Likert scoring treats attitude as one continuous dimension, which complex attitudes often are not

What it is

A Likert scale pairs an attitude statement with ordered options and often sums several items as a position on one latent dimension. That practice assumes attitude is single and continuous: people can be lined up from more agree to more disagree, and the items are interchangeable indicators of that line. Complex attitudes often are not. Someone can like a product’s look and refuse its permissions at the same time, or jump between “usable / not usable” without a meaningful middle. Forcing those into one sum yields an average that matches no psychological object.

Why it happens

Sums and means require items to indicate the same construct interchangeably, and intervals between options to be at least roughly equal. If items actually sit on separate axes—safety, aesthetics, efficiency—high and low scores cancel, and a middling total can come from entirely different mixes. Treating ordered options as a continuous quantity then treats adjacent gaps as equal, while the subjective distance from “slightly agree” to “agree” is often not the distance from “agree” to “strongly agree.” When attitude is a type rather than a degree (trust / distrust, grant / refuse), a continuous dimension paints category boundaries as a gradient, and later correlations treat that false continuum as real variance.

Studying it

Test unidimensionality before forming a total: exploratory or confirmatory factor analysis, parallel analysis, item–total correlations. If several factors stay, score them separately; do not publish one “attitude.” Cognitive interviews on doubtful items ask whether options are heard as degrees or as different objects. When attitude may be typological, compare continuous scoring with categorical models (latent class, threshold models) against the response pattern. A scale report should state why a single dimension was assumed, and how scoring changes if that reason fails.

Where it stops holding

Short, highly specific constructs already shown to be unidimensional—effort on one task, for example—can reasonably use a Likert format. Unidimensionality does not grant equal intervals, and equal intervals do not grant measurement equivalence across groups. Likert-formatted answers can still be read as distributions and as shares choosing extremes without ever being summed. Switching to semantic differentials or ranking changes the scale’s assumptions; it does not automatically get closer to the truth, and the new format has dimensional premises of its own.

Applying it

• Write down how many independent axes the attitude has; if more than one, write separate items and do not pre-sum them.
• After a pilot, check whether high scorers’ spoken explanations point at the same thing; if someone says appearance pulled the score up and permissions pulled it down, split the scoring.
• Lead results with per-axis scores and the share choosing extremes; do not let a single total stand as the full attitude.
• Check: drop any one item and see whether rank order on the total moves sharply; if it does, the total is not a stable dimension.

Related

• Same group: Q3.15.2 Option order produces primacy and recency · Q3.15.3 Mixed reverse items detect careless answering at a cognitive cost · Q3.15.4 Cross-language scales need back-translation, not a word-for-word rendering
• Adjacent: Q3.01 Questionnaire design · Q3.16 Single-item and short scales
• Search terms: Likert scale · unidimensionality · attitude structure