MaxDiff / Best-Worst Scaling
Also known as: MaxDiff, Best-Worst Scaling, BWS, Maximum Difference Scaling
MaxDiff, also known as best-worst scaling (BWS), measures the relative importance or preference of a set of items by repeatedly asking respondents to identify the best (most important or most preferred) and worst (least) item within small subsets. Introduced by Jordan Louviere and formalized by Marley and Louviere's 2005 probabilistic models, the method exploits the fact that people are far better at picking extremes than at rating many items on a scale. Each best-worst judgment reveals the maximum-difference pair in a set, and across many balanced subsets the choices pin down a single interval scale of item utilities. Because every respondent is forced to make trade-offs, MaxDiff sidesteps the scale-use bias and lack of discrimination that plague rating grids, where respondents often call everything important. Item scores can be computed by simple best-minus-worst counts or, more rigorously, by fitting a multinomial logit choice model, with hierarchical Bayes yielding individual-level, probability-scaled importances. The result is a clear, discriminating ranking of items that supports prioritization, segmentation, and feature selection.
Key highlights
- Forces trade-offs, eliminating the scale-use bias and lack of discrimination common in rating grids.
- The best-worst task is simple and intuitive, and choosing both extremes yields more information than a best-only choice.
- Produces a single interval scale of item importance, with hierarchical Bayes giving interpretable individual-level, probability-scaled scores.
- Robust across cultures and languages because it asks for relative judgments rather than absolute ratings.
Intuition
This section is available to Pro members. Upgrade to Pro
How it works
This section is available to Pro members. Upgrade to Pro
When to use it
Use MaxDiff when you need to prioritize or rank a list of items, features, benefits, messages, attributes, by relative importance or preference, and you want to avoid the flat, undiscriminating data that rating scales typically produce. It excels when there are too many items to compare all at once but few enough to cover in a balanced design, and when forcing trade-offs is desirable. MaxDiff is not the tool when you need absolute judgments rather than relative ones, when items are not comparable on a single dimension of preference, or when you must model how features combine into whole products with price, which calls for choice-based conjoint instead. It also presumes respondents can meaningfully choose extremes within each set, so items should be concrete and distinct.
Strengths & limitations
- Forces trade-offs, eliminating the scale-use bias and lack of discrimination common in rating grids.
- The best-worst task is simple and intuitive, and choosing both extremes yields more information than a best-only choice.
- Produces a single interval scale of item importance, with hierarchical Bayes giving interpretable individual-level, probability-scaled scores.
- Robust across cultures and languages because it asks for relative judgments rather than absolute ratings.
- Measures only relative importance among the items included, not absolute importance or value outside the set.
- Requires a balanced design and enough tasks to cover all items, which can lengthen the questionnaire when the item list is long.
- Cannot model how items combine into products or interact with price, so it does not replace conjoint for product configuration.
- Counting scores ignore set composition effects that the choice model accounts for, so naive counts can mislead with unbalanced designs.
Common pitfalls
This section is available to Pro members. Upgrade to Pro
Applications
This section is available to Pro members. Upgrade to Pro
Frequently asked
Why is MaxDiff better than a simple importance rating scale?
Rating scales let respondents call many items important at once, producing flat, undiscriminating data and inviting scale-use and acquiescence biases that differ across people and cultures. MaxDiff forces a trade-off in every task by requiring a single best and single worst from each set, so respondents must reveal which items truly matter more. As Marley and Louviere showed, the best-worst response also captures more information than a best-only choice because it constrains both ends of the ordering. The payoff is a sharply discriminating, comparable interval scale of item importance rather than a clump of high ratings.
What is the difference between counting analysis and the logit model?
Counting analysis computes each item's score as the number of times it was chosen best minus the number of times chosen worst, divided by appearances; it is transparent and great for communication. The logit (maximum-difference) model instead estimates item utilities by maximizing the likelihood of the observed best-worst choices under the model Marley and Louviere derived, properly accounting for which items competed in each set and supporting statistical inference. With a balanced design the two agree closely, but the choice model is preferred for individual-level hierarchical Bayes estimation and whenever set composition might bias raw counts.
Can MaxDiff tell me the absolute importance of an item?
No. MaxDiff yields relative importance among exactly the items you include; an item's score depends on the company it keeps. Adding, removing, or changing items shifts the scores, and the scale has no natural zero of 'unimportant.' This is why MaxDiff is ideal for prioritization but cannot, on its own, say how much someone would pay for a feature or whether any item clears an absolute threshold. For absolute or monetary judgments, anchored MaxDiff variants or choice-based conjoint with a price attribute are the appropriate tools.
Sources
- 1.Louviere, J. J., Flynn, T. N., & Marley, A. A. J. (2015). Best-Worst Scaling: Theory, Methods and Applications. Cambridge: Cambridge University Press.
- 2.Marley, A. A. J., & Louviere, J. J. (2005). Some probabilistic models of best, worst, and best-worst choices. Journal of Mathematical Psychology, 49(6), 464-480.
- 3.Orme, B. K. (2020). Getting Started with Conjoint Analysis: Strategies for Product Design and Pricing Research (4th ed.). Madison, WI: Research Publishers LLC.ISBN 9780972729772
You have read it. What now?
Cite this page
ScholarGate. (2026, June 23). MaxDiff / Best-Worst Scaling. ScholarGate. https://scholargate.app/marketing-research/maxdiff-best-worst-scaling