How to reduce bias in interviews: the steps research supports

To reduce bias in interviews, move decisions away from impressions and into a process. Agree the criteria before you see any candidate, ask everyone the same job-related questions, have each interviewer score alone against written anchors, and discuss candidates only after the scores are in. Each of those steps has research behind it, and each one works by changing what interviewers do, not only what they intend.

This page covers what the research found, where bias gets into a hiring process stage by stage, the control for each stage, what to say in a debrief, and a structure audit you can run on one role this week. How interviewers write up evidence is covered in evidence-based interview feedback; how to decide between finalists is in how to compare candidates after interviews.

What the research supports

Five sources explain most of the advice below. Each has limits as well as findings.

The 15 components in the McCarthy study come from Campion, Palmer and Campion (1997), and the audit at the end of this page is a plain-language version of them. Two cautions: lab studies with students are not hiring decisions, and one field study in one organization is not a guarantee for yours. What the sources agree on is direction. The less an interviewer has to improvise, the less room bias has.

Where bias gets in, stage by stage

Ask where in the process each bias shows up, because that is where the control goes.

The rating errors in the middle rows are covered in detail in evidence-based interview feedback.

Before the first interview: lock the criteria

Uhlmann and Cohen's finding is the most practical one here: people shift what a job "really" needs toward whoever they already prefer, and fixing the criteria first stopped it. So the single most useful document in a fair process is the one written before anyone applies.

• Agree four to six competencies with the hiring manager at intake, each defined as behavior you could observe, not as a trait. The intake meeting questions get you there.
• Mark which are must-haves. A must-have is a gate, not a heavy weight.
• Write the questions and the anchors for each competency before the first screen. The interview scorecard template has a 1–4 anchored scale to start from.
• Name what is not a criterion. Specific schools, specific former employers, "presence", "polish". If one of these genuinely matters, rewrite it as a behavior and add it for everyone.
• Log every later change, with the reason, and apply it to every candidate, including those already declined.

Copy this into the job folder and fill it in at intake:

CRITERIA LOCK: [Role]          Agreed: [date]   By: [names]

Must-haves (a 1 on any of these is a no)
1. [Competency]: [observable behavior, e.g. "has run month-end close
   for a team of 3+ without a manager checking each entry"]
2. [Competency]: [ ]

Weighted competencies
3. [Competency] [weight %]: [behavior]
4. [Competency] [weight %]: [behavior]
5. [Competency] [weight %]: [behavior]

Not criteria for this role (do not score or discuss)
- [e.g. specific schools or employers, industry background, "polish"]

Questions and anchors: interview kit version [1], dated [date]

Changes after screening began
- [date] [what changed] [why] [approved by] [applied to all: Y/N]

In the interview: same questions, same conditions

Structure in the interview means every candidate gets the same chance to show the same things. OPM's guide asks for the same amount of time for every candidate and a consistent setting; the rest follows from that.

• Ask the planned questions in the planned order. Follow-up probes are fine if they are written in the kit and used for everyone, not invented for the candidate you find more interesting.
• Read the resume before, not during. Pull any questions you need from it in advance, then close it. OPM warns that looking at documents mid-interview can tilt ratings toward candidates with strong resumes.
• Separate rapport from assessment. A few minutes of warm-up are good for the candidate. Tell yourself, or say aloud, when the structured part starts, and score nothing from before it.
• Write down what was said. Verbatim fragments, numbers and examples, not adjectives. How to take interview notes covers doing this while still listening.
• Offer accommodations the same way to everyone, in the invitation, so nobody has to ask twice. The EEOC notes that employers must provide reasonable accommodations to applicants with disabilities who need them (EEOC).
• On video, score the answers, not the room. Backgrounds, lighting, a child in the hall or a weak connection say nothing about the competencies. How to run a video interview covers setting this up fairly.

Scoring: alone, against anchors, before anyone talks

Independent scoring is the step most often skipped under time pressure, and the one that protects every other step. Once an interviewer has heard a colleague's view, their own rating is no longer an independent piece of evidence.

