Our experiment is conducted in the entirety of an online job platform in Pakistan. Below are the overall procedures.
1. A half of all employers are randomly assigned as treatment employers, and the other half are control employers. As explained in 3 below, the female-friendliness information is publicized only for the jobs of treatment employers.
2. When treatment and control employers advertise jobs, both of them are required to provide the information about female-friendly job characteristics.
3. Job ads are posted in the online job portal.
[Intervention 1] The job ads of treatment employers include the female-friendliness information, which is public to everyone. The job ads of control employers do not include it.
4. Job alerts: Job alerts are emailed to jobseekers. Based on an algorithm of the online job portal, jobseekers to whom to send a job alert are identified for each job.
[Intervention 2] Intervention 2 randomizes the types of information included in the job alerts. There are seven types of job alerts, as explained in the section of Interventions. For each of the jobs of treatment employers, the seven types are randomized across the identified jobseekers to whom to send a job alert for the job. For each of the jobs of control employers, two types (types 1 and 2) are randomized across the identified jobseekers to whom to send a job alert for the job.
5. Jobseekers apply for jobs, and employers screen applications. Note that jobseekers can apply for any jobs regardless of whether they receive job alerts for the jobs or not.
Note. Whether the gender-friendliness information is included in a job ad or not depends on an employer's treatment status, regardless of how many jobs the employer advertises.
We will use the following data.
(1) Administrative data of the online job portal: It consists of employer data, job data, jobseeker data, application data, and job alert click data. While the application data includes application outcomes (such as whether an applicant is shortlisted, offered a job, and hired), the information about application outcomes is incomplete in the sense that it is voluntarily provided by employers and therefore that it is available only for a fraction of jobs. Our analysis on the effect on the probability of a jobseeker being shortlisted will focus on the jobs for which application outcomes are available.
(2) Baseline online surveys of employers and jobseekers: All active employers and jobseekers in the job portal are invited to take the online surveys.
(3) Endline telephonic surveys of employers (twice): Employers who post job ads during the experiment will be interviewed over the phone. The first endline survey will be conducted between December 2021 and February 2022. The second endline will be conducted in fall 2022.
(4) Endline online surveys of jobseekers (twice): All active jobseekers will be invited to the endline online surveys. The first endline survey will be conducted in February-March 2022; the second endline will be in fall 2022.
To estimate the effects of publicizing the gender-friendly job characteristics, we will regress employer/job-level outcomes on the employer treatment status dummy, the female-friendly job characteristics, and the interactions between the dummy and the female-friendly job characteristics. Since the effects are expected to be different depending on gender-friendly job characteristics, we will capture the differential effects by including the interaction terms.
The effects of informing jobseekers about gender-friendly characteristics through job alerts (i.e., intervention 2) will be estimated.
For outcome variables W1-W3, the unit of analysis is a job-jobseeker pair. The sample will be the job-jobseeker pairs for which job alerts are created and sent. We will regress the outcome variables on the female-friendly job characteristics, the dummies whether female-friendly job characteristics are included in the job alert, and the interactions between them (e.g., supervisor gender, whether supervisor gender is included in the job alert, the interaction between them).