Matthew Gray
United States
Reading
Massachusetts
flag msg tools
admin
Avatar
mbmbmbmbmb
I've done a bunch of recent analysis of BGG data, centering around the notion of gamer clusters. This post summarizes the analysis and the results and discussion those produced.

Executive Summary

By using a k-nn clustering algorithm on user "Top 10" data, 5 distinct groups of gamers emerge. These groups each have their own characteristic preferences, and provides some interesting insight into the gamer demographics of BoardGameGeek.

The groups are...

Euro-1: The core Eurogamers, composing roughly half of BGG users. Puerto Rico, Power Grid, Caylus, and Settlers are the most loved games in this group.
Euro-2: The family Eurogamers, about a fifth of BGG users. Carcassonne, Settlers, Ticket to Ride and Lost Cities are the most loved games in this group.
Euro-3: The complex Eurogamers, about an eighth of BGG users. Tigris & Euphrates, Ra, El Grande and Age of Steam are the most loved games in this group.
Ameri-1: The Amerigamers, about 10% of BGG users. War of the Ring, Memoir '44, TI3e, and Arkham Horror are the most loved games in this group.
Ameri-2: The Wargamers, under 10% of BGG users. Hannibal: Rome v Carthage, ASL, Paths of Glory and Up Front are the most loved games in this group.

Other than the general interest in categorization and classification that tends to be attractive to board gamers, these clusters are useful in identifying which demographics particular games appeal to.

Qualifications

Before going into more detail, it's worth adding some qualifications.

The majority of BGG users fall fairly strongly into a particular cluster, but a large minority either straddle multiple clusters, or don't match any cluster particularly well. These are high-level characterizations with statistical foundations, not a precision delineation of everyone's preferences.

The clusters are clusters of gamers, based on their preferences, not games. Many games are well liked by gamers in multiple clusters, often quite strongly. The games listed in the characterization above are those that are distinctive to that cluster. Some games, such as Ra, appear in the Top 10 for each of several clusters. All of the clusters are fairly eclectic in tastes, particularly Euro-1, with games from many genres and categories being well liked, but each has its own distinct "flavor".

One final important qualification is to recognize that these clusters are driven around what games people like, not at all what they dislike. While a particular cluster may not like a particular game or category of games, what unifies the clusters is which games they particularly like.

Data Details and Sample Selection

Just shy of 3500 BGG users have completed Top 10 lists, and while this is a significant number, it is sufficiently lower than the population of BGG at large, that it raises an important question: Are these 3500 a reasonable cross section, or is there a substantial bias toward certain kinds of gamers?

It's hard to know for sure, but one validation point is to compare non-Top10 measures to the BGG population at large. The answer appears to be that the users who have entered top 10 lists are a fairly reasonable cross section of BGG users. In fact, it's remarkably good. Looking at collections, the clustering sample (the 3500) represents roughly 25% of games owned for games across almost all genres. The only category that seems even slightly overrepresented is the "heavy Euro" category, which seems to come in consistently closer to 30%. Looking at ratings, the ratings line up very closely with the BGG population at large, with possibly a slight bias toward heavier games and away from lighter "family" games. Overall, the users who have filled in top 10 represent a fair cross section of BGG.

History and the GeekLists

I had been experimenting with clustering of game and gamer data for some time. I used things like user ratings and collections and found some interesting, but not particularly illuminating. For examples, there were clusters of "people who rate games highly on average" and "people who rate games lowly on average". Attempts at normalization simply changed the clusters to things like "people who only own/rate games in the top 100" and "people who own/rate games that are bad".

Eventually, I tried doing clustering based on Top 10 lists. Immediately, the results were more promising. In October, 2006 I did a 10-ways clustering (http://www.boardgamegeek.com/geeklist/17062). The main things learned from that list were 1) Be very careful about naming clusters. People are touchy. 2) The clustering approach yields results that people find interesting.

In January, 2007, I decided to run the clustering again. More people had entered Top 10 lists and I wanted to try a regression where an optimal number of clusters could be found. So, I ran with every number of clusters from 2 to 12 and a few higher than that. While there was no obvious singular "elbow" one hopes for in this kind of analysis, there were plateus at 3, 5 and 9 clusters. I posted them all in a GeekList (http://www.boardgamegeek.com/geeklist/18879).

After some discussion and reflection, it became pretty clear that the 5-way clustering was the most descriptive, interesting and illuminating. Of course, as I point out above, not every person falls neatly in one cluster, and any particular game is often liked, strongly or weakly, by multiple clusters. Interpreting these numbers, while not overly onerous, seemed to scream out for a way of visualizing it. To this end, I created a self-classifier and "gamer grapher" and posted some of the results in another GeekList (http://www.boardgamegeek.com/geeklist/18930).

