Expected Discrimination and Worker Sorting

We study the behavior of job seekers when they anticipate ethnic discrimination in the labor market. To do this, we conduct a field experiment where we vary the level of anticipated ethnic discrimination in the hiring process by randomizing job seekers into three groups: Ethnicity condition (mimics the status quo condition), Tolerant condition (where we signal diversity) and Blind condition (blind review of application materials).

2024-09-27

2024-10-11

The two primary outcomes are:

1. Decision to proceed with the job application: This variable will be a dummy taking value 1 if a job seeker proceeds with the application after receiving the email with the treatment message and 0 otherwise.

2. Qualities of those who proceed (not proceed) with the application. We will look at the following qualities of those who will proceed (not proceed) with the application:

i. Highest level of education.
ii. Years of work experience
iii. Whether one is currently a student
iv. Productivity/effort. This will include:
• Average length of response to an optional interview-style question (word count)
• Percentage score for the number of correctly answered maths questions
• Quality of the response to the optional interview-style question as assessed by HRs (out of 100)
v. Availability: Number of hours one is available to work in a week (Monday to Sunday)
vi. Skillset.
vii. Grit: This is the total score for 12 questions developed by Duckworth et al. (2007).

1. Amount effort put into the application. We will have four measures for effort put into application:

i. Whether an applicant formats their CV to our preferred format. This outcome will be a dummy taking value 1 if the applicant formats their CV to our preferred format and 0 otherwise.
ii. The average length of the responses to the three optional interview-style questions.
iii. The average quality of the responses to the three interview-style questions as blindly assessed by HR professionals. The HR professionals will record the score on a scale from 0 to 100.
iv. Performance in an optional data entry task. This will be a continuous variable measured as the number of rows correctly entered.

2. Job choice: Applicants will be asked to choose whether they want to be considered for data entry or data validation position. This variable will be a dummy variable taking value 1 if an applicant chooses data validation and 0 otherwise.

3. Ranking by the HR professionals: We will remove all identifiable information and use codes before forwarding the application materials to the HR professionals, who will then review and record the overall score of each applicant on a scale from 0 to 100.

4. Whether the job seekers are willing to engage with the “employer” in the future. This will be a dummy variable taking value 1 if willing to be contacted in case of future job opportunities and 0 otherwise.

The experimental have five stages:

Stage 1: Job Advertisement
The first stage of our experiment involve sending out a general job advertisement. In the advertisement, we state the salary associated with the position, the starting date, and that the job will be a short-term remote job.

Stage 2: Expression of Interest
Those interested in the job express their interest by registering using the link provided. At this stage, we collected baseline characteristics of job seekers.

Stage 3: Randomization and Treatment
In this stage, we will randomly assign the job seekers to either ethnicity condition, tolerant condition, or blind condition. We will then send them a follow-up email containing the treatment messages varying the level of anticipated ethnic discrimination.

Stage 4: Application and Job Choice
The applicants will have a chance to proceed with the application and choose the position they want to be considered for.

Stage 5: Hiring
We will offer a total of 10 part-time jobs for one month. These will be equally distributed among the top applicants in each category.

Not available

Randomization is done at the individual level using the “randtreat” command in STATA. We stratify randomization by ethnicity, gender, and work experience.

Unit of randomization is individual

No

no cluster sampling

6,816 job seekers

Ethnicity condition: 2,271 job seekers,
Tolerant condition: 2,268 job seekers,
Blind condition: 2,277 job seekers.