Experimental Design Details
Participants are invited to the laboratory and split in two groups: the "workers" and the "advisors". The experiment consists of two parts. In part one, half the participants ("the workers") work on a real-effort task that consists of adding five two-digit numbers for five minutes and receive a piece-rate payment as well as a bonus if their performance meets a certain performance threshold. The advisors do not do anything in this part of the experiment.
In part two, we tell workers that they will have to work on some other real-effort task, and can choose either an “(A)dvanced” or a “(B)asic” task. Workers know nothing else about the task. Instead, we match them to an advisor who is given more detailed information on the two tasks and can therefore give valuable advice to the worker. This mimics real life situations, where “advisors” (teachers, supervisors) may have more information about different career paths than workers or students. To help the advisor give informed advice, we ask the worker to prepare a statement describing their prior experience with mathematics tasks, and indicate their confidence in their ability on such tasks on a scale from 0-10.
We then ask all worker-advisor pairs to, one by one, verbally confirm their presence in the laboratory. As part of this process, each advisor will get to hear the voice of his/her matched worker (and vice versa). We use this method (originally due to Bordalo, Coffman, Gennaioli and Shleifer; AER, forthcoming) to allow advisors to reliably identify their worker's gender, without making it salient to participants that we are interested in gender differences in advice.
We then send the statement and confidence (which jointly refer to as the "motivation letter") to the advisor along with the motivation letters of four other workers in the experimental session. We ask the advisor to recommend task A or task B for all five workers, though only the advisor's actual matched worker will receive the advice he/she sends. The advisor does not know which of the five worker is his/her matched worker. This implies that it is incentive compatible for each advisor to treat each letter as if it was written by his/her worker.
Workers receive advice from their advisor, choose between the two tasks and subsequently work on their chosen task for 10 minutes. The actual task is the same as in part 1; the only difference is that in the basic task (B) participants were paid $13 if they solved at least $13 math problems (and zero otherwise), whereas they were paid $26 if they solved at least $26 problems in task A (and zero otherwise). While workers are working on the task, we ask advisors to go through all five letters again and tell us how confident they are that each worker will reach the respective threshold in each task on a scale from 0 to 10. We also ask them to specify the threshold for a minimum score achieved in part 1 for which they'd recommend the advanced task to a hypothetical worker. The experiment concludes by giving workers and advisors feedback on their earnings, having them go through a brief questionnaire asking them for their gender, what gender they thought their matched worker/advisor had, as well as a number of questions related to how they came to their advice (advisors only) and their risk preferences and self-confidence. The final part of the questionnaire is an implicit association test aimed at testing gender stereotypes.
Our initial sessions will have only this single baseline treatment. In future sessions, we expect to run two additional treatments that either (a) eliminate the procedure that reveals the gender of their partner to each participant or (b) additionally present advisors with the worker's score in part 1 when they are deciding what advice to give.