Given these groups, it's interesting to see what games would be the top rated games for each cluster (http://www.boardgamegeek.com/geeklist/18977) and which games have the greatest disparities in ratings between clusters (http://www.boardgamegeek.com/geeklist/19033).

What good is it?

One of the ways in which game ratings are useful is to help narrow down an often very large field of games one might be interested in to a smaller set, based on the opinions of other BGG users, under the assumption that other BGG users have opinions somewhat like ours.

Of course, one of the ways which game ratings are not useful is that there are many BGG users who's opinions you may disagree with. As such, BGG ratings in general are a coarse guideline as to game quality. It is a rare person who likes more games with a rank worse than 3000 than they do games with a rank above 400. If we could somehow get ratings only from the people "like us", that would be more useful. Obviously, there aren't only 5 types of gamer, but for the majority of BGG users, they fit in one of these groups reasonably well, and better in that group than BGG as a whole. So, the top ranked games for each cluster (http://www.boardgamegeek.com/geeklist/18977) are hopefully useful, especially to those who strongly correspond to a particular cluster.

See my recent post about ratings (http://www.boardgamegeek.com/thread/146451) for more on this topic.

Further, I hope to incorporate some of the analysis here into the recommendation algorithms, including possibly moving towards personalized recommendations.

Other things the clusters have shown

While I doubt anyone ever questioned the fact that "wargamers" and "eurogamers" made up distinct preference groups, there has been some skepticism as to the degree to which some of the other groups genuinely existed as distinct. The clustering makes it clear that the "Amerigamer" group is real, though perhaps the boundaries that the recorded preferences reveal are not always the ones people described. Further, the "Family Eurogamer" emerged as more clearly distinct than I (and others) would have expected.

The clusters also show that some preference groups one might expect to exist either do not have any substantial number of users, or have sufficiently broad tastes in other categories so as to not be meaningfully discernible. These groups include "party gamers", "abstract strategy gamers", "classic American gamers", "train gamers", and "miniatures gamers". This is not to say these groups aren't present, just that there aren't many of them, and more often they fall well into the other clusters (for example, the "abstract strategy gamers" fit in "Complex Eurogamers", mostly, and "classic American gamers" fall in "Family Eurogamers", mostly).

The clusters show that while there are distinct preference groups on BGG, those groups don't disagree much on what constitutes a good game or a bad game. They have different views about which ones are the best. But, the "best" game for any given cluster is well liked by the other clusters, they just don't consider it to be the best. The difference between a "good game" and a "bad game" is far greater than the difference between "the kind of games I like" and "the kind of games you like".
  • [+] Dice rolls
Paul DeStefano
United States
Long Island
New York
flag msg tools
designer
badge
It's a Zendrum. www.zendrum.com
Avatar
mbmbmbmbmb
Is your wife aware of this cluster obsession?
12 
 Thumb up
 tip
 Hide
  • [+] Dice rolls
Matthew Gray
United States
Reading
Massachusetts
flag msg tools
admin
Avatar
mbmbmbmbmb
Geosphere wrote:
Is your wife aware of this cluster obsession?


Oh, yes.

And she's been a real dear to put up with my ongoing ranting and pondering in order to figure out how to use it.
6 
 Thumb up
 tip
 Hide
  • [+] Dice rolls
Christopher Dearlove
United Kingdom
Chelmsford
Essex
flag msg tools
SoRCon 11 23-25 Feb 2018 Basildon UK http://www.sorcon.co.uk
badge
Avatar
mbmbmbmbmb
mkgray wrote:
Euro-1: The core Eurogamers, composing roughly half of BGG users. Puerto Rico, Power Grid, Caylus, and Settlers are the most loved games in this group.
Euro-3: The complex Eurogamers, about an eighth of BGG users. Tigris & Euphrates, Ra, El Grande and Age of Steam are the most loved games in this group.


The clustering may be valid (I haven't finished reading yet) but I'd dispute the names. I don't see the games in the third list being significantly more complex than those in the first. In fact the simplest game on either list is probably Ra on the third list.

Of course I'm not in any of your clusters - I haven't rated any games. If I did you'd find the games on all three of your Euro lists high on my ratings. What that means I've no idea.

Edit: Wrote this before reading "Be very careful about naming clusters. People are touchy." Well, I don't think it's touchiness, after all I've not identified with any of the clusters.
2 
 Thumb up
 tip
 Hide
  • [+] Dice rolls
Richard Pardoe
United States
flag msg tools
badge
Avatar
So....where is the "What Cluster am I" tool to go with this analysis?
7 
 Thumb up
 tip
 Hide
  • [+] Dice rolls
Mark McEvoy
Canada
Mountain
Ontario
flag msg tools
badge
Avatar
mbmbmbmb
^ http://mkgray.com:8000/cgi/gamercluster2

(think you're gonna have to cut and paste the link, I believe BGG breaks it at the colon)

Fascinating stuff. I was playing around with your 'classifier' and have one question and some comments.

