The Impacts of Employee Excellence Rewards on Workers and Firms: The Role of Managerial Discretion vs. Worker Empowerment

Our partner firm sources carpet orders from many small firms, each with around 5-50 carpet weavers (the workers) and one manager. The intervention we study will be a clustered randomized control trial, with 150 small firms. We will randomize whether a firm gets a workplace excellence reward program or not (the control group will also get rewards, but allocated at random). The recipient of the reward would either be decided by the manager, or via worker voting (the allocation mechanism would be randomly chosen). There will be one bonus for every 10 workers. Every two weeks, the recipients will be announced, and the reward will be given out in a small ceremony. The intervention will continue for 3 months. The random allocation (for the control group) would be done through a public lottery to ensure workers knew the allocation was truly random.

2025-01-13

2025-04-08

1) Worker productivity (measured as knots woven, adjusted for carpet complexity)

2) Worker attendance

3) Worker earnings

4) Perceptions about type of workers who won bonus (from endline question asking workers what sort of workers generally won the bonuses in their firms)

5) Actual type of worker who won bonus (constructed using baseline data and workplace observations made by our surveyors)

6) Workplace culture. We will look at the following three sub-indices, constructed using data from endline surveys with workers:
6a) Meritocracy, including reports that hard work/skill matter for compensation, and that the relationship with the manager matters for compensation (x -1)
6b) Recognition, including reports of feeling appreciated by manager, and by other workers
6c) Collegiality, including reports of workplace competitiveness (x -1), that worker asked another worker for help, that worker gave another worker help, the number of other workers the worker is close to, and the number of other workers the worker would ask for help

7) Complementary measures of collegiality
7a) Workers helping one another with work, as observed by our surveyors
7b) Tokens workers send to other workers in dictator game

8) Preferences, of both workers and managers, over bonus allocation mechanism

9) Workers’ mental well-being and perceived agency

10) Defects in carpets produced for our partner firm

We would randomize each of 150 loom centers to one of the following treatments (each arm will have 50 loom centers):
1. The manager chooses which worker the bonus is paid to (managerial discretion treatment).
2. Workers choose via voting which worker the bonus is paid to (worker empowerment treatment).
3. The bonus is allocated at random (control group).
Bonuses will be given fortnightly for three months, and will be roughly equal to 10% of workers’ average monthly wages.

We will use pairwise stratification to maximize power. Specifically, we will form groups of three firms, matched on gender and average worker attendance at baseline, and randomize within these groups. Firms may attrit, leaving their strata without observations from each treatment group for many outcomes. In such cases, we will pool the incomplete strata with the strata of the same gender that has the closest baseline attendance level when estimating treatment effects.

Outcomes will be measured both during and after the three months of bonuses.

We will estimate two main specifications: one that pools both treatments and compares them to the control, and one that estimates effects of the two treatment arms separately. We will use baseline controls and LASSO to improve precision.

We will test for heterogeneity by: firm gender (these are single-gender workplaces), whether worker is a worker the manager said at baseline they were close to, and worker’s baseline preference for bonus allocation mechanism.

We will also record the workers who came in second place in each treatment arm. In the control arm, all workers not randomly selected for the reward will be considered second place. This will allow us to estimate the impact of winning a bonus on workers' outcomes (described above), and also how that impact differs based on how the bonus was allocated.

Note: we revised the outcomes and proposed analyses described here on Feb 7, 2025. This was about a third of the way into our intervention phase but before we estimated any treatment effects.

Stratified randomization done on a computer

Firm

Yes

150

2250 workers, 150 managers

50 firms in each cluster