Applications of Experiments in Economics

Racial Bias in Callbacks

Paper Summary

  • Big question: is there racial discrimination in the labor market?
    • Correlations uninformative: legacy of institutional racism leads to differences in education, opportunities, and human capital
  • This paper: measures a specific kind of discrimination. Callbacks based on name (which can signify race)
  • Methodology: Randomize names on job application to infer effect of changing “black sounding” name to “white sounding”
    • Often referred to as a field experiment

Design of the Experiment

  • Start by collecting resumes from two job search sites
    • restrict to cities and time of experiment
    • restrict to four occupations (sales, clerical, customer services, admin)
    • classify subjectively into two categories (high vs low quality)
  • Generate a pool of names
    • Use all birth certificates from MA: 1974-1979
    • Tabulate frequency by race to get most uniquely white/black names
    • Validate with survey
  • Survey job ads and manually pick two high and two low quality resumes that best fit job description
    • Randomly assign one white name and one black name to each resume of same quality
    • Measure callbacks for each resume
  • Respond to 1300 job ads with nearly 5000 resumes. What next? What would it tell us? Why?

Results

Thoughts?

  • How to interpret this table?
  • What does it tell us?
  • What doesn’t it tell us?

Results

Thoughts?

  • How to interpret this table?
  • What does it tell us?
  • What doesn’t it tell us?
  • What role does randomization play? Why do we need it?

The Effect of Ban the Box on Racial Discrimination

Application 2: Ban the Box

  • Ban the Box: law that makes it illegal for employers to ask job applicants if they have ever been convicted of a crime.
  • Intention: reduce labor market discrimination against individuals with criminal priors
  • Due to mass incarceration and other discriminatory policies, black men have higher rates of prior conviction than white men
  • Possible consequence: increase discrimination against black men without prior conviction (form of statistical discrimination)
  • Link to paper

Paper Summary

  • Main question: is there labor market discrimination against males with prior conviction? Does BTB increase racial discrimination?
  • Another “resume study”
  • Randomized resumes in NY and NJ before and after introduction of BTB in both states

Methodology

  • Population:
    • private and for-profit employers on indeed.com and snagajob.com
    • jobs requiring no post-secondary education, special skills, or work experience (e.g. retail and restaurant)
    • Pre and post BTB introduction in 2015
  • On otherwise identical resumes, randomized along four dimensions:
    1. race (by name);
    2. a non-violent and minor felony conviction;
    3. employment gap in resume; and
    4. GED or High school diploma
  • 15,220 applications to 4,291 stores among 293 chains
  • Now what?

Interpret

Interpret

Regression Framework

Store \(j\) for applicant \(i\): \[ Callback_{ij} = \alpha + \beta_{1} Box_{j} + \beta_{2}White_{i} + \beta_{3}Box_{j}\times White_{i} + \Gamma \mathbf{X}_{i} + \epsilon_{ij} \]

  • Which coefficient captures the effect of box info on racial gap?
  • What values would suggest that employers use race to proxy for criminal conviction?
  • Cross-sectional and temporal sources of variation in \(Box_{j}\)
  • Limit either to pre BTB (cross-sectional) or stores that initially have box (temporal)

Results

Thoughts? How do you think standard errors should be calculated?