My question is, what do the games in red mean, on the match lists?

100 Eurogamer (Type 2) (822 13 18 3 118 25417 10)
65 Eurogamer (Type 1) (13 822 118 18 3 215)
52 Eurogamer (Type 3) (3 118 215 18 521 13 822 25417)
21 Amerigamer (Type 1) (25417 18 3)
6 Amerigamer (Type 2) ()



What seems at least a little funny to me is that Roborally, a heavily-themed combattive spatial miniatures race/combat game designed by an American... strongly correlates me to all three Eurogamer groups (and in red, which I suspect is some sort of especially-high correlation, in Euro 2).

It also seems funny that Crokinole, a Canadian dexterity game, hits on Euro 3 (alone). And Tichu, a climbing card game rooted deeply in traditional Asian card games, hits high for Euro as well.

 
 Thumb up
 tip
 Hide
  • [+] Dice rolls
Matthew Gray
United States
Reading
Massachusetts
flag msg tools
admin
Avatar
mbmbmbmbmb
Dearlove wrote:
I'd dispute the names. I don't see the games in the third list being significantly more complex than those in the first. In fact the simplest game on either list is probably Ra on the third list.


Yeah, the label for Euro-3 is not the best. I wouldn't fixate on Ra's presence. Everyone (ie, all clusters) loves Ra. I'm not sure what a better name would be, but I definitely feel that the games in Euro-3 have something that is less present in the Euro-1 list. That said, Euro-1 and Euro-3 are the3 closest clusters.

Quote:

Edit: Wrote this before reading "Be very careful about naming clusters. People are touchy." Well, I don't think it's touchiness, after all I've not identified with any of the clusters.


Picky?

Maybe I'll stick with Euro-3.
 
 Thumb up
 tip
 Hide
  • [+] Dice rolls
Susan F.
Canada
Lethbridge
Alberta
flag msg tools
Avatar
mbmbmbmbmb
Quote:
So....where is the "What Cluster am I" tool to go with this analysis?


Graph yourself here: http://mkgray.com:8000/cgi/gamercluster2 Be prepared to enter the gameIDs from your Top Ten games.

I could have sworn that link wasn't there when I clicked "Reply"
1 
 Thumb up
 tip
 Hide
  • [+] Dice rolls
Matthew Gray
United States
Reading
Massachusetts
flag msg tools
admin
Avatar
mbmbmbmbmb
thatmarkguy wrote:
^ http://mkgray.com:8000/cgi/gamercluster2


I'll add this to the original post along with the description of the colors, but your guess is essentially correct.

Quote:

What seems at least a little funny to me is that Roborally, a heavily-themed combattive spatial miniatures race/combat game designed by an American... strongly correlates me to all three Eurogamer groups


If you look at just RoboRally alone, it's Euro-2, Ameri-1, Euro-1, Euro-3. I'm slightly surprised Ameri-1 isn't on top, but I've found a lot of the Euro-2's are "introducable to non-gamers" and RoboRally has fit that bill, for me...

Quote:

It also seems funny that Crokinole, a Canadian dexterity game, hits on Euro 3 (alone). And Tichu, a climbing card game rooted deeply in traditional Asian card games, hits high for Euro as well.


Yeah, I suspect that's a side-effect of the gaming-groups which Crokinole has become popular in, but who knows...
2 
 Thumb up
 tip
 Hide
  • [+] Dice rolls
Christopher DeFrisco
United States
Ashland
Oregon
flag msg tools
Ah, negative. I am a meat popsicle.
Avatar
mbmbmbmbmb
I love the work your doing. Keep at it.
I'm sure this has been expressed before (in fact I know it has... and I should probably read past posts), but is there any plan to show us how each of us falls into these groups???
 
 Thumb up
 tip
 Hide
  • [+] Dice rolls
Christopher DeFrisco
United States
Ashland
Oregon
flag msg tools
Ah, negative. I am a meat popsicle.
Avatar
mbmbmbmbmb
DOH.
Just saw the link. shake
 
 Thumb up
 tip
 Hide
  • [+] Dice rolls
!
United States
Brooklyn
New York
flag msg tools
badge
Avatar
mbmbmbmbmb
I find all this fascinating, but since my best cluster match is only 0.17 (i always knew I was a freak), not very promising in terms of personal usefulness. Any thoughts on how clusters might be useful to those of us who don't cluster well?
 
 Thumb up
 tip
 Hide
  • [+] Dice rolls
shumyum
United States
flag msg tools
badge
♒♒♒♒♎♒♒♒♒ sloooowly sinking
Avatar
Awesome job. I really like this stuff.

