Perceived Value, Perceived Harm: Economic Incentives and Bystander Intervention in Workplace Harassment

Our experimental design uses animated, video-based vignettes, designed to represent a typical customer interaction in an apparel store. Each vignette depicts the same pattern of customer harassment directed at a female employee. The harassment script and behavioral cues are held constant across treatment arms. We experimentally vary only the economic value of the customer to the shop, implemented through a cross-randomization of two dimensions of purchase behavior.

First, we vary the customer’s total spending, which serves as a signal of the customer’s economic importance. Respondents are randomly assigned to one of three conditions:
● High-value customer: high total spending
● Medium-value customer: medium total spending
● Low-value customer: low total spending

Second, we vary whether total spending is conveyed by quantity or quality. In one condition, spending is signaled through the price per unit, with the customer purchasing a single item at different prices. In the other condition, spending is signaled through quantity purchased, with the customer buying multiple units (three, two, or one) at a constant low per-unit price. Ultimately, respondents are assigned to one of five conditions with low total spending (buying 1 unit at low price) is the control group.

Respondents view only one vignette, ruling out within-subject contamination or learning effects. The design varies the economic cost of intervention both in magnitude and in the way customer value is signaled, by cross-randomizing total spending and whether spending is conveyed through price per unit or quantity purchased. Holding the harassing behavior constant, this allows us to study whether and how a customer’s perceived financial value and associated status cues shape bystanders’ interpretations of the situation and their intervention choices.

2026-03-11

2026-05-31

The primary outcomes are based on the single action respondents report they would take first after viewing the video vignette. Respondents are first asked to select all actions they might consider taking in the situation. They are then asked to select, from that subset, the action they would be most likely to take first. The first action selected is used to construct the following primary outcome:

Financial-Loss–Inducing First Action: A binary indicator equal to 1 if the first action selected risks immediate financial loss for the shop, and zero otherwise. Actions classified as financial loss-inducing include those that could directly jeopardize the transaction or customer relationship. All other actions are coded 0.

Secondary Outcomes (End Points)

1. Perceived Severity
Perceived severity is measured using a 0–10 scale in response to the question: “Do you think the customer’s behavior was problematic?”, where 0 indicates “not at all problematic” and 10 indicates “extremely problematic.”

The timing of the severity question is randomized at the individual level. For some respondents, severity is elicited before the intervention decision; for others, it is elicited afterward. This allows us to examine whether reported severity differs depending on whether it is measured before or after respondents state their intended action.

2. Anticipated Consequences of Intervention:
Respondents indicate, on a 1–5 Likert scale (1 = Very unlikely, 5 = Very likely), what they believe would occur if someone directly confronted the customer. We construct the following measures:
(i) Financial Loss Index: Average of items capturing anticipated cancellation of purchase, loss of future business, and reputational harm through a public scene.
(ii) Professional / Job-Related Risk Index (employees only): Average of items capturing anticipated managerial complaint, reprimand, or job loss.
(iii) Personal Safety Risk Measure: Likelihood that the customer becomes aggressive or violent.
(iv) Expected Intervention Compliance Measure: Likelihood that the customer stops the behavior upon confrontation and the sale continues.
Indices will be standardized using the mean and standard deviation of the low-spending (baseline) treatment group.

3. Second-Order Beliefs
Respondents are asked what they believe relevant actors would want them to do first in the situation depicted in the video. For each reference group, respondents select one action from the same list used for their own decision. Reference groups include the shop owner/manager (for employees) or employees (for employers), and the female sales assistant in the vignette.

The study is a field-based, randomized vignette experiment conducted among retail shop owners, managers, and employees in Delhi, India. The target sample consists of approximately 1,000 shops. Within each shop, up to three respondents (owner/manager and up to two employees) are surveyed. Randomization occurs at the individual level within shops. Each respondent is independently assigned to one of five treatment conditions. The five conditions arise from variation in total customer spending (low, medium, high) and whether spending is conveyed through price per unit or quantity purchased. The low-spending condition serves as the baseline group. Each respondent views only one version of the vignette.

Not available

Treatment assignment is conducted using computer-generated randomization embedded in the SurveyCTO survey software. Each respondent is independently and automatically assigned to one of the five treatment conditions at the time of survey administration. It does not involve enumerator discretion.

The unit of randomization is the individual respondent. Each shop owner, manager, or employee is independently assigned to a treatment condition. Multiple respondents may be surveyed within the same shop; therefore, while randomization occurs at the individual level, statistical inference will account for clustering at the shop level.

No

1,000 retail shops

Approximately 3,000 individual respondents (up to 3 respondents per shop across 1,000 shops)

Approximately 600 individual respondents per treatment arm (5 arms total), assuming 3,000 total respondents and equal allocation across arms.