Likert-scale items with 5 categories would typically be treated as ordinal variables. Ordinal variables, by definition, are discrete and cannot be normally distributed. This shouldn't be a problem as long as you apply statistical techniques that are appropriate for ordinal data.

It would be different if you used the sum (or average) score across multiple such items (i.e., an item composite/sum score). Such composite scores are often treated as (quasi-)continuous (metrical/interval scale) variables, and normality assumptions may be relevant (depending on which statistical technique you are planning to employ).

Christian Geiser and I have both commented on this issue a number of times. I personally advocate what is the most widely used technique for dealing with a number of Likert-scored items, which is to create a scale (or set of scales) by adding together the relevant items.

This process begins by using Coefficient Alpha to determine whether your variables can be represented by a single scale. Note that all you items must be scored in the same direction so that you have all positive correlations.

If the score on Coefficient Alpha is less that .7, then you should conduct an Exploratory Factor Analysis to find out if the items represent two or more factors.

Yes, correlating and adding together does violate the ordinal nature of Likert-scored data, but the reality is that this gets done all the time in peer-reviewed publications. Other approaches that attempt to rely directly with original measurement techniques generate much more complicated analyses. So, this is a case where the "perfect" is the enemy of the "good," Ot, perhaps the "good enough."

You ask an important question:

Does a Likert Scale Questionnaire require a Normality Test?

I have been working with case studies in survey analysis for almost half a century, and my method remains relevant. I always begin with data observation, where the initial phase focuses on examining and visualizing the data.

You may find it useful to review my two articles (see attached).

Short Answer:

• No, a normality test is not strictly required for Likert scale data.
• Spearman's correlation is indeed appropriate and widely accepted for analyzing non-parametric ordinal data like Likert responses.
• In fact, many statisticians argue against applying normality tests to Likert scale data at all because these data do not meet the assumptions needed for parametric testing in the first place.

Why You Generally Don’t Need a Normality Test for Likert Data:

Nature of the Data:

Likert scales (e.g., 1–5) produce ordinal data—ranked but without guaranteed equal intervals. Normality tests are used for continuous (interval or ratio) variables—so applying a Shapiro-Wilk or Kolmogorov-Smirnov test to Likert items may be statistically inappropriate.

Distribution Visuals:

It’s common for Likert items to show non-normal distributions (e.g., skewed toward agreement or disagreement), especially on polarizing topics.

Best Practices:

Instead of forcing Likert data into parametric assumptions, researchers often use non-parametric tests like:

• Spearman’s rho (✓ for correlation)
• Mann-Whitney U or Kruskal-Wallis (✓ for group comparisons)
• Chi-square (✓ for frequencies)