Course · 2. Choice-based conjoint (CBC)

Designing a CBC study

Choice-based conjoint (CBC) is the most common form of conjoint: each task shows a few products and the respondent picks one, or none.

Attributes and levels

• Pick the attributes that drive the decision you care about, typically 3 to 7. Brand, a few product features, and price.
• Keep levels to what you would actually sell: 2 to 5 per attribute. Every extra level needs more data.
• Make price numeric (actual amounts across the range you would consider). Chosŏn then treats it as a slope, which is what turns part-worths into willingness to pay.

Tasks

• Products per task: 3 to 5 is usual; a full-shelf study can show 20 or more.
• Tasks per respondent: about 8 to 15. Fewer and estimates are noisy; more and attention drops.
• None option: include it if "buy nothing" is a real outcome. It lets the model say whether a launch grows the category or only moves share around.
• Let products share levels within a task. Two products with the same flavor on the same screen are how the model learns whether flavor drives switching. Fully balanced designs that never repeat a level inside a task make that impossible to learn. Chosŏn's data check reports how often this happens.

How many respondents

A common rule of thumb: respondents × tasks × products per task ÷ (most levels of any attribute) should be at least 500. The tutorial study (250 people × 10 tasks × 4 products ÷ 4 levels) scores 2,500. Chosŏn computes this on upload and warns below 500.

Try it in the tutorial: the data check

Next: Your data

Tutorial links open the example project in the app; it is free and needs a free account.