Premium Writing, Design & Advisory · Dubai, UAE

Blog / Academic Support

Academic Support

Likert-Scale Questionnaires: From Design to SPSS Analysis for Your Dissertation (2026)

Likert-type questions are common in survey dissertations, and several analysis choices are fixed by the questionnaire before data collection begins. This guide covers how to build the questionnaire, code it in SPSS, check reliability and validity, and choose the right tests. To size your sample first, use our free sample size calculator.

Researcher and statistics mentor mapping Likert questionnaire responses to analysis charts

Design

Design the questionnaire for the analysis you will run

  • Measure constructs, not single questions. Each concept in your framework (for example job satisfaction or perceived usefulness) should be measured by several items, ideally adapted from a published, validated scale that you cite.
  • Keep the response format consistent. Use the same number of points and the same anchors (for example 1 = Strongly disagree to 5 = Strongly agree) across a construct.
  • Plan reverse-worded items deliberately. They can reduce acquiescence bias, but they must be reverse-coded before analysis.
  • Write one idea per item. “My manager is supportive and fair” asks two things; split it.
  • Pilot it. A small pilot can reveal ambiguous wording and broken survey logic; it does not by itself establish the final scale’s reliability or validity.
  • Translate carefully. For Arabic and English versions, plan translation, independent review and cognitive testing with both language groups. Back-translation alone does not demonstrate that items carry the same meaning.
  • Get ethics approval first. Most universities require approval before data collection; build consent and anonymity into the first page.
Sample

How many respondents do you need?

It depends on the analysis. Start with your primary research question, planned model, target effect, significance level and desired power. A power analysis can then inform the sample target for a specified test. Factor analysis and structural equation models need separate planning based on model complexity, item quality and anticipated missing data; no fixed respondents-per-item rule works for every design. Document the assumptions, likely non-response and the actual usable sample in your methodology. Discuss departures from the plan with your supervisor before running a different test.

Data preparation

Coding and cleaning the data in SPSS

  1. Variable View: give each item a short name (JS1, JS2…), a label with the full wording, and value labels (1 = Strongly disagree…). Set the measure to Ordinal for items.
  2. Reverse-code negatively worded items with Transform > Recode into Different Variables (on a 5-point scale, 1→5, 2→4, 3→3, 4→2, 5→1).
  3. Screen the data: check frequencies for out-of-range values, look for straight-lining (the same answer to every item) and decide how you will handle missing data, then report that decision.
  4. Compute composite scores only when the items can reasonably be interpreted as one construct. Record the scoring rule, reverse-coded variables and how many answered items are required; do not average missing values silently.
Reliability and validity

Check the scale before testing hypotheses

Reliability: run Analyze > Scale > Reliability Analysis for each proposed construct. Cronbach’s alpha is a measure of internal consistency, not proof that the scale is valid or unidimensional. A value around .70 is a common rule of thumb, not a universal pass mark. Inspect item wording, item-total relationships, missing responses and the construct definition before deciding whether any item should be removed. IBM documents the available SPSS reliability outputs.

Validity: first check whether the items cover the intended concept and make sense to the target respondents. EFA can explore an uncertain structure; CFA or another prespecified measurement model can assess a proposed structure when the design and sample support it. KMO, Bartlett’s test and loadings are diagnostics, not certificates of validity. Choose a method from the research question, scale history and model assumptions, then explain its limits. A published scale-development primer separates content, structural and reliability evidence.

Choosing tests

Which test for Likert data?

A single ordered response category is ordinal; frequencies and distributions are a useful starting point. A multi-item score may sometimes be analysed with methods for approximately continuous outcomes, but only after its construction and relevant assumptions are checked. The table below offers possible tests for common designs, not an automatic menu: independent versus paired data, group size, distribution, missingness, scale quality and the hypothesis can change the choice. The Likert analysis review explains why the treatment of individual items and combined scales is debated.

Your question Composite scores Single items
Do two groups differ? Independent t-test Mann-Whitney U
Do three or more groups differ? One-way ANOVA Kruskal-Wallis
Are two constructs related? Pearson r Spearman rho
Do several factors predict an outcome? Multiple regression, or SEM in AMOS/SmartPLS Ordinal regression

To read and report each test’s output, use our table-by-table guide to SPSS output and APA 7 reporting, and see how to choose the right statistical test for other designs.

Stuck on the analysis? Our data analysis support covers SPSS, AMOS, SmartPLS and NVivo: we check your coding, reliability and validity, run or review the tests, and explain every output so you can defend it. Contact Labeeb about your dataset.

Worked example

From five survey items to one defensible result

Suppose a dissertation measures perceived supervisor support with five adapted 1–5 agreement items. Before collection, define the construct, cite the original scale, document any wording changes and specify the primary comparison or association. After collection, inspect each item’s response distribution and missing values, reverse-code any deliberately negative item into a new variable, then examine whether the five items appear to measure one construct. Only then calculate a documented mean or sum score.

For example, if the research question asks whether that score differs between two independent groups, a two-sample comparison might be appropriate after checking group sizes and assumptions. If the question concerns one individual item, an ordinal approach may be more defensible. Report the actual statistic, sample used, uncertainty and limits; do not describe a p-value as proof that a programme worked. Your university’s methods guidance and the study design take precedence over this illustration.

For implementation, keep a data dictionary with variable names, question wording, value labels, reverse-code rules and missing-value codes. Save the syntax or transformation log so the result can be reproduced. If you need an independent check of the design or outputs, see our statistical analysis consultation and share the research question for scope review. Labeeb can explain analysis and edit the report; the researcher remains responsible for the data and conclusions.

Questions

Likert questionnaire questions

Is Likert data ordinal or interval?+

A single Likert item is ordinal. Composite scores averaged across several items are widely treated as approximately interval, which is why t-tests, ANOVA and regression are commonly used on them. Justify your choice in the methodology chapter.

Should I use a 5-point or 7-point Likert scale?+

Both are common. Keep the format of the validated scale you are adapting where possible, and use the same format within each construct.

Do I need CFA if I use an existing validated scale?+

Not automatically. A validated scale may need fresh measurement checks after translation, adaptation or use in a different population. Whether CFA is appropriate depends on your question, model, sample and university requirements; discuss the analysis plan with your supervisor.

How do I reverse-code items in SPSS?+

Use Transform, Recode into Different Variables, and map each value to its opposite (on a 5-point scale: 1 to 5, 2 to 4, 3 to 3, 4 to 2, 5 to 1). Recode before running reliability or computing composite scores.

Next step

Want your questionnaire analysis checked?

We review questionnaire design, coding, reliability and validity, and run or check SPSS, AMOS and SmartPLS analyses so you can explain every result yourself.

See data analysis supportContact Labeeb