Fairness & compliance
How to reduce bias in performance reviews: recency, halo, leniency and calibration checks
Eight types of performance review bias with an example of each, a manager checklist, how to spot bias in rating data, and how check-ins, 360 feedback and calibration help.
By the CLEAR Talent team9 min read
Key takeaways
- Bias thrives where evidence is thin and rating standards are vague. Fix those two conditions before relying on training.
- Recency bias in performance reviews is the most common and the most preventable: a record kept through the year beats memory at year end.
- Look for bias in the data, not only in individual reviews: compare rating distributions by manager, team and gender within comparable roles.
- A pattern in the data is a question to investigate, not proof. Calibration is where it gets asked, before ratings are final.
Most performance reviews are written by a manager working from memory, under time pressure, about people they work with every day. Bias in that setting is not a character flaw. It is what judgement does when the evidence is thin and the standard is vague, which is why telling managers to be objective rarely changes the ratings.
This guide names eight common forms of performance review bias with a one-line example of each, gives managers a checklist to work through before they submit, and shows HR how to look for bias in the rating data — by manager, team and gender. It then covers how continuous check-ins and 360 feedback reduce the problem at source, and where calibration and the EU Pay Transparency Directive come in.
Why performance review bias survives good intentions
A rating is a judgement made under conditions that invite shortcuts. The manager is summarising months of work in an afternoon, often for several people at once, against descriptors such as “meets expectations” that each reader pictures differently. Where evidence is missing, the mind fills the gap with whatever is vivid, recent or familiar.
That is why awareness training on its own tends to have a modest effect. Managers who can name every bias still rate from memory if memory is all they have. The fixes that work change the conditions instead: a shared standard written down in advance, evidence collected through the period, more than one perspective on the same person, and a check of the results before they are final.
Eight types of performance review bias, with examples
Each of these leaves a recognisable signature in a review. The examples are hypothetical; the right-hand column is the question to ask before a rating goes in.
Reference
Eight rating biases and how to check for them
| Bias | One-line example | Check before you submit |
|---|---|---|
| Timing and memory | ||
| Recency bias | A strong final quarter turns an uneven year into “exceeds expectations”. | Cite evidence from every part of the period, not only the last few weeks. |
| First-impression bias | Someone who struggled in their first month is still described as “finding their feet” a year later. | Ask what has changed since the earliest evidence you hold. |
| One trait colours the rest | ||
| Halo effect | A confident presenter is rated highly on planning, which they rarely do well. | Rate each competency against its own anchors, one at a time. |
| Horns effect | One missed deadline drags down the ratings for collaboration and quality. | Separate the incident from the competencies it does not touch. |
| How the scale is used | ||
| Leniency and severity | One manager rates the whole team above the standard; another rates nobody there. | Compare your spread of ratings with comparable teams before submitting. |
| Central tendency | Everyone lands in the middle because the manager has no evidence to defend anything else. | Make the evidence for a middle rating as specific as for any other. |
| Who the rater compares with | ||
| Similarity bias | The colleague who shares the manager’s background and working style is rated more generously. | Would the rating hold for someone with the same results and a different style? |
| Contrast effect | A solid performer is rated down because they sit beside the team’s strongest person. | Rate against the written standard, not against the person beside them. |
Gender and other protected characteristics can enter through any of these routes — in which behaviours are read as leadership, or in how flexible working is treated — which is why the data checks below matter.
Guide: performance rating scales, and how scale design reduces rating errors
A manager checklist for fairer ratings
Work through this before submitting any rating. It is printed in full so you can copy it into your own review guidance.
Checklist
Before you submit a rating
Gather the evidence
- Re-read the goals and rating definitions agreed at the start of the period.
- Collect notes, check-in records and feedback from the whole period, not just recent weeks.
- Find at least one specific example for each competency you are rating.
- Note context outside the person’s control, such as leave, a role change or a team vacancy.
Rate against the standard
- Rate goal results and each competency separately before forming an overall view.
- Rate what was delivered in the time actually worked, including for part-time staff and people returning from leave.
- Describe behaviour and outcomes, not personality: “resolved escalations within the agreed time”, not “a natural leader”.
- Check that you would use the same words for anyone in the role with the same results.
Test before you submit
- Ask what this person would have needed to do to be rated one level higher, and one level lower.
