Creating a CBC

Choice-Based Conjoint (CBC) studies how people make choices and is the premier tool for optimizing features and pricing for products and services. CBC exercises present respondents with realistic scenarios where they choose among different product concepts, each made up of two or more attributes. When making choices, respondents make tradeoffs similar to those they make in the real world, revealing what drives their decisions.

By observing these choices, you can build models that capture the preference weights (utilities) for each attribute level. CBC outperforms traditional rating or ranking approaches because respondents make realistic tradeoffs rather than evaluating features in isolation — the same way they'd decide in a real purchase situation. With that information, you can simulate what people in a given market would choose across a range of potential scenarios using the market simulator.

In the real world, buyers aren't required to purchase — they can walk away if better options aren't available. The "none" option mimics that experience by providing an opt-out choice. There are two types: traditional and dual-response. Traditional is used by default.

Select the none type in the question settings.

Shown as an additional option in each task with the default text "None, I wouldn't choose any of these." Respondents can select a product concept or opt out entirely. The traditional "none" can also be used for a fixed alternative, such as a status quo choice or "I would prescribe diet and exercise to the patient."

A traditional none option is shown below a CBC task asking respondents to select a vacation package.

An additional question presented after each forced-choice task. By default it asks "Given what you know about current offerings and your budget, would you actually purchase the option you chose above?" with yes and no response options.

A CBC with a dual-response none asking respondents

Consider substituting "option" in the default text with "product," "package," or whatever best reflects what respondents are choosing in your study.

While yes and no are the defaults, the wording and number of options can be changed. A common alternative is "How likely are you to purchase the product you selected?" with a five-point scale:

Definitely would
Probably would
Might or might not
Probably would not
Definitely would not

Many researchers are familiar with how a five-point scale relates to actual purchase behavior in their product categories. That said, using a scale requires specifying which options indicate a "buy" or "no buy" — for example, the top two boxes as "buy" and the bottom three as "no buy."

In most marketing-related CBC studies, the dual-response none substantially increases the likelihood that respondents choose "none," which many researchers argue better reflects actual purchase intentions than the traditional "none." Importantly, no information is lost when respondents select "none" because they first choose the best concept in the task. The dual-response "none" acts as a safety net, allowing a none parameter to be estimated without sacrificing precision in the utilities for attributes and levels, especially when "none" usage is high.

Dual-response none exercises export additional tabs for use with Sawtooth's desktop products in the design and choices download. Learn about the design & choices export.

Tells utility estimation which response options count as "would buy" and which count as "would not buy." This can be changed after data collection but requires republishing the survey.

For two-item lists, select which item is the "would buy.

The UI showing the numeric input to enter which item will be considered

For lists with more items, define a range.

The UI showing two numeric inputs to define the range.

A counter showing respondents their progress through the exercise, separate from the survey progress bar. The counter can be turned off and the label is optional.

The task counter is right below the question text.

Controls how attribute labels appear to respondents. Three options are available:

Left of concepts: Attributes appear to the left of the concepts.

Inside concepts: Attribute labels appear inside the concepts, above the levels.

Not shown: Labels are hidden entirely.

On screens narrower than 834 pixels, the CBC layout adjusts automatically — attributes appear inside concepts and concepts are displayed one at a time. Three layout options are available:

Swipe horizontally: Concepts are displayed in a carousel and swiped through horizontally. Dots indicate which concept is being viewed.

Swipe vertically: Concepts are swiped through vertically. Dots on the left indicate which concept is in view.

Concepts are swiped through vertically. Dots are shown on the left to indicate which concept is in frame.

Maintain desktop: Keeps the desktop layout regardless of screen size.

When a concept is selected in either swipe mode, its dot changes to a radio selection indicator.

The select button has three states: default, hover, and selected. Default and hover states show the same text but differ in button color — white for default, blue for hover. The selected state appears after a concept is clicked.

A view of the default, hover, and selected states for the select button.

Before publishing, a valid design must be generated. Discover automatically recommends design settings that balance precision and respondent effort, reduce order bias, and ensure attribute levels are distributed evenly across respondents. To customize the defaults, toggle on Override recommendations in the Design tab.

Tasks: The number of choice tasks shown to each respondent. Recommended range is 6–12.
Concepts per task: The number of product concepts shown per task, excluding the "none" option. Recommended is 3.
Design versions: The number of unique survey versions generated, ensuring respondents across the study see different combinations of concepts and improving overall design efficiency.

The Design tab also includes tools for managing prohibitions — preventing specific level combinations from appearing together — and for setting up nested attributes for defining conditional relationships between attributes. See the design status article for details on generating and validating your design before fielding.

For purely visual adjustments to how concepts are displayed, see merged attributes, which lets you combine similar attributes into a single row without affecting utility estimation or the design.

Each part of the CBC design has its own limits:

100 attributes
100 levels per attribute
100 tasks
100 concepts
100 versions

In addition, there is an overall limit of 30,000 attribute-level pairings per version of your design.

There is also an overall limit of 30,000 attribute-level pairings per version. For example, 8 attributes, 4 concepts per task, and 12 tasks produces 8 × 4 × 12 = 384 pairings per version — well within the limit.

This limit accommodates a wide range of designs, whether that means many attributes with a typical number of tasks, or many tasks with fewer attributes. If you need to exceed these limits, contact our support team.

Discover Help Guide

Creating a CBC

Introduction

"None" option

Traditional

Dual-response

The “Would buy” threshold

Exercise formatting

Show exercise progress & task label

Attribute labels

Mobile layout

Select button

Exercise design

Limitations

Related Content