Experimental Design Details
On Prolific, I recruit two different groups of participants, one to play the role of worker and one to play the role of employer.
Workers complete an initial study with two tasks: a piece rate quiz and a tournament quiz. In the piece rate quiz, the worker earns a piece rate per correct answer. In the tournament quiz, the worker earns a prize of $X if they beat the score of another random worker, and 0 otherwise. In the initial study, the workers agree to complete a follow-up study where they'll do similar tasks. Employers' main job is to decide how to assign the workers to the tasks in this follow-up.
The employer study works as follows. The employer is randomly assigned to one of two treatments. In both treatments, the employer makes decisions about a subset of workers (their "group"). The treatments vary what the workers in the group do in the follow-up, as well as details of the employer's decision.
Treatment 1: Pay Only. All of the workers in the employer's group complete another piece rate quiz and another tournament quiz in the follow up study. In the piece rate quiz, the worker earns a piece rate per correct answer. For the tournament, the worker is randomly partnered with another worker in their employer group; they earn a prize $X if their score on the tournament beats their partner's score on the 1st piece rate quiz and $0 otherwise. For each worker, the employer provides their willingness to pay to decide themselves between two options. In Option A, the tournament counts for the employer and worker: the employer and worker earn $X if the worker beats their partner's score on the 1st standard quiz. In Option B, the piece rate quiz counts for the employer and worker: the employer earns $X if the worker's score is weakly greater than a randomly chosen threshold and the worker earns a piece rate per correct answer. If the employer does not purchase decision rights, the computer chooses an option randomly.
Treatment 2: Pay and Task. The workers in the employer's group complete only one task in the followup study: either another piece rate quiz or another tournament. The rules for each task are the same as in Pay Only. As in Pay Only, the employer provides their willingness to pay to choose between two options. Under Option A, the worker does the tournament, and the employer and worker are paid for this: they receive $X if the worker beats their partner's score on the 1st piece rate quiz and $0 otherwise. Under Option B, the worker does the piece rate quiz and the employer and worker are paid for this: the employer earns $X if the worker's score is weakly greater than a randomly chosen threshold and the worker earns a piece rate per correct answer. In other words, the pay consequences of each option are the same as in Pay Only. The only difference is that under Option A, the worker only does the tournament. Under Option B, the worker only does the piece rate quiz. In Pay Only, the worker does both tasks regardless of the employer's choice. Again, if the employer does not purchase decision rights, the computer chooses an option randomly.
Sequence of experiment: Each worker will choose a nickname when completing the 1st study, and this nickname will be used to identify them throughout the employer study. Before beginning, employers will observe two pieces of information about every worker in their group: performance on the 1st piece rate quiz and chosen nickname (implicitly revealing gender). The employer will make three decisions about each worker: they will 1) predict the percent chance they'll earn $X under each option, 2) choose their preferred option, and 3) report their willingness to pay to receive this option for sure.
Payment: There will be more employers than workers. So, for each worker, one employer will randomly be selected. This employer's choices will decide what happens for that worker, and also count for the employer's payment. This means each employer has at most one assignment decision count. Additionally, the employer's predictions will be incentivized via binarized scoring rule. A random subset of the employers' predictions will be paid. In Pay Only, this is straightforward: if a prediction about task j for worker k is selected for payment, the employer will be compensated based on worker k's performance on j. In Pay and Task, however, the worker may not actually complete task j. In this case, I will select what to the employer would be an observationally equivalent worker (same gender and score on 1st piece rate quiz) in Pay Only. They will be paid according to the performance of the observationally equivalent worker.
A given worker will only be observed by employers in one treatment. The workers will complete the follow-up study according to the assignment decision of an employer. The workers will not be informed that employers made decisions about them: they will simply receive directions for their assigned task (in Pay and Task) or both tasks (in Pay Only). The employers will be told that the workers will not learn about them.