A Likert scale measures an attitude by asking how strongly people agree or disagree with several statements, usually on five or seven labelled points. Each answer gets a number, and a person’s score is the total (or average) across the statements.
Rensis Likert introduced the method in 1932. One statement with its answer options is a Likert item; the Likert scale is several items added together. That difference decides how you can describe and analyse your data.
What is a Likert scale?
A Likert scale is a set of statements that together measure one attitude or trait, such as confidence, belonging or wellbeing. For each statement, respondents choose from ordered options such as ‘strongly disagree’ to ‘strongly agree’. Their answers are then added up, which is why Likert scales are called summated scales.
The answer options are ordinal: they have an order, but the gaps between them aren’t known to be equal. Even if you code ‘sometimes’, ‘often’ and ‘always’ as 3, 4 and 5, the step from ‘sometimes’ to ‘often’ needn’t be the same size as the step from ‘often’ to ‘always’. Keep that in mind when you reach the analysis.
Many of Likert’s 1932 statements asked students to ‘Strongly Approve’, ‘Approve’, be ‘Undecided’, ‘Disapprove’ or ‘Strongly Disapprove’. In his simpler scoring method, he gave the five positions values of 1 to 5 and summed them, which is the approach this guide uses below.
| Likert item | Likert scale | |
|---|---|---|
| What it is | One statement and its answer options | Several items measuring the same thing, added together |
| Type of data | Ordinal | A total or mean score across the items |
| Describe it with | Counts and percentages in each option; the median | Mean and standard deviation if scores are roughly normal; otherwise median and interquartile range |
| Tests often used | Chi-square, Mann-Whitney U, Spearman’s rho | t-test, ANOVA, Pearson correlation, regression |
Our summary, using the tests Sullivan and Artino (2013) list. Their own conclusion is broader: parametric tests can be used with Likert responses (see the analysis section below).
Likert scale examples: response options you can use
Pick labels that match your question: a question about how often something happens needs frequency labels, and a question about satisfaction needs satisfaction labels. These five-point sets come from Iowa State University Extension’s Likert scale examples for surveys, a free guide you can copy from:
| Type | 5-point labels (high to low) |
|---|---|
| Agreement | Strongly agree, Agree, Undecided, Disagree, Strongly disagree |
| Frequency | Almost always, Often, Sometimes, Seldom, Never |
| Importance | Very important, Important, Fairly important, Slightly important, Not important |
| Satisfaction | Very satisfied, Satisfied, Neither, Dissatisfied, Very dissatisfied |
| Quality | Very good, Good, Fair, Poor, Very poor |
| Comparison | Much better, Somewhat better, Stayed the same, Somewhat worse, Much worse |
Source: Iowa State University Extension, Likert scale examples for surveys (2010).
For a seven-point satisfaction scale, the same guide gives: very dissatisfied, moderately dissatisfied, slightly dissatisfied, neutral, slightly satisfied, moderately satisfied, very satisfied. It also lists four-point sets, such as strongly agree, agree, disagree, strongly disagree.
Label every point, not just the two ends. TASO’s guidance on designing Likert scales, written for higher education evaluators, recommends this because labels discourage respondents from relying on the middle option, and respondents prefer them.
The same guidance also suggests asking a direct question with its own answer labels rather than a statement to agree with. For example, ask ‘Are you satisfied with yourself?’, with options from ‘not at all satisfied’ to ‘extremely satisfied’. That reduces people’s tendency to agree regardless of content.
5-point or 7-point, and do you need a neutral middle?
There’s no single right number of points, but most researchers recommend five to seven. Fewer points are quicker to answer but give you less information. More points give finer detail but can be harder for respondents to interpret.
One recommendation is five points for the general population, and seven for respondents with strong verbal skills who are used to questionnaires. So for a broad public sample, a 5-point Likert scale is the safer choice; with postgraduates or professionals who are used to surveys, seven is worth considering.
Whether to include a midpoint (‘neither agree nor disagree’) depends on your respondents. The evidence suggests:
- Include a midpoint when respondents know the topic and may genuinely be neutral, or when you plan to treat the scale as interval data in your analysis.
- Leave it out when respondents are unfamiliar or uncomfortable with the topic, haven’t formed an opinion, may answer to look good (social desirability bias) or are likely to put in little effort.
Dropping the midpoint gives you an even-numbered scale, such as four or six points. It forces a lean one way, which is useful when fence-sitting would hide the answer you need. But it can also push people into a view they don’t hold.
