Pay progression · Performance ratings
Many pay systems move people through the band by rating. The rating looks objective because it is a number. The research shows how much of that number comes from the person giving it, and where gender enters.
On average feedback improves performance, but in more than a third of studies it made performance worse, mostly when people felt it threatened their self-worth (Kluger and DeNisi).
How fair employees find the process strongly shapes whether appraisal improves future performance (CIPD and CEBMa).
Two managers rating the same person often disagree, and agreement drops for complex jobs and for soft dimensions such as leadership (CIPD and CEBMa).
Ratings shift with the manager's power, the manager's own last appraisal, whether the manager hired the person, and whether the manager believes people can change (CIPD and CEBMa).
Helping, volunteering and visible initiative raise overall ratings beyond the actual job tasks (CIPD and CEBMa).
When ratings decide pay or promotion, managers focus on negatives and apply thresholds; in developmental reviews they weigh the whole picture (CIPD and CEBMa).
Annual appraisal is often run as a ritual. The evidence asks a simpler question first: which problem is it supposed to solve? (after Briner)
The bias rarely sits in the form. It sits in what managers notice, how they value it, and which number counts.
“Taking charge” went with the highest ratings for men, not for women. Women were more often criticised as too aggressive, men as not assertive enough.
What it means for your design: Anchor levels in behaviour and check whether the same behaviour earns the same score.
Written comments favoured women or showed no difference; the numerical ratings that decided promotion favoured men.
What it means for your design: Compare narratives with scores; the number is what counts.
Women got higher performance ratings but lower potential ratings, which explain 30–50% of the promotion gap. Women then outperformed men with the same potential score.
What it means for your design: Audit potential ratings by gender, or stop using them for promotion.
Part-time hours and family-friendly arrangements get read as less commitment, and ratings follow.
What it means for your design: Rate on time worked; never use availability as a proxy.
Men whose care duties caused absences got lower ratings and lower reward recommendations; women's ratings were not affected. The penalty is on care, not only on mothers.
What it means for your design: Treat any care-related absence neutrally, for everyone.
Art. 6 · EUR-Lex Art. 9 · EUR-Lex Art. 10 · EUR-Lex Art. 18 · EUR-Lex
A merit matrix on biased ratings doesn't create fairness. It documents unfairness with great precision.
Decide whether the process serves development, pay or promotion, and keep them apart in time and process.
Anchor each level with observable behaviour; drop words like “presence”, “drive” or “too aggressive”.
Define it, tie it to evidence and audit it by gender, or don't use it for promotion at all.
Frame-of-reference training improves accuracy most.
Panels compare ratings across teams with an explicit bias checklist.
A second reader checks that written comments and scores tell the same story.
Rate on the time actually worked, pro-rate objectives for part-time and leave, and never use availability as a proxy for commitment.
Few, clear objectives and minimal paperwork beat complex forms.
Run the rating audit below before ratings go into pay.
Step models, time paths with written rules, or anchored fit scores may serve better.
Six checks to run before any rating touches pay – starting inside one job. Example results for one fictional organisation.
| Check | What to compare | Example result | Status |
|---|---|---|---|
| Same job, different spread | Distribution of ratings by gender within one job | Customer advisor: top-two share 22% women vs 41% men | ✕ The rating is a finding, not an explanation |
| Rating distribution | Share in the top two ratings, women vs men | 28% vs 36% | ✕ Investigate |
| Part-time and leave | Top-two share, part-time vs full-time; returners vs others | 19% vs 35% | ✕ Investigate |
| Potential vs performance | Potential rating at the same performance rating, by gender | Women 0.4 points lower | ~ Review definitions |
| Rating to pay | Average increase at the same rating, by gender | 2.4% vs 2.9% at rating 3 | ✕ Fix before payout |
| Rating to promotion | Promotion rate at the same rating, by gender | 11% vs 12% | ✓ Within range |
Start inside one job. If the same work gets a different spread of scores, the rating is a finding, not an explanation – it cannot be used to explain the gap until the difference has a reason.
Ten questions. Answer yes, partly or no.