Rating scale: types, examples and how to design one

What is a rating scale? Types and examples, how many points to use, midpoint and "don't know", unipolar vs. bipolar scales, labels, and tips for analysis.

A rating scale is a response format that lets respondents judge an object, a statement or a characteristic on a fixed set of ordered points, for example from "Extremely dissatisfied" to "Extremely satisfied" or from 0 to 10 (Jacob et al. 2019, p. 284). Rating scales can use words, numbers, sliders or symbols such as stars, and the Likert scale is only their best-known form. For answers you can analyze well, use five to seven evenly graded points in most surveys. On bipolar scales, use the same number of points on each side. Include a midpoint when a neutral position is possible, and place "don't know" next to the scale, not in the middle of it.

This article answers five questions in turn:

1. What is a rating scale, and how does it differ from a Likert scale and a ranking?
2. What types of rating scales are there, with examples?
3. How many points, with or without a midpoint, and where does "don't know" go?
4. Unipolar or bipolar, how do you label the scale, and in which direction does it run?
5. How do you analyze a rating scale: mean, median or top box?

In the 2006 to 2014 waves of the ALLBUS, the German General Social Survey, 48 to 71% of all questions came with a scale or a list of response categories (Schnell 2019, p. 75, n. 22). So it's no surprise that many of the mistakes I see in questionnaires aren't in the question text at all. They're in the scale underneath: seven long labels side by side on a smartphone, "don't know" as the middle point, numbers from -3 to +3 for a question about importance. Design flaws like these can influence the answers, and they easily go unnoticed in the analysis.

What is a rating scale?

A rating scale is a response format with ordered points on which respondents themselves indicate where they stand on a given dimension. In survey practice, you'll also see it called a response scale or an answer scale. Möhring and Schlütz describe rating scales as questions that measure degrees of attitudes, opinions or behavior (Möhring and Schlütz 2019, p. 86). As a rule, the question is about the degree of importance or agreement (Jacob et al. 2019, p. 284), and just as often about satisfaction, likelihood or frequency.

Respondents first form their judgment and then map it onto one of the points (Schnell 2019, p. 75). For this to work the same way for everyone, a good rating scale covers the whole range of possible answers, neighboring points barely overlap, and all respondents understand each point in the same way (Krosnick et al. 2005, p. 36, cited in Schnell 2019, p. 76).

Rating scale, Likert scale, ranking: the difference

In everyday use the terms are often mixed up, but the textbooks keep them apart (Malhotra and Birks 2007, pp. 341-342; Marx 2026, slide 24):

Ranking and constant sum are comparative techniques: their results only hold relative to the other objects on the list (Malhotra and Birks 2007, p. 341). On a rating scale, a respondent rates each object on its own (Malhotra and Birks 2007, p. 342). Jacob et al. explain the difference with three ice cream flavors. If you ask for a grade, vanilla, chocolate and strawberry can all get a "good". In a ranking, one flavor has to take third place, and whether third place means "inedible" or "almost as good as the others" stays open. If several objects are allowed to score equally well, which according to Jacob et al. is probably the normal case, or if you want to know the absolute level, use a rating scale (Jacob et al. 2019, pp. 184-185).

For the history of the Likert scale, see Likert scales: everything you ever wanted to know. Combining several items into one score is covered in Aggregating multi-item (Likert) scales.

Types of rating scales and examples

The textbooks distinguish rating scales by how the points are presented: verbally, numerically or graphically, often in combination (Jacob et al. 2019, p. 185). Malhotra and Birks also separate continuous scales (a position on a line) from itemized scales with fixed points, among them the Likert scale, the semantic differential and the Stapel scale (Malhotra and Birks 2007, pp. 342 and 347; Marx 2026, slide 56).

One rule applies to every type: the scale has to fit the question. Nobody can sensibly answer "How satisfied are you with ...?" with "Agree".

Verbal and numeric scales

The verbal scale is the standard case. You'll find ready-made wordings for 14 areas, from agreement to frequency, in A list of common Likert scales. On a numeric scale, only the endpoints have verbal labels, with numbers in between (Jacob et al. 2019, p. 192). Möhring and Schlütz tend to advise against school grades as a scale, because in the German grading system, which runs from 1 (best) to 6 (worst), the distance between 1 and 2 is usually felt to be larger than the one between 5 and 6 (Möhring and Schlütz 2019, p. 88).