How to write good Likert items
A good Likert scale questionnaire starts with a clear definition of what you’re measuring. Read how others have measured it first: adapting a published, validated scale can be more reliable than writing your own, and it makes your results comparable with earlier studies. Check the terms of use early, because some scales need a licence. One example is the Warwick-Edinburgh Mental Wellbeing Scale’s page, which sets out its licence options.
If you write your own items, a few habits make the biggest difference:
- Keep each item to one idea. ‘The module was interesting and well organised’ asks two questions in one, so you can’t tell which one people answered.
- Use plain words. Avoid acronyms and complicated grammar.
- Test the wording. Cognitive interviewing (asking a few people to think aloud as they answer) shows whether they read each item the way you meant.
- Plan the number of items. Some researchers put the sweet spot at about five items per construct (the attitude or trait you’re measuring). Another approach is to pilot 10 to 12 and keep the best six to eight.
- Think twice about negatively worded items. They were long recommended to stop people agreeing with everything (acquiescence bias). More recent studies suggest they may cause more validity and reliability problems than they solve. Attention-check items (‘Select "Strongly disagree" if you’re reading this’) are one alternative.
How to score a Likert scale
To score a Likert scale, give each option a number (1 to 5 on a five-point scale), with the high number always at the same end. Reverse-score any items worded the other way (a published scale you adopt may include some, even if you avoid them in your own). Then add the item scores, or average them, for each respondent.
Reverse scoring isn’t new. In Likert’s own instrument, ‘Strongly Approve’ scored 5 on a statement backing co-operation with the League of Nations. On ‘The United States should have the largest military and naval air fleets in the world’, it scored 1.
Worked example (our own illustration). Say you’ve written four items on confidence with statistics, scored 1 (strongly disagree) to 5 (strongly agree). Item 3, ‘I avoid statistics whenever I can’, runs the opposite way, so you reverse it. On a five-point scale the reversed score is 6 minus the answer; on seven points, it’s 8 minus the answer.
| Item | Answer | Score used |
|---|---|---|
| 1. I can choose a suitable statistical test | Agree (4) | 4 |
| 2. I can interpret a p-value | Strongly agree (5) | 5 |
| 3. I avoid statistics whenever I can (reversed) | Disagree (2) | 6 − 2 = 4 |
| 4. I can explain my results to others | Agree (4) | 4 |
| Total | 17 (possible range 4 to 20) |
Forget to reverse item 3 and this respondent scores 15 instead of 17, which understates their confidence. Across a whole sample, the total would mix answers running in two directions.
Is the simple 1-to-5 sum good enough? Likert tested it against his more laborious ‘sigma’ method of scoring. The scores from the two methods ‘correlated almost perfectly’, and he wrote that the results seemed to justify the simpler method.
Published scales spell out their scoring, and it’s worth copying that level of detail. The Warwick-Edinburgh Mental Wellbeing Scale (WEMWBS) has 14 positively worded items. Respondents rate each one for the past two weeks: none of the time, rarely, some of the time, often or all of the time. Items score 1 to 5 and are added with equal weights, so totals run from 14 to 70, and a higher score means better wellbeing.
How to analyse Likert scale data
How you analyse Likert data depends on whether you’re looking at a single item or a summed scale. For a single item, report how many people chose each option. For a summed scale, parametric tests are generally defensible, and the shape of your scores decides whether a mean or a median describes them better.
Analysing single Likert items
Single items are ordinal, so a mean can mislead. What does an average halfway between ‘never’ and ‘rarely’ mean? If answers bunch at both extremes, the mean can land on the neutral middle even if few people chose it. A mean reported to two decimal places also usually suggests more precision than ordinal answers have.
A frequency table (the count and percentage choosing each option) is usually more informative, and some researchers argue for the median.
Experts who treat Likert answers strictly as ordinal recommend non-parametric tests, and these suit single items well. A chi-square test compares how answers are spread across groups, and the Mann-Whitney U test compares two groups. Spearman’s rho measures the association between two items.
Analysing Likert scales
Whether you can use parametric tests on Likert data is a long-running debate. Some experts argue that, strictly, parametric tests need interval data. Others say they can be used with Likert data if each group has at least five to ten observations and the data are normally (or nearly normally) distributed.
Sullivan and Artino point to a review by Geoff Norman (2010) that used real and simulated data. They sum up its finding: parametric tests give answers acceptably close to ‘the truth’ with Likert data, even when their assumptions are violated.
