Insuring gig workers

Last registered on November 17, 2016

Pre-Trial

Trial Information

General Information

Title
Insuring gig workers
RCT ID
AEARCTR-0001555
Initial registration date
August 30, 2016

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
August 30, 2016, 11:10 AM EDT

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

Last updated
November 17, 2016, 8:52 AM EST

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

Locations

Region

Primary Investigator

Affiliation
University of Cologne

Other Primary Investigator(s)

PI Affiliation
Partner firm
PI Affiliation
University of Cologne
PI Affiliation
University of Tuebingen

Additional Trial Information

Status
In development
Start date
2016-08-31
End date
2017-08-30
Secondary IDs
Abstract
Individuals working on a freelance basis are instrumental to the on-demand economy. These "gig workers" are people tasked with and paid for specific work assignments. Many gig workers offering their services via online platforms are formally independent contractors. To align incentives, platforms typically compensate gig workers based on the profits resulting from their service to a client. It is not clear, however, whether gig workers, though legally entrepreneurs, behave like firms - for instance whether they are best thought of as risk neutral and selfish economic actors. While purely commission-based compensation of gig workers creates non-distortionary incentives for both quality and quantity provision, such pay schemes may lead to insufficient insurance against income risks. As a consequence, alternative contractual arrangements that provide some insurance against income uncertainty may be superior from the perspective of both the gig workers and the platforms acting as intermediaries. Such alternatives may work particularly well if – as many platforms claim – their gig workers are enthusiastic about their work and thus have an intrinsic desire to do a good job.

In this project we collaborate with an online platform to study how introducing some insurance to their compensation affects gig worker behaviour and the profitability of the platform. In our setting, the platform acts as the intermediary between clients and gig workers, whose task it is to provide remote shopping advice to clients. This service may result in the online sale of physical goods, again handled by the platform. Gig workers choose the quantity offered, i.e., the number of clients they want to serve. Their efforts also determine the quality achieved, i.e., the usefulness of their advice affects the sales to each client. At the outset, gig workers' compensation is commission-based, paying them a fraction of net sales to the clients they advised.

We start from a stylized formal model in which agents choose quantity (number of clients served) and average quality (sales per client) and compensation contracts may reward both. We first show that when agents are risk neutral, a pure sales-based commission is optimal. When agents are risk averse, however, this may no longer hold. In this case it may be preferable to introduce an order bonus - a piece rate based on the number of client consultations independent of final sales or returned items. This provides insurance and an incentive to serve more clients. [Note: the formal model from which we derive our experimental predictions has been uploaded as a supplementary document to this pre-registration].

We implement a natural field experiment (RCT) to test the following predictions: (a) as compared to purely commission-based pay (control group), compensation including costly insurance (treatment) increases gig workers' desired number of jobs; (b) this effect on quantity is larger for more risk averse gig workers; (c) the intervention decreases gig workers' average quality per job; (d) this effect on quality is smaller for gig workers with higher intrinsic motivation [Note: we added this hypothesis as a consequence of developing our theoretical model further. We also added questions on intrinsic motivation to the baseline survey (see below) shortly before entering the field]; (e) for sufficiently risk averse gig workers, platform profits are higher in the treatment group.
External Link(s)

Registration Citation

Citation
Butschek, Sebastian et al. 2016. "Insuring gig workers." AEA RCT Registry. November 17. https://doi.org/10.1257/rct.1555-3.0
Former Citation
Butschek, Sebastian et al. 2016. "Insuring gig workers." AEA RCT Registry. November 17. https://www.socialscienceregistry.org/trials/1555/history/11846
Sponsors & Partners

There is information in this trial unavailable to the public. Use the button below to request access.

Request Information
Experimental Details

Interventions

Intervention(s)
We create exogenous variation in the way in which the online platform compensates collaborating gig workers. The intervention only involves new gig workers who have no previous experience with the platform and its compensation policies. The intervention randomly varies whether new gig workers are subject to the existing commission-based compensation scheme or to the new compensation scheme that includes order bonuses as an insurance component.

