Grouping Wargame Periods

In one of the many analyses from WSS' 2022 Great Wargaming Survey (GWS), I explored aggregated survey respondent tendencies based upon wargaming period choice.  Do particular periods tend to group together was the question I asked.  This analysis utilized cluster analysis (see Game Period Choice: A Cluster Analysis). 

Why revisit this topic two years in a row?

Well, two motivations came to mind.

• Unranked vs Ranked and Top 5 vs Top 3.  The 2022 GWS asked respondents to choose their top five wargaming periods in an unranked manner. The 2023 GWS asked the same question but responses were limited to the top three in rank order.  What effect will these changes have upon the results?
• Reliability of Data.  While questions often surface about the data collection methods and data reliability, I regularly offer analyses in an attempt to mitigate these reservations.  Do groupings remain reasonably consistent across these two years of survey responses?

Today as in last year's analysis, we switch from descriptive analytics to predictive analytics.

Cluster analysis is again the tool of choice.  Cluster analysis is a statistical technique used to identify and classify homogeneous groups of similar objects or data points into clusters based upon their characteristics or attributes.  Such objects within the same cluster are more similar to each other than to those data points in nearby clusters.  Cluster analysis is an unsupervised machine learning technique.  "Unsupervised" denotes that the results do not rely on any predefined labels or categories. Rather, the machine learning technique discovers patterns and structures present within the data itself.  Cluster analysis can provide a powerful exploratory data analysis tool that can reveal hidden structures and patterns even within complex datasets.

What questions can be answered from this analysis?  For me, a few questions to consider are:

• Using only gaming period choice, do distinctions between historical and non-historical gamers emerge?
• Do some game periods tend to cluster together? Which ones?
• If distinct groups emerge from clustering, are these distinct groups intuitive?

The first step in cluster analysis (after wrangling the data into shape for analysis) is figuring out an optimal number of clusters.  There are 9,282 respondents having 26,741 data points (each respondent could have up to three period choices) used in this analysis.  Using only respondent wargaming period preference from the survey, these data are aggregated and classified using cluster analysis.  Only respondents and their choices are utilized in building this model.  Figure 1 illustrates the initial dendrogram showing how each of the twenty wargaming periods groups.

Figure 1

Starting from the position of hierarchical clustering, we begin on the righthand side of Figure 1 and identify the two-cluster solution by drawing a vertical line through the dendrogram bisecting the graph at the two-cluster solution.  

Figure 2

The two-cluster solution clearly and cleaning bifurcates the twenty wargaming periods into two, distinct groups. The two groups identified, with no ambiguity, are Historical periods and Non-Historical periods.  Well, perhaps a bit of ambiguity.  Pulp groups into the non-historical grouping but perhaps that makes sense since Pulp identifies with a wide genre of adventure/RPG gaming. 

The results demonstrate that historical wargamers generally tend toward historical gaming while non-historical gamers tend to remain in the non-historical genres.  Notice within Non-Historicals that Warhammer periods show distinct separation from the Fantasy/Sci-Fi/Pulp periods.  What emerges if the clustering is taken down to the three-cluster solution?

Moving the vertical cut to the left, the dendrogram is bisected across three branches to identify a three-cluster solution.  Note that the three-cluster solution keeps Non-Historicals separate and intact but Historicals are further split.  As we saw in the two-cluster solution, the split in the three-cluster solution is intuitive in that the Historical groupings are clearly split between Pre-1700 and Post-1700 wargaming periods.  

Figure 3

Note that while Pike & Shotte clusters with the Ancients/Dark Ages/Medieval groupings, that clustering does not occur until much earlier in the process. While Ancients/Dark Ages/Medieval periods join at the ten-cluster point, Pike & Shotte does not join until the four-cluster point.  Pike & Shotte is its own, distinct entity for a long time.

Let's cut the dendrogram one more time to examine the four-cluster solution.  What happens to the clustering solution as we move from the three to four-cluster dendrogram?

Figure 4

In a four-cluster solution (Figure 4), Non-Historicals and Pre-1700 Historicals remain unchanged. This time, the four-cluster solution splits Post-1700 Historicals into two groupings.  This bifurcation breaks out Napoleonic Wars, World War 2, and American Civil War from the rest of the Post-1700 Historical group.  I will classify this subdivision as the Big 3 Post-1700 Historicals cluster.  Naturally, my cluster naming conventions are subjective but I reckon they give a sense for the periods within each grouping.  Other naming conventions are possible.

Did I manage to answer some of the questions I originally set out at the beginning of this analysis?

Let's recap...

• Top 5 Unranked vs Top 3 Ranked.
• While there was slight movement in the ultimate clustering solutions, almost all results showed similar results between 2022 and 2023.  Perhaps any variation can be attributed to the addition of two more choices in the 2022 survey which may have dampened the distinctions between groups.  
• Reliability of Data.
• With the exception that WWII grouped with ACW and Napoleonics in the 2023 survey (perhaps due to the reduction of choices from Top 5 to Top 3), results were almost identical between the two surveys.  This suggests a level of stability and robustness present in the survey even across years.   
• Using only gaming period choice, do distinctions between historical and non-historical gamers emerge?
• Of course!  The two-cluster solution identifies this bifurcation early on.
• Do some game periods tend to cluster together? Which ones?
• In the 2023 survey as well as in the 2022 survey, wargaming period preferences tended to group within the same clusters.  See the figures above to verify.
• If distinct groups emerge from clustering, are these distinct groups intuitive?
• The two, three, and four-cluster groupings identified were given (what I consider) intuitive names.  Each name represents its component periods well.

Like in 2022, the results are fascinating and quite clear given that the only data inputs consist of up to three period choices for each survey respondent.  No other inputs are needed for formulating the inferences highlighted in this analysis.  

Hope you find these results of interest as well.