AEA RC T Registr y

The American Economic Association's registry for randomized controlled trials

The implementation partner, a coalition of multinational retail and apparel firms, is responsible for the intervention. The coalition is a group of firms that are working together to improve the safety performance of their shared supplier base in Bangladesh. The coalition has several programs designed to bring its suppliers into compliance with local labor laws. I study one of these programs using a randomized-controlled trial.

2017-02-15

2018-06-30

The outcome variables for this study will be disclosed via a Pre-Analysis Plan prior to the study completion. I will publicly post the outcome variables following the completion of the trial. Outcome variables are included for factories, workers, and managers.

I am targeting having 80 factories in the intervention group, with 40 factories randomly assigned to the treatment group and 40 factories randomly assigned to the control group. The randomization procedure blocks factories based on size (binary distinction between factories with less than or equal to 3,000 employees and factories with greater than 3,000 employees) and on “readiness for SC Program,” or when the factory met Alliance prerequisites for participation in the SC Program (factories are randomized to treatment and control conditions in batches). The study duration is for nine months. The randomized assignment to treatment and control statuses will be used to identify the intervention’s effects.

I am implementing independent data collection, with a data collection partner, at factories selected to be in the evaluation group. I am also collecting data from several other sources.

The evaluation group of factories is being selected through a two-step process. The pool of eligible factories is limited to factories that supply exclusively to the Alliance (and do not supply to a separate coalition of multinational firms). This is due to difficulty coordinating evaluation implementation with the other coalition (the eligible pool of factories remains a very relevant population to study). From this pool of about 300 factories, the Alliance verifies that a factory has a separate committee that is formed in compliance with Bangladeshi labor law (more details provided in the Pre-Analysis Plan). This other committee must be elected by workers and is responsible for appointing worker representatives to the SC. Once the Alliance verifies that this committee is compliant, a factory is eligible for the SC Program. Each time the Alliance has a batch of verified factories, it sends the list to the researcher. The researcher randomly assigns half of the factories to the treatment group and half to control (after blocking on factory size). The non-random selection of factories for the evaluation raises an external validity concern. I have access to numerous data sources that will allow me to compare my sample to the broader populations of Alliance-covered and exporting factories in Bangladesh along many relevant characteristics. I am confident that the factory sample is a relevant population that allows me to make generalizable conclusions. I am collecting several types of data at varying frequencies. First, I administer baseline and 9-month factory questionnaires to collect factory human resource and production-related data at the monthly frequency, so I will have 9 observations during the evaluation period. Second, factory visits are conducted at baseline, 4 months, and 9 months. At factory visits, surveys are conducted with workers, lower-level managers, SC members, and the factory manager. At factory visits, safety-related documentation required by law and/or the Alliance are also verified. Finally, at eight factories, focus group discussions (FGDs) are conducted with workers and with lower-level managers. Third, I collect Alliance data, including (1) building safety audit performance; (2) building safety remediation progress; (3) number of Alliance members purchasing from the supplier; (4) call logs for the Alliance’s worker helpline; (5) factory escalation and suspension statuses. Data types (2)-(5) are available monthly.

Randomization done in office by a computer.

Randomization to treatment status is conducted at the factory level. The main level of interest is factory-level outcomes. I will also analyze worker- and manager-level outcomes, for which participants are cluster-randomized into assignment (where the factory is the cluster).

No

80 factories.

Repeated cross sections will be taken at each factory. The sample size per factory depends on the number of employees at the factory (due to production disruptions caused by taking workers and managers out of their tasks). At most factories, 20-25 workers and 20-25 managers will be surveyed each data collection round. Under the assumption that all 80 factories can contribute 20-25 worker and 20-25 management participants, the total number of observations will be between 1,600-2,000 observations for workers and managers, respectively, for each round of data collection. Continuing under the same assumption, across all rounds of data collection, the total number of observations of each type will be between 4,800-6,000 observations for workers and managers, respectively. The total number of observations for each group (in particular managers) may be below the estimated amount if a significant number of the evaluation group factories are smaller than I anticipate.

40 treatment factories and 40 control factories.