What is a Likert scale survey?
A Likert-type question asks respondents to choose an ordered response to a statement, often from strongly disagree to strongly agree. A Likert scale commonly combines several related items intended to measure the same underlying concept. In everyday usage, people also use the term for an individual agreement question.
A satisfaction or frequency rating is related but should use labels suited to what it measures. The important task is to define the concept, choose understandable response options and plan how the answers will be interpreted.
This guide covers wording, five- and seven-point options, a worked analysis and the role of comments. It does not treat every score as a diagnosis or assume that a comment proves what caused the rating.
Examples of Likert-type items
The following are illustrative items, not a validated scale. Use them to think through a question’s purpose before adapting the wording.
Scroll horizontally to see all columns →
| Context | Illustrative statement | What it describes |
|---|---|---|
| Employee listening | I know where to find the information needed for my work. | Reported clarity about information access |
| Training | I can explain the main steps of the method. | Self-reported confidence or perceived capability, not a direct skill test |
| Member experience | The registration instructions were easy to follow. | Reported ease of the instructions |
| Service experience | I understood what would happen after my request. | Reported understanding of the next step |
Ask one thing at a time. “My manager communicates clearly and recognizes good work” combines two ideas. A respondent may agree with one and disagree with the other. Split them if both answers are needed.
Consider whether a direct question is clearer than an agreement statement. “How easy were the instructions to follow?” can use ease labels rather than asking people to agree or disagree about ease.
Five-point and seven-point response options
A five-point agreement item might use strongly disagree, disagree, neither agree nor disagree, agree and strongly agree. Seven points add finer distinctions, such as somewhat disagree and somewhat agree.
More response options do not automatically produce better measurement. Choose options that the audience can distinguish meaningfully and use them consistently. Test wording and presentation rather than selecting a point count only because it looks more precise.
A neutral midpoint is different from “do not know” or “not applicable.” If respondents may lack the experience needed to answer, give them an appropriate way to express that instead of forcing a neutral or negative judgment.
Do not randomize the order of an ordered scale as though its categories were an unordered list. Keep the direction and labels understandable, especially across languages and devices.
Start analysis with the response distribution
Show how many people selected each category before compressing the result into one number. The distribution reveals whether answers cluster, spread or divide between opposing views.
Responses are ordered, but the distance between labels is not automatically equal. The methodological article Analyzing and Interpreting Data From Likert-Type Scales discusses this distinction and the choices involved in analysis. A suitable method depends on the measure and research question; do not treat either “always average” or “never average” as a complete rule.
For a practical descriptive report, include the item wording, response labels, valid-answer count and missing or not-applicable count. If you use a coded mean or a positive-response percentage, state the rule.
Worked example: a five-point item
A fictional team receives these 100 valid answers to “The instructions were easy to follow”:
Scroll horizontally to see all columns →
| Response | Code | Count |
|---|---|---|
| Strongly disagree | 1 | 10 |
| Disagree | 2 | 15 |
| Neither agree nor disagree | 3 | 20 |
| Agree | 4 | 35 |
| Strongly agree | 5 | 20 |
If “positive” means agree or strongly agree, the positive share is (35 + 20) ÷ 100 = 55%. The coded mean is (10 + 30 + 60 + 140 + 100) ÷ 100 = 3.4. Describe it as a mean of the assigned codes, alongside the distribution.
Suppose 120 people were eligible to answer but 20 supplied no valid rating. Report that coverage separately. Do not assign zero to missing answers or include “not applicable” as though it were a point below strongly disagree.
Now compare two small fictional groups: one has ten neutral answers, while the other has five strongly disagree and five strongly agree answers. Both coded means are 3.0, but the experiences are very different. Comments may add context, yet the distribution itself already shows the difference.
When should you add an open-ended question?
Add a focused follow-up when an explanation would help the decision and the team can review it. For the instructions item, “What, if anything, would make the instructions clearer?” may produce useful detail.
Do not require a long explanation after every item. A short survey may work better with one or two relevant open questions. If only some respondents are asked for comments, explain that selection when reporting themes.
Keep comments connected to the relevant response context, even within an anonymous submission. Review recurring themes, differing accounts and ambiguous answers. A theme can suggest an issue to investigate; it does not establish a causal explanation for a score.
For writing those questions, see open-ended survey questions.
Be careful when combining several items
Several agreement questions do not automatically form a meaningful total score. Decide what the combined measure is intended to represent and whether there is appropriate evidence for using it that way.
If using an established instrument, follow its scoring and missing-data guidance and understand the implications of changing wording, response options or audience. Do not call a locally assembled set of items validated without evidence.
When an item is negatively worded, check the coding direction before combining it with positively worded items. Reverse coding is a scoring operation, not permission to hide a confusing question. Pilot the instrument and inspect unexpected response patterns.
Compare groups and waves carefully
Keep wording, response labels, eligible population and relevant collection conditions comparable. If something changes, record it and assess what it means for the trend.
For identified follow-up, distinguish matched participants from everyone who answered either wave. For anonymous surveys, describe group-level change and check whether the respondent mix changed. A movement from 3.4 to 3.8 does not prove each person improved or that an intervention caused the change.
Across branches or locations, agree a small shared set of measures and a data dictionary. Local teams can add relevant items. Do not combine different scales under one label simply because all responses were converted to numbers.
The baseline survey guide explains planning the comparison before collecting the first wave.
Use group reporting thoughtfully
For employee or sensitive feedback, define who can see individual answers, comments and subgroup results. Small groups and recognizable quotations can reveal identity even without a name.
Use the organization’s agreed reporting and access rules, and test how filters combine. A view that seems sufficiently broad may become identifying after several filters. Do not promise anonymity unless the actual collection arrangement supports it.
For a wider listening program, see employee survey software requirements.
What should the software make easy?
Test the survey and analysis together: clear labels, mobile presentation, appropriate missing-value handling, distributions, segment definitions, exports and relevant source context. If AI helps review comments, inspect its coverage and interpretation.
A personal identifier is not required for every Likert survey. Answers within an anonymous submission can remain connected. Persistent matching becomes relevant when the design follows the same people or organizations over time.
Sopact’s focus is self-managed collection, analysis and governance of continuing evidence. Evaluate the specific measurement and access requirements in a realistic pilot, rather than assuming a new platform resolves question design or guarantees accurate interpretation.
Check one item from wording to report
- Write the decision the item should inform.
- Choose one clear concept and suitable response labels.
- Test the wording with intended respondents.
- Define missing values and the eligible population.
- Produce a distribution and any stated summary measure.
- Review relevant comments without overstating what they prove.
- Record the definition and any future changes.
For alternative formats, use survey question types. For the full reporting process, read how to analyze survey data.
Frequently asked questions
Is every five-point rating a Likert scale?
No. Five points describe the number of options. Agreement items, satisfaction ratings and frequency ratings measure different concepts, and a multi-item scale requires a meaningful scoring approach.
Should I use five or seven points?
Choose labels your audience can distinguish and that fit the intended measure. Test comprehension and keep the approach consistent where comparison matters.
Can I report the average?
A coded mean may be used in some analyses, but explain the coding and show the distribution. The appropriate method depends on the measure and research question.
What does top-two-box mean?
It combines the two highest response categories. Define which categories are included and use the correct valid-answer denominator.
Does a comment explain the cause of a rating?
It provides the respondent’s account or related context. That can guide investigation, but it does not by itself establish causality.
Should missing answers be coded as zero?
No. Keep missing, not asked, not applicable and valid scale values distinct according to the analysis plan.
