Semantic differential scale (SDS)

A semantic differential scale asks people to rate a brand, product or idea on a row of points between two opposite adjectives, such as "cheap ... premium" or "boring ... exciting". It measures the connotative meaning of a concept, which makes it a standard tool for brand image, product perception and attitude research.

The method comes from Charles Osgood, George Suci and Percy Tannenbaum, who set it out in The Measurement of Meaning (1957). Their original scales had seven points between each pair of adjectives.

What a semantic differential item looks like

The respondent sees the concept once and then a list of adjective pairs:

How would you describe Brand X?
Unreliable ○ ○ ○ ○ ○ ○ ○ Reliable
Modern ○ ○ ○ ○ ○ ○ ○ Old-fashioned
Weak ○ ○ ○ ○ ○ ○ ○ Strong
Slow ○ ○ ○ ○ ○ ○ ○ Fast

Only the ends carry words. The points in between show distance from each pole, and the middle point means "neither" or "equally both".

The three dimensions: evaluation, potency, activity

Osgood and colleagues ran factor analyses on ratings of many concepts and found that most adjective pairs cluster into three dimensions, known as EPA:

In applied research you rarely need all three. Brand studies usually mix evaluative pairs with category-specific ones such as "basic–premium" or "for experts–for beginners".

How to score the answers

1. Code the points. Use 1 to 7, or −3 to +3 if you want zero in the middle.
2. Align the direction. To limit position bias, good designs put the positive pole on the left for some pairs and on the right for others. Before averaging, reverse the pairs where the positive pole is on the left: on a 1–7 scale the new score is 8 minus the old one.
3. Average per pair. The mean of each pair across respondents gives the profile of the concept.
4. Combine pairs into indices if they hang together. Average the pairs of one dimension only after checking that they correlate, for example with Cronbach's alpha or factor analysis.

Worked example: two brands

An illustrative study (not real data) asks 150 people to rate two brands on four pairs, scored 1–7 after alignment, where 7 is the positive pole:

Plotted as two lines across the pairs (the "snake plot"), the profile shows Brand A as reliable and strong but dated, and Brand B as modern and fast but less trusted. That is the typical use: the shape of the profile matters more than any single number. Because the same people rated both brands, the ratings are paired. Before you claim a gap such as 5.9 against 4.8, take each respondent's difference between the brands on that pair and test whether the mean difference is zero with a paired t-test.

Semantic differential, Likert and rating scales

A Likert scale gives a statement ("Brand X is reliable") and asks how much you agree. A semantic differential gives two opposite words and asks where the concept sits between them, so there is no statement to agree with and less room for acquiescence bias. A general rating scale scores one attribute from low to high and may label every point. All three produce ordinal data per item; see scale for the wider family.

When to use it and common mistakes

Use a semantic differential for brand and product image, comparing a concept with competitors, tracking perception before and after a rebrand, and rating names, logos or packaging.

• Pairs that are not opposites. "Friendly–professional" forces a false choice. Test each pair by asking whether the two words exclude each other.
• All positive words on one side. People then tick down one column. Alternate the poles and reverse-score later.
• Too many pairs. Long grids invite identical answers down the page. Keep only the pairs you will analyze.
• Jargon in the adjectives. Every respondent must read both words the same way, and in multi-country studies each pair needs a natural equivalent in every language.
• Adding up unrelated pairs. An index built from pairs that do not correlate mixes different things.
• Reading the midpoint as "neutral" by default. It can mean "both", "neither" or "don't know". Offer a separate "can't say" if that matters.

Building a semantic differential in a survey

SurveyNinja has a Semantic differential question type built for adjective pairs (from the Basic plan). You can also build the scale with the Opinion scale question, a numbered scale with labels at the ends: write the negative adjective as one end label and the positive one as the other, and use the same number of points for every pair. The Opinion scale is on the Free plan. Put the concept (an image or a short description) right before the pairs, alternate which side the positive word is on, and export the answers to CSV, XLSX or SPSS on any plan for scoring.

For labels and point counts, see the rating scale examples. The t-test calculator compares two independent groups, so use it when different people rated each brand; when the same people rated both, run a paired t-test on the per-respondent differences. Check an index with the Cronbach's alpha calculator.