Sliders and continuous scales

On a continuous rating scale, respondents mark a position on a line between two poles (Malhotra and Birks 2007, p. 348; Marx 2026, slide 47).

The precision is often only apparent. In general population surveys with scales from 0 to 100, more than half of the answers are divisible by 5, and most of those by 10 (Schnell 2019, p. 81, n. 35). According to Jacob et al., even more than seven points overwhelm most people, let alone 101 scale points, and the metric level is purely an artifact of the instrument (Jacob et al. 2019, p. 190). So use a slider when respondents should express a feeling without fixed steps, not in the hope of getting finer data.

Stars, smileys and other symbols

Symbol scales replace the words with pictures. The most common form is the smiley scale (Schnell 2019, p. 84), known in research as the faces scale or Kunin scale, usually with five or seven faces (Jacob et al. 2019, p. 189). These scales are considered especially intuitive, above all for children (Malhotra and Birks 2007, p. 354; Möhring and Schlütz 2019, p. 89). Studies comparing them with verbal scales are rare, and older respondents tend to respond less well to the format (Schnell 2019, p. 84).

Stars are handy for a quick overall rating, but nobody defines what three stars mean. So label at least the ends, for example "1 star = very poor, 5 stars = very good". Don't mix many different graphic scales in one questionnaire, because variety quickly turns into overload (Jacob et al. 2019, p. 192).

Semantic differential and Stapel scale

The semantic differential, also known as a polarity profile, goes back to Osgood (1952) (cited in Jacob et al. 2019, p. 193). It consists of several scales, usually with seven points, that have opposite adjectives at the ends, such as "modern" and "old-fashioned". To draw the profile, you first code all pairs in the same direction and then connect the means or medians with a line (Malhotra and Birks 2007, p. 350; Marx 2026, slide 52). The variability in ratings for each pair shows where respondents agree with each other and where they don't. Two rules matter: the pairs have to be true opposites, and the positive words appear sometimes on the left and sometimes on the right, so that nobody ticks the same side every time without thinking (Jacob et al. 2019, p. 194).

The Stapel scale is an alternative when you can't find a meaningful pair of opposites. A single term sits in the middle, and respondents use numbers from -5 to +5 without a zero to indicate how well it applies (Malhotra and Birks 2007, pp. 350-351; Marx 2026, slide 55).

A special case: the NPS scale from 0 to 10

The best-known 11-point rating scale belongs to the Net Promoter Score: respondents say on a scale from 0 to 10 how likely they are to recommend a company (Reichheld 2003, cited in Marx 2026, slide 77). For the score, the answers are assigned to three groups: promoters, passives and detractors. How that works is explained in the article Net Promoter Score (NPS).

How many points does a rating scale need?

The following sections go through the parts of a scale one by one:

There's no single correct number of points. Traditionally, five to nine are considered appropriate (Malhotra and Birks 2007, p. 351; Marx 2026, slide 58). For general population surveys, Jacob et al. regard five points as the best solution in many cases (Jacob et al. 2019, p. 193).

My recommendation for practice:

• Five points as the default. They're enough for comparing groups and for sum scores built from several items (Malhotra and Birks 2007, p. 352).
• Seven points for a knowledgeable target group, or if you want to analyze relationships between variables, because with fewer categories the correlation tends to come out smaller (Malhotra and Birks 2007, p. 352).
• Eleven points (0 to 10) only if you want to stay comparable with an established metric such as the NPS.

How respondent involvement, what is being rated, the survey mode and the planned analysis affect this choice is explained in Likert scales: everything you ever wanted to know, and there's a short decision table "5 or 7 points?" in A list of common Likert scales. For an overview of the five decisions, see Likert scales: five aspects to consider.

Rating scale with or without a midpoint?

A scale with an odd number of points has a midpoint, and one with an even number doesn't. The textbook rule: if a neutral answer is possible for at least some of the respondents, use an odd number. An even number fits when you deliberately want to force a decision or when there's no neutral position (Malhotra and Birks 2007, p. 352; Marx 2026, slide 60). The trade-offs are discussed in detail in Likert scales: everything you ever wanted to know.

One reason to include a midpoint is what happens when it's missing: on paper, some respondents add their own, and online, some abandon the survey. That's why Schnell considers it advisable (Schnell 2019, p. 83). In a meta-analysis, Scherpenzeel and Saris (1997) found no effect of the midpoint on reliability, but better validity with a midpoint (cited in Möhring and Schlütz 2019, p. 91).

