Conjoint Segmentation is a Choice-Based Conjoint variant with the aim to provide detailed information on the sample with several segments of respondents presented as part of the analysis.
The experimental design is created using methods that employ Federov's exchange algorithm so that the final design is D-optimal and nearly orthogonal. The algorithm explores the complete space of possible designs for the best given specified parameters. It is expected that each option (variable level) is present approximately equal number of times compared to other options. The total number of versions varies depending on the setup, but having more than a hundred versions for large setups is common.
The core analysis of respondents' preferences is performed with help of Hierarchical Bayesian Multinomial Logit model. Bayesian model is estimated with Hybrid Gibbs Sampler with a random Metropolis step MCMC. The number of burn-in iterations is determined automatically when there's enough evidence for convergence.
The model takes into account properties of other packages presented in a task when respondent makes a choice. The choice probability correspond to the logit transformation of the linear combination of utility scores of the packages in the task. Respondents are analysed individually, with their preference scores being a realization of pooled "average" opinion which follows a Normal distribution, at the same time reflecting their individual preferences. As a result, raw Logit coefficients are available for every respondent.
Segmentation analysis consists of two stages. First, Gaussian Mixture Clustering model is applied over conjoint raw coefficients to determine a number of clusters with respondents that share similar opinions. Secondly, a number of tests is performed using available data on respondents' traits and answers to other survey questions, to find the most relevant features that would explain dissimilarity of one cluster from the others.
The statistics page shows several elements: segmentation results, market simulator, Attribute Importance values, Preference Likelihood, and Utility scores impact.
In market simulator, Preference Likelihood mode assumes a state of the market when two packages exist: an average one, and the chosen one. With such settings it is possible to see how market share of the chosen package change with a change of the attributes.
Utility scores mode uses raw Logit coefficients of the chosen package. In this setting an average package gets a score of 0.
Importance values for each attribute, calculated as shares of ranges of utility scores. The higher importance, the higher impact a decision on the attribute has on the package's market share.
- Raw coefficients export: Logit coefficients and cluster assignment for each respondent
- Summary Excel: Attribute Importance; Preference Likelihood and Utility Score impact
- Raw data Export: data on what each respondent saw, and what decision was made on each task;