Likert Scale Analysis: 7 Do's and Don'ts for Your Thesis

Q: Can I calculate the mean of a single Likert item?

You can, and many theses do, but it is an ordinal variable, so the mean has no clear interpretation and should be accompanied by the median and the percentage distribution. For inferential tests on single items, use non-parametric methods.

Q: Is a 5-point or a 7-point scale better for analysis?

Seven points produce composite scores with more distinct values and slightly higher reliability; five points are easier to answer on a phone. Either works for parametric analysis at the scale level. If you adapt an existing scale, keep its original number of points so results remain comparable.

Q: What if my composite score is not normally distributed?

Check how severe the departure is. Mild skewness with groups of about 30 or more is tolerated by t tests and ANOVA (Norman, 2010). With severe skewness, small groups or ceiling effects, use the non-parametric alternative and report medians, or run both and show that they agree.

Q: Should I include a neutral midpoint?

Including it lets genuinely undecided respondents answer honestly; removing it forces a direction and can inflate agreement. Both are defensible if you state the choice. Do not code "don't know" as the midpoint; treat it as missing.

Q: Can Likert composites be used in regression or SEM?

Yes. Composite scores are routinely used as continuous predictors and outcomes in regression, and item-level data are used as indicators of latent variables in confirmatory factor analysis and structural equation modeling, where estimators for ordinal indicators are available if needed.

By the Mastermind PhD team · Last updated: 3 November 2026 · Editorial policy

Can you calculate a mean of “strongly agree”? Ask three statisticians and you may get three answers, which is why the Likert question keeps appearing in viva reports. The disagreement is smaller than it looks. Almost all of it comes from mixing up two different objects: a single Likert-type item, which is ordinal, and a Likert scale built from several items, which behaves closely enough to an interval measure for the usual statistics. This guide separates the two, summarizes what the main contributors to the debate actually wrote, and then gives the practical sequence of Likert scale analysis: coding, reverse coding, reliability, composite scores, interpretation bands, test selection and reporting.

This article is part of our guide How to Write a Master’s Thesis by the Mastermind PhD team and relates to our Statistical Analysis service. Sources last checked: November 2026.

What is a Likert scale? A Likert scale, named after Rensis Likert (1932), is a measurement instrument composed of several statements to which respondents indicate their degree of agreement on a symmetric set of response options, typically five or seven. The individual statement with its response options is a Likert-type item; the scale is the sum or mean of the items that measure one construct. The distinction matters because the item yields ordered categories while the scale yields a score with many possible values.

Likert scale analysis: What is a Likert scale? — In short – the Mastermind PhD team

Table of Contents

Item or scale? The distinction that decides everything

Feature	Likert-type item	Likert scale (composite)
What it is	One statement, one response from 1 to 5 (or 1 to 7)	Sum or mean of 4 or more items measuring one construct
Level of measurement	Ordinal: the distance between “agree” and “strongly agree” is unknown	Approximately interval: many possible values, near-continuous distribution
Central tendency	Median and mode; mean reported with caution	Mean
Dispersion	Interquartile range, frequency of each category	Standard deviation
Display	Frequencies and percentages per category, stacked bar chart	Mean and SD, histogram
Tests	Non-parametric (Mann-Whitney, Kruskal-Wallis, Spearman, chi-square)	Parametric (t test, ANOVA, Pearson, regression) when assumptions are reasonable

If your research question is about a construct (job satisfaction, attitude to e-learning), analyze the composite. If it is about one specific statement (“I would recommend this hospital”), analyze the item as ordinal data.

What did Jamieson, Norman and Carifio and Perla actually say?

Jamieson (2004) warned that Likert data are ordinal, that means and standard deviations are therefore inappropriate, and that non-parametric tests should be used. Her short paper is correct about single items, and it is still the most cited piece on the topic.

Carifio and Perla (2007, 2008) replied that the debate confuses the response format with the scale. They argued that a Likert scale made of several items can be analyzed with parametric methods, that the F test in particular is robust to the kind of departure Likert data show, and that analyzing individual items is itself a methodological mistake when the construct is what matters.

Norman (2010) reviewed the simulation evidence and concluded that parametric statistics give correct conclusions with Likert data even with small samples, unequal variances and non-normal distributions, because the tests are robust to these violations; he regarded the “ordinal, therefore non-parametric” rule as unsupported by the evidence for scale scores.

The practical consensus in methods teaching, which Sullivan and Artino (2013) summarize for health education research, is therefore: treat items as ordinal, treat multi-item scales as interval, run the parametric test on the scale, and report descriptive statistics appropriate to each level.

📱 Ask a statistician about your data — Send a short description of your instrument and sample size on WhatsApp and get a suitable analysis plan. Every analysis comes with the raw SPSS file and a results report.

What must you do before any Likert scale analysis?

