Signaling Purpose

We test the predictions in an online experiment conducted on Prolific. In a first step, we consider a “non-profit” setting where the output only has a social but no material value and vary whether agents have precise or imprecise knowledge about this social value – which in the experiment is the amount donated to a charity per unit of effort exerted by the agent. A control group where principals and agents are symmetrically informed about the social value of output is compared to a treatment group where only the principals know the value of output precisely before deciding on the bonus rate.
In a second step, we study a “for-profit” setting where the principals also receive personal material returns, i.e. their profits depend on the agents’ efforts. Here we compare again a control group where both are symmetrically informed to the social returns to two treatment groups. In both treatment groups agents are uninformed about social returns but while in one of them agents know the principal’s material returns ex-ante, they do not know these material returns in the other treatment.

2025-07-30

2025-12-31

The primary outcome variables in this study are the bonus rate set by the principals and the effort levels chosen by the agents. The study aims to determine whether bonuses can act as a signal of a task's social purpose, particularly when agents lack precise knowledge about the task’s exact contribution, and whether this signaling effect influences effort beyond the bonus's direct incentive role. Moreover, we test this mechanism both in a “non-profit” and a “for-profit” setting.

The experiment will be conducted on Prolific and proceeds as following: in the beginning, participants start with a survey, where we elicit the information about their willingness to donate, their beliefs about the willingness to donate of others, their attitude towards risk, their willingness to give money for a good cause, and their willingness to return favors.

Participants are then randomly assigned to one of two roles—Employer (principal) or Employee (agent) and to one of the five treatments. Within each treatment Employers and Employees receive identical instructions. Each participant is given an endowment and informed that the Employee will choose an effort level, which the Employer can incentivize by offering a bonus deducted from their own endowment. The effort contribution is costly, and both parties are made aware of the underlying cost function. Participants can explore different choices on their screens to understand their potential payments under various scenarios. Additionally, they learn that each effort point contributes to both a donation to charity and (depending on the treatment) a financial return to the principal. The rates at which efforts generate donations and profits for the principal are randomly drawn from a normal distribution with a mean of 10 and a standard deviation of 3.5 (rounded to integer values and capped at 0 and 20). The exact donation and return values are disclosed only on the task screen depending on the treatment.

As described above we study five treatments:

Non-Profit – Baseline: Efforts only produce social returns which are known to Employer and Employee.

Non-Profit – Unknown Social Returns: Efforts only produce social returns, and these returns are only known to Employers (Employees know the distribution from which they are drawn).

For-Profit – Baseline: Efforts produce both social returns and material profits for the Employer. Both material and social return rates are known to Employer and Employee.

For-Profit – Unknown Social Returns: Efforts produce both social returns and material profits for the Employer. Material returns are known to both, but social returns are only known to the Employers (Employees know the distribution from which they are drawn).

For-Profit – Unknown Social and Material Returns: Efforts produce both social returns and material profits for the Employer. Both, material and social returns are only known to the Employers (Employees know the distribution from which they are drawn).

Following the instructions, participants answer six comprehension questions to ensure understanding; those who fail twice are excluded from the study. The participants then proceed with their tasks, after which they complete a survey to provide demographic information (age, gender, education, race). Finally, participants are informed of their payments. Since Employees’ decisions follow the Employers’ tasks, Employers receive details of their final payments via direct message through Prolific.

Roles in the experiment are assigned sequentially: the first participant to enter becomes a principal. Once the principal completes their task, the next participant to join becomes an agent. The agent inherits the treatment assignment, donation rate, and bonus chosen by the principal. This setup ensures that every agent has a predefined bonus value, and if an agent drops out, a new participant can take over as an agent with the same inherited values. Additionally, principals and agents do not need to participate simultaneously, significantly reducing waiting times for participants.

Treatments are randomly assigned to participants. To ensure an equal number of participants in the treatments, participants are first assigned to batches. Within each batch, participants are then randomly assigned to their respective treatment condition.

Individual subject

No

-

We invite 2000 individual subjects to participate in the study.

We plan to have 400 individual subjects in each treatment.