A Likert item is a statement followed by an ordered set of answers, usually running from strongly disagree to strongly agree. Put several items about the same idea together, score them, and you have a Likert scale. It is named after Rensis Likert, the psychologist who published it in 1932, and it has lasted because it is quick to answer and easy to summarize.
It is also easy to break. A statement that asks two things at once, labels that lean positive, a neutral option doing the job of "not applicable": each produces tidy numbers that mean less than they seem to. Most of the craft is in the writing, not the analysis.
What a Likert scale is, and what it is not
The name covers three different things.
A Likert item is one statement and one response scale: "The onboarding made it clear what to do first", answered from strongly disagree to strongly agree. It gives you one ordinal answer per person, an order without a fixed distance between the steps.
A Likert scale, in the strict sense, is a set of items that measure the same underlying attitude, answered on the same options and combined into one score. Eight statements about workload, summed, make a workload scale, and the combined score is steadier than any single item.
A Likert-type question is what most people mean in practice: any single question with an ordered, labeled set of answers, whether the labels measure agreement, frequency, importance, satisfaction or likelihood. It is written and analyzed the same way.
What a Likert scale is not: a star rating, a 0 to 10 recommendation score, or a slider. Those are rating scales, where a number is the answer, and the rating scale questions guide covers them. The defining feature of a Likert item is that each point carries words. For how it sits next to choice, ranking and open questions, see the survey question types guide.
4 →
The onboarding made it clear what to do first.
How much do you agree?
5 points or 7?
It matters less than the debate suggests, provided you pick one and keep it. The practical differences:
| 4 points | 5 points | 7 points | |
|---|---|---|---|
| Middle option | None, so everyone has to lean | Yes | Yes |
| Label wording | Easy | Easy, the standard set | Needs "somewhat" steps that people read differently |
| Suited to | Decisions where you need a side | General audiences and most surveys | Experts and research scales |
| On a phone | Fits easily | Fits easily | Fine one per screen, cramped in a grid |
| Typical failure | Undecided people pick at random | Heavy use of the middle | Extra steps add noise, not detail |
Five points is the right default. Everyone understands the labels without reading them twice. Seven points gives more resolution in theory, and it earns its place when respondents know the topic well: clinicians rating a protocol, engineers rating a tool they use daily. For a general customer audience, seven points mostly spreads the same opinion over more boxes.
The seven-point agreement set adds "somewhat disagree" and "somewhat agree", and people disagree on how far "somewhat agree" sits from "agree". If the steps cannot feel even, use five.
Two rules outrank the choice itself. Keep the same number of points on every item in a scale, because you cannot sum a 5-point answer with a 7-point one. And keep it the same between survey waves, or every trend line breaks.
Five labeled points for most surveys, seven for expert respondents, never both inside one scale.
Label every point, and keep the labels balanced
You can label only the ends ("1 = strongly disagree, 5 = strongly agree") and leave the middle as bare numbers, or put a word on every point. Label every point. People read a bare "3" in different ways, but they read "Neither agree nor disagree" the same way, which makes the answers easier to interpret and fairer to compare between groups.
Labels also have to be balanced: as many positive options as negative ones, mirrored around the middle. "Poor, fair, good, very good, excellent" looks like a five-point scale, but three of the five options are positive and the midpoint is already favorable. Everything you learn from it leans positive, and nobody notices, because the numbers look normal.
A balanced set mirrors its words: very dissatisfied and very satisfied, dissatisfied and satisfied, a true neutral in the middle. That is a bipolar scale. A scale that measures an amount (how important, how often) is unipolar: it runs from none to a lot, and "moderately important" is a real middle amount rather than an absence of opinion.
Finally, pick one direction, negative to positive or the reverse, and keep it for the whole survey. Switching halfway guarantees that somebody answers the second half backwards.
The neutral midpoint, and when to add an opt-out
A neutral option is meant to say "I have no leaning either way". Respondents also use it for "I don't know", "this doesn't apply to me" and "I'd rather not say". All four land in the same box, and your data cannot tell them apart.
The answer is not to remove the midpoint, which only moves a neutral person's noise into the agree or disagree columns. Give the other meanings their own exit instead.
- Keep the neutral point on bipolar scales, labeled in words ("Neither agree nor disagree"), not "3".
- Add "Not applicable" or "Don't know" as a separate, last option when some people genuinely cannot answer, such as a question about the mobile app sent to website-only users.