Code consistently. Strongly disagree = 1 through strongly agree = 5. Enter the labels in SPSS Variable View or as factor levels in R so that tables print words, not numbers.
Reverse-code negatively worded items. For a 5-point item the new value is 6 minus the old value (1 becomes 5, 2 becomes 4); for a 7-point item it is 8 minus the old value. In general: (k + 1) − x, where k is the number of points. Do this in a new variable and keep the original.
Handle missing items. Decide the rule before you look at the data, for example compute a composite only when at least four of five items are answered, and report how many cases that excluded.
Check internal consistency. Compute Cronbach’s alpha for each construct; a value of .70 or higher is the common threshold for research use (Nunnally & Bernstein, 1994). If an item’s deletion raises alpha noticeably and the item correlates poorly with the rest, consider removing it and say so.
Check the distribution of the composite. Look at the histogram and the skewness statistic for each group. A moderately skewed composite with groups of about 30 or more is acceptable for t tests and ANOVA (Norman, 2010); with severe skewness or very small groups, use the non-parametric alternatives or report both.

How do you build the composite score?

Two options exist, and the mean is usually better for a thesis.

Method	Example (five items: 4, 5, 3, 4, 4)	Advantage	Drawback
Sum	20 (range 5 to 25)	Traditional Likert scoring; more distinct values	The range depends on the number of items, so sections cannot be compared
Mean	4.00 (range 1 to 5)	Stays on the original 1 to 5 metric; sections comparable; missing items handled easily	Fewer distinct values with few items

In SPSS use Transform > Compute Variable with MEAN(item1 TO item5), or MEAN.4(item1 TO item5) to require at least four answered items. In R, rowMeans(df[, items], na.rm = TRUE) does the same without the minimum; add a count of non-missing items if you need the rule.

Likert scale: what this guide covers - Item or scale? The distinction that decides everything، What did Jamieson, Norman and Carifio and Perla actuall — What this guide covers – the Mastermind PhD team

Which interpretation bands should you use?

In Likert scale analysis, interpretation bands turn a mean into a verbal level. The most common convention for a 5-point scale divides the range of 4 (from 1 to 5) into five equal classes of 0.80.

Mean score	Level (five bands)	Mean score	Level (three bands)
1.00 to 1.80	Very low	1.00 to 2.33	Low
1.81 to 2.60	Low	2.34 to 3.67	Moderate
2.61 to 3.40	Moderate	3.68 to 5.00	High
3.41 to 4.20	High
4.21 to 5.00	Very high

Bands are a presentation convention, not a statistical test. State which one you use and why in the methodology chapter, apply it consistently, and never use it as a substitute for a comparison test: two means of 3.9 and 4.3 fall in different bands, but only a t test can say whether they differ beyond chance.

Which test fits which question?

Research question	Data analyzed	Test when analyzing single items (ordinal)	Test when analyzing composite scores (interval)
Is the level above or below the midpoint (3)?	One variable vs a constant	One-sample Wilcoxon signed-rank	One-sample t test against 3
Do two groups differ?	Two independent groups	Mann-Whitney U	Independent-samples t test (Welch if variances differ)
Do three or more groups differ?	Three or more independent groups	Kruskal-Wallis H, then pairwise comparisons with correction	One-way ANOVA with Tukey or Games-Howell
Did scores change after an intervention?	Two related measures	Wilcoxon signed-rank	Paired-samples t test
Are two constructs related?	Two variables	Spearman’s rho	Pearson’s r
Does X predict Y?	Predictors and an outcome	Ordinal logistic regression	Linear regression

Report an effect size with every test in your Likert scale analysis: r or d for two-group comparisons, eta squared for ANOVA, and the correlation coefficient itself for relationships; our article on effect size explains the small, medium and large conventions.

How do you report Likert results in APA style?

For items, a table with the percentage choosing each option, the median or the mean, and the item’s rank reads well and lets a reader see the distribution. For composites, report the mean, SD, number of items and alpha, then the test. A results sentence for a composite: “Job satisfaction was high (M = 3.84, SD = 0.62, 16 items, α = .88). Nurses on night shifts scored lower than nurses on day shifts, t(214) = 2.71, p = .007, d = 0.37, a small to medium effect.” A sentence for a single item: “Most respondents agreed or strongly agreed that handovers were clear (71%); the median response was 4 (agree).”

Avoid p = .000 (write p < .001), avoid reporting the mean of a single item to three decimals, and avoid describing a 0.2-point difference as “large” without a test and an effect size.

Likert scale analysis do’s and don’ts at a glance

Do	Don’t
Reverse-code negatively worded items before computing anything	Average items with mixed directions
Compute a composite per construct and report alpha	Report alpha for the whole questionnaire across unrelated constructs
Use the mean (1 to 5) for composites so sections are comparable	Mix sums and means across sections
Report medians and percentages for single items	Report a mean of 3.47 for one item and call it “interval”
Choose the test by the level of the variable you analyze	Choose the test by habit or by whichever gives p < .05
State your interpretation bands once and apply them consistently	Invent new bands to make a result look “high”
Report effect sizes and confidence intervals	Report p-values alone

📱 Ask on WhatsApp now — 5 years of experience across all disciplines and 14 academic services. Two-stage payment (80/20); revisions are free until supervisor approval, within 30% of the chapter.

Frequently asked questions

Can I calculate the mean of a single Likert item?