What Is a Likert Scale? How to Choose Levels and Analyze It in SPSS
What Is a Likert Scale
A Likert scale measures respondents' level of agreement, perception, or evaluation of a statement. Instead of choosing only yes or no, respondents select an ordered response level, such as a scale from 1, strongly disagree, to 5, strongly agree. Likert is the name of the attitude measurement technique introduced in the original work by Rensis Likert (Likert, 1932).
In a quantitative thesis, you usually use several items to measure an abstract construct. For example, the construct “satisfaction” may be measured with SAT1, SAT2, SAT3, and SAT4. Each item is a statement in the questionnaire, while the Likert scale is the response-level system respondents use for those statements.
A 5-point Likert scale is commonly coded as follows:
| Code | Response level |
|---|---|
| 1 | Strongly disagree |
| 2 | Disagree |
| 3 | Neutral |
| 4 | Agree |
| 5 | Strongly agree |
You need to distinguish between two uses of the term. A “Likert item” usually means one statement with ordered response levels. In research, a “Likert scale” usually means the full group of items measuring the same construct. When you run SPSS, Cronbach's Alpha is calculated for the group of items, not for one statement by itself.
If you are building a data collection instrument, you can also read the guides on what a questionnaire is, survey forms, and the topics in Surveys.
The Meaning of a Likert Scale in Quantitative Research
A Likert scale turns a subjective evaluation into data that can be coded, summarized, and analyzed. After data collection, you can calculate Mean, standard deviation, Cronbach's Alpha, EFA, regression, or use the items in a SmartPLS model. Each step answers a different question, so you should not look at one number and decide whether the entire scale passes or fails.
Mean shows the response tendency of the sample. For example, SAT1 with a Mean of 4.12 indicates that respondents tend to agree with the statement, but this number does not prove that the scale has good reliability. Reliability needs to be assessed through Cronbach's Alpha and Corrected Item-Total Correlation. Convergent validity and discriminant validity require different tests.
A Likert scale also helps you test hypotheses. If your hypothesis states that service quality affects satisfaction, you first need to check the items measuring both constructs. You can then create a representative variable or use latent variables in the model. A high Mean for one item does not mean that the item has a strong effect on the dependent variable.
A 5-point scale works well when the questionnaire needs to be short, respondents have varied backgrounds, and you want the response options to be easy to understand. A 7-point scale allows more subtle distinctions between response levels, but respondents also have more options to process. Choose one format and use it consistently within the same group of items unless you have a clear methodological reason to do otherwise.
A Likert scale is different from a nominal scale. A nominal scale only classifies groups, such as gender or region, and codes 1, 2, and 3 do not represent higher or lower levels. You can read more about what a nominal scale is before coding your data in SPSS.
What Counts as an Acceptable Likert Scale
There is no single threshold for every stage. Separate three tasks: designing the number of response levels, checking each item, and evaluating the scale as a whole. The table below uses only thresholds with sources in the permitted reference list.
| Check | Reference threshold | Source |
|---|---|---|
| Cronbach's Alpha for the scale | 0.70 or above | (Nunnally, 1978) |
| Cronbach's Alpha for a new or exploratory scale | May be 0.60 or above | (Hair et al., 2010) |
| Corrected Item-Total Correlation | 0.30 or above | (Nunnally and Bernstein, 1994) |
| KMO when performing EFA | 0.50 or above | (Kaiser, 1974) |
| Bartlett's Test in EFA | Sig. below 0.05 | (Kaiser, 1974) |
| Eigenvalue when deciding the number of factors | Above 1 | (Kaiser, 1960) |
| Total Variance Explained | 50% or above | (Hair et al., 2010) |
| Factor loading in EFA | 0.50 or above | (Hair et al., 2010) |
| Observations per item | Approximately 5 to 10 observations | (Hair et al., 2010) |
These thresholds are reference points for reading your output, not reasons to delete items mechanically. A scale with a high Alpha can still measure the wrong construct if the statements do not fit together. A new scale in an exploratory context may be interpreted with an Alpha from 0.60, but you need to state the context and the reason.
For a 5-point Likert scale, do not ask “what Mean is acceptable” in the same way you ask about Cronbach's Alpha. Mean reflects respondents' evaluation, not a direct reliability criterion. If you want to interpret Mean intervals, state the interval rule you selected and apply it consistently throughout the results chapter.
