Reducing Hiring Bias: A Field Guide
Bias audits, structured interviews, blind sourcing, calibration sessions, and the metrics that prove your changes are working.
You can't reduce what you can't measure. This guide covers the four bias-reduction interventions with the strongest empirical support, the disparate-impact math every recruiter should know, and the dashboards to monitor.
The four interventions with the strongest evidence
Decades of organizational psychology research have produced a depressingly short list of bias-reduction interventions with replicable effect sizes. The good news: each is cheap and can be deployed in your current ATS without a re-platform.
- Structured interviews. Same questions, same order, same rubric, calibrated across interviewers. Moves interviewer-explained variance from ~60% to ~15% across replicated studies.
- Anonymized resume review. Strip name, school, dates, and graduation year for the first screen. Drops gender-of-name and prestige-of-school effects without materially affecting downstream pass-through.
- Work samples. A 60-minute task tied to the actual job outperforms every behavioral signal we know how to measure. The catch: pay candidates for their time.
- Calibration sessions. Once a week, every interviewer scores the same recorded interview, then debates the deltas. This is where your rubric stops being theatre.
The math every recruiter should know
Disparate impact under EEOC Uniform Guidelines is measured by the “four-fifths rule”: the selection rate for any protected group must be at least 80% of the selection rate for the highest group. If white-male applicants pass screening at 50% and Black-female applicants pass at 35%, the ratio is 0.70 — below the threshold, and your screening process is prima facie discriminatory unless you can defend it as job-related and consistent with business necessity.
This isn't theoretical: this is the math a plaintiff's lawyer runs first. Run it yourself, quarterly, at every funnel stage. If you can't pull the data, your ATS is the problem.
The dashboards to monitor
Four reports, weekly:
- Funnel pass-through by demographic group. Application → screen → onsite → offer → accept. Show the rate at each stage, not the cumulative number.
- Source-of-hire by group. If one source (referrals, a particular bootcamp, a particular agency) is producing a skewed pipeline, you'll see it here.
- Interviewer-level pass rates. Outliers in either direction need a conversation. Not punishment — a conversation.
- Time-in-stage by group. If one group spends materially longer in each stage, the loop is slower for them and they'll churn.
What stops working in a year
Two things drift: rubrics and interviewers. Rubrics drift because hiring managers add “just one more” criterion per loop until the rubric is unscorable. Interviewers drift because they get tired, get promoted, or stop attending calibration. The fix is the same as everywhere else: scheduled reviews. Once a quarter, your TA lead, your DEI lead, and a sample of interviewers re-audit the rubric and the calibration artifacts. Bias-reduction without ongoing maintenance is a photograph of a posture, not a posture.
Build a consent-first hiring pipeline today.
Most of what you just read is built into SourceHire's free plan. Try it without talking to a salesperson.