Discouraging Network-Based Hiring Without Compromising Productivity: Experimental Evidence from Low-Wage Workers in Pakistan

In the experimental phase, we introduce three treatment arms by randomly varying aspects of the manager-laborer interaction. The treatment arms are:
- Information treatment: For the subset of above-average productivity workers from baseline measurements, we will randomly select a subset of labourers and provide their names and contact information to managers, informing the manager that these were high-performing labourers. The information treatment is randomised at the manager- labourer level.
- Incentive treatment: Managers are provided with an incentive payment for their team’s average productivity for a given day. The incentive treatment is randomised at the manager-day location level.
- Monitoring treatment: We vary whether managers play a monitoring role for the team they hire on a given day. This is to be done by removing managers from the worksite and having their team monitored by a different manager. The monitoring treatment is randomised at the manager-day level.

2019-06-14

2027-12-31

The study will understand the relevance of five mechanisms using the following primary outcomes:
Worker’s productivity on a given day
Hiring status (whether a worker is hired by the manager on a given day)

In order to quantify workers' productivity, we utilise a real-world setting where employers and laborers are doing their normal tasks. After a tremendous amount of piloting, we developed an laborer observation tool to collect detailed information for tasks on an hourly basis. Our productivity measure incorporates both the rate at which labourers complete their tasks and the total time spent at work each day, controlling for site-date level and a range of other sources of productivity shocks. Mainly, we measure workers’ output by the rate of bricks or tiles or blocks (and plaster) placed per minute times the total minutes spent working (end time – start time – breaks).

The study will also assess how the treatment condition influences the effort or time allocation, job search effort, and any resulting changes in their income.

We partnered with several major construction firms operating across urban and suburban areas of Lahore, Pakistan, employing hundreds of workers. As is typical in the construction and agriculture industries, these firms rely on managers (contractors) who act as intermediaries between the firm and the labourers. The managers are responsible for hiring, overseeing, and supervising the laborers. The laborers are hired on short-term, renewable contracts that typically last for a few days or weeks.
To identify potential construction firms for our study, we contacted the relevant government department and requested a list of construction firms involved in upcoming projects in Lahore. Firms were identified for selection if they had brick, tile, or block work (possible paintwork) for a significant number of months, operational sites in Lahore or its surrounding areas, hired workers on short-term contracts, and consented to allow observations and treatment roll-out on-site. All contractors and workers involved in brick, tile, block, and painting tasks are eligible to participate in the study.
The experiment is structured into two phases. In the first phase, we conduct laborers observations and surveys. The second phase involves the implementation of three treatment arms—information, incentive and monitoring. During the RCT phase, we continue to collect data through observations and surveys with labourers and managers.

Phase 1: Laborer Observation
First, we measure the daily productivity of labourers for on average three days on real tasks using a laborer observation tool developed by the research team. This includes a sample of labourers who have previously worked with the firm, along with labourers who are unknown to the hiring managers.

Phase 2: Information, Incentive and Monitoring Treatments:
Then, during the experimental phase, we introduce three treatment arms by randomly varying aspects of the manager-laborer interaction. In the information treatment, we share the names and contact information of a subset of above-average productivity workers with managers, informing the manager that these were high-performing labourer. The incentive treatment provides incentives to managers based on their team’s performance at a randomly selected location on a given day. Lastly, we implement monitoring treatment by varying whether managers play a monitoring role for their team.

To test five mechanisms, we collect data from three primary data collection exercises: labourer workday observations, labourer surveys, and manager surveys.
- Laborer workday observations: enumerators observe workers during the entirety of their day and record the tasks conducted and the output of that task on a minute-by-minute basis. This data allows us to have an accurate picture of laborers productivity. Laborer workday observations are conducted before and during the RCT phase.
- Laborer survey: Laborer surveys capture comprehensive information on worker demographics, employment, job search, contractual arrangements, supervisor monitoring, teammates, and social networks. Laborer surveys are conducted at the baseline and at the follow-up.
- Manager survey: We conduct a survey with managers to capture their beliefs about workers' productivity, the search process for labourers, social networks, and demographics.

Not available

The treatments were randomly assigned using a computer-generated number (except for a few pilot or initial rounds of some treatments).

Information treatment: The information treatment is randomised at the manager-labour level.
Incentive treatment: The incentive treatment is randomized at the manager-day level.
Monitoring treatment: The monitoring is randomized at the manager-day level.

Yes

Laborers: 1,700
Managers: 85

Laborers: 1,700 Managers: 85 Laborers Daily-Observation: 7,000

Information Treatment: Laborers: 450 (Control: 150 and Treatment: 300)
Incentive Treatment: Observations Days: 740 (Control: 370 and Treatment: 370)
Monitoring Treatment: Observations Days: 100 (Control: 50 and Treatment: 50).
We will use different approaches to test for monitoring treatment arm. This will be a combination of randomly varying managers presence or using natural variation in managers presence on site. The sample will be based on which approach will be used. We expect that we have 100 manager days from treatment variation.

Given the sample specified, baseline-endline correlation, 3 pre-observations, 5 post-observations for the output outcome and 3 post-observations for the hiring outcome, using an ANCOVA and taking into account the treatment interaction structure (Muralidharan, Romero, and Wüthrich, 2022), we expect to detect the following effect sizes with power of 80% and alpha of 0.05. Effect size for hired: Information (0.13 sd), Incentive (0.16 sd), and Monitoring (0.17 sd) Effect size for output: Information (0.11 sd), Incentive (0.13 sd), and Monitoring (0.14 sd) We assume a 0.5 sample test correlation for the hired outcome using 200 units as the treatment and 800 control group. We use a 0.6 sample test correlation for the output using 200 units as the treatment and 800 control group.