Should you use MaxDiff or conjoint analysis?
TL;DR
- MaxDiff ranks items. Conjoint models trade-offs between attributes of a whole product.
- MaxDiff is quicker to design and answer; conjoint needs careful attribute and level design but can simulate market share and price effects.
- Conjoint entered marketing in 1971 with Paul Green and Vithala Rao's paper in the Journal of Marketing Research.
- A common sequence: MaxDiff to cut 20 features to 5, conjoint on those 5 plus price. Neon Blue can field both on a simulated audience. Run your first study free
Neon Blue builds simulations of human behavior. Use MaxDiff when you need to rank a list of separate items, such as features, claims or benefits, by how much people value them. Use conjoint when you need to know how people trade off attributes that come together in one product, especially price. MaxDiff answers "which of these matters most?". Conjoint answers "which combination would people pick, and what would they give up to get it?". If price is part of the decision, you almost always want conjoint, or both in sequence.
Where did conjoint analysis come from?
In 1971, Paul Green and Vithala Rao published "Conjoint Measurement for Quantifying Judgmental Data" in the Journal of Marketing Research. They brought a method from mathematical psychology into marketing: instead of asking people how much each attribute matters, show them whole product profiles and infer the value of each attribute from their choices. That inference is still the core of every conjoint study. Best-worst scaling, the method behind MaxDiff, came later; its first marketing publication, by Adam Finn and Jordan Louviere, appeared in 1992. Its use in message work is covered in message testing survey questions.
How do MaxDiff and conjoint compare?
| MaxDiff (best-worst scaling) | Choice-based conjoint | |
|---|---|---|
| What people see | Sets of 4 or 5 items | Two to four full product profiles |
| What they do | Pick the most and least important | Pick the one they'd buy, or none |
| Output | Ratio-scaled preference for each item | Part-worth value for each level of each attribute |
| Handles price? | Poorly; price is not an item | Yes, and yields willingness to pay per attribute |
| Can simulate market share? | No | Yes, against defined competitors |
| Design effort | Low: a list of items | Higher: attributes, levels, prohibitions |
| Typical use | Features, claims, benefits, messages | Product configuration, pricing, packaging tiers |
The short rule: if the items could all be true at once (several claims on one pack), MaxDiff. If the items are levels of the same thing (12 or 24 hours battery), conjoint.
When is MaxDiff the right call?
- You have a long list (10 to 30 items) and need a ranked shortlist.
- The items don't combine into one product: claims, reasons to buy, perks.
- You need a result the whole team can read without a model. A ranked bar chart is hard to misread, which helps when the product marketing team has to defend the list.
Feature lists are the most common case; the feature prioritization guide pairs MaxDiff with an importance and satisfaction gap.
When is conjoint the right call?
- Price is one of the levers.
- You need to know what people would give up: a shorter battery for a lower price, fewer colors for a faster ship date.
- You want to simulate what share you'd take against named alternatives at different prices, the question behind a go or no-go call.
For price questions without a full product design, simpler methods work too; see willingness to pay methods and measuring willingness to pay.
What does running both look like on a simulated audience?
Hypothetical: a consumer electronics brand is planning a new pair of wireless earbuds and has 16 candidate features plus three price points in mind.
- Audience: US adults who bought wireless earbuds in the past two years, split by gym and commute users, heavy and light phone-call users, and people who paid under or over the category midpoint.
- Stage 1 (MaxDiff): all 16 features in sets of 4, on a simulated audience of earbud buyers.
- Stage 2 (conjoint): the top 5 features as attributes with 2 or 3 levels each, plus price, against two current market options described generically.
Questions fielded:
- Of these four features, which matters most to you in earbuds, and which matters least?
- If these were the only options available, which would you buy? [Three profiles plus "none"]
- Which of these would you give up first to get a lower price?
- What did you picture using these for most?
Example output layout (illustrative, not a real result)
| Feature (from MaxDiff) | Preference share | Carried to conjoint? | Value per level (from conjoint) |
|---|---|---|---|
| Active noise cancelling | __% | Yes | Off / basic / strong: __ / __ / __ |
| Battery life | __% | Yes | 6h / 8h / 10h: __ / __ / __ |
| Sweat resistance | __% | Yes | Yes / no: __ / __ |
| Spatial audio | __% | No |
The conjoint stage then shows how share shifts at each price. That is the input the team needs for setting the price for a new product, and the segment cuts show which buyers carry the premium tier.
Questions about MaxDiff and conjoint
Is MaxDiff a type of conjoint?
They come from the same family of choice models, and some people call MaxDiff "best-worst conjoint". In practice they answer different questions: MaxDiff ranks separate items, conjoint values combinations.
How many respondents does a conjoint study need?
It depends on the number of attributes, levels and segments. Many studies plan at least a few hundred per segment you want to read. MaxDiff can work with smaller samples for a total ranking.
Can you include price in a MaxDiff?
Not well. Price isn't an item people weigh against features one at a time. Use conjoint, or a separate price ladder such as Van Westendorp.
Can Neon Blue run MaxDiff and conjoint studies?
Yes. Neon Blue fields MaxDiff tasks and choice-based conjoint profiles on a simulated audience of your buyers and returns rankings and level values by segment, labeled as simulated. Start a free first study.