Choice modeling

Understanding how and why people make choices among competing products and services is critical to the design and positioning of all products and services, whether they be new televisions, highways, park management programs, or transit services.  

For over 30 years, RSG has been a pioneer in the application of innovative and powerful methods for modeling choice behavior. We develop, refine, and apply discrete choice modeling methods to create robust forecasting models of people’s choices among alternative products and services. These models provide rich descriptions of how products’ features, designs, and attributes, such as price and positioning, affect choices, and are derived from specially constructed surveys that include choice-based conjoint or stated-choice exercises, and/or data on actual choices observed in the marketplace. 

RSG’s choice exercises* and choice modeling methods** go well beyond what can be done with off-the-shelf software. Our methods are fully customizable and take into account aspects such as a respondents’ past choices, different underlying distributions and model structures, attribute interactions, and efficient experimental designs. We have also developed and applied complementary methods to model the myriad other considerations that are involved in fully understanding consumers, including: how people decide what products and services to consider; what influences how much/how often they will use the product or service; how the set of alternatives influence choices; and what drives satisfaction, loyalty, and re-purchase/re-use. 

*Selected Choice-based Conjoint Exercises 

• Discrete Choice 
• Choice-based conjoint 
• Adaptive choice-based conjoint 
• MaxDiff (Maximum Difference Scaling) 

**Selected Choice Modeling Approaches 

• Mixed logit models 
• Latent choice/latent variable (LCLV) models 
• MDCEV (Multiple Discrete and Continuous Extreme Value) models
• Classical, simulation-based, Maximum Composite Marginal Likelihood (MACML) and hierarchical Bayesian estimation methods 
• TURF (Total Unduplicated Reach & Frequency) models