Back

Rating Scale Questions: Types, Examples & Analysis

Matthieu Saussaye

A rating scale turns an opinion into a number. Get the scale wrong and the number is wrong too, however many people answer.

Rating scale questions are the workhorse of every survey: satisfaction, agreement, likelihood, importance, quality. They are easy to write badly. The number of points, the labels, the direction and the way you summarise the answers all change the result. This guide covers each choice with examples you can copy and a worked calculation. If you want to skip the drafting, SmartInterview writes the questionnaire from your objective, scores it, and returns analysed results, usually in about 48 hours.

The Short Answer

  • A rating scale question asks respondents to place an answer on an ordered set of points, for example 1 to 5 from "Very dissatisfied" to "Very satisfied".

  • The main types are numeric (0-10, 1-5), Likert (agree / disagree), semantic differential (two opposite adjectives), star ratings and sliders.

  • Use 5 points for unipolar ideas (how satisfied, how important) and 7 points for bipolar ideas (agree to disagree). Use 0-10 when you need a benchmark such as NPS.

  • Label every point with words when you can. Fully labelled scales are more reliable than scales with only the end points named.

  • Report top-2-box and the full distribution, not only the mean. Two groups can share a mean of 3.7 and still be very different.

Rating Scale Types Compared

Type

What it looks like

Best for

Watch out for

Numeric

"Rate from 0 to 10" with only the ends labelled

Benchmarks (NPS), quick overall scores, mobile

Middle numbers mean different things to different people

Likert

Strongly disagree ... Strongly agree, usually 5 or 7 points

Attitudes, statement batteries, brand image

Acquiescence: people tend to agree with any statement

Semantic differential

Cheap 1 2 3 4 5 6 7 Premium

Brand and concept perception, positioning maps

Both adjectives must be true opposites

Star rating

1 to 5 stars

Product and service reviews, app feedback

Strong skew to 4 and 5 stars; hard to compare with research norms

Slider

Drag a handle from 0 to 100

Fine-grained estimates, price perception

Start position anchors answers; more effort on phones

Verbal frequency / intensity

Never, Rarely, Sometimes, Often, Always

Behaviour, usage, frequency

Words must be ordered and spaced evenly

All of these are closed-ended questions with ordered options. The Likert scale is the best-known family; a grid of several rated items on the same scale is a matrix question.

How Rating Scales Work, and How to Read Them

Unipolar vs bipolar

A unipolar scale measures the amount of one thing, from none to a lot: "Not at all important" to "Extremely important". A bipolar scale runs between two opposites with a neutral middle: "Very dissatisfied" to "Very satisfied", or "Strongly disagree" to "Strongly agree". Mixing the two (for example "Not satisfied" to "Very satisfied") is a common error: "not satisfied" can mean neutral or unhappy, and respondents will split on it.

How many points?

Research on scale length is consistent. Jon Krosnick and Leandre Fabrigar recommend 5 points for unipolar scales and 7 for bipolar ones. Preston and Colman (2000, Acta Psychologica) compared 2 to 11 points and a 101-point scale: reliability and validity were lowest for 2 to 4 points and rose up to about 7 to 10, and respondents rated the 10-point scale easiest to use for expressing their views. Beyond 11 points you gain nothing but noise. Our rule: 5 for amount, 7 for agreement, 0-10 when an industry benchmark requires it.

Odd or even?

An odd number gives a true midpoint. That is honest when people really are neutral, but it also gives a home to people who do not want to think. An even number forces a lean. Prefer odd scales and add a separate "Don't know / not applicable" option outside the scale, so that people with no opinion do not hide in the middle.

Worked example: mean, top-2-box and the trap between them

Two customer groups of 200 answer "Overall, how satisfied are you with our service?" on a 5-point scale (1 = Very dissatisfied, 5 = Very satisfied).