In the commission-based compensation scheme, earnings are a fraction of the net sales realized as a consequence of the gig worker's consulting service to a client. In the new compensation scheme, this commission is reduced and an order-based piece-rate component is added, paid out as long as the client places an order and irrespective of final sales or returned items. This insurance is costly for gig workers in the sense that their upside from a particularly successful trade is reduced: they receive a smaller share of net sales. The treatment is calibrated such that in expectation - assuming no quality adjustment - earnings per average job are approximately equal under the two compensation schemes.
Intervention (Hidden)
Intervention Start Date
2016-09-20
Intervention End Date
2017-02-20

Primary Outcomes

Primary Outcomes (end points)
Our primary outcome variables are desired quantity, quality achieved and platform profits realized. Our secondary outcome variables include individual earnings and job satisfaction.
Primary Outcomes (explanation)
Desired quantity is measured as the average daily number of slots a gig worker makes available in his calendar for client consultations. For quality achieved, we measure sales per order and per consultation; we also use a measure developed by the platform that adjusts net merchandise value for the number of items recommended and the client's expected willingness to spend.

Secondary Outcomes

Secondary Outcomes (end points)
Secondary Outcomes (explanation)

Experimental Design

Experimental Design
The online platform recruits gig workers on a rolling basis. As new gig workers are admitted to the platform, they are randomly allocated to one of two groups, stratifying by predicted activity level (to increase power). Group 1: control - purely commission-based pay. Group 2: treatment - lower commission plus order bonus. [Note: we initially planned to also stratify based on risk aversion elicited from our baseline survey (see below). We realised in a pre-test that our measure of risk aversion in the baseline survey needed to be changed. We were able to make last-minute adjustments to the survey, but could not use the pre-test sample to define risk-aversion strata and thus decided to stratify only on predicted activity level].

The treatment starts with HR informing new gig workers of their compensation rules. For any given new gig worker, the treatment ends after two months. Thereafter, both groups are compensated according to the same scheme (which is used for all other existing gig workers on the platform). Treated gig workers know ex ante that the insurance component their compensation entails is limited to two months.

We collect baseline data using an online survey among all new applicants to the platform. The survey is conducted prior to allocation into treatment or control group. Detailed administrative data on gig workers' activities during and after the treatment period are provided by the platform. We complement this with a post-intervention online survey. [Note: at the time that this amendment to the preregistration is uploaded, the authors have had no access to or received information about performance or activity data on participants of the field experiment.]
Experimental Design Details
Randomization Method
Stratified randomization (by predicted-activity quantiles) is performed in office by tossing a fair coin. (In the medical literature on clinical trials, where sequential randomization is commonplace due to patients trickling in, this method is known as permuted block randomization). Predictions of gig worker activity use information from the baseline survey and predictive regressions estimated on a pre-test sample of existing gig workers.
Randomization Unit
Individual (gig worker)
Was the treatment clustered?
No

Experiment Characteristics

Sample size: planned number of clusters
n/a
Sample size: planned number of observations
200 (280) individuals. [See section on power calculation for details]
Sample size (or number of clusters) by treatment arms
100 (140) individuals control, 100 (140) individuals treatment. [See section on power calculation for details]
Minimum detectable effect size for main outcomes (accounting for sample design and clustering)
The online platform expects to recruit around 40 new gig workers per month. Based on power simulations using past gig worker data, we requested that the experiment be left in the field for at least seven months, giving us an expected 280 observations. With 257 observations, our simulated MDE for desired quantity with 77% power is .3 standard deviations. Due to planning constraints, however, the online platform was able to make a commitment for only five months in the field (yielding an expected 200 observations). This reduces power: with 200 observations, our simulated MDE for desired quantity (with 79% power) is .35 standard deviations.
Supporting Documents and Materials

There is information in this trial unavailable to the public. Use the button below to request access.

Request Information
IRB

Institutional Review Boards (IRBs)

IRB Name
Ethikkommission der Wirtschafts- und Sozialwissenschaftlichen Fakultaet, Eberhard Karls Universitaet Tuebingen
IRB Approval Date
2016-08-09
IRB Approval Number
A2.5.4_065_aa

Post-Trial

Post Trial Information

Study Withdrawal

There is information in this trial unavailable to the public. Use the button below to request access.

Request Information

Intervention

Is the intervention completed?
No
Data Collection Complete
Data Publication

Data Publication

Is public data available?
No

Program Files

Program Files
Reports, Papers & Other Materials

Relevant Paper(s)

Reports & Other Materials