How to Read a Likert Scale in the Output
In SPSS, the process usually starts with a reliability check at Analyze > Scale > Reliability Analysis. Move the items measuring the same construct into the Items box, choose Alpha for Model, then open Statistics and select Scale if item deleted, Item, and Scale. You will read at least two tables, Reliability Statistics and Item-Total Statistics.
The table below is illustrative output, not the result of a real study. The column names are kept as SPSS displays them so you can compare them with the file open on your screen.
| SPSS table | Variable or index | Cronbach's Alpha if Item Deleted | Corrected Item-Total Correlation | Cronbach's Alpha |
|---|---|---|---|---|
| Reliability Statistics | SAT1 to SAT4 | 0.842 | ||
| Item-Total Statistics | SAT1 | 0.801 | 0.681 | |
| Item-Total Statistics | SAT2 | 0.817 | 0.624 | |
| Item-Total Statistics | SAT3 | 0.833 | 0.548 | |
| Item-Total Statistics | SAT4 | 0.790 | 0.716 |
In this example, the group's Alpha is 0.842, above the 0.70 reference point in (Nunnally, 1978). All Corrected Item-Total Correlations are above 0.30 according to (Nunnally and Bernstein, 1994). The Cronbach's Alpha if Item Deleted column shows how Alpha would change if each item were removed. If deleting an item increases Alpha, that is a signal to investigate, not an automatic instruction to delete it.
After Cronbach's Alpha, you can run EFA at Analyze > Dimension Reduction > Factor. In the dialog box, move eligible items into Variables, select Descriptives and tick KMO and Bartlett's test, select Extraction and check Eigenvalues greater than 1, then select Rotation, often Varimax when you are using an independent-factor approach. The rotation method needs to fit your model and your supervisor's guidance.
If the output shows KMO and Bartlett's Test, Total Variance Explained, and Rotated Component Matrix, read them in that order for data suitability, the number of factors, and factor loadings. Do not look only at one item with a high loading and ignore whether it loads on two factors or changes the theoretical meaning of the group.
How to Handle a Likert Scale That Does Not Pass
Work from data checks toward theory. This order helps you avoid deleting items just to make the output look better.
Check Coding and Missing Data
Open Analyze > Descriptive Statistics > Frequencies to check the codes that appear. If the questionnaire uses 1 to 5 but some rows contain 0, 6, or 99, determine whether these values are data-entry errors or codes for missing responses. Declare missing codes correctly. Do not let SPSS treat 99 as a level of agreement.
You also need to check reverse-coded items. If SAT3 is a negatively worded statement but has not been recoded, its correlation may be low and Alpha may decrease. For a 1 to 5 scale, reverse coding usually changes 1 to 5, 2 to 4, keeps 3 unchanged, changes 4 to 2, and changes 5 to 1. Create a new variable or save syntax so you can trace the operation later.
Review Corrected Item-Total Correlation
If an item is below 0.30, reread the statement, check the coding, and ask whether the item really belongs to the same construct. If the data are correct, deleting the item may be acceptable when you have a methodological and theoretical reason. Record which item was deleted, at which round, and the results before and after deletion.
Compare Alpha if Item Deleted
If deleting an item increases Alpha slightly, a small increase is not enough to conclude that you must remove it. An item that is important to the content may be worth retaining if the index remains acceptable. The committee may ask why the item was deleted, so your answer needs to rely on the scale content and the specific index, not only on a desire to increase Alpha.
Rerun the Analysis with a Coherent Item Group
After each deletion, rerun Reliability Analysis and EFA if needed. Do not delete several items at once and present the process as if nothing changed. If several items fail together, the problem may be in the translation, wording, survey sample, or incorrect grouping of constructs.
When the data genuinely do not support defending a scale, the most honest approach is to report the result and the study's limitation. Software can help you check the scale, but it cannot turn a group of statements measuring different constructs into a good scale.
Distinguishing a Likert Scale from Mean and a Nominal Scale
A Likert scale is a way to collect responses by level. Mean is a descriptive statistic calculated after you have the data. Students often mix these concepts when they write that “the scale has a Mean of 3.8.” A more precise sentence is, “The items were measured with a 5-point Likert scale and had Mean values ranging from ...”.