Answer

Group A

Group B

1 Very dissatisfied

10

45

2 Dissatisfied

20

10

3 Neither

40

15

4 Satisfied

80

20

5 Very satisfied

50

110

Mean

3.70

3.70

Top-2-box (4+5)

65%

65%

Bottom-2-box (1+2)

15%

27.5%

Group A mean: (10×1 + 20×2 + 40×3 + 80×4 + 50×5) / 200 = 740 / 200 = 3.70. Group B gives the same 740 / 200 = 3.70. Top-2-box is (80 + 50) / 200 = 65% for A and (20 + 110) / 200 = 65% for B. Both headline numbers are identical, yet Group B has almost twice as many unhappy customers. B is polarised; A is broadly content. This is why you always look at the bottom box and the full distribution.

Precision matters too. For Group A, the sample standard deviation is 1.10, so the 95% confidence interval for the mean is 3.70 ± 1.96 × 1.10 / √200 = 3.70 ± 0.15. For top-2-box, the margin is 1.96 × √(0.65 × 0.35 / 200) ≈ ±6.6 points. A move from 65% to 68% next quarter is inside the noise.

Is a rating scale ordinal or interval?

Strictly, rating scale answers are ordinal: the gap between "Satisfied" and "Very satisfied" is not guaranteed to equal the gap between "Neither" and "Satisfied". In practice, means are widely used on 5-point and longer scales, and they work well for comparing groups. Use medians and box scores when you need to be strict, and see correlation in survey data for relating two scales. To put scales of different lengths on one footing, rescale to 0-100 with (score - min) / (max - min) × 100: a 3.70 on a 1-5 scale becomes 67.5.

Ready-to-Use Rating Scale Questions

Copy these, change the object, keep the labels.

1. Overall satisfaction (5-point, bipolar)

"Overall, how satisfied or dissatisfied are you with [product]?" Very dissatisfied / Dissatisfied / Neither satisfied nor dissatisfied / Satisfied / Very satisfied. Report top-2-box. More wording options in our customer satisfaction survey questions.

2. Importance (5-point, unipolar)

"How important is each of the following when you choose a [category]?" Not at all important / Slightly important / Moderately important / Very important / Extremely important. Keep the list to 8 items or fewer, or switch to a rank order question when you need a clear winner.

3. Agreement (7-point Likert)

"How much do you agree or disagree: This product is worth its price." Strongly disagree / Disagree / Somewhat disagree / Neither agree nor disagree / Somewhat agree / Agree / Strongly agree. Write some statements in the negative direction to catch straight-lining.

4. Likelihood to recommend (0-10)

"How likely are you to recommend [brand] to a friend or colleague?" 0 = Not at all likely, 10 = Extremely likely. NPS = % promoters (9-10) minus % detractors (0-6). See the Net Promoter Score guide or the NPS calculator.

5. Concept perception (semantic differential, 7-point)

"Where would you place this concept between each pair?" Cheap ... Premium; Ordinary ... Innovative; Hard to understand ... Easy to understand; Not for me ... Made for me. Show the pairs in a fixed order across concepts so they are comparable, as in monadic and sequential concept tests.

6. Purchase intent (5-point)

"If this product were available at CHF 29, how likely would you be to buy it?" Definitely would not buy / Probably would not buy / Might or might not buy / Probably would buy / Definitely would buy. Top-box ("Definitely") is usually a better predictor than top-2-box, because stated intent overstates real buying.

7. Ease of use (5-point)

"How easy or difficult was it to [complete the task]?" Very difficult / Difficult / Neither easy nor difficult / Easy / Very easy. Pair it with an open "why?" for anyone who answers 1 or 2.

8. Doing it in SmartInterview

In SmartInterview, you type the objective ("find out why trial users do not convert, and which feature matters most") and the AI drafts the full questionnaire: rating scales with labelled points, screen-outs, quotas and the open questions that explain the scores, with a quality score before launch. You send it to your own customer list through a reusable link, to an audience recruited in Switzerland and Europe, or both, with every response tagged by source. When someone gives a low rating and writes a vague reason, the AI asks one follow-up; in our studies that brings up to 4x longer answers and 2.2x more topics. The open answers are coded into themes linked to verbatims, and you cross-tab every scale by segment or source, then export to Excel or SPSS. Pharmalp, a Swiss natural health brand, used this approach to validate the pricing and positioning of four new supplement concepts (see the Pharmalp case study). Plans start from CHF 45 per month for 150 respondents; details on the pricing page.

