Investigate the institution
Compare randomized name signals with interview callbacks, then distinguish an absolute gap, a relative gap and an unmeasured hiring outcome. What does a callback difference reveal about an institutional gate?
Institutional gatekeeping can create unequal opportunities before an interview occurs. An audit isolates a signal at one gate rather than measuring every source of labor-market inequality.
Random assignment reduces systematic differences in the assigned signal groups. It does not make a historically and geographically bounded set of employers representative of every labor market.
A callback is neither a job offer nor a wage. Similar statistical disparities can arise through different decision processes; distinguishing those mechanisms needs additional evidence.
Distinguish authentic evidence from the practice ledger
Randomized resume experiment
Researchers submitted fictional resumes to newspaper job advertisements in Boston and Chicago, randomly assigning names intended to signal perceived race. The report describes approximately 50% more interview callbacks for White-associated names.
This primary research summary concerns differential treatment at a specific hiring stage in the study setting. Names are signals, not a biological measure or complete account of identity. The short summary does not supply individual records or a full uncertainty analysis.
Marianne Bertrand and Sendhil Mullainathan — NBER working paper 9873, July 2003; published AER 2004. This is an attributed summary, not reproduced article text.
Published article and replication reference
The published article identifies its response as callbacks for interviews and links replication materials. Its design manipulates a perceived signal while varying resume characteristics; it does not follow real applicants through wages or job tenure.
The working-paper and journal versions report the same study, not two independent replications. Read methods and coding documentation before using the replication package.
Bertrand and Mullainathan, American Economic Review 94(4), 991–1013 — September 2004. This is an attributed summary, not reproduced article text.
Fictional calculation packet
For arithmetic practice only, condition A receives 12 callbacks from 100 resumes and B receives 8 from 100. A second fictional employer set gives 6 callbacks from 50 in A and 8 from 100 in B.
These invented counts are not the study counts. Hold the outcome definition fixed, label both denominators and calculate differences before interpreting them.
Inspect the fictional ledger
Fictional practice data, not the study results. Scroll horizontally on narrow screens.| Fictional condition | Callbacks | Resumes | Rate to calculate |
|---|
| First set A | 12 | 100 | Divide callbacks by resumes |
|---|
| First set B | 8 | 100 | Use the same unit |
|---|
| Second set A | 6 | 50 | Do not compare counts alone |
|---|
| Second set B | 8 | 100 | Identify the reference rate |
|---|
Label each arrow in assigned name signal → screening decision → callback → interview → offer as manipulated, observed or not observed. The arrow represents a possible process, not proof that every step occurred.
Work the comparison
Fictional A: 12/100 = 12%; B: 8/100 = 8%. The absolute difference is 4 percentage points. Relative to B, (12−8)/8 = 50%. Reversing the reference gives (12−8)/12 ≈ 33.3% fewer for B. Neither statement changes the observed counts.
Make an inspectable artifact
Draw signal → screening → callback → interview → offer. Mark what was manipulated, measured and not observed. Calculate both fictional comparisons, then write a bounded claim about the actual study and a test of a rival mechanism.
Record your diagram or table as a written trace in the claim field. Number each link, identify the relevant source ID and distinguish measured evidence from a proposed mechanism. Save a capture before revising your interpretation.
Test a stronger claim
Cite the real research and distinguish it from the invented arithmetic packet. Explain the gate, randomization, denominator, absolute/relative comparison, one unmeasured outcome and the extra evidence needed to identify a mechanism.
Independent extension
Inspect the linked research methods and replication documentation. Propose how to account for multiple applications to the same employer and audit signal validity. Explain why reanalyses of the same records are not independent replications.
Use the supplied records and public research descriptions. Do not conduct deceptive hiring or housing applications, collect classmates’ sensitive information, or treat the fictional ledger as a real effect estimate.
