How many points should a performance rating scale have? A performance rating scale is the fixed set of grades a manager chooses from when summing up one person’s performance for a review period, and for most companies of 100 to 1,500 people the practical answer is five points, each with a written definition and example behaviours, checked in calibration. Choose four points if you deliberately want managers to make a call above or below the bar, and three only if the rating does not drive pay or promotion. Seven or more points rarely help with performance grades, because managers cannot defend the extra distinctions. The number matters less than two things around it: whether every manager reads each point the same way, and whether the grades actually spread out when performance does.

Key takeaways

  • Five anchored points is the sensible default. It leaves a clear grade for solid work at the bar, one step below it and two steps above it, which is what pay and promotion decisions usually need.
  • Very short scales lose information. In a study of 149 people rating services on scales from 2 to 11 categories, Preston and Colman (2000) found that 2, 3 and 4-point scales performed relatively poorly, with reliability and validity indices rising up to about 7 points.
  • More points will not fix accuracy on their own. Matell and Jacoby (1971) found reliability and validity were broadly unaffected by the number of response options, so extra points are no substitute for clear definitions.
  • Segmentation can affect motivation. In a laboratory experiment with 305 participants, Bartol, Durham and Poon (2001) found a 5-category system produced higher self-efficacy, higher improvement goals and larger rating improvements than a 3-category one.
  • The real risk is compression. Bol (2011) found that centrality bias, where ratings bunch in the middle, rises when gathering evidence is costly and manager–employee relationships are close, and that it was associated with weaker performance improvement.

3, 4, 5 or 7 points: the options compared

Every scale length is a trade-off between how finely managers can grade and how consistently they can do it. The table below sets out what each option gives you and where it tends to break in a growing company. The five levels themselves, with example behaviours for each, are on the employee performance rating scales guide.

ScaleWhat it gives youWhere it breaksFits when
3 points (below, at, above the bar)Quick to teach and easy to explain to employees.Solid and strong performers share one middle grade, so pay and promotion decisions need a second, informal ranking.Ratings are for development only, or the company is small enough that leaders know every person’s work.
4 points (no middle grade)Forces every manager to say whether someone is above or below the bar.No grade for reliable “at the bar” work, so many people feel pushed into a label that does not fit.You have a serious central-tendency problem and want a reset for one or two cycles.
5 points (anchored)A grade for solid work, one step below, and two steps above for strong and exceptional work.Without written anchors and calibration, managers avoid the ends and five points behave like three.Ratings feed pay, promotion or talent decisions across several teams. The usual default.
7 points or moreFiner grading on paper.Two managers rarely defend the gap between a 5 and a 6 the same way, and calibration time grows.Rarely, for performance grades. Better suited to surveys than to decisions about people.
No overall ratingKeeps the conversation on development rather than a number.Pay and promotion decisions still need a judgement, which tends to move into private spreadsheets.Pay is not linked to performance and promotions run as a separate process.

If your ratings feed a pay matrix, the scale length also sets the number of rows in that matrix. The merit matrix guide shows how a calibrated rating and position in range become a recommended increase, and the post on whether performance reviews should be linked to pay covers when that link helps and when it hurts.

What the research says about scale length

Very short scales lose information. Preston and Colman (2000) asked 149 people to rate a recent shop or restaurant visit on scales that differed only in the number of categories, from 2 to 11, plus a 101-point scale. Scales with 2, 3 or 4 categories performed relatively poorly on several indices of reliability, validity and discriminating power, and the indices were higher for longer scales up to about 7 categories. Test–retest reliability tended to fall above 10 categories, and respondents preferred the 10-point scale, followed closely by the 7 and 9-point scales. The caveat matters: these were customers rating services, not managers grading colleagues against a written standard.

More points do not create accuracy. Matell and Jacoby (1971) tested Likert items with different numbers of alternatives and found that reliability and validity were essentially independent of the number of options. Read together with Preston and Colman, the message for a People team is modest: avoid very short scales if you need to separate people, but do not expect a longer scale to make managers agree.

