inconvenience (d_inc) disutilities respectively, and d_single is the direct single-ride distance. Expected utility E[Δu] conditions on other users’ destinations and sharing decisions because detour and inconvenience depend on matches. City topology and matching: A stylized city network with one central origin and multiple destinations on two concentric rings is used. Users originate at the center and select destinations uniformly among branches. Shared requests can be matched pairwise (up to two riders) along the same or adjacent branches if the combined trip plus return is shorter than the sum of individual trips, consistent with a profit-maximizing provider. Optimal pairwise matching is computed via maximum-weight matching (Edmonds’ Blossom algorithm, Blossom V), using distance savings as edge weights. Matching assumes the provider returns to origin after trips to quantify routing and detour consistently. Demand and decision dynamics: In a one-to-many setting, S users arrive within a short window τ, each choosing single or shared. Users observe outcomes over repeated rides and adapt sharing probabilities per destination π(d) via discrete-time replicator dynamics to converge to equilibrium strategies where expected utilities balance. Simulations initialize low sharing adoption and iterate until fluctuations fall below a threshold, using repeated random request sets and current π(d) to estimate E[U_share|share] and expected matching probabilities. Heterogeneity: The model is extended to multiple user types with heterogeneous ε/ζ ratios (inconvenience tolerance vs. discount), mixed within the same origin. Equilibria per type are computed jointly; macroscopic adoption is the aggregate across types. A synthetic city scenario superimposes many origins with different demand levels and distributions of user types to emulate urban heterogeneity. Empirical analysis: The authors analyze >250 million High Volume For-Hire Vehicle (HVFHV) trips in NYC (2019) and >110 million Transportation Network Provider (TNP) trips in Chicago (2019). Data include origin/destination zones, timestamps, and indicators of shared requests (NYC) or sharing authorization/matching (Chicago; note the matching flag includes consecutive non-empty trips). Average request rates are computed assuming 16 hours/day active demand. For each origin, the fraction and count of shared requests are computed to compare S_share versus total request rate s across zones. Very low-volume OD pairs (<100 trips/year) are excluded from spatial visualizations to avoid noise. Results are contrasted with model predictions of high- and low-sharing regimes.

• Two adoption regimes: The model reveals a high-sharing regime where financial incentives compensate expected inconvenience (ε/ζ > 1), leading to near-universal sharing, and a low-sharing regime where the number of shared requests saturates as total requests increase, causing the fraction of sharing to decline with demand (S_share ≈ const while S grows).
• Discontinuous transition: In the high-demand limit (S → ∞), the transition between regimes becomes discontinuous at ε = ζ; a small increase in financial incentives or reduction in perceived inconvenience can abruptly shift adoption from low to high.
• Anti-coordination and spatial patterns: As S increases, users’ expected detour and inconvenience increase due to higher matching probability, producing an anti-coordination effect. Equilibria exhibit alternating sharing/non-sharing patterns across destinations; with further demand growth, mixed strategies emerge where discounts exactly balance expected inconvenience.
• Robustness: The qualitative transition persists under heterogeneous demand across destinations, asymmetric topologies, noisy information, alternative matching strategies, and heterogeneous preferences.
• Empirical consistency: Analysis of ~250M NYC and ~110M Chicago requests shows sharing decisions distributing between high- and low-sharing branches predicted by the model. At low request rates, S_share increases roughly linearly with S, but adoption is below 100% (about 20% in NYC and 40% in Chicago). At higher request rates, NYC zones like East Village and Crown Heights North maintain relatively high adoption consistent with high-sharing behavior, while JFK and LaGuardia airports show lower, saturated sharing consistent with low-sharing. In Chicago, several high-demand zones display approximately constant S_share, indicating low-sharing; large downtown zones fall into partial-sharing states.
• Overall low adoption: Despite high-demand efficiency potential, NYC overall had less than 18% of requests flagged as shared in 2019, underscoring a large gap between operational shareability and user adoption.

The findings explain why ride-sharing adoption may decline with increasing demand despite improved matching efficiency: higher demand raises the probability of actual sharing, increasing expected inconvenience and detour costs, which, without sufficient discounts, deters users (an anti-coordination effect). The ratio ε/ζ acts as a control parameter, separating low- and high-adoption regimes; in homogeneous populations the transition is abrupt in the high-demand limit. In heterogeneous populations, user groups separate into distinct regimes, producing macroscopic partial-adoption patterns that align with empirical observations in NYC and Chicago. These insights suggest that modest adjustments to financial incentives or service quality (reducing inconvenience and detours) can trigger large increases in adoption among specific user segments currently in the low-sharing regime. The model’s qualitative predictions are robust to many modeling choices and matching strategies, providing a conceptual framework for designing incentives and anticipating unintended collective responses. The empirical patterns across zones (e.g., airports vs. residential neighborhoods) align with differences in perceived inconvenience and price sensitivity, emphasizing the role of socio-economic context in shaping adoption.

This work introduces a user-centric, game-theoretic model of ride-sharing adoption that integrates core incentives—financial discounts, detour costs, and inconvenience—into a collective decision framework on a stylized urban network. The model uncovers a discontinuous transition between low- and high-sharing regimes controlled by the balance of discounts and inconvenience. Analyses of 360 million trips in NYC and Chicago exhibit adoption patterns consistent with model predictions, with zones clustering into high- and low-sharing behaviors and a broad range of partial-adoption outcomes due to heterogeneous preferences. The results imply that carefully calibrated incentives and service improvements could disproportionately increase adoption among low-sharing segments, aiding sustainability goals. Future research should refine quantitative predictions by measuring user preference distributions, extend utility specifications beyond linear forms (e.g., threshold detour sensitivities), incorporate correlated demand and multi-origin matching, consider inter-provider dynamics and modal competition with public transit, and assess rebound effects to fully evaluate system-wide sustainability impacts.

• Utility specification: The model uses a linear utility form and first-order scaling with distance/time; precise quantitative predictions would require detailed empirical estimation of user preferences and potentially nonlinear or threshold effects for detours and inconvenience.
• Stylized topology and pairing: The city is abstracted to a central origin with ring-branch destinations and pairwise matching (max two riders). Real cities involve multiple origins, complex networks, and multi-passenger pooling which may alter detour/inconvenience dynamics.
• Matching assumptions: Optimal pairwise matching with assumed return-to-origin may differ from real provider routing policies. Alternative matching strategies can change the level of incentives needed for high adoption, though qualitative regimes persist.
• Information and learning: Replicator dynamics approximate user learning over repeated rides; actual user adaptation and information availability may differ.
• Data constraints: NYC shared flag indicates shared requests; Chicago’s matching flag includes consecutive non-empty trips and may not precisely capture simultaneous sharing. Request rates assume 16 active demand hours/day; low-volume OD pairs were excluded from maps, potentially biasing spatial visualizations.
• External factors: Socio-economic differences, airport-specific contexts, availability of public transit, and inter-provider competition can influence adoption beyond modeled incentives. Potential rebound effects (increased total travel demand) are not explicitly modeled.