What Is a Likert Scale? How to Choose Levels and Design a Questionnaire

What Is a Likert Scale in Quantitative Research

A Likert scale records a respondent's level of agreement, satisfaction, frequency, or evaluation of a statement. In a quantitative questionnaire, you present an item and ask participants to choose one response level arranged in order. For example, for the statement “I find this application easy to use,” a respondent might choose from 1, strongly disagree, to 5, strongly agree.

The name Likert comes from the attitude measurement technique introduced by Likert (1932). Each level needs a clear meaning, and all levels must follow the same direction. If a higher score always means stronger agreement, data entry, mean-score calculation, and running Cronbach's Alpha become easier to control.

You should distinguish three concepts that are often mixed together. A Likert item is a single statement, such as “Employees respond quickly.” A Likert scale may refer to the response levels for one item or to a group of items measuring the same construct. The questionnaire is the complete instrument, including the introduction, screening questions, scale sections, and demographic information. If you are confusing the questionnaire with the scale, see what is a questionnaire.

In practical terms, a Likert scale converts a subjective evaluation into ordered data for analysis. In a thesis, you usually use several items for one latent variable, then assess reliability with Cronbach's Alpha, run EFA, or evaluate the measurement model in SmartPLS. Choosing the Likert levels is therefore only one part of scale development. It does not replace scale assessment.

How to Choose a Likert Scale for Your Questionnaire

There is no single number of levels that fits every study. Consider the construct being measured, the education level of the respondent group, the questionnaire length, and how previous studies designed the scale. In practice, a 5-point scale is common in Vietnamese theses because respondents can understand it easily, the questionnaire stays compact, and the data are convenient to analyze in SPSS. A 7-point scale captures finer distinctions, but respondents may find it difficult to distinguish between adjacent levels.

For an attitude scale, clearly label both endpoints and keep the meaning of each level evenly ordered. A common format is: 1 = strongly disagree, 2 = disagree, 3 = neutral, 4 = agree, 5 = strongly agree. For a satisfaction scale, you can use 1 = completely dissatisfied to 5 = completely satisfied. Do not mix “agree” and “satisfied” within the same group simply because both formats have five levels.

The table below is a selection guide, not a mandatory rule. (Likert, 1932) supports the attitude measurement technique using ordered response levels. (Hair et al., 2010) concerns scale assessment in quantitative research, including the use of multiple items and the evaluation of reliability.

You also need to decide whether to include a “no opinion” option. If respondents may genuinely never have used the product, place a screening question first, such as “Have you used this service in the past six months?” A neutral option should not conceal the fact that a respondent did not understand the item.

For sample size, (Hair et al., 2010) is often used as a reference for the rule of 5 to 10 observations per item. This is only a reference rule for survey design. You still need to consider the number of items, the research model, the analysis method, and your actual ability to collect data. A questionnaire with too many items can increase careless responding even when the sample size appears sufficient on paper.

Designing a Questionnaire for This Objective

Before opening Google Forms or another form builder, list the latent variables, items, and hypotheses served by each item group. For example, if your model proposes that “service quality” affects “satisfaction,” each construct needs its own group of items. Do not write items merely because they sound reasonable and then try to fit them into the model later.

Writing Items

Each item should express one reasonably clear idea and use words that the respondent group understands. The statement “This application is easy to learn and processes transactions quickly” contains two ideas. Respondents may not know which part they are evaluating. Split it into two items if both aspects need to be measured.

Avoid double negatives such as “I do not think this service is not reliable.” Statements like this often create data-entry errors and cause respondents to select the wrong direction. If you must use a reverse-coded item, record the coding rule and reverse the score before calculating a total score or running the analysis. Do not reverse-code items merely to make the questionnaire look more varied.

Choosing the Number of Response Levels

Use the same number of levels across the main scale groups when you need to compare scores and calculate combined scores. Five levels often work well for questionnaires aimed at respondents with different ages and education levels. Seven levels can be useful when respondents are comfortable making detailed evaluations, but check how the options appear on a phone so they do not sit too close together.