Segmentation changes what people aim for. Bartol, Durham and Poon (2001) compared a 3-category and a 5-category appraisal system in an experiment with 305 student volunteers doing a search task. The 5-category system led to higher self-efficacy about reaching the next category, higher goals for improving the rating and larger actual improvements, and both the system and the rating received affected perceptions of fairness. A laboratory task is not a workplace, but the mechanism is familiar: a next step that looks reachable is more motivating than one that looks a long way off.

Compression is the failure to watch. Bol (2011), studying the incentive plan of a financial services firm, separated centrality bias (ratings squeezed towards the middle) from leniency bias (ratings pushed up). Both rose when gathering performance information was costly for the manager and when the manager–employee relationship was strong. Centrality bias was associated with weaker performance improvement. A five-point scale where nearly everyone gets a 3 carries the same information as a one-point scale.

Removing ratings is a live debate, not a settled answer. Adler and colleagues (2016) recorded a debate among industrial–organisational psychologists on dropping ratings altogether. The case for removal included disagreement between raters and the weak link between ratings and actual performance; the case for keeping them included the point that performance is always evaluated somehow, and that changing the rating format alone is unlikely to fix a performance process. That is the strongest reason to treat scale length as a design detail and evidence quality as the main event. For the wider list of rating errors and how to counter them, see the performance review bias guide.

Five or four: the middle-grade question

The real choice for most companies is between five and four points, and it turns on the middle grade. Survey research by Weijters, Cabooter and Schillewaert (2010), with 1,207 online respondents across eight scale formats, found that including a midpoint increased agreeing responses and reduced extreme ones. In a performance review, the equivalent is a middle grade that becomes a safe harbour for the manager who has not gathered the evidence.

That does not mean the midpoint should go. A middle grade defined as “fully meets the bar for this level” describes most people in a healthy company, and taking it away pushes managers into labels they do not believe. A better fix is to define the middle grade with observable behaviours, require a short rationale for every grade, and watch the distribution in calibration. Use four points only as a temporary reset if your last two cycles put most people in the middle with no evidence to support it.

Why scale length becomes a problem at 100 to 1,500 people

In a small company, founders and a few senior leaders usually know everyone’s work, so the scale is a formality. Past 100 people that stops being true. Ratings start to come from a dozen or more managers who have never compared notes, the same grade means different things in sales and engineering, and the rating begins to drive money through a merit budget. At that point HR teams tend to see three symptoms at once: most grades sit in the middle, the top grade is given out by one or two generous managers, and the calibration meeting turns into a debate about what a “4” means instead of who earned one.

Changing the number of points will not cure any of these on its own. What helps is a written definition for every point at every level, a required rationale tied to goals and evidence, and a calibration session that looks at each manager’s spread as well as each person’s grade. The calibration meeting guide covers the checks for generous graders and thin evidence, and the stack ranking vs calibration page explains why a forced distribution is the wrong answer to compression.

Our perspective

When a People lead asks me whether to move from five points to four, or from four to six, I ask to see last cycle’s distribution by manager before we talk about the scale. More often than not, the problem is visible in one view: a handful of managers who never use the ends, and a handful who use nothing else. A new scale would give both groups a fresh set of numbers to use in exactly the same way.

So my advice is to keep the scale boring and make the definitions sharp. Five points, each described in terms of the work someone at that level actually delivers, with a sentence of evidence required for every grade. Then spend the effort you would have spent on redesign on the calibration conversation, because that is where the scale either starts to mean something or quietly stops.

— Ash Taherian, Product Sales Executive, Revolut People

Where Revolut People fits, and where it doesn’t

Read this section knowing I work at Revolut People. In Revolut People, a review cycle runs on scorecards: managers answer the scorecard questions, and grades are calculated from them rather than typed in, which keeps every manager on the same scale definitions. In calibration, dashboards compare each person against their last four cycles and flag line-manager and functional-manager leniency, as well as grades not supported by deliverables, so calibrators know where to challenge. The CHRO overview summarises this, and the short films on the product tour show a review and a calibration session.

