Training Experiment in the Latin American Quick-Service Restaurant Industry. Enhancing Managerial Skills and Workforce Development.

2025-01-29

2025-12-31

Performance (sales, units sold, and tickets)
Personnel variables (hiring, turnover, absenteeism, scheduling, training)
Manager report indices (control, personality, experience & education, soft & hard skills, metrics, broad skilling, preemptive planning, rapport, pyramid organization, autonomy, and kaizen)
Worker report indices (rapport, psychological safety, harassment, and satisfaction)

All managers at the store level were selected as training participants, with a total of 460 managers in 74 restaurants. Multiple managers can work at the same time in the same store, and managers are divided into shift managers, area managers or restaurant managers. Randomization was thus done at the store level, and stratified on three variables. The first is store location, grouping all stores in the capital city, Bogotá, and all stores in other cities. Out of the 74 stores, 34 are located in Bogotá and 40 in other cities. The second variable used for the stratification is the gender balance of the store.Following Adhvaryu et al. (2023a), we construct a measure of Manager-Worker gender balance. The measure identifies "imbalanced" stores as stores where the share of managers of a certain gender exceeds the share of workers of the same gender by at least 15% (stores where the share of female managers exceeds the share of female workers by at least 15% are denoted as Female-heavy Management or FHM, and stores where the share of male managers exceeds the share of male workers by at least 15% are denoted as Male-heavy Management or MHM). We also classify "balanced" stores into Male-balanced or Female-balanced, with stores where the share of male managers exceeds the share of male workers by less than 15% and where the share of female managers exceeds the share of female workers by less than 15% respectively. We do not observe stores where the share of male or female managers is exactly equal to the share of male or female managers. The third variable is a binary variable that outlines stores that are in the process of closing down. While implementing the deployment of our intervention, our firm partner made us aware of the plans to close down 3 stores sometime between the planned start of our intervention and the end of the year. As such, we stratify for this stores so that after they are closed, no group ends up with a larger number of stores. By stratifying our sample this way, we can also expect to to identify differential treatment effects according to the gender composition of the store managers are at.

Not available

We randomize into 3 treatment arms, one for each curriculum and control, and all stores from our firm partner are selected into one of the three arms. As such, stores are randomized into 3 possible groups. While the control group has no access to any curriculum, the managers at these stores were given access to the hosting platform but were only able to see and fill out the pre-post testing survey for each module of the curriculum, and were told that the survey is intended for a managerial characteristics study. Additionally, since managers in the control only had to fill the 4 pre-post surveys, we expected that they would be able to complete this in around 15-20 minutes, and were likely to not return to the platform to fill the surveys again later. As such, control managers were only asked to fill the survey once rather than twice.

In order to make the three intervention arms as similar as possible and improve balance, we run a re-randomization exercise. Following Banerjee et al. (2020), we first run 5000 iterations of the randomization, followed by performing balance tests between groups for several relevant measures: store sales, both monthly and quarterly, employee hiring and turnover, and manager indexes, using the indexes described in section 3. After we have run these balance tests, we select 5% of the randomization that performed better, and randomly select one of these to be our final randomization. This ensures that our randomly assigned groups are balanced and are as comparable as possible while still being assigned randomly.

Restaurant/store

Yes

71

700 workers 400 managers

C=23
T1=24
T2=24