A field experiment on non-financial incentives to improve parking by users of shared bicycle systems

Last registered on November 22, 2021

View Trial History

Pre-Trial

Trial Information

General Information

Title

A field experiment on non-financial incentives to improve parking by users of shared bicycle systems

RCT ID

AEARCTR-0008391

Initial registration date

October 17, 2021

Initial registration date is when the trial was registered.

It corresponds to when the registration was submitted to the Registry to be reviewed for publication.

First published

October 18, 2021, 2:22 PM EDT

First published corresponds to when the trial was first made public on the Registry after being reviewed.

Last updated

November 22, 2021, 10:51 AM EST

Last updated is the most recent time when changes to the trial's registration were published.

Locations

Country

China

Region

Hangzhou & Nanning

Primary Investigator

Name

Duan Su

Affiliation

Beijing Jiaotong University

Contact Primary Investigator

Other Primary Investigator(s)

PI Name

Yacan Wang

PI Affiliation

Beijing Jiaotong University

Contact Investigator

PI Name

Adriaan Soetevent

PI Affiliation

University of Groningen

Contact Investigator

Additional Trial Information

Status

In development

Start date

2021-09-15

End date

2021-12-18

Keywords

Behavior

Additional Keywords

Field experiment, transport decisions, bike-sharing.

JEL code(s)

C93, D12, H23, L9, R41

Secondary IDs

National Natural Science Foundation of China and the Joint Programming Initiative Urban Europe, National Natural Science Foundation of China

Prior work

This trial does not extend or rely on any prior RCTs.

Abstract

Dockless bike-sharing is popular and provides the commuting service for "the last mile of trip". Users often park improperly which causes negative externalities to other users, the public space, and the circulation of shared bicycles. Users often continue to park bicycles even at parking locations that are already very crowded. Company employees need to spend efforts to transfer these bicycles to less crowded areas in order to balance local supply and demand. It is more difficult for other users to find bicycles. In this field experiment, we use machine learning to predict the destination of users' trips. Users who are predicted to ride to very busy areas will via the ride-sharing app be exposed to information treatments encouraging them to avoid parking in these areas.

External Link(s)

Registration Citation

Citation

Soetevent, Adriaan , Duan Su and Yacan Wang. 2021. "A field experiment on non-financial incentives to improve parking by users of shared bicycle systems ." AEA RCT Registry. November 22. https://doi.org/10.1257/rct.8391-1.1

Sponsors & Partners

Partner

Name

Hello Inc.

Type

private_company

URL

https://www.hello-inc.com/

Experimental Details

Interventions

Intervention(s)

1. Reminder: Use app information to remind users to avoid parking in crowded areas.
2. Reciprocity information: Use app information to emphasize the indirect reciprocity relationship among bike-sharing users.
3. Circularity information: Use app information to emphasize the bicycle circularity among bike-sharing users.
4. Specific reciprocity information: Use mobile message to inform the average time window when the bicycle that the last user ride is used again after parking in crowded areas or non-crowded areas.
5. Specific circularity information: Use mobile message to inform the time window when the bicycle that the subject ride is used again after his or her parking.

Intervention Start Date

2021-10-18

Intervention End Date

2021-11-18

Primary Outcomes

Primary Outcomes (end points)

Change in fraction of bicycles in each treatment group that is parked in crowded areas

Primary Outcomes (explanation)

Each user i will complete Ti trips (Ti =0,1,…) in the experimental period. Each trip t will end in a crowded area (xit = 1) or not (xit = 0). Per individual with Ti >0 , we compute the average value (x_i ) ̅= 1/T_i ∑_(t=1)^(T_i)▒x_it The prime outcome variable is the fraction of trips that ends in crowded areas, averaged over all users in a treatment group.

Secondary Outcomes

Secondary Outcomes (end points)

The fraction of times user i parks in crowded areas for the k’th trip in the experimental period (with k = 1,2,…): x_ik.

Secondary Outcomes (explanation)

The primary outcome variable focuses on the fraction of orderly parkings in the full experimental period. It is conceivable that the treatment effect wears off. Hence we are also interested in a between treatment comparison of the average fraction of orderly parkings on the k’th trip.

Experimental Design

During the treatment period, when a subject unlocks a shared bicycle mobile, the company will predict the destination of his or her riding. If we forecast that the subject will ride to a very crowded area, the app sends treatment information to them.

Experimental Design Details

Users are assigned to the control or treatment group randomly. During the treatment period, for each trip, we use the machine learning algorithm to predict the user's destination when she scans QRs based on the lock point corresponding to her historical lock point. If the user is forecasted to ride to a crowded area, the app will show the treatment information to her. In the specific groups, we also send the mobile messages in the Monday if she is intervened by app infomration last week.

Randomization Method

One AI platform to make complete randomization

Randomization Unit

Individual user

Was the treatment clustered?

Experiment Characteristics

Sample size: planned number of clusters

10000 users.
Explanation: In Hangzhou, Hellobike has about 9 million 650 thousand users (2019 data). There are also many users in Nanning. Based on machine-learning techniques, Hellobike will identify 6,000 users for treatment in Hangzhou and 4000 users in Nanning. These users have a history of parking outside the designated areas and are likely to do so again on future trips. The investigators will randomize these 10,000 users into the control and treatment groups T1-T6.

Sample size: planned number of observations

10,000 users (trip level information)

Sample size (or number of clusters) by treatment arms

T1 Control: about 2,000 users
T2 Reminder: 1,600 users
T3 Reciprocity: 1,600 users
T4 Specific reciprocity: 1,600 users
T5 Circularity: 1,600 users
T6 Specific circularity: 1,600 users

Minimum detectable effect size for main outcomes (accounting for sample design and clustering)

Unit of measurement: the fraction of trips ending in crowded areas at the individual user level. For a type I error probability of alpha = 0.05 and a power of 1- k = 0.8, and N =1,000 per treatment arm, the standardized minimum detectable effect size is 0.126 standard deviations. The standard deviation of out outcome variable is highest when half of the users always parks in a crowded area and the other half never, in which case s.d. = 0.5. In this most extreme case, the minimum detectable effect size (MDE) would be a change in parking in disorderly areas of 6.3 percentage points (=0.5x0.126). If only half of the selected individuals takes up treatment (that is: completes at least one trip in the experimental period), the MDE would increase to 8.9 percentage points. Increasing the power to 1-k =0.9 increases the MDE to 7.3 percentage points. In the experimental area, the benchmark percentage of parking in crowded area’s is over 40%.

Supporting Documents and Materials

IRB