The limits are real. Software will not choose the number of points for you or write the definitions behind them; you still need to agree what each grade means at each level and who signs off. The Performance plan has a public list price from £10 per employee per month, billed annually, with a minimum annual spend of £5,000 in the UK and EEA (see the pricing estimator), so a smaller company with one review a year is often better served by a well-anchored review template and a spreadsheet. If your current tool lets you set the scale and report on the spread by manager, you may only need to fix the definitions.

If you want a second opinion on your scale and last cycle’s distribution, book a 15-minute review. I will tell you whether the scale or the process behind it is the problem, whichever tool you use.

Step plan: choose and test your scale in one cycle

  1. Pull last cycle’s distribution by manager. Count how many grades each manager gave at each point. If most managers use only two or three points, compression is your problem, not scale length.
  2. Decide what the rating has to do. If it feeds pay, promotion or talent reviews, choose five anchored points. If it is for development only, three may be enough, or no overall rating at all.
  3. Write the definitions at each level. Describe each point in terms of observable work for each seniority level, using the five-level examples as a starting draft.
  4. Test the definitions on real cases. Give three managers the same two anonymised summaries and ask each for a grade and one sentence of evidence. Where they disagree by more than one point, rewrite the definition. The bias checker helps tidy the wording.
  5. Require a rationale for every grade. Add a required line that links the grade to goals and evidence, as set out in what should be included in a performance review form.
  6. Review the spread in calibration. Look at each manager’s distribution before discussing individuals, using the calibration meeting template. After the cycle, keep the scale unless the definitions failed; changing it every year makes trends unreadable.

To check where your wider review process stands before you start, the free performance health check takes a few minutes.

Bottom line

Use five points with written definitions at each level unless you have a specific reason not to. Research suggests very short scales lose information and that more points do not make ratings more accurate, while ratings bunching in the middle have been linked to weaker performance improvement. Fix that with sharper definitions, a required rationale and calibration that looks at each manager’s spread, not by changing the number of points.

FAQ

What is a 5-point performance rating scale?

A 5-point performance rating scale gives managers five grades to choose from, typically one below the bar, one for solid work at the bar and two above it, plus a lowest grade for serious underperformance. Each point should have a written definition and example behaviours for each seniority level so that a 4 from one manager means the same as a 4 from another.

What does exceeds expectations mean in a performance review?

Exceeds expectations usually means the person delivered more than the role and level require during the period, for example more complex work, faster or with less support. It should be backed by specific evidence against agreed goals. On the five-level scale in this playbook it maps to Advanced, the fourth of five levels.

Should a performance rating scale have a middle option?

Usually yes. A middle grade defined as fully meeting the bar describes most people in a healthy company. Survey research shows that a midpoint attracts more cautious answers, so define it with observable behaviours, require evidence for every grade and check each manager’s spread in calibration. Drop the midpoint only as a temporary reset if most grades sit there without evidence.

How do you stop managers rating everyone in the middle?

Make the evidence easy to gather and hard to skip. Keep goals and notes in one place through the period, require a one-sentence rationale for every grade, show each manager their distribution against peers before calibration, and ask calibrators to challenge both bunched and generous spreads. Avoid forced quotas, which swap one distortion for another.

Sources

About the author: Ash Taherian is a Product Sales Executive at Revolut People, based in Dublin, working with People leaders on rating scales, calibration and review cycles for companies of 100 to 1,500 people.

Disclosure: I work for Revolut People, which sells performance management software. The research cited here is independent of Revolut People, and the advice on scale length applies whichever tool you use. Several studies cited used survey respondents or laboratory tasks rather than workplace reviews, as noted in the text. Prices are indicative public list prices; your actual price is confirmed in a written quote.