STOCHASTIC REWARDS AND WORK PERSISTENCE

2021-09-15

2021-12-31

Our primary outcome is the likelihood of quitting the work task depending on the various conditions.

This experiment is conducted online with subjects registered at the University of Nottingham.
This is a 2x2 factorial design with two dimensions and two levels per dimension. The first dimensions regards the uncertainty of rewards (certain or stochastic) and the use of an attention popup (present or absent).

All treatments use a real-effort task in which participants have to make summations of five one-digit numbers for up to 30 minutes. They can quit the task at any moment in time by clicking the corresponding button. After quitting the task, participants are not able to go back to the task screen and can terminate the experiment. Before the experiment ends, all subjects are asked questions about self-reported stress, attention to the task and enjoyment of the task. In addition, they are also administered the IMI (Intrinsic Motivation Inventory) questionnaire. In a companion trial, we independently measures subjects' cognitive skills (cognitive reflection and inhibitory control) along with their addiction to work/study (Bergen work addiction scale).

In the certain treatment (baseline), participants receive exactly 2 tokens per correct answer. In the stochastic treatment, the number of tokens received by participants for each correct summation follows a random pattern. The token value is an integer number between 1 and 23 with equal probability. The token value was drawn after each correct summation, and participants knew the stochastic process generating the token value. In all treatments, the reward is divided by an exchange rate equal to 1, 2, 4, 8, 80, and 500 in the six 5-minute periods of the experiment. The two treatments entail the same expected value for the rewards thus only differing on the dispersion of rewards.

The other dimension of the design varies the presence of a popup warning subjects of a change in the exchange rate. This second features implies that we conduct four treatments: certain (with and without an attention popup) and stochastic (with and without an attention popup).

In a fifth treatment, we ask subjects before starting the experiment whether they want to participant in the certain or stochastic treatment (without a popup). The rest of the experiment is the same as described above.

On September 15th 2021, we decided to add three new conditions to the study after observing the results of the previous conditions.
We have added a sixth treatment, which is the same as the fifth treatment but with a popup announcement.

We have also added two new conditions which are the same as the certain and stochastic treatments with the popup, except that no real-time feedback is provided to the participants on the adding task.

Because of COVID restrictions, the experiments cannot be conducted at the University of Nottingham. The study is thus conducted online via MTurk. Because our previous results were based on lab findings, power calculations was redone and so we need to recruit about 100 people per treatment instead of 70.

In December 2021, we also decided to contact the participants of the MTurk study that completed the four main treatments to administer them a questionnaire on gambling habits and attitudes. We also ask them to bet on lotteries with varying levels of entropy with and without feedback, following the design of a previous preregistered study (see here: https://aspredicted.org/see_one.php).

Computerized random draws.

At the individual level.

No

We aim at recruiting 70 subjects per treatment.

We aim at recruiting 70 subjects per treatment so 350 in total.

n=70 by treatment

Analysis