Most review forms use one rating scale for every question, and that’s a big reason the ratings come back muddy. A better approach is to match the scale to what you’re asking: a frequency scale for how often someone does something, an agreement scale for a statement about working with them, a proficiency scale for a skill, and an expectations scale for the manager’s overall call. Below are rating scale examples for each, with the labels written out, a 5-point performance rating scale with definitions, and a short rule for choosing.
I’ve argued before that a score is only a pointer to the feedback, and I haven’t changed my mind. Most teams will use a scale anyway, so it’s worth getting the scale right. The good news is that the part people argue about most, five points or seven, matters far less than what each point is tied to.
Key Takeaways
- Match the scale to the question: frequency for “how often”, agreement for a statement, proficiency for a skill, and an expectations scale for the manager’s overall rating.
- The number of points matters much less than a one-line definition for every level that two raters would read the same way.
- On a 5-point performance rating scale, say plainly that 3 means good, expected work, or ratings drift up and bunch at 4.
- In a 360 review, peers and direct reports rate what they’ve seen, so frequency and agreement scales fit them better than “exceeds expectations”.
- Every score needs a written reason next to it, and the spread of scores across raters tells you more than the average.
A score points at something; the written comment next to it is what a person can act on. That was the argument in the earlier piece, and this page doesn’t walk it back.
What it adds is the practical half. If your review form has rating questions, three things decide whether the numbers mean anything: each point has a written definition, each score has a comment next to it, and you read how the scores spread before you look at the average. The scale type is the first decision that makes those three possible, because a definition only works when the scale fits the question.
| Scale | Labels, low to high | The question it answers | Best for |
|---|---|---|---|
| Frequency | Never, Rarely, Occasionally, Often, Always | How often does this happen? | Behaviour anyone around the person can observe |
| Agreement | Strongly disagree, Disagree, Neutral, Agree, Strongly agree | Do you agree with this statement? | What it’s like to work with someone |
| Proficiency | Novice, Developing, Competent, Advanced, Expert | How skilled is this person at this? | A specific craft, rated by someone who knows it |
| Expectations | Well below, Below, Meets, Above, Well above expectations | How did the period compare with what was agreed? | The manager’s overall rating |
| Numeric | 1 to 10 | Where does this sit, roughly? | A quick position, read alongside the comments |
A frequency scale asks how often something happens, which is the easiest thing for a rater to answer from memory. It works best with questions about a specific behaviour:
The endpoints are “Never” and “Always”, and that’s fine on a scale, where they mark the two ends of a range. The trouble starts when the same words show up in the comment box as a verdict (“she never listens”), which is the problem with always and never in written feedback. Ask raters to back a 1 or a 5 with one example.
An agreement scale, also called a Likert scale, asks raters how much they agree with a statement. It suits questions about the experience of working with someone:
The middle point, “Neutral”, is useful when a rater’s view sits in the middle. It also becomes the answer for raters who haven’t seen the behaviour or don’t want to commit. Add a separate “I haven’t seen this” option so Neutral stays a real answer.
A proficiency scale rates skill, so each level needs a description for that skill. Here’s a ladder for writing project proposals:
| Level | What it looks like |
|---|---|
| Novice | Needs a template and close review to produce a proposal |
| Developing | Writes a usable proposal for familiar work and needs help on scope and risk |
| Competent | Writes proposals others can approve without major edits |
| Advanced | Handles unfamiliar or cross-team proposals and spots risks others miss |
| Expert | Their proposals become the team’s template, and colleagues ask them to review drafts |
Only people who know the craft can use this scale well. A peer from another discipline asked to rate someone’s proposal writing from Novice to Expert will guess, so keep proficiency questions for the manager and colleagues in the same field.
This is the classic manager’s scale: did the person meet what was agreed for the period? It comes in three-, four- and five-point versions, and the five-point one is covered in detail below. It needs a rater who knows the expectations, which usually means the manager.
Ten points feel precise, but nobody can write ten distinct definitions, so each rater invents their own meaning for a 7 versus an 8. A 1 to 10 scale works when you treat it as a rough position and let the comment carry the detail. If you want labels on every point, five is the practical limit.
A 5-point performance rating scale usually runs from “well below expectations” to “well above expectations”, with 3 as “meets expectations”. The definitions are what make it usable:
| Rating | Label | Definition |
|---|---|---|
| 5 | Well above expectations | Went well beyond the agreed goals and raised the standard for the team, with results people outside the team would recognise |
| 4 | Above expectations | Met every agreed goal and went further in at least one area that mattered |
| 3 | Meets expectations | Delivered what was agreed, at the expected quality, with the usual support |
| 2 | Below expectations | Missed some agreed goals or needed more support than expected, with the gaps named in the comments |
| 1 | Well below expectations | Missed most agreed goals, and a support plan is needed |
Say plainly, in the form itself, that a 3 means good, expected work. If people read 3 as average or as a warning, managers will start giving 4s to anyone they don’t want to upset, and the scale stops separating anyone.