One drawback is ambiguity: the midpoint can represent a neutral attitude, "partly agree, partly disagree", uncertainty, lack of interest or silent protest (Möhring and Schlütz 2019, p. 91). It can also hide a socially acceptable "don't know" (Sturgis et al. 2012, cited in Schnell 2019, p. 83). So decide on the number of points, the midpoint and "don't know" together.

"Don't know" and "not applicable": next to the scale, not in the middle

The midpoint is a position on the scale. "Don't know" or "can't judge" isn't a position. It says that the respondent has no basis for a judgment. Schnell advises providing a separate answer, especially on rating scales, for those who can't rate an object because they have no experience with it (Schnell 2019, pp. 81-82). This answer goes at the end of the scale and is visibly set apart from it by a gap or a line (Schnell 2019, p. 112).

The HR department handles requests promptly.
Strongly disagree · Disagree · Neutral · Agree · Strongly agree
(set apart, only for questions that not everyone can judge) Can't judge

Offer this answer only where someone may really lack the knowledge, experience or observation, not for questions about their own satisfaction. The rule and the reasoning behind it are in the article Levels of measurement.

"Not applicable" means something else again: the question doesn't concern me. Online, it's better to replace this answer with a filter question that shows the rating only to those who can give it. What happens when an opt-out answer is included in the analysis as a number is worked through in the article Levels of measurement: depending on the coding, the same data set gives a mean between 3.5 and 4.1.

Unipolar or bipolar? And why bipolar scales have to be symmetric

A unipolar scale measures how strongly a characteristic is present, from "not at all" to "very". A bipolar scale stretches between two opposite poles, for example from disagreement to agreement or from dissatisfied to satisfied (Möhring and Schlütz 2019, p. 89).

Unipolar scales are more common and easier to construct and analyze, but there's no clear-cut recommendation (Möhring and Schlütz 2019, pp. 89 and 92). Importance has no opposite, only "less". A scale from "very unimportant" to "very important" with -2 to +2 implies an opposite pole that doesn't exist.

Minus-to-plus scales suggest a bipolar construct, and the negative range may then hardly be used. That's why Jacob et al. recommend positive numbers (Jacob et al. 2019, p. 192).

A bipolar rating scale should be balanced, that is, have as many positive as negative points (Malhotra and Birks 2007, p. 352; Marx 2026, slide 59). Möhring and Schlütz use an eBay survey from 2009 to show how it goes wrong. Its scale, translated from German, read: "very poor, poor, adequate, good, very good, exceptionally good". Three points lie above "adequate" and only two below, and that distorts the results (Möhring and Schlütz 2019, p. 88). Malhotra and Birks allow an exception: "If the distribution of responses is likely to be skewed, however, either positively or negatively, an unbalanced scale with more categories in the direction of skewness may be appropriate. If an unbalanced scale is used, the nature and degree of imbalance in the scale should be taken into account in data analysis" (Malhotra and Birks 2007, p. 352). I advise against it: a scale with more positive than negative points pulls responses toward the positive side (Marx 2026, slide 59), and you can no longer distinguish respondents' opinions from the effects of the scale itself. The agreement, satisfaction and importance scales under "Common Scale Labels" in QUESTIONSTAR are evenly graded, and the bipolar ones among them are symmetric.

A unipolar scale doesn't have two sides. What counts here is that the steps look equal in size. The words in the example above are a common English wording for an intensity scale of the kind Jacob et al. show in German (Jacob et al. 2019, p. 189).

Label every point or only the ends?

According to Malhotra and Birks, labeling every point doesn't necessarily improve accuracy or reliability, but it can reduce ambiguity (Malhotra and Birks 2007, p. 353). The course sees no clear evidence that labeling all points is better than labeling selected ones (Marx 2026, slide 62). Some textbooks still recommend words on every point, Schnell for example (Schnell 2019, p. 81).

Four variants are common (Marx 2026, slide 63):

The only thing that's mandatory is that both poles are labeled unambiguously. With an odd number of points, the midpoint is often worth labeling too, but it isn't always necessary, because the midpoint is often self-evident. Whether you also label the points in between isn't decisive according to the research: a clear advantage of fully labeled scales hasn't been demonstrated (Marx 2026, slide 62). In practice, that means: with five points, take a ready-made, fully labeled scale from A list of common Likert scales. For scales with seven or more points, including scales from 0 to 10, label the poles, add a label for the midpoint where it helps, and number the points. Place the labels as close as possible to the response fields (Malhotra and Birks 2007, p. 353). There's more on this in Likert scales: everything you ever wanted to know.

