Experimental Design
[ Work task ]
We are working with Karya—a digital work platform committed to ethical AI practices. Unlike many lower-paying , exploitative jobs in the sector, Karya offers attractive wages (starting at $5 per hour, 20 times the Indian minimum wage,) with the potential to earn up to $1,500 per person. Karya gig workers use smartphones to perform simple microtasks, the outputs of which are often used to build training datasets for AI and machine learning algorithms.
Our study tasks are accessible to workers with only basic smartphone literacy and educational attainment: they require workers to read and record simple sentences displayed on the app, earning piece-rate wages for accurate recordings. These tasks are conducted in local languages (Hindi) and the app itself is in Hindi, to ensure it is accessible to lower-education populations. Before the job, workers are given in-depth training on setup, job execution, payment, and troubleshooting.
[ Sample and Recruitment ]
Our proposed work is in Bihar, one of the lowest-income states in India, with a female labor force participation (FLFP) rate of around 10% (Periodic Labour Force Survey, 2023). We conduct our study across 175 Gram Panchayats (GP) in Bihar. The study population are low-income married women and male decision makers from the same households. To identify eligible households, we will first conduct a mapping exercise of the sample communities using a random walk method. Based on the mapping exercise, we will identify eligible households and enroll a random subset (up to 20 households) from each community.
Eligible households must have a married female and male decision maker in the household that meet the following eligibility criteria:
- Access to a smartphone and a bank account,
- Can perform basic functions on smartphones
- Able to read in Hindi
- Be of prime working age (18-60)
- Available in the community during the study period.
- In addition, the eligible female participant should not have a regular paid job.
[ Randomization ]
Prior to beginning fieldwork, we will randomize sample communities into one of the following conditions:
- Control: No jobs offered
- Treatment 1: Female Reservation—one job offer is made to the selected female respondent in the household
- Treatment 2: Male Reservation—one job offer is made to the selected male respondent in the household
- Treatment 3: Open—one job offer is made to the household, and either the selected male or female respondent can take the job
Communities in the three treatment arms will be cross-randomized into either a high or low wage condition (thus creating 7 experimental cells in total). In all treatment communities, enrolled households will be offered 4-week-long digital jobs using the Karya mobile app; the work is part time, and should take up to 1-2 hours if workers complete all assigned tasks. While one woman and one man are surveyed from each household, only one member of the household will be allowed to sign up for the job. In communities assigned to the high wage condition, experimental job offers will have daily pay slightly above the market unskilled daily wage for men; in low wage communities, job offers will have daily pay below the market daily wage for unskilled female labor.
[ Study Procedures ]
Once households are enrolled into the study, we will conduct three surveys. In the first survey (baseline), we ask men and women to answer questions related to the following topics: basic demographics, baseline attitudes towards women’s work and phone use, household income and expenditures, female decision-making power, and women’s work status.
At the end of the baseline survey, the respondents in the treatment groups are presented with the information about the Karya job and told they will have the opportunity to sign up for the job when we return and conduct a follow-up survey with the respondent approximately five days later. We elicit initial interest in the job and predictions about household take-up decisions. After baselines are completed in the community, we hold a job information session at a central location in the village to provide more detailed information about the job.
The second survey (pre-work endline) is conducted after the job information session. During the pre-work endline survey, the participants in the treatment groups are first asked to indicate the household’s final decision about the job sign-up. Following this, the survey asks a battery of norms questions and questions about suitability of Karya work for women.
The pre-work endline survey is followed by a community-level job-offer lottery and an onboarding procedure: If fewer than 10 households sign up for the job in a given community, all of them are onboarded to the job. If more than 10 sign up, we will conduct a lottery to select 10 to onboard.
The final (post-work endline) survey will be conducted over the phone with all study participants to capture additional changes in attitudes after actual work is completed (note that these outcomes confound treatment effects of job reservation with access to work, and therefore have a different interpretation; we may exploit the job offer lottery to assist in unbundling these effects). Here we will also measure attitudes using two real-stakes situations. First, respondents can vote to influence the future reservation policy in their community; (we will implement future work and the chosen reservation policy in a randomly chosen community). Respondents are allowed to cast a vote for their preferred reservation policy—open, reserved for men only or reserved for female only—or receive a small compensation instead of voting. The second measure involves taking a costly action (i.e., filling out a digital form) to access information related to women’s work opportunities. We also ask questions about household spending, decision-making, and mental health.
[ Main Hypotheses ]
The main outcomes of interest include personal views surrounding women’s work and smartphone use, as well as beliefs about the suitability of Karya work for women (collected at the pre-work endline). Questions are designed to capture three prominent norms dissuading female work: work as a threat to women’s purity, work interfering with women’s household responsibilities (domesticity), and the norm that the man should be the breadwinner. Some questions are designed to capture second-order beliefs about the community norms for these same dimensions.
We test the hypothesis that men will express more conservative norms under the open treatment, especially in the high wage arm (we discuss alternative hypotheses for how gender reservations could affect norms in our PAP). If this is due to the desire to reserve jobs for men, we would expect the effects to be larger if men lack access to other good job opportunities. We also explore whether the effects may differ depending on how constrained women are in terms of bargaining power and access to work at baseline. More details on these dimensions of heterogeneity are provided in the PAP.
[ Supplementary Experiment ]
As an add-on intervention with the Karya workers, we include some stories in the sentences that the workers have to read as part of their work. A random set of workers see treatment stories that are related to female empowerment, e.g., a female protagonist who is able to work and contribute to her household. Another set sees placebo stories, which are the same stories with a male protagonist. The remaining workers belong to the pure control group which only sees disjoint sentences. This allows us to assess whether reading the treatment stories also affects the post-job survey outcomes. A set of questions checking for story comprehension and measuring gender attitudes will be directly administered through the app.