Definitions like these describe the result. For a single competency, you can write the same five levels as behaviour. Here is communication:
| Rating | Communication, in practice |
|---|---|
| 5 | Others reuse their written updates as the reference, and stakeholders outside the team rarely need a follow-up meeting |
| 4 | Shares decisions and risks in writing without being asked, clearly enough to act on |
| 3 | Keeps the team informed about their own work, and updates sometimes need a follow-up question |
| 2 | Updates usually arrive when someone asks, and important changes reach others late |
| 1 | People regularly hear about changes to this person’s work from someone else |
A 4-point version drops the middle (well below, below, above, well above), which forces every rating to lean one way. A 3-point version (below, meets, above) is simple to use but can’t tell a strong year from an exceptional one.
An anchor is a short description of what a rating means for a particular question. “Communication: 4” leaves each rater to decide what a 4 is, while “4: shares decisions and risks in writing without being asked” tells them.
Imagine a 40-person company where two managers rate the same analyst on communication. One gives a 4 because the analyst’s updates are clear. The other gives a 2 because the updates only arrive when someone asks. Both are right about different things, and an average of 3 tells the analyst nothing.
With an anchor on each level, both managers are rating the same behaviour, and if they still disagree, the gap points at something worth a conversation.
The formal version of this is a behaviorally anchored rating scale (BARS), where every level of every question is described by an example of real behaviour. Full BARS takes a long time to build. One line per level, for the handful of questions that matter most, gets you most of the benefit. If you’re writing those questions now, start with 360 feedback questions that ask about a specific moment, or use the 360 feedback question generator, which gives every question a short anchor for what a high rating means.
Write the question first, then pick the scale that answers it:
| If the question asks… | Use |
|---|---|
| How often someone does something | Frequency |
| Whether the rater agrees with a statement about working with them | Agreement |
| How skilled someone is at a specific craft | Proficiency |
| How the period compared with agreed goals (manager only) | Expectations |
| A quick overall position, with the detail in the comment | Numeric, 1 to 10 |
Picture a team lead building a first review form with 12 questions, all on an agreement scale, including “I agree that [Name] is an expert in SQL”. Half the raters have never seen the person’s queries, so they click Agree to be kind, and the report shows a team of experts. Moving the skill question to a proficiency scale rated by the two people who review that code, and asking everyone else how often [Name] helps them get unstuck, gives two answers that each mean something.
On the number of points: in my experience the five-versus-seven debate is the least important choice on the form. An odd number gives raters a middle; an even number makes them lean. For a first round, five labelled points work well. Keep similar questions on the same scale, so ratings can be compared and nobody has to stop and work out which way a 4 points. If your tool uses one scale per round, pick the one that fits most of your questions and phrase the rest to suit it: “How often does [Name] explain decisions clearly?” works on a frequency scale as well as the agreement version does.
In a manager review, one person who knows the goals rates the whole period. In a 360, peers and direct reports each see a slice of the work. They can say how often someone shares context in meetings with them, and they can’t say whether the person exceeded expectations they never saw. That’s why frequency and agreement scales suit 360 raters, and why the expectations scale belongs to the manager.
The other difference is that a 360 gives you several scores for the same question. If two peers give a 5 and two direct reports give a 2, the person probably behaves differently with each group, and that difference is the most useful thing in the report. An average of 3.5 hides it, which is also how the average hides rater bias. Read the spread first.
The scale should be the last decision on a review form, and most teams make it first. Decide what you want to learn about the person, write the question so it asks about something a rater has seen, and the right scale usually becomes obvious. Then write one line for each level, add a comment box, and you have ratings that point at something specific.
If you run your reviews in Lynxify.me, you pick one of four scales for each review round: frequency (the default, Never to Always), agreement, proficiency (Novice to Expert), or a plain 1 to 10. Every question carries a line describing what a high rating means, and each score sits next to the reviewer’s written answer. There’s no expectations scale among them, so a manager’s overall rating against goals stays in your company’s own review process. You can start free with Lynxify.me and sign in with Google or Microsoft, with no setup call.
What is a good rating scale for performance reviews?
For most teams, a 5-point scale with a one-line definition for every level. Match the type to who is rating: an expectations scale (below, meets, above) for the manager’s overall rating, and a frequency or agreement scale for colleagues, who can say what they’ve seen but not what was agreed. Put a comment box next to every score.
What does a 5-point performance rating scale look like?
5 is well above expectations, 4 above expectations, 3 meets expectations, 2 below expectations, and 1 well below expectations. Each level needs a one-line definition of what it means in practice. Tell people that 3 means good, expected work, or most ratings will drift up to 4.
Should a rating scale have a neutral middle option?
It can, as long as it means something. A middle point lets raters say the behaviour is about where it should be. It also gives them somewhere to hide when they don’t want to commit, so add a separate “I haven’t seen this” option and keep the middle for raters who have a view.
How many points should a rating scale have?
Four or five labelled points work for most review forms, and a 1 to 10 scale is fine if you treat it as a rough position. The point count matters much less than a written definition for each level, a comment next to every score, and reading how the scores spread across raters.
What is a behaviorally anchored rating scale (BARS)?
A rating scale where each level is described by an example of real behaviour for one specific skill. For communication, a 4 might read “shares decisions and risks in writing without being asked”. It takes longer to write than a generic scale, so most teams do it for the handful of questions that matter most.
Dmytro Shtapauk
Practical tips and experience from top-notch experts