• No discussion between interviews. This is one of Campion's components of structure. A hallway "how did it go?" between the second and third interviewer is enough to break it.
• Every score has evidence under it, and "not assessed" is allowed. A competency nobody asked about must not get a polite 3.
• Submit before the debrief, with a timestamp, and do not open anyone else's scorecard first.
• Run the swap test on your own ratings. Picture the same answer, word for word, from a candidate with a different name, age, accent or school. If your score would move, the score is not about the answer.

Interview Signal, for example, attaches evidence quotes checked against the transcript to each score, and on a panel shares only which questions were covered, not anyone's notes. The judgment stays with the interviewer.

Panels: diverse, consistent, and not a substitute for structure

OPM's guide recommends a panel of two or three interviewers because they document and interpret more, reduce the risk of rating bias, and allow a diverse range of interviewers, which shows candidates the organization values fair treatment. That is a good reason to build mixed panels.

The McCarthy study adds a useful nuance. In its highly structured interviews, the racial composition of the panel did not affect scores, and the authors note that diverse panels may still help attract diverse candidates. Read together: diversify panels for the candidate's experience and for the range of perspectives in the debrief, but do not expect a mixed panel to fix an unstructured interview.

• Keep the same interviewers for every candidate on a role where you can. OPM and Campion both list this; a different panel for each candidate makes scores harder to compare.
• Give each interviewer their own competencies so nobody rates a skill they did not ask about. The panel interview plan template sets this out.
• Do not ask one person to represent a group on every panel. It burns out the same few colleagues and puts the fairness of the process on them rather than on the design.

In the debrief: question the words, not the person

Debriefs are where scores that were independent become a group view. Reveal all scores at once, let the person with the least seniority speak first, and use the interview debrief template for the agenda. Then listen for shorthand. The question that works on every comment, positive or negative, is "what did they say that shows that?"

If a score changes during the debrief, it changes because of evidence someone missed, and the change is recorded with the reason. OPM's panel process has members initial any changed rating. Comments that should never be written down at all are listed in what not to write in interview notes.

A structure audit for one role

Pick one open role and answer each line honestly. Every "no" is a place where the interviewer is improvising. The list adapts the 15 components of structure from Campion, Palmer and Campion (1997), as listed in the McCarthy study.

STRUCTURE AUDIT: [Role]                          Date: [ ]

Content of the interview
[ ] 1. Competencies come from a written analysis of the job
[ ] 2. Every candidate is asked the same questions
[ ] 3. Follow-up probes are planned and limited
[ ] 4. Question types are chosen on purpose (past behavior,
       situational, experience)
[ ] 5. Enough time to ask every planned question
[ ] 6. Extra information (resume, referrals, test scores) is
       kept out of the interview itself
[ ] 7. Candidate questions come after the structured part

Evaluation
[ ] 8. Each competency is rated on its own scale
[ ] 9. Every point on the scale has a written anchor
[ ] 10. Interviewers are expected to take detailed notes
[ ] 11. More than one interviewer rates the candidate
[ ] 12. The same interviewers see every candidate
[ ] 13. No discussion of candidates between interviews
[ ] 14. Interviewers are trained on the kit and the scale
[ ] 15. Ratings are combined by a rule agreed in advance,
        not by the loudest voice

Score: [ ] / 15    Fix first: [the "no" that affects most candidates]

Items 6, 12 and 13 slip quietly under time pressure. Item 14 is the one teams assume they have covered; an interview calibration session is the practical version.

Check the outcomes, not only the process

A fair-looking process can still produce lopsided results, so look at who moves through each stage. Track, per role and per quarter, how many candidates reached each step: applied, screened, interviewed, offered.

U.S. federal agencies use a rule of thumb from the Uniform Guidelines on Employee Selection Procedures: a selection rate for any race, sex or ethnic group below four-fifths (80 percent) of the rate for the group with the highest rate is generally regarded as evidence of adverse impact (29 CFR 1607.4(D), as of September 2026). It is a starting signal, not a verdict, and the same section notes that small numbers need care.

If any tool in your process scores or ranks candidates automatically, local rules may require more than this. New York City, for example, requires a bias audit for automated employment decision tools; see NYC Local Law 144.