Discretion in Referrals: Evidence from a Tanzanian Export Garment Factory

Our partner firm is offering an on-the-job training program for promising workers, focusing on technical production, communication, leadership, and management skills. The program will be designed and implemented by our partner firm.

In a follow up phase of the project, our partner firm will be choosing workers to internally promote for a new facility they are opening. As part of the selection process, they will give a leadership test measuring a worker’s technical knowledge and leadership potential. Prior to the test, they will provide all workers with study materials with relevant information that will be tested. They will also ask supervisors to provide performance evaluations of workers, but as each supervisor supervises a large team and there is not enough time to evaluate all workers, in collaboration with the research team it will be randomly selected which workers are evaluated by their supervisors. Workers will be informed whether they were randomly selected for a performance evaluation and that if they were selected then the factory will assess their readiness for promotion by both their performance on the leadership test and their supervisor’s evaluation of their performance and that if they were not selected then the factory will assess their readiness for promotion by both their performance on the leadership test and their performance record.

2023-06-22

2023-08-31

The key outcome variables of interest are the types and quality of the workers that apply and that are referred under the different referral experimental conditions, including gender, religion, past performance, and training assessment scores.

In a follow up phase of the project, our key outcomes of interest will also include whether workers demand performance evaluation information, how much workers study for the leadership test (as measured by how many questions related to the study materials they answer correctly on the test), how well they perform on the leadership test, whether they chose to apply for promotion, and whether they invest in their own skills development (as measured by whether they choose to attend a voluntary training session).

Part 1 - Supervisor and coworker referrals: Supervisors and a random sample of workers will be given the opportunity to refer other workers for the training program. All supervisors and referring coworkers will receive a referral bonus if the worker(s) they recommend for the training successfully completes the training and scores within the top 50% of the training cohort in an assessment of their technical knowledge and soft skills. Prior to giving referrals, supervisors and referring coworkers will be randomized into one of two groups:
- Ex-ante bonus announcement: Respondents will be informed about the performance-based referral bonus prior to giving their referrals
- Ex-post bonus announcement: Respondents will be informed about the performance-based referral bonus after they have given their referrals

Part 2 - Worker applications: All eligible workers will be given the opportunity to apply for the training program. Specifically, by production teams, workers will be invited to an application session to learn about the training program and receive applications where they can indicate whether they want to apply to the training program or not. In these information sessions, workers will be randomly assigned to receive either one of two versions of the application form. The application forms are identical except for a difference of two words. Specifically, in the "Status Quo" version, the application form states that selection for the training program will be based on many criteria, including supervisor referrals. In the "Meritocratic" version, the application form states that selection for the training program will be based on many criteria, including past performance records.

In a follow up phase of the project, we will randomly select some workers to receive a performance evaluation from their supervisor. Analogously, we will inform workers whether they were randomly selected to receive a performance evaluation from their supervisor and, if they were selected for a performance evaluation, that the factory will assess their promotion readiness based on their performance on a leadership test and their supervisor’s evaluation of their performance; or, if they were not selected for a performance evaluation, that the factory will assess their promotion readiness based on their performance on a leadership test and their performance record.

Not available

For the referrals experiment, supervisors will be invited to attend a session for the referral process by facility. Within each facility meeting, supervisors will be randomized in person, into two groups, with assignments alternating consecutively based on the order in which supervisors are seated during the meeting. Within each facility, the referral process will then be conducted separately for each group, with one group receiving the ex-ante bonus announcement and the other group receiving the ex-post bonus announcement. The exact same randomization process will be used for the coworker referrals. For the application experiment, application forms will be pre-sorted in an alternating pattern (C/T/C/T/...etc) and handed out to workers consecutively based on how they are seated during the application information session for their production team. Both randomization methods are comparable to the randomization procedure used in Bursztyn and Jensen (2015).

The randomization for the follow-up phase of the project will follow a similar randomization strategy to the one used for the referrals experiment. All workers will be invited to attend a lunch session, upon arrival they will be randomized into either group A or group B consecutively in the order that they arrive. One group will be randomly assigned to receive performance evaluations from their supervisors, while the other group will be randomly assigned to not receive performance evaluations from their supervisors.

For all experimental parts, randomization is conducted at the individual level.

No

Sample sizes:
(1) Worker applications: 1,200 workers
(2) Coworker referrals: 250 randomly selected workers
(3) Supervisor referrals: 120 supervisors
(4) Performance evaluations: 2,000 workers

(1) Worker applications: 1,200 workers (2) Coworker referrals: 250 randomly selected workers (3) Supervisor referrals: 120 supervisors (4) Performance evaluations: 2,000 workers

(1) Worker applications: 600 in the "Status Quo" form group; 600 in the "Meritocratic" form group
(2) Coworker referrals: 125 in the "Ex Post Bonus" group; 125 in the "Ex Ante Bonus" group
(3) Supervisor referrals: 60 in the "Ex Post Bonus" group; 60 in the "Ex Ante Bonus" group
(4) Performance evaluations: 1,000 in the "no supervisor evaluation" group; 1,000 in the "supervisor evaluation group"

The main power constraints come from the coworker and supervisor referrals. A sample size of 120 supervisors will allow us to detect a 0.5 standard deviation (SD) difference in the average performance of workers referred by supervisors in the "Ex-Ante Bonus" and "Ex-Post Bonus" groups, with an alpha level of 0.05 and a power of 80%. Assuming the average performance of the referred workers in the "Ex-Post Bonus" group is at the 50th percentile with a standard deviation of 20%, a 0.5 SD difference means that the average performance of the referred workers in the "Ex-Ante Bonus" group would be at the 60th percentile. Similarly for coworkers, a sample size of 250 workers has been chosen to ensure that a 0.35 SD difference in the average performance of workers referred by coworkers in the "Ex-Ante bonus" and "Ex-Post bonus" groups can be detected. Assuming the average performance of the "Ex-Post bonus" group is at the 50th percentile, a 0.35 SD difference would correspond to a shift of 7 percentile points to the 57th percentile.