Regency EF
Interpreting Cluster Analysis Results: Using Relative Importance Methods as a Decision Aid (304063)
*Joseph Nicholas Luchman, Fors Marsh Group
Keywords: Cluster Analysis, Data Interpretation, Relative Importance, Segmentation
Cluster analysis methods such as k-means or latent class/mixture model clustering are common methods applied in market segmentation, person-centered behavioral analysis, and other unsupervised learning. Despite their common application in practice, few methods are available to assist analysts in the interpretation of clustering results. This poster’s goal is to provide a methodology which can serve as a decision aid for analysts to identify variables in the cluster analysis that best sort observations into different clusters. This decision aid can be used as the basis for interpreting the cluster solution’s conceptual meaning by identifying mean and proportion differences across clusters on a variable that appear to be most influential. To do so, cluster membership categories or probabilities are predicted by variables that are used as separating factors in the cluster analysis. Relative importance metrics are then used to identify influential variables among the separating factors. This poster outlines illustrative, reproducible examples using publicly accessible data.