GWS 2023: Location, gaming, and the Magic Quadrant
By Jon Freitag
The last time we looked at the Great Wargaming Survey, we examined the possibility that survey respondents' location is an influencer in driving gaming preferences. That analysis focused on a series of pairwise comparisons of location against table size, game venue, group size, gaming frequency, game length and favorite wargaming period. Almost more than a handful of attributes! Each attribute had multiple levels (treatments) as well. Only those survey respondents who classified themselves as primarily historical wargamers were included. As a refresher, these pairwise comparison analyses are found at Does Location Influence How We Game?. The survey results demonstrated that location not only plays a role in how we game, but also what we play.
The blog post summary suggested that a further "next step" in the analysis would be to combine these pairwise comparisons into a multi-dimensional, holistic examination of group association. Today's analysis is an extension of the pairwise comparisons seen earlier and addresses the multi-dimensional question.
Multiple Correspondence Analysis (MCA)
Multiple Correspondence Analysis (MCA), briefly, is a statistical technique from which any underlying structure in the categorical response survey data may be detected. MCA is an extension of Simple Correspondence Analysis. The results are then presented in graphical form in two-dimensional, Euclidean space. Interpretations are understandable and intuitive without knowing the underlying technique. Interpretation, though, can be as much art as science. Interpretations are only tendencies and associations based upon the survey data.
Data Reduction
Plotting the analysis outcomes over seven attributes, each having multiple levels makes for a messy plot. In order to make the resulting plot more manageable and (hopefully) more intelligible, data reduction techniques are introduced. The main technique utilized is simply re-binning levels of each attribute to reduce the number of levels under consideration.
After re-binning, the levels for each of the attributes are:
- Location: UK (and Ireland), USA (and Canada), Europe (Continental Europe and Scandinavia), Australasia (Australia and New Zealand).
- Wargaming Periods: Only the top periods remain. They are: WWII, Napoleonics, Ancients, Medievals, 18th Century, ACW.
- Game Frequency: Weekly, Monthly, Less Than Monthly.
- Game Length: 1-2 hours, 2-3 hours, 3+ hours.
- Game Type: Scenario, Campaign, Pick-up.
- Game Venue: Solo, At home, At club, At game store, Friend's home.
- Table Size: Small, Average, Large.
Now consider the interactions between the seven wargaming attributes. When analyzed using MCA, the seven categorical attributes with their accompanying levels are plotted in two-dimensional space as illustrated in the graphic below.
![A Multiple Correspondence Analysis plot](https://cdn.shopify.com/s/files/1/0296/7067/0384/files/gws2023_location_mca01.png?v=1709657275)
Some of the levels are clustering near others while some levels are seemingly plotted as outliers. Like correspondence analysis, attributes plotted along the origins exhibit little differentiation. Those levels farther from the origin demonstrate more separation and distinction. At this stage, inferences can be difficult to make. Adding in the origin reference lines helps to accentuate these differences. Reference lines also allow distinctions between each of the four quadrants.
![Improved plot of Multiple Correspondence Analysis](https://cdn.shopify.com/s/files/1/0296/7067/0384/files/gws2023_location_mca02.png?v=1709657380)
Magic Quadrant
Once the plot is divided into quadrants, each quadrant is examined to discover any associations between the levels of the various attributes. For ease of visual identification, each quadrant is color-coded.
![Colorised improved plot of Multiple Correspondence Analysis](https://cdn.shopify.com/s/files/1/0296/7067/0384/files/gws2023_location_mca03.png?v=1709657467)
Does each quadrant present a specific collection of levels and attributes? Are levels of an attribute unique to a particular quadrant? Is there overlap? Can any useful inferences be gleaned from the MCA?
Since gamer location was the motivation for these studies, let's examine Location. The Magic Quadrant shows that each location level appears in one quadrant only. There is no overlap of two in the same quadrant. That result suggests that location does influence how gamers game. Do other general tendencies surface from the analysis? Are there tendencies peculiar to a specific region? Summarizing by quadrant produces the following tendencies and associations:
- UK (Orange Quad): Tend to meet weekly at a club for two to three hours per session. Clubs tend to have ten or more gamers. Perhaps a slight tendency to game 18th Century and WWII more than other regions.
- Australasia (Yellow Quad): Tend to game monthly at either a friend's house or at a game store. Games tend to be played on large tables with between five and nine players. Game sessions last three or more hours. May feature Napoleonics more than other regions.
- USA (Green Quad): Tend to game less often than either UK or Australasia, at home, and with one to four players. Tend to game ACW more often than other regions.
- Europe (Blue Quad): Tend to game solo, on small tables, with games only one to two hours in duration. They tend to play Ancients and Medievals more than other regions.
Interesting results. MCA can provide a data-driven dimension to uncovering tendencies of gamers across the globe. Does location influence what is played and under which conditions? The survey results suggest, that yes, it does.
Of course, these are broad generalizations of the tendencies provided by respondents to the survey. We are individuals. Still, does this present a picture of the wargaming hobby that is familiar to you?