Conjoint Analysis Basics: Definition, Types, Terminology, and More

Getting started in conjoint analysis might seem overwhelming, as often is the case with beginning any new skill or acquiring new knowledge. 

Becoming familiar with the basic concepts of conjoint analysis makes it easier to follow along and begin to acquire the skills to establish proficiency. Sometimes technical terminology can make it harder to understand or how to do a conjoint analysis study on your own, unless you have a guide to help you understand what the terms mean.

That’s why we’ve created this conjoint analysis basics guide, a cheat sheet, with basic terminology and key principles you need to understand to get up and running with conjoint analysis. 

Good software makes the process easier (and we’ve got great software at Sawtooth!); but we also provide excellent training resources (videos, white papers, in-person training), free technical support, and paid consulting services if you need greater dedicated support from our experienced professionals.

Definition of Conjoint Analysis

We’re all familiar with what a conjoined twin is. The word conjoint refers to multiple things joined or combined. For marketing and economics topics, those multiple things combined are things such as product features (brand, color, warranty, price).

For example, a product concept (electric vehicle) made up of conjoined product features could be:

Telsa, Model S, red, 250-mile range per charge, $45000

Conjoint analysis is a research methodology that involves creating different product concepts for respondents to evaluate, then analyzing respondents’ reactions to those product concepts to understand which product features are more important in driving those judgements, with the goal of optimizing the product feature combinations and price to maximize success in the marketplace.

Getting Started with Conjoint Analysis: A Simple Example

Let’s give a simple example involving electric vehicle (EV) choice to show how conjoint analysis works.

Imagine we are trying to learn about how to design an optimal EV for Ford to compete with Tesla. We are interested in variations of:

• Brand (Tesla or Ford)
• Driving range (150 miles per charge, 200 miles per charge, 250 miles per charge)
• Color (Red, Blue, Silver, Grey)
• Price ($40000, $50000, $60000)

A user-friendly software tool like Sawtooth Software’s survey platform can take this product feature list and automatically (using smart algorithms) create comparisons involving different vehicles for respondents to compare and choose. For each product profile to be evaluated, we assign one and only one level from each attribute.

Each respondent typically completes 8 to 12 such choice scenarios (called Choice Tasks), where each choice task varies the features of the EVs. For example, Task 2 may ask this question:

If you were buying an EV today, which of these options would you choose?

(Task 2 of 8)

Tesla 150 miles per charge Red $40000

Ford 200 miles per charge Silver $50000

None: If these were my only options, I would not buy an EV today.

Over hundreds of respondents, we will typically have collected choices for over 1000 choice tasks, like these two shown above.

The software intelligently varies the features across these choice tasks in a fair and balanced experiment. Each feature level appears (nearly) an equal number of times, and each feature level appears with every other feature level (nearly) an equal number of times.

For example, it wouldn’t be a fair experiment if Ford always appeared at the lowest price point; or at the highest miles per charge! The software makes sure that Ford shows an equal number of times at both low and high prices, and at both low and high range per charge.

A conjoint analysis model examines respondents’ choices to the 100s or 1000s of choice tasks for determining the product features that are more important to driving (explaining) product choice. 

The statistical modeling approach for finding the preference weights (scores) that drive consumer choices for conjoint analysis studies like the one described above is called logistical regression. 

Sawtooth Software’s tools automatically estimate the logistical regression model to find the preference scores that explain respondents’ choices. With those preference scores (also known as utilities) in hand, we then understand respondents’ preferences for brand, driving range, color, and sensitivity to prices. We are in a position to design the Ford EV that can compete with Tesla, while at the same time being profitable for Ford to sell.

Sawtooth Software’s conjoint analysis tools lead to an intuitive what-if market simulator, wherein the analyst or manager can specify different versions of the Ford EV to compete with Tesla’s offering and immediately simulate the percent of respondents who would prefer each Ford offering to Tesla. By changing the features and price, Ford can find the optimal product that maximizes the expected demand, revenues, or profits.

And, though we’ve shown this example above with just four attributes and no more than four levels per attribute, conjoint studies can cover a dozen or more attributes involving some attributes that have a dozen or more levels. It is, thus, able to be extended to a variety of problems, including those with graphically represented attributes.

Types of Conjoint Analysis

Discrete Choice Analysis (Experiment)

Conjoint questions in which respondents are asked to choose one of available options (or often involving None).

Common uses for Discrete Choice Analysis

• Product feature optimization for new product research, product repositioning, or product line extensions
• Price sensitivity measurement, including own-price and cross-price elasticity
• Brand equity measurement, including willingness to pay
• Market segmentation (needs-based segmentation)

What’s the Difference Between Discrete Choice and Conjoint Analysis Studies?

Nowadays, these terms are usually applied interchangeably. But, originally a line of distinction was drawn regarding whether respondent evaluations were discrete choices (Discrete Choice) or the earlier ratings/rankings of product profiles (conjoint analysis) that was popularized in the 1970s and 1980s.

Choice-Based Conjoint (CBC)

Just another term for Discrete Choice Analysis (Experiments), emphasizing that respondents provide choices among the profiles rather than rating/ranking the individual product profiles.

Common uses for Choice-Based Conjoint (CBC)

Same uses as for the synonymous term (Discrete Choice Experiments):

• Product feature optimization for new product research, product repositioning, or product line extensions
• Price sensitivity measurement, including own-price and cross-price elasticity
• Brand equity measurement, including willingness to pay
• Market segmentation (needs-based segmentation)

Adaptive Choice-Based Conjoint (ACBC)

A computerized interview form of Choice-Based Conjoint in which later questions are designed based on respondents’ earlier answers. In this way, the interview adapts to each respondent’s preferences, providing a more engaging and customized interview experience.

