Online marketplaces have adopted new quality control mechanisms that can accommodate a flexible pool of providers. In the context of ride-hailing, we measure the effectiveness of these mechanisms, which include ratings, incentives, and behavioral nudges. Using telemetry data as an objective measure of quality, we find that drivers not only respond to user preferences but also improve their behavior after receiving warnings about their low ratings. Furthermore, we use data from a randomized experiment to show that informing drivers about their past behavior improves quality, especially for low-performing drivers. Lastly, we find that UberX drivers exhibit behavior comparable to that of UberTaxi drivers, suggesting that Uber’s new quality control mechanisms successfully maintain a high level of service quality.
Authors
Related Organizations
- Acknowledgements & Disclosure
- We are grateful for funding from the Sloan Foundation and the Stanford Cyber Initiative. The paper was initiated while Bharat Chandar was an employee of Uber. He no longer retains equity in the company. The views expressed herein are those of the authors and do not necessarily reflect the views of the National Bureau of Economic Research. Bharat Chandar Chandar is a former employee of Uber. He no longer retains equity in the company.
- DOI
- https://doi.org/10.3386/w33087
- Pages
- 63
- Published in
- United States of America
Table of Contents
- Introduction 3
- Uber, telemetry, and ratings 8
- Driving behavior and rider preferences 12
- Rider preferences 12
- Driving scores 17
- What determines driving behavior? 20
- Ratings, incentives, and nudges 21
- Response to ratings and notifications 21
- Driver deactivation 25
- Information on past behavior 28
- Response to individual rider effects 30
- UberX versus UberTaxi 33
- Conclusions 38
- Construction of trip characteristics groups 1
- Variable selection and score construction 1
- Selecting driving metrics 1
- Score model 1
- Imputing duration and distance metrics for UberTaxi trips 1
- Alternative samples for score construction 1
- Additional results 1
- Response to riders preferences on scores 1
- Heterogeneity in UberX effect on scores 1
- Distribution of distance metric for UberX and UberTaxi trips 1
- Robustness checks 1
- Rider fixed effects 1
- Traffic conditions 1
- Alternative trip characteristics fixed effects 1
- Heterogeneity by time of the week 1
- Response to car prices 1
- Response to last rating 1
- Excluding trips before warnings 1
- Further details about the deactivation process 1
- Additional experimental results 1
- Balance of experimental sample 1
- Effect of Dashboard Experiment on Metrics 1
- UberNudgesAppendices.pdf 1
- Introduction 1
- Uber, telemetry, and ratings 1
- Driving behavior and rider preferences 1
- Rider preferences 1
- Driving scores 1
- What determines driving behavior? 1
- Ratings, incentives, and nudges 1
- Response to ratings and notifications 1
- Driver deactivation 1
- Information on past behavior 1
- Response to individual rider effects 1
- UberX versus UberTaxi 1
- Conclusions 1
- Construction of trip characteristics groups 43
- Variable selection and score construction 44
- Selecting driving metrics 44
- Score model 45
- Imputing duration and distance metrics for UberTaxi trips 46
- Alternative samples for score construction 46
- Additional results 50
- Response to riders preferences on scores 50
- Heterogeneity in UberX effect on scores 50
- Distribution of distance metric for UberX and UberTaxi trips 50
- Robustness checks 51
- Rider fixed effects 51
- Traffic conditions 51
- Alternative trip characteristics fixed effects 54
- Heterogeneity by time of the week 54
- Response to car prices 54
- Response to last rating 55
- Excluding trips before warnings 59
- Further details about the deactivation process 59
- Additional experimental results 62
- Balance of experimental sample 62
- Effect of Dashboard Experiment on Metrics 62
