We conduct a survey experiment with workers at two of the apparel manufacturer's factories, which together employ over 8,700 people. First, we conduct a stratified random selection of workers to participate in the survey experiment. Using the entire list of employees in the two factories, we sample workers from four types of production teams. Among these teams, we chose teams with a sufficient large number of workers to accommodate this survey and a separate survey that we are running for a different project (approximately above 15). We are left with 112 eligible teams and a total of 5,948 eligible workers, out of a workforce of 8,727 people, including 1,000 managers and administrative staff members.
We next stratify workers on eligible teams by their sex, which we identify based on name (male, female, uncertain). We are able to categorize names as male or female for 5,929 out of 5,948 eligible workers. In some cases, there are teams with very small numbers of one group; in this case, we aggregate these workers to the smallest level that yields a sufficiently large groups size (e.g., production section-floor).
We conduct this RCT in coordination with another survey experiment that we are running with the apparel manufacturer (RCT ID: AEARCTR-0007103). Consequently, we randomly sample workers from strata as follows:
1. For workers in the sewing section, for sewing line & female strata, randomly select 11 workers per stratum. For sewing line & male strata in the sewing section, which are smaller, randomly select men in proportion to the strata size. This produced an initial target sample of 574 workers.
2. The remaining 5,374 workers were eligible to be recruited for the other survey experiment, AEARCTR-0007103. For this survey experiment, within strata, the research team randomly selected workers as target participants. We randomly ordered the remaining workers in each stratum so that workers who declined to participate could be replaced. We conducted the second survey experiment, for which we recruited or attempted to recruit 3,638 workers. This left 1,736 eligible workers from the list of 5,374 workers who had not been recruited or attempted to be recruited for AEARCTR-0007103.
3. To arrive at the final target sample for this survey experiment, we appended the list of 574 workers and the list of 1,736 workers, for a maximum total sample size of 2,310. Given our successful recruitment rate of 62% in AEARCTR-0007103, we anticipate achieving a realized sample size of around 1,432.
We randomly assign workers to treatment arms in Stata. We randomly assign workers to each treatment arm with equal probability. To address misfits across strata, we use the randtreat package by Carril (2017). We conducted 10 randomizations and selected the one that performed best in terms of balance on two covariates available to the research team (tenure and skill group).