Which direction? The order of the points

Whether a scale starts with the negative or the positive end isn't a matter of taste. The direction can influence the answers (Lamp 2004, cited in Möhring and Schlütz 2019, p. 90). In a telephone survey, the means were shifted toward the end of the scale that was mentioned first (Yan and Keusch 2015, cited in Möhring and Schlütz 2019, pp. 91-92). In Western countries, placing small values on the left or at the top and large values on the right or at the bottom matches the reading direction (Tourangeau et al. 2004, cited in Möhring and Schlütz 2019, p. 90).

Three rules follow from this:

1. A high value for a high level. Code "Strongly disagree" as 1 and "Strongly agree" as 5 (Möhring and Schlütz 2019, p. 92). When numbers and words appear together, they point in the same direction: "very often (5) to never (1)", not "very often (1)" (Schnell 2019, p. 81).
2. One direction throughout the questionnaire. Someone who reads "Extremely satisfied" on the left in question 3 and "Extremely dissatisfied" on the left in question 4 easily ticks the wrong answer. The exception: in a semantic differential, you alternate the side of the positive word on purpose (Jacob et al. 2019, p. 194).
3. Never randomize the points. Random order is meant for unordered answer lists. A scale loses its order that way. More on this in Order effects in surveys.

Rating scales on a smartphone

Overall, data quality hardly differs between smartphone, tablet and PC, but some question formats, sliders among them, are harder to use on a smartphone (Antoun et al. 2017; Tourangeau et al. 2017, both cited in Möhring and Schlütz 2019, pp. 147-148).

What that means for the scale:

• The whole scale has to be visible without scrolling. When in doubt, arrange the points one below the other, which is what Schnell recommends almost always. Only a few short points fit side by side. All points go consistently in one column or in one row, with equal spacing (Schnell 2019, p. 112).
• Use matrix questions sparingly. A matrix, that is, several statements with the same scale in one table, saves space but tempts respondents into mechanical patterns ("two left, two right"). That's why Schnell generally advises against it (Schnell 2019, p. 114). Keep matrices short, because long batteries are tiring (Möhring and Schlütz 2019, p. 87).
• Use a slider only with good reason. If a scale with fixed points does the same job, it's the safer choice on a smartphone. In QUESTIONSTAR a slider only counts as answered once it has been moved. In other tools, check in a test where the handle sits when the question opens and whether a slider that hasn't been moved is saved as an answer.
• Test on several devices. What looks flawless on the team's screens can look quite different on other devices (Möhring and Schlütz 2019, p. 147).

All the decisions from this article at a glance:

Analyzing a rating scale: distribution, median, mean and top box

Strictly speaking, a rating scale produces ordinal data (Jacob et al. 2019, p. 35). Whether and how you can still calculate means is covered in the article Levels of measurement.

One metric that often works better than the mean in company reports is the top box: the share of respondents who chose the highest point. The top-2 box combines the highest and the second-highest point, and the bottom-2 box mirrors it with the two lowest. Linke uses these terms for employee surveys and recommends looking at the distribution of the individual values alongside the means (Linke 2018, p. 105).

Worked example: top box or top-2 box? Two training courses, two winners

The example of the two bakery stores in A list of common Likert scales shows why the mean alone isn't enough. Here we look at the next question: which "box" do you report?

An invented example. A software vendor asks 50 participants after each of two training courses: "How satisfied are you with the training?" The scale runs from 1 (Extremely dissatisfied) to 5 (Extremely satisfied).

By the top box, training A is ahead. By the top-2 box, it's training B. With 50 answers per group and independent groups, however, only the difference in the top box is significant at the 5% level in a two-sided test for two proportions (|z| = 3.46, critical value 1.96). The top-2 box differs by four people, or 8 percentage points. This difference is not significant at the 5% level and is compatible with random sampling variation (|z| = 0.94).

The top box shows how many are extremely satisfied. The top-2 box shows how many are satisfied overall, and it hides the difference between "Somewhat satisfied" and "Extremely satisfied". So define the metric before the analysis, always state the number of answers in the report, for example "top-2 box 72% (n = 50)", and show the distribution as well.