- Look at your own spread of ratings: if everyone sits in one band, ask whether the evidence says so.
- Flag any rating you would struggle to explain to the employee with evidence.
If you cannot find evidence for a rating, that is a finding in itself: record the gap rather than filling it with an impression.
How to detect bias in rating data
Individual reviews rarely reveal bias on their own; patterns across many ratings do. HR can run these checks on each cycle’s draft ratings before calibration.
- Distribution by manager
- Place each manager’s ratings side by side for comparable roles. A manager whose team sits consistently above or below everyone else’s may be applying a different standard, or may genuinely lead a stronger team. The evidence decides which.
- Distribution by team and level
- Compare how ratings fall across departments and grades. If the top band goes mostly to senior staff or to one function, check whether the standard is applied consistently or whether visibility to leadership is doing the rating.
- Distribution by gender
- Compare ratings, and the pay or promotion outcomes linked to them, by sex within comparable roles. Look at working patterns too: part-time staff and people returning from leave are where gaps often hide.
- Changes made in calibration
- Track which ratings moved in calibration, in which direction and for whom. If adjustments consistently favour one group, the meeting is adding bias rather than removing it.
- Language in the written comments
- Read a sample of comments attached to the same rating across groups. “Abrasive” for some people and “direct” for others with the same results is a signal worth discussing.
Two cautions. Small groups produce noisy numbers and can identify individuals, so set a minimum group size and handle the analysis within your data protection obligations. And a gap is a question, not a verdict: the next step is to examine the evidence behind the ratings, not to adjust them until the numbers look even.
How continuous check-ins and 360 feedback reduce bias at source
Most rating bias is an evidence problem. Two practices change the evidence a manager holds when the review opens.
- Regular check-ins counter recency
- A short record from each one-to-one — progress on goals, feedback given, obstacles raised — means the year-end rating is written from twelve months of notes rather than six weeks of memory. It also challenges first impressions, because the record shows change over time.
- A running journal keeps the detail
- Observations noted when they happen, tagged to the competency they show, are more accurate than anything reconstructed later. They also help the quiet contributor whose good work was never presented to leadership.
- 360 feedback widens the view
- Peers, direct reports and other stakeholders see behaviour the manager does not. Collected against the same competencies as the review, their input tests a halo or horns judgement rather than repeating it.
- Anonymity and thresholds keep it honest
- Raters are franker when their feedback is grouped and cannot be traced, so set a minimum number of raters per group. Treat 360 input as evidence to weigh, not a score that overrides the manager.
See continuous performance management: check-ins, journals and feedback between reviews
Where calibration and the EU Pay Transparency Directive come in
Calibration is the organisation’s last check on bias before ratings become final. Managers compare how they have applied the scale across comparable roles, starting with the outliers, and every changed rating is recorded with its reason. Bring the distribution checks above into the calibration pack so the room sees the patterns as well as the individual cases.
The stakes are rising in the EU. Directive (EU) 2023/970 expects pay and pay progression to rest on objective, gender-neutral criteria that employers can explain, and where performance ratings feed raises, bonuses or promotion, they become part of that explanation. Where gender pay gap reporting shows a gap of at least 5% in a category of workers that the employer cannot justify and has not remedied within six months, a joint pay assessment with worker representatives follows. Ireland has confirmed it will miss the June 2026 transposition deadline and is phasing implementation, so check the current position before relying on any date.
The practical response is the work this article describes: written criteria, evidence behind every rating, a check of outcomes by sex within comparable roles, and a record of what changed and why. This is general information, not legal advice.
Where CLEAR Talent fits
CLEAR Talent is built to keep evidence attached to ratings. Managers record 1-on-1s, development notes and kudos tagged to competencies, and log observations in a performance journal as they happen, so the formal review opens with the period’s evidence already there. 360° feedback runs against the same competency framework as the review, with raters anonymised by group and minimum-rater thresholds enforced.
At review time, Ava, the platform’s AI assistant, flags rating patterns that may indicate gender, recency or leniency bias and prompts the manager before submission. A prompt is not a conclusion; the manager decides. Calibration views use heatmaps and outlier flags to show inconsistent ratings across managers before anything is final, every change is logged with who, what and when, and a roll-up Excel export supports your own analysis.
See 360-degree feedback software mapped to your competencies