Conjoint analysis: trade-offs and willingness to pay

Customers rarely choose a product for one reason. A smartphone buyer weighs brand, storage, battery, warranty and price, and accepts less of one to get more of another. Conjoint analysis measures these trade-offs by asking people to evaluate complete product profiles, then decomposes their preferences into the value of each feature level.

Attributes and levels

A conjoint study describes products by attributes (brand, storage, warranty, price) with two to five levels each (Apex, Nova, Zenit; 64, 128, 256 GB). Choose attributes that matter to the decision and that customers understand, with realistic levels that span the range you might offer. Keep the number manageable: five to seven attributes is typical, because respondents simplify when profiles get too complex.

Rating-based conjoint

In traditional (rating-based) conjoint, respondents rate a set of profiles generated by a fractional factorial design, for example on a 0–10 scale. A regression with dummy-coded levels then estimates the part-worth utility of each level.

In the example, price is the most important attribute (34.5% of importance), followed by battery life (26.3%), brand (20.2%) and camera (19.0%). Attribute importance is the range of an attribute’s part-worths divided by the sum of all ranges: price ranges from +0.79 ($299) to −0.72 ($499), the widest spread. The most preferred profile combines Nova, $299, three days of battery and the 108 MP camera. Because part-worths are estimated per respondent in rating-based conjoint, you can also segment respondents by what they value.

Choice-based conjoint (CBC)

Rating profiles is unlike real shopping. In choice-based conjoint, respondents see several choice tasks, each with three or four alternative products (often plus a “none” option), and choose the one they would buy. This mimics real decisions, handles price better and is now the most common form of conjoint. The choices are analysed with a conditional (multinomial) logit model.

Reading the CBC results

The model, estimated on 1,200 choice sets, is significant (LR χ²(6) = 501.2, p < .001; McFadden R² = .190, a good value for choice models). Coefficients are relative to the first level of each attribute: Nova adds 0.544 utility compared with Apex, 256 GB adds 0.917 compared with 64 GB, and a two-year warranty adds 0.481. Price enters as a numeric attribute with a coefficient of −0.0094 per currency unit.

Willingness to pay

Because price is in the model, any utility difference can be converted into money: WTP = utility difference ÷ |price coefficient|. One unit of utility is worth 1 / 0.0094 ≈ 106 currency units. So customers would pay about 57.79 more for Nova than for Apex, 97.37 more for 256 GB than for 64 GB, and 51.08 more for a two-year warranty. These are the numbers product managers love: they tell you whether a feature’s price premium is justified by what it costs to provide.

Attribute importance in this CBC is price 49.2%, storage 24.0%, brand 14.2% and warranty 12.6%. Note that importance depends on the level ranges you chose: a wider price range would make price look even more important. Always report the levels with the importances.

Market simulation

The real power of conjoint is simulation. Define the products in a market (yours and competitors’), compute each product’s total utility and convert the utilities into predicted shares with the logit formula. Then ask what-if questions: what happens to our share if we add a two-year warranty, or cut our price by 50? Simulated shares are preference shares, not market share forecasts; they ignore distribution, awareness and availability, so use them to compare scenarios.

Design tips

• Use 8 to 15 choice tasks per respondent, with three or four alternatives each.
• Generate a balanced, efficient design (survey platforms such as Qualtrics, Sawtooth or Conjoint.ly do this), and randomise the order of tasks.
• Include a “none” option if not buying is a realistic choice.
• Plan for at least 300 respondents for aggregate models, more for segment-level analysis.

Reporting

A choice-based conjoint study (1,200 choice sets) was analysed with a conditional logit model, LR χ²(6) = 501.2, p < .001, McFadden R² = .19. Price accounted for 49.2% of attribute importance, storage for 24.0%, brand for 14.2% and warranty for 12.6%. Respondents were willing to pay 97 more for 256 GB than for 64 GB storage, 58 more for the Nova brand than for Apex, and 51 more for a two-year warranty.