Each level should have a label, or at least both endpoints should be explained. If the questionnaire shows only 1, 2, 3, 4, and 5 without explaining 1 and 5, respondents may interpret the numbers differently. In an electronic questionnaire, place the instructions immediately before the item group rather than on a distant page.

Arranging Screening Questions

Place screening questions before the Likert groups. For example, ask whether participants have bought something on the platform, how long they have used it, and how frequently they use it. Someone who has never had the relevant experience should not continue to items evaluating delivery quality unless your research design gives a clear reason.

After screening, arrange the item groups in a logical order, moving from general experience to specific evaluation. Demographic information is usually placed at the end to reduce the feeling that respondents are being asked for personal information immediately. You can also see survey forms to check how to divide sections and write response instructions.

Testing the Questionnaire Before Release

Complete the questionnaire yourself on both a phone and a computer. Check that variable codes are consistent, required items work correctly, question branches send respondents to the right section, and the exported data create one column for each item. If possible, test it with a small group resembling the survey population to identify unclear wording and an excessive completion time.

A pilot test helps you revise the instrument, but it does not automatically prove that the scale is acceptable. Once you have the official data, you still need to check Cronbach's Alpha, Corrected Item-Total Correlation, and the analysis steps appropriate for your model.

Collecting the Data

Before sending the link, define the target population, sampling criteria, access channel, and collection period. “Survey everyone” is not enough to explain your sample. Record which groups were invited, who was excluded, and the start and end dates, because this information belongs in the methodology chapter.

DoThesis has fillform.info, our own form tool for creating questionnaires and collecting responses in Vietnamese. You can also use Google Forms, which is a common option when you need a familiar tool. Whichever tool you use, download a test .csv or .xlsx file before releasing the questionnaire. Check that column names are not truncated, Vietnamese characters display correctly, and each option is coded properly.

When you distribute the survey, the introduction should briefly state the research purpose, the eligible respondents, the completion time, and how information will be kept confidential. Do not lead respondents with a statement such as “Please rate this excellent product.” That wording can push responses in one direction. You can also review what is a survey and Google survey before choosing the collection channel.

Track the response rate and completion time during collection. Some records completed unusually quickly may need checking, but do not delete them only because the completion time is short. Consider several signals together: many unanswered items, the same option selected throughout the questionnaire, contradictory answers to attention checks, and screening information that does not meet the criteria.

Cleaning Data Before Importing It into SPSS

After closing the survey, save an untouched copy of the original data. Create a duplicate for cleaning and record every change in a log file. This practice lets you explain why the number of observations in the analysis file differs from the number of original responses.

First, check the column names. Each item should have a short code, such as PU1, PU2, PU3 or HL1, HL2, HL3. Do not use a long question as a column name because SPSS has limitations and you may confuse variables when selecting them. Separate submission time, email address, respondent name, or internal notes from the variables used to run the model if they are not part of the analysis design.

Next, check missing data. Review whether each row leaves too many items blank and whether any column has an unusually high missing rate. If someone did not meet the screening criteria but still appears in the main section, handle the record according to the exclusion rule you documented earlier. Do not automatically replace every blank cell with the mean. That approach can change the distribution and hide a questionnaire problem.

You also need to check straight-lining, meaning that a respondent selects the same level for most or all items. Straight-lining is a signal to investigate, not conclusive proof that the record should be removed. Compare completion time, attention-check items, and screening responses. If you remove the record, record the row number or record code and the reason.

For reverse-coded items, recode them before calculating scores. In a 5-point scale, the usual formula is new score equals 6 minus old score. Use this formula only when the item has exactly five levels and the reverse direction was established in the design. Then check the result with frequencies and descriptive statistics so you do not reverse an entire group by mistake.

Finally, save the file as .sav if you will work mainly in SPSS, or as .csv if you need to transfer it to SmartPLS. Check that Likert columns are numeric rather than string variables. If SPSS reads “5” as text, you cannot calculate means or run the analysis correctly until you convert the data type.

How to Report This in Chapter 3

In Chapter 3, report the source used to build the scale, how the wording was adapted, the number of Likert levels, the survey population, the collection procedure, and the data-cleaning criteria. If the scale was translated from an English source, explain that you adapted it to the research context. Do not describe the translation as a validated scale if you have not yet analyzed the data.

