To parse and interpret this large body of text, machine learning techniques are introduced. Specifically, a variety of cluster analysis techniques are applied to the survey data. These statistical methods help uncover underlying data associations and reduce thousands of unique words (tokens). The aim of this data reduction step is to transform unstructured text into a manageable set of representative tokens while preserving essential meaning.
After routine preprocessing and tokenization, the dataset produced 2,719 unique terms with associated frequencies. A further data step involved removing near-zero variance terms thereby reducing this set dramatically to just nineteen key word tokens. These nineteen tokens are:
games, people, painting, miniatures, history, research, collect, model, terrain, table, fun, creative, army, build, aspect, good, time, hobby, like.
With the dataset reduced from over 4,000 terms to fourteen, cluster analysis provides the next layer of insight. The resulting dendrogram (see Figure 1) reveals a clear and intuitive structure that I break down layer by layer.
| Figure 1 |
The dendrogram organizes the “most liked” aspects of wargaming by how closely respondents associate the tokens. Joining at a lower height in the dendrogram indicates a stronger relative relationship with higher splits indicating more separation.
At the highest level, "games" stands apart from all other terms, joining the rest of the terms only at a relatively large distance. This indicates that respondents view gaming as distinct reinforces the notion that gaming, itself, is the central facet of the hobby. From the dendrogram (see Figure 2), "games" is not intertwined with the other activities.
The next major split separates "people" from the remaining terms (see Figure 3). Together, this creates a clear hierarchy in a three-cluster solution of:
- Games (most distinct)
- People (second most distinct)
- Everything else (more tightly interrelated)
Slicing across the dendrogram at a practical level (around mid-height) yields four meaningful groupings (see Figure 5):
- Games (standalone)
- People (standalone)
- Painting/miniatures (craft-focused sub-cluster)
- All remaining hobby activities (a blended cluster of building, modeling, and researching)
| Figure 5 |
What can we infer from this analysis of survey respondents' likes about wargaming?
To begin, analysis supports the notion that gaming, social interaction, and painting emerge as primary drivers. All other activities form a supporting ecosystem. Analysis also suggests a structure to the wargaming hierarchy. That is, wargaming should not be viewed as a single, unified, activity but as a combination of semi-independent domains. If we return to the three-cluster solution, the clusters are:
- Gaming is central but conceptually separate.
- Social interaction is nearly as important and also distinct.
- Everything else forms an interconnected hobby engine of crafting, researching, and building.
| Figure 6 |
- Games (standalone)
- People (standalone)
- Painting/miniatures (craft-focused sub-cluster)
- All remaining hobby activities (a blended cluster of building, modeling, and researching)