- Leave the opt-out out of the math. Report how many chose it, then calculate everything else without them.
A forced-choice scale with four or six points has a place when you need people to commit, such as choosing between two policies. For measuring how people feel, keep the middle.
Likert scale examples: 5 label sets and 20 statements
Almost every Likert scale question uses one of five label sets. Pick the set that matches what you are measuring, then word the statement to fit it. Force a frequency question onto an agreement scale ("I use the dashboard often") and "disagree" could mean sometimes or never.
| Label set | 1 | 2 | 3 | 4 | 5 |
|---|---|---|---|---|---|
| Agreement | Strongly disagree | Disagree | Neither agree nor disagree | Agree | Strongly agree |
| Frequency | Never | Rarely | Sometimes | Often | Always |
| Importance | Not at all important | Slightly important | Moderately important | Very important | Extremely important |
| Satisfaction | Very dissatisfied | Dissatisfied | Neither satisfied nor dissatisfied | Satisfied | Very satisfied |
| Likelihood | Very unlikely | Unlikely | Neither likely nor unlikely | Likely | Very likely |
Agreement
The classic Likert item: a first-person statement the respondent agrees or disagrees with.
- The onboarding made it clear what to do first.
- I have the tools I need to do my job well.
- The course materials prepared me for the final assessment.
- The pricing on the website was easy to understand.
Frequency
For behavior. Name a period when you can ("in the last month") so that "often" means the same thing to everyone.
- In the last month, how often did you use the reporting dashboard?
- How often do you get feedback on your work from your manager?
- How often do you finish a lesson in one sitting?
- How often do you need to contact support to finish a task?
Importance
For priorities. People call most things important, so if everything comes back above four, switch to a ranking question.
- How important is offline access to you?
- How important was price in your decision to buy?
- How important is flexible working when you consider a new job?
- How important is it that the course includes live sessions?
Satisfaction
For experiences that already happened. Name the specific thing, not the company as a whole.
- How satisfied are you with the speed of replies from support?
- How satisfied are you with the venue for this year's event?
- How satisfied are you with your first week of onboarding?
- How satisfied are you with the delivery time of your last order?
Likelihood
For intentions. Treat the answers as a direction, not a forecast; people overestimate what they will do.
- How likely are you to renew at the end of your current plan?
- How likely are you to attend next year's event?
- How likely are you to recommend this course to a colleague?
- How likely are you to try the new reporting feature in the next month?
For the questions that sit around a Likert block (screeners, open follow-ups, demographics), the survey questions guide has 120 more, written for copying.
Six mistakes that quietly ruin Likert data
None of these throws an error. They produce plausible numbers, which is worse.
Double-barreled statements. "The support team was fast and friendly." Somebody who got a friendly, slow reply has no honest answer, so they pick the middle or guess. One idea per statement: split it into two items.
Unbalanced labels. More positive options than negative ones, or a "neutral" that is really mild approval. It is the most common flaw in real surveys because the output looks normal. Mirror the labels word for word.
Leading statements. "Our redesigned dashboard makes reporting easier" tells people what you hope they think. Write it plainly, in the respondent's voice: "The dashboard makes reporting easier for me."
Agreement bias. People agree more readily than they disagree, so a long block of agree-or-disagree items drifts upward. Where you can, use an item-specific scale: "How easy was it to build a report?" from very difficult to very easy.
Careless negations. A negatively worded statement ("I do not feel supported") inside an agree scale creates double negatives that careful respondents misread. If you use reversed items, keep them simple and reverse their scores (5 becomes 1) before you combine anything.
Changing the scale between waves. New labels, a different number of points or a reworded statement means this quarter's result is not comparable with last quarter's. Freeze the wording of anything you track.
The survey design guide applies the same discipline to a whole questionnaire.
How to analyze Likert scale data
Likert answers are ordinal. Agree is more than neutral, but the gap between neutral and agree is not guaranteed to equal the gap between agree and strongly agree. That decides the order in which to read the results.
1. Look at the distribution first. For each statement, the percentage who chose each label, plus how many opted out. A diverging bar, disagreement left of zero and agreement right, shows the whole shape in one line.
2. Report the median. The label where half the answers sit on each side. It respects the ordinal nature of the data and a handful of extreme answers cannot distort it.
