Experimental Design
Recruitment: We conduct door-to-door sampling in urban neighborhoods in Patna, India. Survey teams are allocated to streets within a neighborhood. We target streets that are close to the call centers we are establishing and contain a higher proportion of low or middle-income households. We target all households on a given street. At each household we approach, we document whether or not we spoke to anyone in the household, who we spoke to, and any reason they gave for not participating in the survey.
Sample: After approaching a house, we explain we are conducting a survey and ask to speak with a married woman, who is aged between 18-60, and has completed 5th grade. If there are multiple such women, the surveyor will prioritize any whose husband is currently present, and then the one who is closest to 35. We recruit a maximum of one woman in any given household. If an eligible woman’s husband is available and willing to participate, we collect informed consent and begin the survey. We complete the survey irrespective of whether husbands are present. However, we take several measures to increase the likelihood that husbands are present for the survey. For this reason, we begin surveying early in the morning, and survey on weekends, when husbands are more likely to be at home. During these times, if a woman’s husband is not available or unwilling to participate, we may reschedule the survey to another time so we can focus on other households where husbands may be present.
Demographics: After consent, we collect basic demographic details from the wife including work history and education. In this section, we ask the wife whether she is interested in new part-time work opportunities using a 4-point Likert scale. We stratify the randomization of mixed-gender and women-only workplaces based on their level of interest, and exclude participants who say they are not at all interested in work opportunities. If the husband is present, we obtain consent from the husband if we have not already done so. We collect basic demographic information from the husband including age, work history and education. If the husband is not present, we collect this information from the wife.
Description of workplaces: We then inform participants of the two nearby call centers we are opening (the mixed-gender center and women-only center) and that we are looking to hire women for the call centers on a part-time basis. We then explain that the tasks, management, amenities, and probability of getting a job are the same across both centers. We also highlight that this is a one-time, two week offer. We then introduce women (and their husbands, if they are present) to both workplaces by handing them two flyers (one for each workplace) which we then explain to them.
Safety treatments: We then tell them about the safety measures implemented at the workplace. We have three experimental arms. The first, (the “full safety” group) receives the full set of safety information, explaining that the company takes women’s safety seriously and does not tolerate sexual harassment of any kind. We outline five specific safety measures implemented at the workplace – reference checks, female security guards, a strict sexual harassment policy, sexual harassment training, and security cameras at the workplace. The second group, (the “signal” group) gets told that the organization does not tolerate sexual harassment of any kind, however, we do not mention any of the specific measures implemented. Thus, this group receives any signal (positive or negative) that comes from discussing safety protocols in the job description. The treatment also closely matches generic language that many organizations use in their job descriptions. The final, group, (the “control” group) does not get told about any safety measures or precautions.
We then explain that there are limited job openings in both call centers, and that we are randomizing which call center women are eligible to apply to. We hand participants an envelope that includes the workplace that we randomly assigned to the participant. We ask them not to open the envelope but explain that it contains the workplace they have been randomly assigned to. We stress that they will only be able to take up a job in that workplace, and there is no way to change the offer in the envelope. We will then ask them, for each workplace, whether they would prefer to apply for the job or receive a small gift (which includes the option of a monetary payment). We explain that after they have made their choices, we will implement their choice based on whatever flyer is in the envelope. We tell participants that if they say they are interested in the job that is in the envelope, we will contact them and their husband to discuss and schedule the job and training day. We then tell them the salary at the workplace. Finally, we elicit their preferences in both workplaces, randomizing the order in which we collect their preferences.
Next, we ask participants three comprehension questions. The first, confirms that they understand their choices will not affect their office allocation. The second, confirms that that they understand how their choices will be implemented. Finally, we confirm that they understand the mixed workplace has men working in it and the women only does not. If they answer any of these questions incorrectly, we correct them, and ask them again. If they answer incorrectly twice, we explain things to them a third time and proceed to the next question. We then give participants the opportunity to update their decisions if they would like to.
Randomization: We stratify treatment into the mixed-gender or women-only workplace and cross-randomize the three safety treatments. We also randomize which version of the recruitment flyer participants see. We use over 20 unique photos that are split across 4 women-only and 4 mixed-gender versions of the flyers to ensure that treatment effects are not driven by any one photo or flyer. We also randomize the wage offer participants receive to measure willingness to accept jobs and to benchmark treatment effects.
Intensive margin preferences: We then ask them which call center they would prefer to work in and why.
Beliefs: We then elicit participant’s beliefs about:
(i) How safe they think both workplaces are for women (as a first-stage for the safety measures)
(ii) The proportion of men working in the mixed workplace
(iii) The quality of the office facilities (to capture spillovers of the safety treatment)
(iv) Whether they think the office has had a problem with sexual harassment in the past (to test whether the safety treatments signal a history of sexual harassment incidents)
We also ask participants whether they would (hypothetically) be interested in applying for the mixed job if the proportion of men in the office were 3, 5, or 7 out of 10 employees.
Norms and spousal jealousy: We measure spousal jealousy and gender norms regarding impropriety using four survey questions. We ask these questions to the wife, in private. The questions ask (i) directly about spousal jealousy, (ii) personal beliefs about how appropriate it is for married women to interact with other men (iii) second-order beliefs, and (iv) a measure of controlling behavior on the part of the husband. We will create an index using these questions. As secondary analysis, we will analyze heterogeneous treatment effects using the index.
Job application form and contact information: If women are interested in applying for the job in the workplace they were assigned to, we ask them to complete a job application form outlining their personal details, motivations for applying as well as professional or personal references.
Training day: Interested participants will be invited for training sessions for the job. Before the training day, we contact participants and their husbands to remind them of the session. If the husband was not present, then we explain the two workplaces to them and the call center their wife has been randomly allocated to. We compensate participants for attending the training day and remind them to attend by visiting their house on the day of their scheduled training session.
Attendance: A subset of the trained candidates will be offered jobs at the workplaces. The two-week long contract will provide the participants with data collection and data entry tasks. We will record participants’ attendance. We are not powered to study this as an outcome variable.
Spillovers: In a random subsample of streets, we will minimize spillovers by randomly skipping some houses on the street. This will allow us to estimate treatment effects in the absence of any spillovers that arise from knowing other women are also taking part in the survey and attending the workplace. We will estimate a second type of spillover effect, from having a neighbor allocated to the same treatment group, by comparing treatment effects among people who have more or fewer neighbors randomly assigned to the same workplace (controlling for the number of neighbors in the study). In a follow-up exercise, we survey participants to measure the impact of spillovers on perceptions of both workplaces on dimensions such as safety, social acceptability, ease of making new friends, and the proportion of men working in the mixed office. We also implement an information treatment in which we re-elicit beliefs after asking participants to imagine they knew more women working in either the mixed or women-only workplace. We also test the whether the social proximity of other women attending the workplace impacts beliefs, by varying whether the additional women attending the workplace live in the same street or same neighborhood.