Sullivan and Artino’s conclusion, in a 2013 editorial for medical education researchers, is that parametric tests can be used to analyse Likert scale responses. But to describe the data, they say means are often of limited value unless scores follow a normal distribution. They add a practical step: decide how you’ll describe and analyse your data when you plan the study, and explain your choice in your methods section.
Our guide to the statistical treatment of data introduces descriptive and inferential statistics. It also explains the Type I and Type II errors to watch for.
Check the scale holds together
Before you add items into a scale, show that they measure the same thing. Cronbach’s alpha checks how well the items correlate with one another, and factor analysis tests whether they share one underlying factor. The WEMWBS developers, for example, looked for an alpha above 0.70.
How two published scales did it
- WEMWBS (UK wellbeing). The authors reported medians and interquartile ranges: 50 (45 to 55) in a sample of 348 students and 51 (45 to 56) in a population sample of 1,749. Cronbach’s alpha was 0.89 and 0.91, and in the student sample, test-retest reliability after one week was 0.83. The population scores were reasonably normal, with a slight skew.
- CLASS (a survey of students’ beliefs about physics). Students answer 42 statements on a five-point agree-disagree scale, and 36 are scored. The authors treat ‘strongly agree’ and ‘agree’ as the same answer, because students interpret the two inconsistently. They report scores as the percentage of favourable answers and say ordinal scoring is preferable for their survey.
Both choices are defensible, and both are stated clearly. That’s the standard to aim for.
How to report Likert results in your dissertation
Your methods section should let a reader rebuild your scale. Your results section should match the description to the type of data. Here’s a template to adapt:
Methods. Attitudes towards [topic] were measured with [name of scale and citation, or ‘a scale developed for this study’], comprising [number] items rated on a [five]-point Likert scale from [lowest label] to [highest label]. Items were scored 1 to [5]; items [numbers] were reverse-scored. Scale scores were calculated as the [sum/mean] of the items (possible range [x] to [y]). Internal consistency was [Cronbach’s alpha = value]. Single items were analysed as ordinal data using [test], and scale scores using [test], because [reason].
Results. For single items, give the number and percentage choosing each option (a table or a stacked bar chart works well). For scale scores, give the mean and standard deviation, or the median and interquartile range if the scores are clearly skewed. Then report the test, its statistic and the p-value.
Put the full questionnaire, with every item and its answer labels, in an appendix. Our guide to appendices in research covers what else belongs there.
If you’re still deciding whether a Likert scale questionnaire is the right tool, our guide to what a research instrument is sets out interviews and observation as alternatives.
Other attitude scales: the Thurstone scale
The main alternative is the Thurstone scale. Before the survey, a panel of judges places each statement on an 11-point scale. A respondent’s score is then the average value of the statements they agree with. A Likert scale needs no judging stage: respondents rate how far they agree with every statement, and the numbers come from their answers. Our guide to the Thurstone scale compares Thurstone, Likert and Guttman scales side by side.
Questions people ask
Is a 1-10 scale a Likert scale? Not in the sense Likert used: his approve-disapprove statements had five labelled positions. There’s no theoretical reason to rule out other numbers of points, but current guidance recommends a label on every point, so describe a 1-10 rating labelled only at the ends as a numeric rating scale in your methods.
Is Likert data qualitative or quantitative? It’s quantitative, but ordinal. The answers are ordered categories that you code as numbers, so you can count and test them, but the gaps between options aren’t known to be equal.
Is there a 4-point Likert scale? Yes. A four-point scale (strongly agree, agree, disagree, strongly disagree) drops the neutral midpoint, so every respondent has to lean one way.
Using a Likert scale in your own research
Before you write a single item, check whether a validated Likert scale already measures your construct, and agree it with your supervisor. Pilot your items, decide your scoring and reversed items in advance, and write your analysis plan before you collect any data. If you’re deciding where your study sits more broadly, our overview of the types of research is a good place to start.
And if your project is turning into a bigger research question, you can find a PhD where survey research could take you next.
Sources
- Sullivan GM, Artino AR Jr (2013): Analyzing and interpreting data from Likert-type scales, Journal of Graduate Medical Education
- Likert R (1932): A technique for the measurement of attitudes, Archives of Psychology 140
- Tennant R et al. (2007): The Warwick-Edinburgh Mental Well-being Scale (WEMWBS): development and UK validation, Health and Quality of Life Outcomes
- Adams WK et al. (2006): New instrument for measuring student beliefs about physics and learning physics, Physical Review Special Topics, Physics Education Research
