Designing Peer Information and Testing Income Targeting in the Gig Economy

How to cite

Lim, Zhi Hao. 2026. "Designing Peer Information and Testing Income Targeting in the Gig Economy." AEA RCT Registry. AEARCTR-0018613. https://doi.org/10.1257/rct.18613-3.0

Download: BibTeX · RIS — for reference managers without a browser translator. Zotero and most others pick up this page's citation metadata directly.

Zhi Hao Lim*

* Primary investigator

Registry ID
AEARCTR-0018613
DOI
https://doi.org/10.1257/rct.18613-3.0
First registered
Last updated
Status
In development
Trial dates
2026-05-29 – 2026-09-30
Countries
Private
JEL codes
C93, D91, J24
Registry record
http://www.socialscienceregistry.org/trials/18613

Abstract

Recent work shows that worker responses to peer information are heterogeneous, depending on workers' information preferences and their relative earnings position from peers (Lim, 2025, 2026). I conduct a follow-up field experiment with the same sample of around 2,900 rideshare drivers to test whether targeting who receives peer information and which peer benchmark each worker observes can amplify the labor supply gains from peer-information disclosure in the workplace. Over a six-week intervention, drivers are randomly assigned to one of three information arms: no peer information, standard peer information based on a randomly drawn reference group, or targeted peer information whose reference group is drawn from lower earnings percentile ranks. The targeting rule is calibrated using estimates from Lim (2026). Orthogonally, drivers receive two transitory wage increases (10% and 20% higher trip earnings during a select time window) on randomly assigned days, with announcement timing of the monetary incentives varied between same-day and advance notice. Using administrative personnel records, I estimate the effects of standard versus targeted peer information, and test for persistence and habituation by leveraging the original study's treatment assignments to peer information. In addition, I leverage the exogenous variation in wages and announcement timing to test for income targeting among rideshare drivers.

Intervention

Refer to experimental design.

Experimental design

Drivers are randomly assigned to one of three information groups:

- Control group (T1): Drivers receive only their own earnings updates each week.

- Standard Peer Info group (T2): Drivers receive their own earnings updates as in T1, and additionally receive peer earnings updates of a reference group comprising 10 non-participating drivers randomly drawn from the same baseline-hours bin.

- Targeted Peer Info group (T3): Drivers receive peer earnings updates of the same form as in T2, but whose reference group comprises 10 non-participating drivers drawn from lower earnings ranks within the same baseline-hours bin.

In addition to the information treatment assignments, which are fixed throughout the six-week intervention period, each driver is randomly assigned to two single-day monetary incentives:

- Low Bonus: On the assigned day, drivers earn X% more on total trip earnings during a fixed time window.
- High Bonus: On the assigned day, drivers earn Y% more on total trip earnings during a fixed time window.
(X and Y are placeholders that will be calibrated based on the firm's existing incentive structure, where X is less than Y)

Each driver is also randomly assigned to one of two announcement conditions for the monetary incentives:

- Same-day Notice: Drivers are only informed of the incentive on the incentive day.
- Advance Notice: Drivers are informed of the incentive in advance, and reminded again on the incentive day.

The assignment of Low and High Bonus days is randomized within driver, so that each driver experiences both bonus levels (X% and Y%). The specific weeks in which the bonuses apply are randomly assigned with at least one washout week between a driver's two incentive days. The announcement condition is randomized between drivers and held fixed across both of a driver's incentive days.

Design details

Randomization method
Randomization is done by a computer using Python.
Randomization unit
Randomization is done at the individual driver level, stratified by baseline utilization hours, tenure with the firm, and the driver's treatment assignment in the original experiment (Lim, 2025).
Clusters
Approximately 2,900 drivers. The sample comprises all drivers who signed up for the original study (Lim, 2025), restricted to those who remain active and in good standing on the platform at the start of the intervention period.
Observations
Over 426,300 driver-day level observations (2,900 workers * 7 days/week * (9 weeks pre-intervention + 6 weeks of intervention + 6 weeks of post-intervention))
Arms
The sample is evenly split across the three information treatment groups, with approximately 970 drivers in each group. The announcement conditions are also evenly split, with approximately 1,450 drivers in each condition.
Primary outcomes
i. Labor supply outcomes (i.e., indicator for working on a given day, total daily earnings, total utilization hours, total trips completed, hourly earnings, total distance travelled, hazard rate of stopping work, indicator for whether driver has attrited from the platform) ii. Search behavior outcomes (i.e., spatial footprint, share of trips picked up in central region, share of auto-accepted trips, mean deadhead time, and hourly earnings ex. incentives at the driver-shift level) iii. Endline survey measures of driver well-being (i.e., work stress, meaning, satisfaction, motivation), self-reported labor supply outcomes (i.e., number of days worked per week, hours per week, multihoming patterns), information preferences and income targets

Cited by

No citation data available for this registration yet.

Keywords