ATLASResearch
methods
Quantitative/ Analysis

Cluster analysis

Discover a defensible grouping of observations

Cluster analysis groups observations according to a chosen similarity or probabilistic model. K-means uses distances to centroids, hierarchical methods build nested partitions, and mixture models estimate probabilistic membership. The number and meaning of clusters depend on variables, scaling and assumptions. A discovered grouping requires stability and external validation before it becomes a typology.

WHEN IT FITS

Use for exploratory segmentation or pattern discovery when group labels are unknown. Select substantively relevant features and a distance or distribution model suited to their scales and expected structure.

Strengths

  • Reveals multivariate structure without predefined labels
  • Probabilistic methods express membership uncertainty

Limitations

  • Algorithms can impose groups on continuous variation
  • Solutions depend strongly on scaling and tuning choices

Know the boundary

Clusters do not establish natural kinds, diagnoses or causally distinct populations.

USED ACROSS
Banking & financeBusiness & MBAPsychologyComputer scienceEducation