You also need to distinguish a single item from a representative score for a construct. SAT1 is an item. The Mean of SAT1 reflects the response to one statement. After the scale passes the reliability checks, you can calculate the average of SAT1 to SAT4 to create the SAT variable, depending on your analysis method and research guidance.
A Likert scale is also different from a nominal scale and a purely ordinal scale. Likert levels have an order, but treating the distance between 1 and 2 as exactly equal to the distance between 4 and 5 needs to be explained in your analysis design. In business theses, Likert data are often aggregated and entered into quantitative analyses, but you still need to describe the coding and variable-creation process clearly.
Do not confuse a Likert scale with a star rating, a ranking scale for alternatives, or a multiple-choice question. If you are preparing to collect responses online, you can read the guides on online survey questionnaires and survey forms. These pages concern the collection tool. Whether the scale passes still needs to be checked using the data.
Common Errors When Using a Likert Scale
The first error is using one item to draw a conclusion about the reliability of an entire construct. Cronbach's Alpha requires a group of items. If a construct has only one item, report the limitation of that measurement approach instead of calculating an artificial Alpha.
The second error is changing the response labels without changing the codes. For example, you may write 1 as strongly agree in the questionnaire but interpret 1 as strongly disagree when entering the data into SPSS. Create a codebook before releasing the questionnaire and check several rows after exporting the CSV or SAV file.
The third error is assuming that a higher Alpha is always better. An excessively high Alpha may suggest that the items repeat the same content. Review the statement content, number of items, item-total correlations, and measurement objective together.
The fourth error is deleting an item only because its Mean is low. A low Mean shows that the sample tends to disagree. It does not show that the item is wrong. Delete an item only when the data, scale design, and research argument support the decision.
The final error is reporting results without stating whether the scale has 5 or 7 levels, what 1 and the highest score mean, which items were deleted, and why. These details help readers reproduce the process and help you answer questions about the output.
How to write this in your thesis
State the scale format, coding, reliability results, item decisions, and the next analysis in the order you actually performed them. Keep the bracketed placeholders visible when you draft the paragraph: [number of levels], [coding], [Cronbach's Alpha], [item deleted], and [next analysis].
Frequently asked questions
Is a 5-point Likert scale mandatory?
No. Five levels are common because they are easy to answer and interpret, but you can use 7 levels when the research design, original scale, or survey population supports that choice. State the label for every level, code the responses consistently, and do not mix two coding formats within the same group of items without a reason.
What counts as an acceptable Likert scale?
If you are asking about reliability, you usually check Cronbach's Alpha from 0.70 and Corrected Item-Total Correlation from 0.30 according to (Nunnally, 1978) and (Nunnally and Bernstein, 1994). For a new scale or exploratory research, an Alpha from 0.60 may be considered according to (Hair et al., 2010).
If you are asking what Mean is acceptable, Mean is not a direct criterion for deciding whether a scale passes. Interpret Mean as the sample's evaluation, while reliability and scale validity require separate tests.
How do you run a Likert scale analysis in SPSS?
Enter each item as a column and each respondent as a row, check missing codes, and recode reverse-coded items first. Then go to Analyze > Scale > Reliability Analysis, select the item group, tick Scale if item deleted in Statistics, and read Reliability Statistics together with Item-Total Statistics.
If your study needs to reduce or confirm the factor structure, you can run EFA at Analyze > Dimension Reduction > Factor. Cronbach's Alpha and EFA answer different questions, so one table should not be used as a substitute for the other.
Should you delete an item to increase Cronbach's Alpha?
Consider deletion only when the item has an unacceptable index, inappropriate content, or a clear problem in the scale structure. The Cronbach's Alpha if Item Deleted column supports the decision, but it is not an automatic command. After deletion, rerun the analysis and record the change in your thesis.
Can you calculate Mean for a Likert scale?
You can calculate Mean for each item to describe the data. After the scale passes the necessary checks, you can calculate a representative construct score using the average or total score, depending on your research method. In the methods section, state the coding, number of levels, reverse-coding procedure, and method used to create the representative variable.
Open your data file now, check the codebook for 1 to 5 or 1 to 7, then run Reliability Analysis for each item group and save the output before deleting any item. If you need to run the analysis on your .sav or .csv file and continue writing the results, see the M4 data analysis module.