Cluster analysis

Groups respondents into clusters based on their values across several variables.

When to use it

To find natural segments in your sample — for example, grouping respondents by a profile of scores.

Inputs

• Variables — the numeric/ordinal columns that define similarity between respondents.

Options

• Method — K-means or Hierarchical.

• Linkage (hierarchical) — Ward, Complete, Average, etc.

• Number of clusters.

• Standardise variables — z-scores each variable first, so variables on larger scales don’t dominate the distance.

• Show assignments — adds a per-respondent cluster table.

• Verbal indicators (in-table columns), Verbal report (dropdown for how much written interpretation), and Plots.

Output

• Cluster sizes and centroids (the average profile of each cluster).

• A silhouette score indicating how well-separated the clusters are.

• A 2-D cluster scatter (via PCA), a dendrogram for hierarchical clustering, and a k-selection plot to help choose the number of clusters.

• Per-respondent assignments (if enabled).

Notes

• Results are reproducible — the clustering and 2-D projection use a fixed random seed.

• Standardising is usually advisable when variables are on different scales.