Common uses for Adaptive Choice-Based Conjoint (ACBC)

• Same uses as stated above for Choice-Based Conjoint, except that Adaptive Choice-Based Conjoint can extend the researcher’s ability to study a greater number of product attributes (features)

Adaptive Conjoint Analysis (ACA)

An early (invented in the 1980s) computerized interview form of conjoint analysis that involved asking respondents to rate pairs of product alternative on typically a 1-9 scale, where “1” meant “Strongly prefer product on left” and “9” meant “Strongly prefer product on right”. ACA is rarely used today.

Common uses for Adaptive Conjoint Analysis (ACA)

• Able to cover more product features than non-adaptive forms of conjoint analysis.
• For product/service decisions not involving price, as ACA tended to understate the true importance of price.

Menu-Based Conjoint (MBC)

A multi-check form of conjoint analysis, where respondents can pick multiple options on the screen, such as when purchasing from a menu.

Common uses for Menu-Based Conjoint (MBC)

• For studying choices in menus, including those involving bundles of items vs. a la carte choices. 

Key Conjoint Analysis Terminology

Conjoint Analysis Study (Experiment)

Marketing research survey questions in which respondents are shown varying products (at different brands, features, and prices) and choose which one(s) they would buy. Each product profile is constructed by assigning one (and only one) level from each attribute. By analyzing the choices respondents make, we learn which features are most driving their choices, as well as their price sensitivity.

Attributes (Features)

Attributes are characteristics (features) of a product or service being evaluated in a conjoint analysis experiment. Examples of attributes for vehicles: make (brand), model, type, year, mileage, price.

Levels

Levels are the degrees (discrete specifications) of an attribute to be evaluated in a conjoint analysis experiment. Examples of attribute levels for vehicle brands: Ford, Dodge, Chevrolet, GMC.

Choice Tasks

Choice tasks are the series of conjoint analysis survey questions presented to survey respondents. Choice tasks are comprised of concepts or product profiles from which survey respondents choose their most preferred option. 

Concepts (or Profiles)

Concepts are hypothetical product profiles or offerings to be considered in choice tasks of conjoint analysis surveys. Concepts are made up of attributes with varying levels. 

None Alternative

The option available to the respondent to say that none of the product concepts in a choice task are acceptable to them. 

Experimental Design Algorithm

For proper and efficient conjoint analysis questions, we need to construct the product profiles such that each level of each attribute is covered (nearly) an equal number of times, and that each level of each attribute appears with each level of different attributes (nearly) an equal number of times. The experimental design algorithm constructs the series of product profiles (the combinations of attribute levels) to show each respondent to achieve this air and balanced experiment. 

Efficient conjoint analysis questions can be submitted to the Utility Estimation Algorithm to estimate respondent preference scores (utilities) with a high degree of precision. Inefficient questions would be those, for example, in which there are correlations between attributes, such as certain colors being correlated with certain prices.

Utility Estimation Algorithm

A statistical procedure for fitting preference scores (weights) to the attribute levels that best explain respondents’ choices to the conjoint questions.

Ordinary Least Squares (OLS) Regression

A statistical procedure for fitting preference scores (weights) to the attribute levels when respondent judgements are on a rating scale (such as 0-100).

Multinomial Logistic Regression (Logit, or MNL)

A statistical procedure for fitting preference scores (weights) to the attribute levels when respondent judgements are choices.

Hierarchical Bayes

A statistical procedure for fitting preference scores (weights) to the attribute levels at the individual level. There are hierarchical Bayes forms of both OLS and MNL, appropriate for when respondent judgements are ratings (HB-OLS) or choices (HB-MNL).

Modeled Choice Probability

The likelihood (on a 0-100% scale) of whether a product option would be preferred to another (or set of others). This is often referred to as “share of preference” or “share of choice”.

Log-Likelihood Measure

In logistic regression models, a measure of fit for the preferences scores (utility weights) of the attribute levels explaining the choices respondents made. It is the natural log of the likelihood, where the likelihood is the joint probability of observing a series of choices on a 0%-100% scale.

Conjoint Experiment Outputs

A conjoint analysis experiment with accompanying respondent data may be analyzed, leading to a number of outputs:

Part-Worth Utilities (or Preference Scores)

Utilities are the mathematical representation of respondent preference (weight or utility value) associated with attribute levels of the conjoint experiment. For a conjoint experiment made up of two or more attributes, the total utility of a product profile is made up of the sum of the part-worths of the utilities across its parts: the brand part, the color part, the price part, etc.

Attribute Importances

Importances are scores or values associated with the impact each attribute can have on choice by varying from its worst level to its best level. The scores are usually normalized to sum to 100%. For example, importances for four attributes can be: Brand 40%, Color 20%, Speed 10%, Price 30%.

Note that importance scores are relative to the range of levels included in the study, so importances are affected by decisions made by the analyst regarding which levels to include in the study, and therefore should be interpreted with caution.

Market Simulator (What-If Simulator)

One of the most practical and intuitive deliverables of a conjoint analysis study, market share simulators (often delivered as interactive Excel files) enable analysts and managers to “predict” the consumer’s choice (market share) for products on the existing market as well as new products that they are looking to develop.

Market simulators are valued by managers and analysts for the ability to set up dozens or hundreds of potential market scenarios to account for varying market conditions and predict reactions to myriad product configurations.

The insights gained from conjoint analysis studies in conjunction with market simulations, when done correctly, truly pave the way for product-market acceptance and business success.