3. Report top-2-box, with bottom-2-box beside it. Top-2-box is the share who chose the two most positive labels. It is the number people remember and it tracks cleanly across waves. Bottom-2-box belongs next to it, because 60% agreement means one thing with 5% disagreement and another with 35%.
4. Use averages with caution. Coding the labels 1 to 5 and averaging is common, and widely accepted for a multi-item scale, where the summed score behaves more like a continuous number. For a single item, never report the mean alone.
A worked example: 200 people answered two statements on the same five-point agreement scale.
| Onboarding was clear | Support replies were fast enough | |
|---|---|---|
| Strongly disagree | 10% | 2% |
| Disagree | 15% | 8% |
| Neither agree nor disagree | 10% | 30% |
| Agree | 35% | 48% |
| Strongly agree | 30% | 12% |
| Mean (1 to 5) | 3.6 | 3.6 |
| Median | Agree | Agree |
| Top-2-box | 65% | 60% |
| Bottom-2-box | 25% | 10% |
Read the mean and the median and the two statements are identical. Read the distributions and they are two different problems. Onboarding has a clear majority in favor and a quarter who are genuinely unhappy: something specific is failing for a specific group, and that is where to look next. Support has one person in ten unhappy, few enthusiasts and nearly a third in the middle. It is not broken, just unremarkable.
Two statements can share an average and have nothing else in common.
A few habits keep the rest honest. Compare subgroups (new against returning customers) on the same statement, not different statements with each other. Show the number of respondents behind every percentage. And before you combine items into a scale score, check that they move together; Cronbach's alpha is the usual statistic.
None of this needs specialist software. In a spreadsheet, add a column that turns each label into a number from 1 to 5, and the distribution, median and top-2-box are a few formulas.
Why one statement per screen beats a grid
Many survey tools present a Likert scale as a matrix: statements down the side, labels across the top, one radio button per cell. On a wide screen it is compact. It also causes three well-known problems.
- Straightlining. Faced with identical rows, some respondents tick the same column all the way down, and the result looks like consistent opinion.
- It does not fit a phone. Five labels across plus a statement will not fit a 375-pixel screen, so the labels shrink, the grid scrolls sideways, or it gets rebuilt as separate questions anyway.
- The labels drift out of view. By the eighth row the column headings are off screen, and people choose by position rather than by word.
Upform has no matrix or grid question type, and for Likert items that is less of a gap than it sounds. Each statement is its own screen, with the labels as full-width options and, on a phone, a sticky back and next bar. It costs more screens than a grid. In return every statement gets read, every label is visible when answering, and nothing scrolls sideways. The matrix questions guide looks at the grid format in more detail.
One statement per screen does limit length. Keep a Likert block to four to eight statements, use an identical label set on every screen, and switch on the progress bar so respondents can see the end coming.
Building a Likert scale survey in Upform
There are two ways to build a Likert item in Upform, both on the free plan with all 21 question types, conditional logic and unlimited responses.
The labeled way: Single Choice
Add a Single Choice question, write the statement as the question, and put the instruction ("How much do you agree?") in the description. Add the five labels as options, in order, plus "Not applicable" as a sixth only when some people genuinely cannot answer. Repeat for each statement, options in the same order every time. Use this version for anything you plan to report: every point carries its label, and the answer you get back is the label itself.
The numbered way: Rating
A Rating question shows numbered buttons from 1 up to a maximum you set, from 3 to 10. Set it to 5 and explain the ends in the description: "1 is strongly disagree, 5 is strongly agree." It is quicker to build and the answer arrives as a number, but only the two ends are defined. Rating always starts at 1; for a 0 to 10 scale, such as a recommendation score, use Slider Scale.
5 →
I have the tools I need to do my job well.
1 is strongly disagree, 5 is strongly agree.
Which to use: labeled Single Choice for attitudes you will report, Rating when a tracked number with two defined ends is enough. Never mix them in one scale.
Follow up on the unhappy answers
The most useful addition to a Likert item is the question after it. With conditional logic, a Show rule can hold two conditions joined by OR: the answer "is exactly" Disagree, or "is exactly" Strongly disagree. The Long Text question it targets ("What got in the way?") then appears only to those people. On a Rating item the operators are equals, greater than and less than, so there the condition reads "less than 3". Everyone else moves straight on.
Answers arrive in the Responses table and export to CSV, or land row by row in a sheet through the Google Sheets integration, ready for the distribution and top-2-box.