Why ratings drift between managers
Two managers can watch the same work and give different ratings. Neither is lying: they apply the scale differently. Four patterns explain most of the gap:
- Leniency: some managers rate everyone high, to motivate or to avoid a hard conversation.
- Central tendency: others put nearly everyone in the middle band, which hides both strong and weak performance.
- Recency: the last two months weigh more than the first ten.
- Halo: one strong trait, or one memorable success, colours every other rating.
Calibration does not remove these patterns from people. It catches them in the ratings before the ratings become pay, promotion or development decisions.
Before the session
A calibration session is only as good as what managers bring to it. Prepare three things.
- Ratings with evidence. Every proposed rating should come with its reasons: the KPI results, the examples, the feedback. A rating without evidence cannot be compared.
- The current distribution. Show how many people sit in each rating band, per team and overall, so the group sees the pattern before discussing individuals.
- The ground rules. Agree who facilitates, who decides when the group disagrees, and what a change needs: a reason written down, and HR's approval if your process requires it.
Keep the group small enough to discuss people the members actually know: usually the managers of one department or one level, with HR facilitating.
Running the session
Start with the distribution, not with names. If one team has twice the share of top ratings of another, ask why before anyone defends an individual.
Then discuss people in an order that keeps the meeting honest: the highest and lowest ratings first, then anyone whose rating moved a long way from last year, then cases a manager asks to raise. Everyone in the middle band does not need a discussion.
For each case, the manager states the rating and the evidence; the others ask questions about the evidence, not the person. When the group agrees on a change, record the new rating and the reason in the meeting, not afterwards from memory.
Lock ratings while the session runs, so nobody changes a rating outside the room, and close the session only when every change has a reason.
Target distributions and the bell curve (forced distribution)
A target distribution says roughly what share of people you expect in each band: for example, 10% in the top band, 25% in the next, 45% in the middle. It is a useful check: a team far from it should be able to explain why.
A forced curve, often called a bell curve, is a different thing: it requires the shares, and moves people to fit them. Its known problems are real. Small teams cannot fit a curve, strong teams are punished for being strong, and people learn that ratings are rationed rather than earned.
The practical middle ground is a target with tolerance. Compare each band with its target, flag bands that are outside an agreed tolerance, and accept the exception when the group records a reason. The reason is what makes the outcome defensible later.
After the session
- Approve the changes. If your process needs HR approval, get it before anything reaches the employee.
- Write the final ratings back to the employee's record, keeping the earlier rating and the reason it changed.
- Tell managers what changed and why, so they can explain the final rating to their people in their own words.
- Review the process once the cycle closes: which teams moved most, and what that says about how the scale is understood.
A short checklist
- Every proposed rating comes with evidence.
- The group sees the distribution before discussing names.
- Extremes and big movements are discussed; the middle band is not.
- Every change has a written reason.
- Ratings are locked during the session.
- Exceptions to the target distribution are recorded, not forced away.
- Final ratings are written back with their history.