The 5-point Likert scale: wording, reverse items and how to analyse it

The Likert scale is the most common answer format in thesis questionnaires: a statement, then options from "strongly agree" to "strongly disagree". Its apparent simplicity hides decisions that shape your results: how many points, how the statements are worded, which ones are reversed, and how the means are interpreted. This article goes through them in order.

What is a Likert scale? {#what}

The American psychologist Rensis Likert introduced it in 1932 to measure attitudes. Respondents rate how much they agree with a set of statements, and the answers are summed or averaged into a score for the construct.

Researchers separate two things that are both called "Likert" in everyday use:

• A Likert item: one statement with its options.
• A Likert scale: several items that together measure one concept and are combined into one score.

The difference matters in analysis: a single item is ordinal data, while the sum of several items is treated in most studies as quantitative.

Five points or seven? {#points}

Five points is the usual choice in theses and is enough for most purposes. Choose seven when the published scale you are using is seven-point, or your supervisor asks for it.

What about the neutral option? Keeping it lets someone with no opinion say so honestly. Removing it (with a 4- or 6-point scale) forces a lean one way, but may push some people into an answer that does not represent them. Keeping it is more common.

Wording the statements {#wording}

• One idea per statement. "The lectures are useful and enjoyable" is two statements in one: what does someone choose who finds them useful but dull?
• Short and clear, in the respondent's language, not the researcher's.
• Don't lead. "Wouldn't you agree that..." pushes towards agreement.
• Avoid double negatives. "I don't think the lectures are not useful" confuses the reader.
• Balance the options: as many agreeing levels as disagreeing ones.

Reverse items: why you write them and why you reverse them {#reverse}

A reverse (negatively worded) item points the opposite way from the concept. In an academic engagement scale, "I miss lectures without a good reason" is a reverse item: agreeing with it means less engagement.

Why write them? To catch people who give the same answer to every statement without reading it, and to break the habit of agreeing automatically.

Why reverse them? Because a 5 on them means the opposite of a 5 on the other items. Before analysis their coding is flipped:

On a 5-point scale the new score is 6 minus the original. On a 7-point scale it is 8 minus the original.

Note forgetting to reverse a single item is enough to lower the Cronbach's alpha of the whole construct, and to make that item's correlation with its construct negative. See our article on Cronbach's alpha.

Don't overdo them: reverse items confuse some respondents, especially when written as negations. One or two per construct is usually enough.

Interpreting means: class width {#class-width}

Once you have the mean of every item and construct, you need a rule for judging it: is 3.4 high or moderate? A common method is the class width:

Class width = (highest score − lowest score) ÷ number of levels

For five levels: (5 − 1) ÷ 5 = 0.80

For three levels: (5 − 1) ÷ 3 ≈ 1.33

Both are in use. What matters is that you state in your thesis which one you used, and why, and apply it to every table.

Relative weight and rank {#relative-weight}

Relative weight turns a mean into a percentage of the highest possible score:

Relative weight = (mean ÷ highest scale point) × 100

An item with a mean of 3.9 on a 5-point scale has a relative weight of (3.9 ÷ 5) × 100 = 78%. Items are then ranked within each construct by their mean, highest first, so one table shows what respondents rate strongest and weakest.

Reporting the results {#reporting}

The usual table for each construct has: the statement, the mean, the standard deviation, the relative weight, the level and the rank, then a last row for the construct as a whole.

Under the table, write a reading of it rather than repeating its numbers: which items came first and last, and what that means for your research question. A large standard deviation deserves a mention, because it means respondents are divided on the item even when its mean is "moderate".

Are Likert data ordinal or interval? {#ordinal}

Some examiners will ask. A single item is ordinal by nature: the distance between "agree" and "strongly agree" is not necessarily the same as between "neutral" and "agree". But most studies take the mean of a construct made of several items and analyse it with parametric tests such as the t-test and ANOVA, after checking normality. When the data are far from normal, the alternatives are tests such as Mann-Whitney and Kruskal-Wallis.

In Pulseform {#pulseform}

• The Rating / Likert Scale question is 5-point, and you write the labels of points 1 to 5 yourself.
• This item is reverse scored flips the answer (6 minus the answer) in every analysis, so you never reverse it by hand.
• The Construct / Study Dimension field groups items into constructs, which construct scores and reliability are built on.
• The APA 7th descriptive table shows each item and construct with its mean, standard deviation, relative weight, level and rank. Its level uses three classes: Low up to 2.33, Medium up to 3.67, High from 3.68.

The details are in Create a survey and Academic surveys.

References {#sources}

The original sources of the formulas, thresholds and rules of thumb in this article:

• Likert, R. (1932). A technique for the measurement of attitudes. Archives of Psychology, 22(140), 1–55.
• Norman, G. (2010). Likert scales, levels of measurement and the "laws" of statistics. Advances in Health Sciences Education, 15(5), 625–632.
• Preston, C. C., & Colman, A. M. (2000). Optimal number of response categories in rating scales: Reliability, validity, discriminating power, and respondent preferences. Acta Psychologica, 104(1), 1–15.

Note class width and relative weight are common conventions in theses, not methods traced to one original source. So state plainly in your thesis which one you used.