Hand someone a list of 20 features and a five-point importance scale, and a familiar thing happens: almost everything gets a 4 or a 5. Nothing is unimportant, so nothing stands out. The scale has no teeth — people don't pay a cost for saying yes to all of it.
MaxDiff (maximum difference scaling, also called best-worst scaling) fixes this by removing the option to like everything. In each small set, respondents must name the single best item and the single worst. Every answer forces a trade-off, and the trade-offs add up to a sharp ranking.
Note
How MaxDiff Works
The full item list is broken into overlapping subsets of four or five. For each subset, the respondent picks the best and the worst, and the design rotates so every item appears many times against different rivals.
| In this set, which is… | Item |
|---|---|
| Best | Faster support |
| (neither) | Mobile app · Dark mode |
| Worst | Custom themes |
Choosing best and worst from a set of four actually implies five of the six possible pairwise judgments at once — which is why MaxDiff extracts so much information from so few clicks.
Reading the Results
Aggregated across respondents and sets, the choices yield a ranking with real spacing between items — like the bars above.
| Output | What it tells you |
|---|---|
| Importance score (0–100) | Each item's relative importance on a common scale |
| Preference share | The share of choices an item would win |
| Net score | Times chosen best minus times chosen worst |
| Gap analysis | How far apart adjacent items really are |
Tip
What MaxDiff Doesn't Do
MaxDiff measures relative preference only. It tells you item A beats item B, and by how much, but not whether customers are satisfied with either in absolute terms, and not what they'd pay.
Watch out
Running a MaxDiff Study
- 1List the items to prioritize — features, messages, benefits (often 10–30).
- 2Let the design build balanced, overlapping best-worst sets automatically.
- 3Respondents pick the best and worst in each set.
- 4Aggregate into importance scores, preference share, and net scores.
- 5Read the ranking and the gaps to decide what makes the cut.
MaxDiff in the SKARI Survey Lab
SKARI's Survey Lab runs best-worst scaling end to end: it builds the balanced sets, collects the best-worst choices, and computes the full set of scores.
- Automatic best-worst set design from your item list
- Importance scores on a 0–100 scale
- Preference share and net scores per item
- Gap analysis showing how far apart adjacent items sit
Takeaway
Frequently Asked Questions
How many items can I test?
Typically 10–30. Each appears in several sets, so more items means more sets — balance coverage against respondent fatigue.
MaxDiff or a rating scale?
MaxDiff whenever you need to prioritize and ratings would cluster at the top. Ratings are fine when you genuinely need absolute levels.
MaxDiff or conjoint?
MaxDiff ranks standalone items; conjoint models trade-offs between bundled attributes and can price them. Use MaxDiff for a clean priority list.
Key Takeaways
Forces
Choice
best & worst
Yields
0–100
importance
Shows
Gaps
real spacing
Measures
Want
not satisfaction
MaxDiff turns a mushy rating survey into a ranking with backbone. By forcing a best and a worst in every set, it separates the items that matter from the ones that merely poll well.
Takeaway
Kano Model
Classify needs by satisfaction type
AHP
Weight named criteria by pairwise comparison
Conjoint Analysis
Trade-offs and willingness to pay