Conjoint Analysis

Why this matters for your decision

Many offers consist of many levers: scope of cover, term, deductible, features, price. Conjoint analysis is strong at weighting these levers at the same time. It shows which attribute brings how much preference and allows simulations: what happens to the share if attribute X is dropped and the price falls?

This makes it well suited to filtering the few sensible combinations out of a very large number of possible ones. For the decision between these final variants, a different question arises: do people also choose them when it is not a questionnaire?

What research shows about accuracy

A comparison with real purchase data reached a nuanced result: incentive-compatible methods, in which the choice has real consequences, performed best. Classic, hypothetical conjoint showed hypothetical bias but could still deliver usable demand curves. In general, hypothetically stated willingness to pay is on average above the amount measured in real conditions.

This is not an argument against conjoint, but for its proper role: weighting and narrowing down in the questionnaire, checking the final variants through behaviour.

Example

An insurer is planning a supplementary dental tariff with five attributes, each with three levels. The conjoint analysis shows that waiting period and reimbursement level drive preference most strongly, and simulates two attractive packages at €19.90 and €24.90 per month.

Both packages then run in a Painted Door Test at Horizon. Measured sign-up intent is 2.4% and 2.2%; the difference is not statistically robust. For the decision, this means: the higher price costs hardly any demand.

When each method fits

If many attributes and levels are still open, conjoint analysis is the more efficient tool: it covers a large space of combinations with one sample. If two to six concrete variants are fixed and a larger investment depends on them, a behavioural test provides the additional data point of whether these variants are also chosen outside the questionnaire.

Horizon works complementarily to conjoint studies. Many teams use the simulation to select candidates and then check them in a Painted Door Test.

How it differs

MaxDiff ranks attributes by importance without combining them into offers. Van Westendorp and Gabor-Granger ask directly about prices. Conjoint analysis combines attributes and price into offers and derives preferences from choices. The Painted Door Test measures choice not in a questionnaire but in the real online environment, though with few variants (up to six at Horizon).

Limitations

Conjoint analysis remains a survey: the choice has no consequences, respondents know the context and decide with more concentration than in everyday life. A very large number of attributes is overwhelming, and very new attributes are hard to evaluate when respondents have no experience with them. The Painted Door Test has the opposite limitations: it checks only a few variants at a time and cannot weight attributes with the same breadth.