Discrimination in labor markets of Ecuador, Colombia, Perú, Bolivia, and Paraguay: field experiments

We show an intervention message at the beginning of the selection process. There will be two intervention messages that will appear randomly to participants. Therefore, the sample will be divided into three groups: Treatment group 1 (intervention message 1), treatment group 2 (intervention message 2), and control group (did not see any message). The content of the message is:
• Nudge 1: “Remember that people who belong to disadvantaged groups must overcome more social and cultural obstacles to achieve the same level of education and work experience as others.”
• Nudge 2: “Our organization depends on you to identify the candidates who will contribute the most to our optimal performance.”

2023-10-15

2024-02-28

Probability of selecting a minority candidate; Differences in wages offered between a minority and a non-minority candidate; Differences in fit for the job assessments between minority and non-minority job applicants.

Research Design and Collaboration with a Consulting Firm:
In our upcoming research spanning five countries (Colombia, Ecuador, Peru, Bolivia, and Paraguay), we intend to partner with a consulting firm that has a physical presence in each of these nations. This firm will be instrumental in implementing our field experiment, given its local expertise and understanding of the diverse labor markets.

Recruitment of Human Resource Recruiters:
We will hire experienced human resource recruiters from the labor markets of each country. To ensure genuine engagement and commitment, the recruitment process will be presented as a real job-hiring exercise. Our strategy to source these recruiters will involve respondent-driven sampling (RDS) referrals and LinkedIn and other media platforms job postings. To acknowledge their expertise and time, recruiters will be compensated competitively, which will also enhance the credibility and impact of our study.

Gathering Preliminary Insights:
To inform our experimental design, we plan to conduct interviews and focus groups with key stakeholders in each country. This will include recruitment agencies, independent agents, LGBTQ+ migrants and women workers, and labor market experts in academia and governments. The insights and feedback from these interactions will be pivotal in tailoring our experiment to closely resemble the authentic job market settings in the respective countries.

Creation of Synthetic Job Applicant Profiles:
We aim to create a controlled experimental environment that mirrors the real-world job application process. To achieve this, we will design synthetic resumes that are representative of actual job applications in the formal labor markets of the five countries. These profiles will be comparable in qualifications and attributes, with the only distinction being the self-identification as a minority for some candidates. This approach will allow us to isolate and measure the potential impact of minority identity on recruiters' evaluations.

Experimental Trials:
Our experiment will consist of up to ten trials per human resources recruiter hired in each country. In some trials, an applicant will be randomly assigned to disclose their minority membership, while others will not. We will also include trials comparing different genders, migrants versus locals, and placebo trials with no distinguishable prejudiced attributes. This design will enable us to explore discriminatory behaviors in various contexts, while also masking the primary research objective.

Development of an Online Platform:
To efficiently conduct our experiment, we will develop a custom web platform tailored to our study's requirements. This platform will serve as the interface for presenting candidates to the recruiters and collecting their evaluations. It will be designed to ensure a seamless experience for the recruiters, while also maintaining the integrity and confidentiality of the experiment.

Recruiter Onboarding and Data Collection:
All participating recruiters will be required to register on our online platform. We will collect information on their socioeconomic background, education, work experience, and knowledge of the local labor market. To gain deeper insights into their evaluation process, we will track metrics like the time spent reviewing applications and their activity on the platform. After registration, recruiters will be presented with job vacancies and corresponding candidate profiles to evaluate.

We will use Ordinary Least Squares (OLS) regressions to compare job candidates with and without physical disabilities, as well as those from Venezuela vs. Ecuador. We looked at three dependent variables:
"Selection" - this tells us which candidate the recruiter chose as their top pick (1 for selected, 0 for not selected).
"Fit for the job" - a rating on a scale of 1 to 10, showing how suitable the job applicant seemed to the recruiter.
The salary offered to each candidate (if they were offered the job), measured on a logarithmic scale.
Our main focus was on whether the candidate had a physical disability or was Venezuelan, which we treated as our main independent variable.
We used three different models to analyze the data:
M1: The first model looked at the relationship between the candidate's background and the outcomes without any conditions.
M2: The second model considered how recruiters behaved when they saw information about the candidate's minority status, as well as the candidates’ demographic characteristics. Fixed effects were added at city level and variations in standard errors were clustered at the recruiter level.
M3: The third model factored in the candidates’ demographic characteristics. Fixed effects were added at the recruiter level, including the recruiter’s sociodemographic characteristics. Variations in standard errors were taken into account at city level.
We calculated the average differences in outcomes using these three models. The estimated models for the analysis follow the general equation number 1:
Y_itr=β_0+ β_1 X_it+ ε_itr (1)
where Y is the dependent variable, which corresponds to synthetic candidate i for position t, evaluated by recruiter r; X is the independent variable of interest; and ε is the error term. The dependent variables, the main independent variable and the control variables are included according to the specification indicated in the estimates.

A random number will be assigned to job applicant candidates within each evaluation set (or trial of the experiment); the online platform will use that random number to assign a minority condition to one of the individuals in each set.

The unit of randomization are the simulated job applicants

No

350 per country

1750 recruiters across all countries

150-200 treated with nudge 1; 150-200 treated with nudge 2; 150-200 untreated