Experimental Design
We conducted identical resume audits in 2016 and 2017. The audits began in March and ended in July each year. Using a popular internet job search board, we submitted 36,880 randomly generated resumes to online ads chosen at random from a bank of jobs constructed by our research team. In lieu of submitting resumes to many different types of jobs, the bank of job ads includes those from the following job categories: account executive, banking, customer service, finance, insurance, and marketing.3 Ads requiring certifications or expertise in a foreign language as well as those requiring company-specific applications were excluded from the job bank. Although it is common in the audit literature for researchers to audit select cities/labor markets (Bertrand and Mullainathan, 2004; Lahey, 2008; Nunley et al., 2015; Kroft et al., 2013), we focus on the US labor market but impose no additional location restrictions regarding which ads enter the job bank. After creating the bank of jobs, our research assistants–using randomly generated resumes–applied to job openings that were randomly selected from the job bank. Four resumes were submitted to each ad selected to be audited, and applications were submitted to 9, 220 unique advertisements. The result is 36, 880 observations (= 9, 220 jobs × 4 applications).
We use the program developed by Lahey and Beasley (2009) to randomly assign a name, address, university, major, and work experience obtained during college to the resumes. In line with the audit literature (e.g., see Bertrand and Mullainathan, 2004), we use names that are distinct along racial/ethnic and gender lines to signal race/ethnicity and gender to prospective employers with the goal of detecting the existence of discrimination as well as the extent of discrimination. The names used in our study are presented in Table 1. Our design does not restrict the combinations of racial/ethnic-sex specific names that could be submitted to a given job ad. For example, consider applications submitted by fictive applicants with distinctively Black names. Of the 9,220 job openings to which we applied, 15 percent did not receive a submission from a Black male or a Black female; 23 percent received an application from one Black female and zero from Black males, and vice versa (also around 23 percent); 22 percent received applications from one Black male and one Black female; Around 5 percent received two applications from Black females and zero applications from Black males, and vice versa (also around 5 percent); 3 percent received three distinctively Black applications (two males and one female, or two females and one
male); and less than 1 percent received four distinctively Black applications (two males and two females). The analogous percentages for the distinctively White and Hispanic names are similar to those for the Black names. The addresses assigned to fictive applicants are tied to the university to which they are assigned. The addresses were chosen such that the fictive applicant lived within a few miles of the university from which they will receive or have received (depending on the month of application) a Bachelor’s degree. Per our IRB agreement, we are unable to provide the names of the universities, but we note that each is a public, flagship university, and the chosen universities span the continental US. Using Census region groupings, three of the universities are located in the Southeast, two in the Southwest, two in the West, three in the Midwest, and two in the Northeast. Eight majors are incorporated into the design and assigned to fictive applicants at equal
probability (12.5 percent): economics, finance, marketing, anthropology, philosophy, chemistry, biology, and psychology. The work experience accumulated during college that is assigned to the applicants includes three different types and was assigned at equal prob-
ability (i.e. = 1/3): retail/sales, restaurant/coffee shop, and university employment (e.g., library, campus recreation, dining services). We are unable to provide the job titles or firm names associated with the applicant’s college work experience due to our IRB agreement. Portions of the fictive applicants are assigned minors, internship experience, grade point averages (GPA), volunteer experience, fluency/proficiency in speaking Spanish, receipt of study-abroad scholarships, and different sets of computer skills (e.g., programming and
data analysis). In terms of minors, two are included in the design: history and mathematics. These minors are available to students from each of the universities used in our experiment. The internships assigned to the fictive applicants’ resumes are grouped into
two categories: analytical and social. For example, we include a number of internships with the titles ”Marketing Analyst Intern”, ”Financial Analyst Intern”, and ”Research Intern” in the ”analytical” category, and ”Marketing Sales”, ”Financial Sales”, and ”General Sales” into the ”social” grouping. Twenty-five percent of applicants report no information regarding their GPA. The remaining applicants report GPAs of 3.0, 3.2, 3.4, 3.6, 3.8, and 4.0. The probability of being assigned no GPA or one of the other six possibilities is 12.5 percent. The volunteer experiences randomly assigned to fictive applicants vary across three different types. We are unable to reveal the types of volunteer experiences due to the prospect of violating our IRB agreement. Even revealing the type of charitable work performed by the organizations would likely reveal their identities. Twenty-five percent of the fictive applicants were assigned a study abroad experience, but these experiences vary across seven countries: Argentina, China, Dubai, Italy, Japan, Mexico, and South Africa. Conditional on being assigned a study abroad experience, the country in which the experience takes place is randomly assigned with equal probability. In terms of computer
skills, applicants report the following on their resumes: no computer-related skills or information (25 percent), basic skills (25 percent), data analysis (25 percent), programming (12.5 percent), and the combination of data analysis and programming (12.5 percent)