Workplace Requests and Negotiation Support (WINS) among Retail Workers in Delhi

We study everyday workplace requests among retail workers and employers in Delhi. While existing research on negotiation often focuses on job offers or salary bargaining, much less is known about how workers already employed in small firms ask for changes to their jobs, how employers perceive these requests, and whether workers correctly anticipate employer reactions. These dynamics may be especially important in small firms, where employment relationships are informal and many aspects of work are adjusted through direct communication between workers and employers. The project has two parts. First, we conduct an in-survey vignette experiment with employees and employers to measure beliefs, expectations, and stated behavior around workplace requests. We refer to this section of the study as Part 1. Respondents evaluate requests for salary increases, role changes, and non-monetary workplace improvements. For the request they are shown, we measure whether it is perceived to be legitimate, how receptive they expect the employer to be, how risky they think it is for a worker to make such a request, the anticipated costs to an employer of conceding such a request, how common they believe such requests to be, second-order beliefs about other workers and employers, and the inferences respondents draw about the worker making the request. This allows us to compare worker expectations with actual employer responses and to identify where beliefs about workplace voice may be miscalibrated.

Second, we test whether structured negotiation support changes workers’ real asking behavior. Employees who express interest in receiving support and agree to be contacted by WhatsApp are randomized to receive either a structured preparation tool or a control condition. The tool helps workers identify a feasible workplace improvement, prepare a concrete request, anticipate the employer’s perspective, and frame the request constructively. We then measure whether support changes whether workers make a request, what they ask for, how they ask, and whether their requests lead to changes in workplace outcomes. We refer to this section of the study as Part 2.

2026-08-15

2026-12-15

Vignette Experiment:

Employee outcomes. After the vignette, employees are asked whether the request was legitimate (Q1); the highest reasonable salary the worker could ask for, in the salary arm only (Q2); how much they would like working with the protagonist (Q3); whether they would personally make the same ask (Q4); how likely they think the employer is to grant the request (Q5); how likely the worker is to face negative consequences for asking (Q6); two incentivized second-order belief questions, one about employers and one about employees (Q7 and Q8); how common they think this type of request is (Q9); and what they would do if the employer initially refused and the worker considered pushing back (Q10). The support-recruitment question (Q11) and salary-typicality item (Q12) follow the vignette outcomes.

Employer outcomes. Employers are asked analogous questions about the same vignette. These include whether the request was legitimate (Q1); the highest reasonable salary in the salary arm only (Q2); how much they would like working with the protagonist (Q3); how likely other workers would be to make the same request (Q4); how likely they themselves would be to grant the request (Q5); two incentivized second-order belief questions, one about employers and one about employees (Q6 and Q7); how common they think this type of request is (Q8); how they would respond if the worker pushed back after an initial refusal (Q9); and how typical they think the stated salary is (Q10). Employers are also asked whether they would be interested in meeting a worker with this profile for a one-month internship through the Small Firm Big Talent program (Q11). The question follows the vignette and tells employers that the worker’s request was not granted and that the worker is looking for another job. We interpret this item as a behavioral-style measure of employer interest in hiring a worker similar to the one described in the vignette, after observing the request.

Incentivized measures. Two belief-elicitation items are incentivized on both the employee and employer side. Respondents estimate, out of 100 employers in their market, how many would grant the request (employee Q7; employer Q6), and, out of 100 employees in their market, how many would judge the request legitimate (employee Q8; employer Q7). After all surveys in a market are complete, the respondent whose estimate is closest to the realized market-level value receives a prize, a reusable water bottle. All other vignette outcomes are unincentivized stated responses.

Negotiation Support Intervention

Approximately three weeks after first engagement, workers in both treatment and control arms are asked the primary outcome: whether they spoke to their employer or manager to ask for something at work in the past three weeks, with response options yes and no. Follow-up questions to those who asked are: type of request; what happened, with response options granted, partly granted, considered, refused, or no answer; quality of ask; whether the ask was repeated; relationship/backlash items, including whether the relationship with the employer changed and whether anything negative happened; and reason for not asking, if the ask was not made. Workers in both arms are also asked their confidence about asking on a 1–5 scale and three worker-wellbeing items: job satisfaction, feeling heard at work, and likelihood of staying in the job.

Process and engagement outcomes. The delivery tool logs process and engagement outcomes, including take-up, lessons read, whether a script was shown, and worker rating of the support. These outcomes are used descriptively to measure engagement with the intervention.

The vignette experiment

All respondents are read one hypothetical workplace scenario. The scenario describes a 26-year-old worker who has been employed for two years at a shop of the respondent’s own type — clothing retail, restaurant, or spa, pre-filled from the main survey — and currently earns Rs. 18,500 (approximately $200) per month. The worker speaks privately to the employer and makes one workplace request. Employees answer the follow-up questions from the perspective of the worker; employers answer from the perspective of the employer.

