MaxDiff vs conjoint analysis: what's the difference and when should you use each?
A side-by-side comparison, a decision table and a worked design example for each method.
MaxDiff ranks one list of items, such as features, messages or claims, by asking people to pick the best and worst from small sets. Conjoint analysis measures how attributes and their levels, often including price, trade off inside whole product profiles. Use MaxDiff to prioritise a list; use conjoint when the decision is a combination of features at a price.
How do MaxDiff and conjoint analysis differ?
MaxDiff puts items from one list in competition with each other; conjoint builds products from several attributes and asks people to choose between whole products.
Sawtooth Software, which makes software for both, describes MaxDiff as measuring "the relative importance or preference of a list of items" and conjoint as quantifying how people value multiple attributes of a product. Conjoint analysis has been applied in consumer research since 1971 (Green and Srinivasan, 1978); MaxDiff was invented by Jordan Louviere in 1987 (Sawtooth manual).
| MaxDiff | Conjoint (choice-based) | |
|---|---|---|
| What people see | Sets of usually four or five items from one list | Two or more product profiles, each a mix of attribute levels |
| What they do | Pick the best and the worst item in each set | Choose the profile they prefer, often with a "none" option |
| What you get | A score for each item on one relative scale | A utility for every level, the importance of each attribute and a market simulator |
| Price | Can't be traded off; at most an item on the list | Usually an attribute, so you can estimate willingness to pay |
| Typical scope | Typically 12 to 50 items | A handful of attributes, each with a few realistic levels |
| Respondent effort | Lower: short sets, simple choices | Higher: whole products to compare |
| Main limit | Scores are relative to the list tested | Results depend on the attributes and level ranges chosen |
When should you use MaxDiff and when conjoint?
Start from the decision: if you are choosing items from a list, use MaxDiff; if you are designing a product or a price, use conjoint.
| Your question | Use | Why |
|---|---|---|
| Which 5 of these 25 features matter most? | MaxDiff | One list, and you need separation that ratings won't give |
| Which three messages should lead the campaign? | MaxDiff | Messages compete for the same slot |
| Which plan should we sell, with which features, at which price? | Conjoint | Features and price trade off inside one product |
| How much more will people pay for faster delivery? | Conjoint, with price as an attribute | Willingness to pay comes from the price trade-off |
| What share might we win against a rival at $40? | Conjoint | Only conjoint feeds a market simulator |
| The list has 40 ideas and conjoint can only take a few | MaxDiff first, then conjoint on the top items | MaxDiff cuts the list cheaply |
| One product, and only the price is open | Neither | Use Van Westendorp or Gabor-Granger |
| You need to know why people chose | Either, plus open questions or interviews | Both measure what wins, not why |
What does a typical design look like for each?
A MaxDiff of 20 items needs at least 15 sets per person; a conjoint is usually sized at around 300 respondents.
MaxDiff, 20 features. Sawtooth's design guidance is to show four or five items per set, never more than half the list, and enough sets that each item appears three to five times per respondent. The minimum number of sets is 3K ÷ k, where K is the number of items and k the items per set. For 20 items shown four at a time that is 3 × 20 ÷ 4 = 15 sets; with five per set it is 12. Showing 20 sets of four puts each item in front of each person four times.
Conjoint, a software plan. Say five attributes (price, storage, support, seats, contract length), the largest with five levels, ten choice tasks of four profiles each. Sawtooth's rules of thumb give three checks:
- Each level should appear at least 500 times: n × t × a ÷ c ≥ 500, where t is tasks, a profiles per task and c the most levels in any attribute. Here n ≥ 500 × 5 ÷ (10 × 4), about 63 respondents. The post calls 500 the bare minimum and advises planning for 1,000, which gives 125.
- As a default, start with 300 respondents.
- Plan at least 200 for every subgroup you will report on its own, so three segments need about 600.
The formula is a floor, not a target: in this example the 300 default and the subgroups set the sample.
What mistakes weaken each method?
Most failures come from the list or the levels, not the statistics.
- MaxDiff: mixed or vague items. Features, benefits and values on one list make choices hard to interpret. Keep items distinct and at the same level of detail.
- MaxDiff: reading scores as absolute. A low-ranked item lost to the others on this list; it may still matter. MaxDiff tells you the order, not whether anything clears a bar.
- Conjoint: unrealistic level ranges. An attribute looks more important when its levels span a wider range, so use ranges buyers would actually meet.
- Conjoint: a missing attribute. Leaving out something that drives the real choice distorts the importance of what remains.
- Both: no pilot. Run the exercise with a few people first to check they understand the items and can finish without fatigue.
PulseLake's articles on MaxDiff without the math and conjoint analysis for product decisions cover design and interpretation in more depth. In PulseLake, both run as traditional research, alongside the study's objectives, method, evidence and decisions.
Frequently asked questions
Is MaxDiff a type of conjoint analysis?
Loosely, yes. Sawtooth Software describes MaxDiff as a one-attribute conjoint with often about 12 to 50 levels. It is also called best-worst scaling, and Sawtooth's documentation credits Jordan Louviere with inventing it in 1987.
Can a MaxDiff study include price?
Only as an item, such as "a lower monthly price", which tells you how a price cut ranks against other benefits. It cannot tell you how much people would pay. For that, use conjoint with price as an attribute, or a pricing method such as Gabor-Granger.
Can you run MaxDiff and conjoint in the same survey?
Yes, and a common sequence is MaxDiff first to cut a long list, then conjoint on the shortlist. Both exercises take effort, so pilot the combined survey and watch its length and drop-off.
Which method is cheaper?
MaxDiff, usually. Sawtooth's comparison rates conjoint as more complex and costlier with longer timelines, because the attributes, levels and design need more work up front and the analysis has more parts.
Sources
Sources for the facts on this page, last checked October 9, 2026.
- Sawtooth Software, What is the difference between MaxDiff and conjoint analysis? (updated 22 October 2024) checked October 9, 2026
- Sawtooth Software, Lighthouse Studio manual: designing a MaxDiff study checked October 9, 2026
- Sawtooth Software, Lighthouse Studio manual: what is MaxDiff? checked October 9, 2026
- Sawtooth Software, Sample size rules of thumb for choice-based conjoint (updated 29 December 2020) checked October 9, 2026
- Green and Srinivasan, Conjoint Analysis in Consumer Research: Issues and Outlook, Journal of Consumer Research 5(2), 1978 checked October 9, 2026
Related guides
- MaxDiff analysis without the math
- Conjoint analysis for better product decisions
- Gabor-Granger vs Van Westendorp: which pricing method should you use?Van Westendorp finds a range of acceptable prices; Gabor-Granger finds the price that earns most within a set you choose. Worked examples of both, with numbers.
- What questions should you ask in a concept test?A concept test asks the same questions about every concept: appeal, relevance, uniqueness, believability, purchase intent, price and why. A ready-to-use set.
Run research end to end. Keep the knowledge working.
One AI-native operating system for market research and insight professionals — from study design and evidence generation to agents, institutional knowledge, delivery and action.