Check three things before every calculation:

• Remove opt-out answers. "Can't judge" belongs neither in the mean nor in the denominator of the top box. See Levels of measurement.
• Reverse-code negatively worded statements. In scales made up of several statements, negatively worded items have to be recoded before you combine them, so that a high value always means the same thing (Malhotra and Birks 2007, p. 349; Marx 2026, slide 49). More on this in Reliability and validity.
• Don't compare scales of different lengths directly. Comparing the same question asked with different numbers of points requires strong assumptions (Schnell 2019, p. 80, n. 31).

Rating scales in QUESTIONSTAR

QUESTIONSTAR has a matching question type for the common kinds of rating scales (Help: Available question types):

• Verbal and numeric scales: the "Single Choice" question type, vertical or horizontal, and in the horizontal layout with labels for the poles if you like.
• Several statements with the same scale: the "Matrix" question type, as a Simple Matrix, as a Side by Side Matrix (importance and satisfaction next to each other, for example) or as a Semantic Differential. On a smartphone, it shows one statement at a time by default and collapses it once it's answered (Help: Matrix question type).
• Sliders and graphic scales: Sliders (from 0 to 100, for example), Gradient Sliders, Bars, stars and faces.
• Ready-made scales: under "Common Scale Labels" in the question editor, you can apply answer scales with one click.
• Codes: the points are coded automatically as 1, 2, 3 ..., and you can set your own values.
• Opt-out answer: set up as an opt-out answer, "can't judge" doesn't count in the mean and is analyzed separately.
• NPS: in the NPS question, you set the start and the end of the scale freely and define which points count as detractors, passives and promoters. QUESTIONSTAR calculates the score automatically.
• Analysis: for every question you see the count, percent and cumulative percent for each point, plus the median and the standard deviation. The top-2 box is the sum of the percentages for the two highest scale points. You also get Pearson and Spearman correlations, t-tests and crosstabs with chi-square. For further calculations, export the raw data as CSV, XLSX or SPSS.

Checklist: check your rating scale before launch

• Type, points, midpoint: a verbal scale with five points as the default, seven for a knowledgeable target group, a midpoint if a neutral position is possible.
• Bipolar scale symmetric: the same number of equally strong points on both sides. Unipolar scale: even steps, no negative numbers.
• Both poles labeled unambiguously, the midpoint wherever it isn't self-evident.
• One direction throughout the questionnaire, a high code for a high level.
• "Can't judge" only for questions that not everyone can judge, set apart from the scale. "Not applicable" replaced by a filter question.
• Tested on a smartphone: whole scale visible without scrolling, matrices short, sliders checked.
• Metric for the report defined in advance (mean, median, top box or top-2 box), with the distribution and n shown alongside.

Frequently asked questions about rating scales

What is a rating scale?

A rating scale is a response format with ordered points on which respondents judge an object, a statement or a characteristic, such as satisfaction, agreement or importance (Jacob et al. 2019, p. 284). The points can be shown as words, numbers or symbols, or as a slider. Response scale and answer scale are often used for the same thing.

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

Rating scale is the umbrella term. A Likert item captures agreement with a statement. A Likert scale in the original sense combines the answers to several such statements into one score per person (Malhotra and Birks 2007, p. 349). Today, single 5-point verbal scales are often called Likert scales as well (Jacob et al. 2019, p. 283).

How many points should a rating scale have?

Five to seven points are usual, and five to nine are traditionally considered appropriate (Malhotra and Birks 2007, p. 351). Five points are enough for most surveys, especially for comparing groups (Jacob et al. 2019, p. 193). Seven are worth it for a knowledgeable target group or if you want to analyze relationships between variables.

What is the difference between a unipolar and a bipolar scale?

A unipolar scale measures the strength of a characteristic from "not at all" to "very", for example importance or frequency. A bipolar scale runs from one pole to its opposite, for example from "Extremely dissatisfied" to "Extremely satisfied" (Möhring and Schlütz 2019, p. 89).

What is a semantic differential?

A semantic differential is a technique that uses several rating scales, usually with seven points, between opposite terms such as "modern" and "old-fashioned". The ratings form a profile, for example the image of a brand (Jacob et al. 2019, pp. 193-194).

What is the difference between rating and ranking?

In a rating, a respondent rates each object on its own, and several objects can score equally well. In a ranking, a respondent puts the objects in order. A ranking doesn't show whether last place is bad or only slightly less good. If you want to know the absolute level, use a rating (Jacob et al. 2019, pp. 184-185).

What does top-2 box mean?

