Evaluation of ride-sourcing search frictions and driver productivity: A spatial denoising approach

Natalia Zuniga-Garcia, Mauricio Tec, James G. Scott, Natalia Ruiz Juri, Randy B. MachemehlTransportation 2019 · DOI · arXiv

Introduction

In an ideal ride-sourcing market, drivers should be indifferent to a passenger's destination. This concept, known as destination invariance, suggests that expected earnings should remain consistent regardless of where a trip ends. However, real-world markets are plagued by "search frictions." When a driver drops off a passenger in a low-demand "dead zone," they often face significant idle time or "dead-heading" (driving without a passenger) to return to active areas. These frictions create spatial disparities in driver productivity, leading to inequitable outcomes and potentially reducing service reliability as drivers attempt to avoid "unlucky" fares.

This paper introduces a novel analytical framework to quantify these frictions and evaluate how current pricing strategies, such as surge pricing, impact driver equity. By combining a two-trip lookahead metric with advanced spatial denoising techniques, the researchers demonstrate that destination invariance is rarely achieved in practice.

Methodology

The study employs a three-pronged methodological approach to isolate the impact of destination on earnings:

  1. Two-Trip Productivity Metric: Rather than analyzing isolated trips, the authors calculate productivity across two consecutive trips. This approach captures the "continuation payoff"—the inherent value of being at a specific drop-off location for the subsequent fare.
  2. Natural Experiment: To control for origin-based variables, the researchers analyzed trips originating from Austin’s Central Business District (CBD). Because these trips are undifferentiated at the start, any variance in total earnings can be attributed directly to the destination and the market conditions encountered there.
  3. Spatial Denoising (Graph-Fused Lasso): High-resolution data at the Traffic Analysis Zone (TAZ) level is often sparse and noisy. To address this, the authors utilized Graph-Fused Lasso (GFL). Borrowed from image processing, GFL treats the urban map as a graph of connected neighbors. It smooths data by penalizing large differences between adjacent zones unless the data indicates a sharp "break," allowing for high-resolution insights without the statistical noise of low-sample areas.

Datasets and Spatial Context

The research utilizes a comprehensive dataset from RideAustin, a non-profit Transportation Network Company (TNC) that operated in Austin, Texas. The dataset includes over 1.4 million rides recorded between September 2016 and April 2017—a period when Uber and Lyft had temporarily exited the market, making RideAustin's data uniquely representative of the city's total ride-sourcing activity. This was supplemented by TAZ-level demographic and density data from the Capital Area Metropolitan Planning Organization (CAMPO).

Spatial characteristics of the Austin study area, including area types, CBD location, and population/employment densities.

Key Insights

The Reality of "Unlucky" Fares

The analysis confirms that drivers starting from the same location experience significant productivity gaps based purely on their passenger's destination. In the CBD natural experiment, productivity varied by approximately $16/hr. A primary driver of this disparity is idle time; average wait times for a subsequent trip ranged from just 5 minutes in the city center to over 30 minutes in suburban fringes like Lakeway or Buda.

The Paradox of Surge Pricing

While surge pricing is intended to balance supply and demand, the study found it actually exacerbated inequality. Because surge multipliers are typically based on the trip's origin, they fail to compensate drivers for the opportunity cost of ending a trip in a low-demand area. Consequently, surge pricing increased the earnings gap between "lucky" drivers (heading to high-demand zones) and "unlucky" ones.

Distance vs. Demand

One of the study's most surprising findings is that trip distance is a more significant factor in productivity than the destination's demand level. Long trips generally yielded higher productivity because the immediate revenue of a long fare outweighed the benefit of a shorter wait time following a short trip, even if that short trip ended in a high-demand area.

Case Study: Austin-Bergstrom International Airport (ABIA)

The airport served as a clear example of market imbalance. The study found a 45% discrepancy between trips arriving at and departing from ABIA. This imbalance forced drivers dropping off passengers to endure long idle times (22–27 minutes), significantly dragging down their hourly productivity compared to city-center trips.

Conclusion

The application of Graph-Fused Lasso significantly improved the model's explanatory power, increasing the R-squared value for the relationship between idle time and ridership from 0.496 to 0.674. The results suggest that current TNC pricing models are insufficient for ensuring driver equity. To stabilize labor supply and ensure reliable service across all urban zones, the authors recommend moving toward "spatio-temporal" pricing models that explicitly account for a destination's future earning potential.

Summary updated

BibTeX
@inproceedings{zunigagarcia2019evaluation,
  author = {Natalia Zuniga-Garcia and Mauricio Tec and James G. Scott and Natalia Ruiz Juri and Randy B. Machemehl},
  title = {{Evaluation of ride-sourcing search frictions and driver productivity: A spatial denoising approach}},
  booktitle = {Transportation},
  year = {2019},
  doi = {10.1016/j.trc.2019.11.021},
}
Figure