MaxDiff vs. rating scales: when to use best–worst scaling
Published November 16, 2026 · 6 minute read
Ask people to rate ten features from 1 to 5 and most will rate most of them “important”. MaxDiff forces trade-offs instead, which reveals what really matters most.
How MaxDiff works
Respondents see a series of small sets of items (typically four or five at a time) and choose the most and the least important in each set. Across the sets, every item appears several times alongside different others. Because people must choose, there is no room for “everything is important”.
When to use it
- Prioritising features, benefits, messages, service improvements or budget items.
- When rating scales come back flat, or when scale use differs between countries or groups.
- Lists of about 8 to 30 items. With fewer, a ranking question is simpler; with more, split the list or use a sparse design.
Designing the study
- Items per set: 4 is a good default (5 for long lists).
- Number of sets: enough for each item to appear about three times per respondent, so roughly 3 × items ÷ items per set.
- Use a balanced design: each item appears equally often and pairs of items appear together about equally often.
- Keep item wording short and parallel, so respondents compare the ideas rather than the phrasing.
Reading the results
The simplest score is best minus worst, divided by the number of times the item was shown, which runs from −100 (always least) to +100 (always most). It is easy to explain and usually agrees closely with model-based scores. For individual-level utilities and segmentation, analysts estimate a hierarchical Bayes or latent class model from the same data.
MaxDiff or conjoint?
MaxDiff ranks a list of separate items. Choice-based conjoint measures how combinations of attributes (price, contract length, support level) drive choice between whole products, and tells you how much each attribute matters. Use conjoint when the question is “which package would they choose?” and MaxDiff when it is “which of these matters most?”.