The top-2 box is the share of respondents who chose one of the two highest points, so 4 or 5 on a 5-point scale. The top box is the share for the highest point alone (Linke 2018, p. 105). Calculate both from the valid answers, without "can't judge".

Sources

Antoun, C., Couper, M. P. and Conrad, F. G. (2017). Effects of Mobile versus PC Web on Survey Response Quality. Public Opinion Quarterly, 81(S1), 280-306. Cited in Möhring and Schlütz (2019), p. 148.

Jacob, R., Heinz, A. and Décieux, J. P. (2019). Umfrage. Einführung in die Methoden der Umfrageforschung [Surveys: an introduction to the methods of survey research] (4th ed.). Berlin/Boston: De Gruyter Oldenbourg. In German.

Krosnick, J. A., Judd, C. M. and Wittenbrink, B. (2005). The Measurement of Attitudes. In D. Albarracín, B. T. Johnson and M. P. Zanna (eds.), The Handbook of Attitudes (pp. 21-76). Mahwah: Lawrence Erlbaum. Cited in Schnell (2019), p. 76.

Lamp, E. (2004). Richtungseffekte bei Bildblattvorlagen im demoskopischen Face-to-face-Interview [Direction effects of picture show cards in face-to-face opinion research interviews]. ZA-Information, 54, 113-146. In German. Cited in Möhring and Schlütz (2019), p. 90.

Linke, R. (2018). Mitarbeiterbefragungen optimieren. Von der Befragung zum wirksamen Management-Instrument [Optimizing employee surveys: from survey to effective management tool]. Wiesbaden: Springer Gabler. In German. DOI 10.1007/978-3-658-17722-5.

Malhotra, N. K. and Birks, D. F. (2007). Marketing Research. An Applied Approach (3rd European edition). Harlow: Pearson Education.

Marx, P. (2026). Principles of Survey Research. Online course, chapter 2, slides 24, 47, 49, 52, 55, 56, 57, 58, 59, 60, 61, 62, 63 and 77. QUESTIONSTAR. Course, chapter 2.

Möhring, W. and Schlütz, D. (2019). Die Befragung in der Medien- und Kommunikationswissenschaft. Eine praxisorientierte Einführung [Surveys in media and communication studies: a practical introduction] (3rd ed.). Wiesbaden: Springer VS. In German. DOI 10.1007/978-3-658-25865-8.

Reichheld, F. F. (2003). The One Number You Need to Grow. Harvard Business Review, 81(12). Cited in Marx (2026), slide 77.

Scherpenzeel, A. C. and Saris, W. E. (1997). The validity and reliability of survey questions: A meta-analysis of MTMM-studies. Sociological Methods & Research, 25, 341-383. Cited in Möhring and Schlütz (2019), p. 91.

Schnell, R. (2019). Survey-Interviews. Methoden standardisierter Befragungen [Survey interviews: methods of standardized surveys] (2nd ed.). Wiesbaden: Springer VS. In German. DOI 10.1007/978-3-531-19901-6.

Sturgis, P., Roberts, C. and Smith, P. (2012). Middle Alternatives Revisited: How the neither/nor Response Acts as a Way of Saying "I Don't Know"? Sociological Methods & Research, 43(1), 15-38. Cited in Schnell (2019), p. 83.

Tourangeau, R., Couper, M. P. and Conrad, F. (2004). Spacing, position, and order: Interpretive heuristics for visual features of survey questions. Public Opinion Quarterly, 68, 368-393. Cited in Möhring and Schlütz (2019), p. 90.

Tourangeau, R., Maitland, A., Riviero, G., Sun, H., Williams, D. and Yan, T. (2017). Web Surveys by Smartphone and Tablets. Effects on Survey Responses. Public Opinion Quarterly, 81(4), 896-929. Cited in Möhring and Schlütz (2019), p. 148.

Yan, T. and Keusch, F. (2015). The effects of the direction of rating scales on survey responses in a telephone survey. Public Opinion Quarterly, 79(1), 145-165. Cited in Möhring and Schlütz (2019), p. 92.

I hope your respondents all understand your scales in the same way, and that your results accurately reflect what those scales can measure.

Related articles:

• Levels of measurement: nominal, ordinal, interval and ratio
• A list of common Likert scales
• Likert scales: everything you ever wanted to know, but never dared to ask

Dr. Paul Marx

Author: Dr. Paul Marx
Dr. Paul Marx is the founder and Managing Director of QUESTIONSTAR and was a professor of marketing at the University of Siegen.