Bayesian analysis is a method of learning based in probability theory, in which a hypothesis is updated as more information becomes available. Bayesian analysis is compelling and powerful in modeling and predicting consumer behaviors based on survey response.
In market research, Bayesian analysis produces results that are potentially more reliable than conclusions based on descriptive statistics (e.g., those frequency percentages of survey respondents saying they agree with a given behavior or sentiment we so commonly feature in market research reports). In effect, we learn from the data as results are processed and we come away with confidence in our results.
Collecting direct responses to survey questions supports insightful analysis, but adding questioning for MaxDiff Choice Exercises and Choice-Based Conjoint yields a rich data set for marketplace projection. These modules of survey questioning complement the direct questions we ask, generating data that can be examined with stronger reliability.
A Quick Example
In surveys, there is often a need to assess consumers’ likelihood to purchase a certain product or service. While we may ask such a question directly with a five-point scale and interpret the percentage indicating “4” or 5″ on the scale, we never really come away knowing if that consumer (or those he or she represents in the sample) is going to follow-up in the real world.
Asking questions that force the respondent to make trade-offs around their preferences and likely behaviors can be more revealing. The specific survey responses to the trade-off questions are not in themselves explanatory. Hierarchical Bayesian analysis (an application of Bayesian analysis) is conducted on the sample’s responses to model consumers’ predicted behavior. Through this process we develop a robust and reliable perspective on the data.
The Somewhat Technical Stuff
Hierarchical Bayesian analysis models consumer choice by using information from many respondents to refine estimates received from each individual. The process utilizes the responses from each individual to learn and reformulate the model (e.g., to develop separate sub-models and the hierarchical overall model).
Bayesian analysis provides a way to update initial hypotheses and emerging probabilities – over thousands of iterations, not just response from the sample of a few hundred respondents – yielding an increased reliability of insight. Predicated behaviors, preference, and intent may be interpreted based on this robust statistical modeling to support the research objectives.
The primary way in which W5 adds reliable predictive perspective to a survey project is by adding Choice-Based Conjoint conducted with Hierarchical Bayesian analysis. When a full conjoint study is not appropriate, a MaxDiff Analysis may be recommended, leveraging the benefits of Hierarchical Bayesian analysis to complement other study insights. We welcome the opportunity to use these exercises and deeper analyses to model and understand consumer sentiment. The results truly complement the direct survey responses we obtain. The results are illuminating, adding insights based upon a combination of robust statistical modeling and expert interpretation.