Where I draw the line

A rating on its own is a thermometer. It tells you the temperature and nothing about the cause. I have seen teams track a satisfaction score for two years, watch it drop by four points, and have no idea why, because the survey never asked. If a score will drive a decision, put an open question right after it, at least for the people at the bottom of the scale.

I also stop people who want a 100-point slider for everything because it "feels more precise". It is not. On a phone it is slower, the starting position pulls answers toward it, and the extra decimals are noise. A well-labelled 5 or 7-point scale and a good follow-up beat a fancy widget every time.

Matthieu Saussaye, Founder of SmartInterview.

Rating Scale Checklist

  1. Decide unipolar or bipolar before writing the labels.

  2. Pick the length: 5 for amount, 7 for agreement, 0-10 only for a benchmark.

  3. Label every point with words that are evenly spaced and clearly ordered.

  4. Keep direction consistent across the survey: negative on the left, positive on the right.

  5. Add "Don't know / Not applicable" as a separate option, not as the midpoint.

  6. Ask about one thing per question. "Fast and friendly" is two ratings.

  7. Follow low scores with an open question so the number has a reason attached.

  8. Plan the analysis first: top-2-box, bottom-2-box, mean with confidence interval, and cross-tabs by segment. Our list of survey analysis tools helps.

Common mistakes

  • Unbalanced options, such as Poor / Fair / Good / Very good / Excellent: three positive points, one neutral-ish, one negative. Results look rosier than reality.

  • Leading wording ("How much did you enjoy...") that assumes the answer. See survey bias.

  • Changing the scale between waves. A 5-point score cannot be compared with last year's 7-point score without rescaling, and even then the label effect remains.

  • Long grids of 15 items on the same scale. Respondents straight-line after the first few rows, especially on mobile.

  • Reporting only the mean, which hides polarisation, as the worked example shows.

  • Using a scale for a yes/no fact. "Did you buy in the last month?" is a dichotomous question, not a rating.

Frequently Asked Questions

What is a rating scale question?

It is a closed question where the respondent chooses a point on an ordered scale, such as 1 to 5 or 0 to 10, to express how much they agree, how satisfied they are or how likely they are to do something. The answers can be counted and compared across groups and over time.

What are the main types of rating scales?

The common types are numeric scales, Likert agreement scales, semantic differential scales with two opposite adjectives, star ratings, sliders and verbal frequency scales. Each suits a different job, as shown in the comparison table above.

Is a 5-point or 7-point scale better?

Use 5 points for unipolar questions such as importance or satisfaction with one feature, and 7 points for bipolar agreement questions. Research by Krosnick and others shows that both are reliable when every point is labelled; going beyond about 10 points adds little.

What is the difference between a Likert scale and a rating scale?

A Likert scale is one type of rating scale: it measures agreement with a statement, from strongly disagree to strongly agree. "Rating scale" is the wider family that also includes satisfaction, likelihood, numeric and star scales.

How do you analyse rating scale data?

Start with the distribution of answers, then report top-2-box (the share choosing the two highest points), bottom-2-box and the mean with a confidence interval. Compare segments with cross-tabs, and read the open answers behind the low scores.

Should a rating scale have a neutral option?

Usually yes. A true midpoint is accurate for people who are really neutral. Add a separate "Don't know" option so that people with no opinion do not choose the midpoint by default.

What is top-2-box score?

It is the percentage of respondents who chose the two most positive points on a scale, for example "Satisfied" and "Very satisfied" on a 5-point scale. With 130 of 200 respondents in those two boxes, the top-2-box score is 65%.

Want rating scales that come with the reasons behind them? SmartInterview drafts the questionnaire, probes vague answers and codes the results, free for your first 50 respondents.

Table of contents

Related articles

Sign up for free

Sign up for free

Sign up for free