The contrast effect in interviews: definition, research and how to prevent it

The contrast effect in interviews is when a candidate's rating depends on who was interviewed just before them rather than on the written standard: an average candidate looks strong after two weak ones and weak after two strong ones. The U.S. Office of Personnel Management lists it among common interviewing mistakes. It was first shown in a 1972 experiment by Wexley and colleagues and has since been measured in tens of thousands of real hiring and admission interviews. The controls that hold up are rating each candidate against written anchors before the next interview, leaving breaks between interviews, and combining several independent ratings.

This page covers the research in more depth than our one-paragraph entry in common interviewer mistakes, how candidate order shifts scores, the signs in scorecards, and which controls the evidence supports, including some that do less than people assume.

What is the contrast effect, exactly?

OPM's Structured Interviews: A Practical Guide (September 2008) puts it in one line: the order in which candidates are interviewed can affect their ratings, and interviewers should not compare and contrast candidates with those interviewed before. The benchmark silently becomes "the last person I saw" instead of the anchor on the scorecard.

What does the research show?

Two details from the field studies are worth knowing. In Simonsohn and Gino's data, a one standard deviation rise in the average score of the day's earlier applicants lowered the next applicant's expected score by about as much as 30 fewer GMAT points would. In the 2024 working-paper version of Radbruch and Schiprowski, candidates who followed someone who got a yes vote were about 15 percent less likely to get a yes in the admission process and about 40 percent less likely in the hiring process, and the effect was strongest early in a sequence and when consecutive candidates were similar.

Read together: the effect is weak when the evidence is on paper and the candidates are clearly different, and strong for the middle-of-the-range candidate in a live interview, which is exactly where most hiring decisions are close.

How does candidate order shift scores in practice?

An invented example: three interviews for a support team lead on one Tuesday, one interviewer, scale 1 to 4.

10:00  Candidate A   Escalation handling  4  (anchor met for 4)
11:00  Candidate B   Escalation handling  4  (anchor met for 4)
13:00  Candidate C   Escalation handling  2
       Note: "Fine, but nowhere near A or B. Less structured."

What C actually said: described taking over a customer escalation,
calling the customer the same day, agreeing a fix date with
engineering and confirming it in writing. The written anchor for 3:
"Owns the escalation end to end, sets expectations with the customer,
coordinates one other team."

Candidate C met the anchor for a 3 and was scored a 2, and the note explains why: the comparison was with A and B, not with the anchor. Had C been first that morning, the same answer would likely have scored 3, or even 4 after a weak start to the day. Nobody involved would describe it as bias; it feels like calibration.

What are the signs of the contrast effect in scorecards?

• Comparative words in the evidence. "Better than this morning's candidate", "not at the level of the others", "the strongest so far". Evidence describes what the candidate said, not how they ranked.
• Scores that track the order. Across an interviewer's scorecards, a candidate's score tends to be lower after a high score and higher after a low one, more often than chance would explain.
• Day-level quotas. "I've already said yes to two today." That is narrow bracketing, and Simonsohn and Gino found its pull grew as the day went on.
• Ranking before scoring. The interviewer orders candidates in their head, then picks scores to fit the order.
• Batch scoring at the end of the day. When all scorecards are written at 6 p.m., every candidate is rated beside the others rather than against the anchor.

To check your own pattern, take a quarter of scorecards, sort them by interviewer and time, and compare each score with the previous candidate's. A handful of interviews proves nothing either way; look for a consistent pattern across many.

How do you prevent the contrast effect in interviews?

1. Write anchors and read them before every score. OPM's guide says interviewers minimize rating errors by comparing what the candidate did with the behaviors anchoring each rating level. Anchors are necessary but, on the evidence above, not sufficient: the 1973 anchoring treatment failed and the 2019 anchored scales did not significantly reduce contrast. Use them actively: read the anchor text, then the evidence, then score. Wording to start from is in behavioral anchors for interview scores.
2. Score each candidate before the next interview. OPM recommends rating immediately after the candidate leaves. Never batch scorecards at the end of the day.
3. Leave real breaks. In Radbruch and Schiprowski's admissions data, the previous candidate's influence roughly halved when an hour or more separated two interviews and approached zero across a change of day. Back-to-back slots with no gap are the worst case. Our suggestion, not a tested threshold: at least 15 minutes after each interview, used for scoring.
4. Do not rely on random order. In their simulations, randomly reordering candidates left the effect unchanged; it only moves the harm to different people. Reordering so that similar candidates were not back to back reduced it by about 40 percent, though that was a simulation, not a trial.
5. Collect independent ratings. Several interviewers who each score alone dilute any one person's sequence. The same simulations suggest the dilution is slow, so treat it as a complement to the steps above. Keep the ratings independent: see independent interview feedback before the debrief.
6. Flag pivotal scores. Before the debrief, mark any decisive score given right after an unusually strong or weak candidate, and look at its evidence again. The authors propose this kind of flag as a cheap safeguard.
7. Train with practice, not a warning. In 1973 an intensive workshop worked where warnings did not, and in the 2024 field study a briefing about the contrast effect did not significantly reduce it. Rating the same recorded answers as a group and discussing gaps, as in an interview calibration session, is the closer equivalent.
8. Compare candidates only at the end. Ranking belongs in the debrief, after every scorecard is in, using the evidence under each score.

Does the contrast effect matter for recruiters and agencies too?

Yes, wherever someone sees candidates one after another. A recruiter running back-to-back phone screens is exposed in the same way as a hiring manager, and the same steps apply: a short written screen against the brief, completed before the next call. When a client reviews a shortlist, sending the candidates together with the same summary format makes it more likely each is read against the brief rather than against whoever arrived first. That is a practical suggestion, not a research finding. For the wider set of rating errors and where each one enters the process, see how to reduce bias in interviews.