Choice Modeling Techniques
“Choice modeling” refers to a family of multivariate statistical techniques that help identify optimal combinations of marketing variables. In choice modeling experiments, consumers are simply asked to make choices, given a set of concepts, ads, package designs, or products on a shelf. Choice modeling simulates the consumer’s natural buying-decision behavior.
The choice-modeling experimental design is tailored to the specific objectives, constraints, and variables of each project. Customization is the key to success because every category/brand has critical idiosyncrasies.
Here are summaries of the major types of choice models:
In a MaxDiff survey exercise, respondents see 15-20 sets of product features (for example) and select which is most important and which is least important to their purchase decision.
- Works well with a variety of items
- Produces a full ranking of attributes
- Easy for respondents to complete
- Repetitiveness of the respondent tasks
- No competitive measurement
- No brand-price or brand-feature interactions
- No cannibalization within a product line
Essentially, MaxDiff is a trade-off exercise that includes one attribute with many levels.
Discrete Choice Models
During a discrete-choice exercise, respondents see several screens of products and select one product to purchase on each screen. For industries like healthcare and pharmaceuticals in which the single-choice technique does not adequately capture the decision-making process, the model can be modified to point-allocation. It allows respondents to allocate points (usually 10 or 100) across the options shown, and a share model is then estimated to predict outcomes.
- Predicts accurate market shares
- Includes the competitive context
- Assesses brand-specific prices and features
- Identifies optimal products and product lines
- Difficult to design with many moving parts
- Requires a larger research budget
Volumetric Choice Models
Volumetric choice models are a compilation of two or more models. One model predicts the likelihood of a consumer selecting a product or service, and the other(s) predict(s) the number of units of each product or service the consumer would purchase. During the exercise, respondents see several screens of products within the category and choose to buy 0, 1, or more of each.
- Predicts accurate market share, especially when the models are calibrated using actual sales
- Includes competitive effects and thereby attempts to mimic the actual purchase decision
- Includes brand-specific price and feature effects
- Enables evaluation of product lines
- Respondents may have a difficult time producing realistic volume estimates on each screen (overestimation of volume is common)
- Can take a long time to complete, thereby increasing respondent fatigue
Menu-Based Choice Models
Menu-based choice models are a compilation of multiple models. One or more models predict the likelihood of selecting the base product or multiple products, and additional models predict the likelihood of purchase and/or the number would purchase of add-on features that would be purchased. During the survey exercise, respondents choose products and then add on other features. Some examples of add-on features include trim packages for vehicles, insurance riders for a policy, or a DVR for to a video-service package.
- Accurate market share and unit sales predictions
- Inclusion of competitive brands
- Assessment of brand-specific prices and features
- Identification of optimal product lines, packages, or bundles
- Respondent burden needs to be carefully managed as product categories can have hundreds of different features
Regardless of the choice modeling method used, the overriding goal is to create realistic “shopping” scenarios that properly represent the buying behavior that we are striving to model. Whenever possible, the choice options are presented virtually, using shelf sets, online websites, apps, or other buying environments. This increased realism results in a better understanding of customers' purchasing behavior, which powers better business decisions.
Analytical Consulting Services
Decision Analyst is a global marketing research and analytical consulting firm with over 40 years of experience in state-of-the-art modeling, simulation, and optimization. A team of Ph.D.’s heads up Decision Analyst’s choice modeling work. They also publish many white papers on advanced analytical methods and speak frequently at marketing research industry conferences. They program choice models in SAS, Sawtooth, and the R-Language.
If you would like more information or would like to discuss a possible project, please contact contact Jerry W. Thomas, President/CEO (firstname.lastname@example.org), or Elizabeth Horn, Senior Vice President of Advanced Analytics (email@example.com), or call 1-800-ANALYSIS (262-5974) or 1-817-640-6166.