A scale-description table should include the item code, item wording, corresponding construct, and reference source. If you adapted a group of items, state clearly that it is a reference scale and will be assessed in the results chapter. Do not put a Cronbach's Alpha value in Chapter 3 if you do not yet have the official analysis result.

How to write this in your thesis: “The study used a [5/7]-point Likert scale, where [1] represented [lowest label] and [5/7] represented [highest label]. The items were adopted and adapted from [scale source] to fit the context of [research population]. After data collection, the scale was assessed using [Cronbach's Alpha/EFA/measurement model] before the hypotheses were tested.”

For the sample, report the number of questionnaires distributed, the number returned, the number of valid questionnaires, and the exclusion criteria. These figures must come from your data file. Do not put illustrative numbers in the actual thesis. If you cleaned the data in several rounds, keep a separate tracking table so you can explain every change when asked.

Common Errors

The first error is calling every set of response options from 1 to 5 a complete Likert scale. A single question can be a Likert-type item, but a scale measuring a construct usually needs a group of related items. Identify the construct first, then decide how many items are needed.

The second error is using different types of labels within the same group. One item asks about agreement, the next asks about frequency, yet both are coded from 1 to 5. When you calculate a combined score, the numbers may no longer have the same meaning. If both types of information are necessary, separate them into two scale groups.

The third error is changing data to increase Cronbach's Alpha without a content-based reason. You can review the Item-Total Statistics table and consider removing an item, but the decision must reflect the construct meaning, the relationship with other items, and the original design. Deleting items repeatedly until the index looks good can strip the scale of important content and leave you unable to defend the decision.

The fourth error is failing to save the questionnaire and the final coding version. If the SPSS file contains PU1 through PU5 but the questionnaire wording for PU3 changed after data collection, you will not know which version the data represent. Save the revision date, form version, and corresponding data file.

You should also avoid placing “do not know” responses in the middle of an agreement scale without a coding plan. “Do not know” does not always belong between “disagree” and “agree.” You can make it a separate option and handle it according to the analysis design instead of assigning an arbitrary score.

Frequently asked questions

What is a Likert scale, and must it use 5 levels?

A Likert scale measures the level of response to a statement through ordered choices. Five levels are common in theses, but they are not mandatory. You can use 7 levels when the study needs finer distinctions and the respondent group can use those levels consistently.

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

There is no universally better choice. A 5-point scale is easier to explain and reduces response complexity, while a 7-point scale captures more shades of opinion. Prioritize the format used by the original scale source, its fit with the survey population, and consistency throughout the questionnaire.

How many Likert levels are acceptable?

The number 5 or 7 does not determine whether the scale is acceptable. After collecting data, you need to assess scale reliability and validity with a suitable method. For example, Corrected Item-Total Correlation of 0.3 or above is commonly referenced under (Nunnally and Bernstein, 1994), while Cronbach's Alpha of 0.7 or above is commonly referenced under (Nunnally, 1978). For a new or exploratory scale, Alpha from 0.6 may be considered under (Hair et al., 2010), but you still need to explain the context.

Can I mix 5-point and 7-point Likert scales in one questionnaire?

You can, if each group has its own objective and instructions, but this increases the risk of respondent confusion and makes mean-score interpretation more complicated. Unless there is a clear methodological reason, keep the same number of levels across the main scales.

Can Likert data be analyzed in SPSS?

Yes. Code the levels as numbers, check the data type, and use them in analyses that fit your research design. Before running SPSS, handle reverse-coded items, missing data, invalid records, and columns that are not used in the analysis. Do not treat coded numbers as proof that the scale is acceptable until you have assessed it.

What should I do when a respondent selects the same level for every item?

Check completion time, screening questions, attention checks, and whether the answers contradict each other. A respondent may genuinely have the same evaluation for every item, so do not delete the record automatically because one level is repeated. If there is enough evidence that the record is unreliable, remove it according to the predefined criteria and record the reason in the cleaning log.

Open your questionnaire file and data file now, create a variable-code table for each Likert group, check the final version, and compare it with your cleaning steps before importing the data into SPSS. If you need to run the analysis on your own .sav or .csv file, see M4 data analysis.