The vignette randomizes three features:
• Request type. Respondents are assigned with equal probability to one of three request categories: a 20 percent salary increase, a non-wage workplace amenity, or a promotion to a supervisory role. Within the amenity category, respondents are further assigned with equal probability to one of three specific amenities that they make a request for: a harassment-reporting mechanism, two fixed 15-minute breaks during the workday, or permission to sit during quiet periods when there are no customers.
• Framing. The request is framed either in rights/fairness-based terms or in practical/instrumental terms.
• Protagonist gender. The worker is described either as Priya, a female worker, or Rahul, a male worker.
The exact wording for each vignette cell is reported in the Appendix, including the Hindi translations. Randomization is at the individual respondent level. Each respondent is independently assigned the protagonist gender, request type, framing, and, when applicable, amenity sub-type. Employees and employers from the same shop are therefore not necessarily assigned to the same vignette. In the main analysis, request type will be coded at the three-category level: salary increase, amenity, and supervisory role. Analyses that separate the three amenity sub-types are treated as exploratory, given the smaller cell sizes.

Link to the intervention. The vignette experiment motivates the content of the negotiation support intervention in Part 2. In particular, it allows us to compare employer stated receptiveness (employer Q5) with employee expectations (employee Q5) by request type. If workers underestimate employer receptiveness to some requests, especially non-wage amenities, this creates scope for light-touch support that helps workers identify feasible requests and prepare how to make them. The belief-correction component of the intervention draws directly on these employer-side responses by informing workers about employers’ reported receptiveness to amenity requests relative to salary requests.

Part 2: Negotiation support intervention

Treatment assignment
Among eligible workers, assignment is a 2×2 randomized design at the individual employee level. Eligible workers are employees who report interest in receiving negotiation support in the survey module (employee Q11) and share a WhatsApp number through which they can receive the support.

Workers are randomized into different treatment arms, with treatment stratified by worker gender × treatment status in the parent project, producing four strata. Within each stratum, eligible workers are assigned with equal probability to:
• Treatment arm vs control arm: full negotiation support (`T_wins`) or control (`C_wins`).
• Belief correction on vs off: Receive belief correction message or not. Workers assigned to receive a belief correction message see a paragraph reporting employers’ stated receptiveness to amenity requests relative to pay raises. This assignment is conducted independently within each treatment arm.

Randomization is conducted in weekly cohorts by the research team on a computer and not on the platform used to deliver the WhatsApp chat support. Each Friday, newly eligible workers surveyed since the previous Friday are randomized into one of the four experimental groups.

Treatment content
Within a few days of randomization, workers assigned to `T_wins` receive a structured WhatsApp flow called “Apni Baat” — translated as “Your Own Say”. The flow first asks workers to choose a request topic from a short menu. Workers assigned to “belief correction on” then receive a one-paragraph message reporting employers’ stated receptiveness to amenity requests relative to pay raises.
The flow then delivers ten short lessons on simple negotiation principles: giving a reason, linking the request to a benefit for the shop, being specific and polite, choosing a calm and private moment, offering a trial, and treating an initial “no” as “not now.” It also presents the employer’s likely perspective, provides an editable example script matched to the worker’s shop type, and offers a line the worker can use if the employer initially says no. The flow is rule-based, uses numbered menus, and relies on fixed, pre-translated English and Hindi content.

Control content
Workers assigned to `C_wins` receive an introductory message identical to that of the treatment group and are asked to choose a request topic from the same menu. They then receive no communication for approximately three weeks. At the end of this period, they are asked the primary outcomes administered to all participants, after which they also receive the full negotiation support flow. The treatment-control comparison therefore compares treated and control participants in the three-week window before the control group receives the support content.

Not available

Computerized. Vignette features randomized within SurveyCTO using once(random()) per interview. Intervention arm assignment conducted externally in Stata using the randtreat package, run in weekly cohorts. Treatment arm is assigned 50/50 within four strata (gender × treated in original RCT). Belief correction (on / off) is assigned 50/50 within eight strata (gender × treated in original RCT × treatment arm here).

Component 1 (vignette): Individual respondent (employer or employee).
Component 2 (intervention): Individual employee.

No

NA

Component 1 (vignette experiment): we anticipate interviewing approximately 2,000 employers and 4,000 employees in the registered sample Component 2 (intervention): sample conditional on take-up. Upper bound: 4,000 employees.

50% of participants in part 2 will be in treatment and 50% in control group. More details in the attached pdf.