Assumptions of the analysis

There are no defined cluster “shapes”

Unlike with K-means clustering analysis, there’s no assumption for the shape of clusters that are defined for your data. Instead, hierarchical clustering simply forms clusters by combining the two “closest” observations (or clusters) at each step using whichever distance metric you selected.

The distance metric matters

Prism offers a variety of different distance metrics when performing hierarchical clustering. Each of these methods calculates the distance between points slightly differently, and so may result in entirely different structures of clustering.

The linkage method matters

Hierarchical clustering works by iteratively combining observations and clusters together by joining the “closest” objects at each step. In addition to defining what distance metric to use to when calculating how “close” these objects are, you must also choose an appropriate linkage method. By definition, clusters consist of more than one observation, and so there are multiple different ways to calculate the distance from this cluster to another cluster or observation.