Vacancy design and applicant selection

We conduct a field experiment in Addis Ababa, Ethiopia, to study the impact of different aspects of vacancy design on the pool of applicants. For this purpose, we collect real vacancies from firms and exogenously vary the content of these job adverts.

We vary the following information on the job adverts:
1. Whether the advert includes information about the offered monthly salary or wage. When collecting the vacancies from firms, we ask about the monthly wage or salary that they expect to pay for this position. We then randomly vary whether we include this information in the vacancy advert.
2. Whether the advert includes a gender inclusivity statement. When collecting the vacancies from firms, we ask which gender they expect the majority of applicants to be off. Based on this information, we then randomly include a gender-specific inclusiveness statement that reads: "We encourage candidates of all genders to apply for this position, including [GENDER].", where [GENDER] is the expected minority gender. If the respondent does not know, which gender will be prevalent, we instead include the following statement: "We encourage candidates of all genders to apply for this position."
3. Whether the advert includes information about the "general skill" most demanded by this vacancy. Based on this information we randomly include a "Desired skills" category in the job advert that lists the top-ranked skill.

For each vacancy, we will create three different versions of each job advert with each of the above-mentioned components cross-randomized (i.e. independently randomized). We will post each vacancy on three different channels (i) offline job boards across the city, ii) Facebook adverts, and iii) the major online job board (www.ezega.com). We then randomize which version of the vacancy is posted on which channel.

The randomization is thus at the vacancy level and all treatment components are cross-randomized. Assuming a total of 400 different vacancies, we end up with the following treatment and control groups in this experiment:

Wage information treatment:
- 200 treated vacancies with wage information
- 200 control vacancies without wage information

Gender inclusivity information treatment:
- 200 treated vacancies with gender inclusivity information
- 200 control vacancies without gender inclusivity information

Skill requirements information treatment:
- 200 treated vacancies with skill requirement information
- 200 control vacancies without skill requirement information

Since the three different information treatments are independently randomized, we will obtain the following distribution of pieces of information over the 400 vacancies:
- 12.5% of vacancies will contain all three information treatments
- 37.5% of vacancies will contain two information treatments (either wage-gender, gender-skill, or wage-skill)
- 37.5% of vacancies will contain one information treatment (either wage only, gender only, or skill only)
- 12.5% of vacancies will contain no information treatments (pure control)

Our main outcomes of interest are the number and composition of applicants to these differentially treated vacancies.

We plan to write separate papers for each of the treatments.

2022-05-01

2022-09-30

(IHS) number of applicants, applicant quality index

Gender composition, beliefs about the vacancy (including beliefs about offered wages), and preferences

We conduct a field experiment in Addis Ababa, Ethiopia, to study the impact of different aspects of vacancy design on the pool of applicants. For this purpose, we collect real vacancies from firms and randomly vary the content of these job adverts. We also include some vacancies created as part of our project in the study.

All job adverts contain phone numbers that lead our experimental call center. Job-seekers who are interested in applying call the experimental call center and will complete a short pre-screening survey. They will then be invited to a screening session in a central location in Addis Ababa at a time of their choosing. To proceed with their application, they will have to come to the screening center and complete a screening survey. Once all applications are received and applicants screened, we pass the list of candidates on to the firms that created the vacancies.

We conduct a field experiment in Addis Ababa, Ethiopia, to study the impact of different aspects of vacancy design on the pool of applicants. For this purpose, we collect real vacancies from firms and exogenously vary the content of these job adverts.

We vary the following information on the job adverts:
1. Whether the advert includes information about the offered monthly salary or wage. When collecting the vacancies from firms, we ask about the monthly wage or salary that they expect to pay for this position. We then randomly vary whether we include this information in the vacancy advert.
2. Whether the advert includes a gender inclusivity statement. When collecting the vacancies from firms, we ask which gender they expect the majority of applicants to be off. Based on this information, we then randomly include a gender-specific inclusiveness statement that reads: "We encourage candidates of all genders to apply for this position, including [GENDER].", where [GENDER] is the expected minority gender. If the respondent does not know, which gender will be prevalent, we instead include the following statement: "We encourage candidates of all genders to apply for this position."
3. Whether the advert includes information about the "general skill" most demanded by this vacancy. Based on this information we randomly include a "Desired skills" category in the job advert that lists the top-ranked skill.

For each vacancy, we will create three different versions of each job advert with each of the above-mentioned components cross-randomized (i.e. independently randomized). We will post each vacancy on three different channels (i) offline job boards across the city, ii) Facebook adverts, and iii) the major online job board (www.ezega.com). We then randomize which version of the vacancy is posted on which channel.

The randomization is thus at the vacancy level and all treatment components are cross-randomized. Assuming a total of 400 different vacancies, we end up with the following treatment and control groups in this experiment:

Wage information treatment:
- 200 treated vacancies with wage information
- 200 control vacancies without wage information

Gender inclusivity information treatment:
- 200 treated vacancies with gender inclusivity information
- 200 control vacancies without gender inclusivity information

Skill requirements information treatment:
- 200 treated vacancies with skill requirement information
- 200 control vacancies without skill requirement information

Since the three different information treatments are independently randomized, we will obtain the following distribution of pieces of information over the 400 vacancies:
- 12.5% of vacancies will contain all three information treatments
- 37.5% of vacancies will contain two information treatments (either wage-gender, gender-skill, or wage-skill)
- 37.5% of vacancies will contain one information treatment (either wage only, gender only, or skill only)
- 12.5% of vacancies will contain no information treatments (pure control)

Our main outcomes of interest are the number and composition of applicants to these differentially treated vacancies.

Randomization done in office by a computer

The randomization for each of the three information treatments is done at the vacancy level and all treatment components are cross-randomized independently, as elaborated under "Intervention" and "Experimental design".

In addition to that, we will create three different versions of each vacancy with each of the above-mentioned information treatments cross-randomized. We will post each vacancy on three different channels (i) offline job boards across the city, ii) Facebook adverts, and iii) the major online job board (www.ezega.com). We then randomize which version of the vacancy is posted on which channel.

No

No clustering: approximately 400 vacancies across 3 posting channels, leading to approximately 4000 (expected) applicants

approximately 400 vacancies across 3 posting channels and 4000 (expected) applicants

The information treatments (salary, gender, skill) are independently randomized over each vacancy, with a 50% chance for each treatment to be included on a given vacancy (see "Experimental design"). This means that pure control vacancies (no information treatment) represent 0.5^3=0.125 of all vacancies.

All treatments are distributed as follows over the sample of vacancies:
- 12.5% of vacancies will contain all three information treatments
- 37.5% of vacancies will contain two information treatments (either wage-gender, gender-skill, or wage-skill)
- 37.5% of vacancies will contain one information treatment (either wage only, gender only, or skill only)
- 12.5% of vacancies will contain no information treatments (pure control)