My humble suggestions for the cluster names:

Family Eurogamer --> Light Eurogamer
Complex Eurogamer --> Heavy Eurogamer

Edit: I see that's how you define them in your cluster tool thingy. Makes sense to me!
 
 Thumb up
 tip
 Hide
  • [+] Dice rolls
David A
United States
Los Angeles
California
flag msg tools
Avatar
mbmbmbmbmb
travistdale wrote:
I find all this fascinating, but since my best cluster match is only 0.17 (i always knew I was a freak), not very promising in terms of personal usefulness. Any thoughts on how clusters might be useful to those of us who don't cluster well?


I was a 0.39 myself and it tells me that my next microbadge should be an "I love all games" badge rather than a Wargamer or AmeriTrash badge. It turns out I lean slightly towards family Euros! Ha! Who knew?

I love this tool, Matthew. Now I'm off to analyze my "hot 10" instead of my top 10.
 
 Thumb up
 tip
 Hide
  • [+] Dice rolls
Matthew Gray
United States
Reading
Massachusetts
flag msg tools
admin
Avatar
mbmbmbmbmb
travistdale wrote:
I find all this fascinating, but since my best cluster match is only 0.17 (i always knew I was a freak), not very promising in terms of personal usefulness. Any thoughts on how clusters might be useful to those of us who don't cluster well?


Yeah, if you're a weak match and you want to get a better sense of where your cluster alignment lies, do the following:

Go through the BGG Top 200 games and pick the 5-15 games that are your favorites, from that set. Then enter those gameids into the tool. You're much more likely to get a strong cluster match, since, unsurprisingly, the games most frequently appearing on top 10 lists are the "top games".

(edited: fixed odd/wrong wording)
 
 Thumb up
 tip
 Hide
  • [+] Dice rolls
Paul Imboden
United States
Evanston
Illinois
flag msg tools
designer
publisher
badge
Avatar
mbmbmbmbmb
RA is a COMPLEX Euro?

1 
 Thumb up
 tip
 Hide
  • [+] Dice rolls
Teh Slipperboy
United States
Allen Park
Michigan
flag msg tools
designer
Have you joined the 2017 Secret Cthanta yet?
badge
Avatar
Opie wrote:
RA is a COMPLEX Euro?



Ra is not complex to play, but it seems to me that it is a game that can benefit from complex analyis while being played.

Besides, this tool isn't about games it's about gamers. So Ra is a game that is liked (in the top 10) by gamers who like complex games.

Edited poor grammar.
1 
 Thumb up
 tip
 Hide
  • [+] Dice rolls
Justin
United States
Michigan
flag msg tools
Avatar
mbmbmbmbmb
According to the cluster machine I am:

100 Eurogamer (Type 2) (50 10630 14996 11170 483 9674)
88 Amerigamer (Type 1) (10630 11170 483 14996 15363)
65 Eurogamer (Type 1) (10630 14996 50 483 11170 9674)
47 Eurogamer (Type 3) (10630 9674 50 483 1115 11170)
26 Amerigamer (Type 2) (483)


A family eurogamer. That is exactly what I am! Cool beans!

Nice tool!
Justin
 
 Thumb up
 tip
 Hide
  • [+] Dice rolls
Eric Brosius
United States
Needham Heights
Massachusetts
flag msg tools
badge
My favorite 18xx game for six players is two games of 1846 with three players each.
Avatar
mbmbmbmbmb
mkgray wrote:
And she's been a real dear to put up with my ongoing ranting and pondering in order to figure out how to use it.


You might even call it "carrying on"...

 
 Thumb up
 tip
 Hide
  • [+] Dice rolls
Robert Wesley
Nepal
Aberdeen
Washington
flag msg tools
designer
badge
Avatar
mb
MarvelNemesis wrote:
According to the cluster machine I am:

100 Eurogamer (Type 2) (50 10630 14996 11170 483 9674)
88 Amerigamer (Type 1) (10630 11170 483 14996 15363)
65 Eurogamer (Type 1) (10630 14996 50 483 11170 9674)
47 Eurogamer (Type 3) (10630 9674 50 483 1115 11170)
26 Amerigamer (Type 2) (483)


A family eurogamer. That is exactly what I am! Cool beans!

Nice tool!
Justin
and according to that same 'dealie' then this is what I 'got'!
According to the "cluster machine"
GROGnads is:

*whir-r-r-r* 'skir-r-r-l-l' "chirp"-"chirp"
"G-R-O-G-n-a-d-s...that-does-not-compute..."
"D-A-N-G-E-R-! Will Robinson!...."
"Dave? what are you doing DAVE!?!?..."
"Dave's not HERE!"..."I KNOW man, I'm DAVE! I got the 'stuff' man! Let me IN!"
"What we have here'r...is a 'failure to corroborate-&-collate'..."


uhh...I might have 'broke'd' this a "tad", ya think?
surprise


15 
 Thumb up
 tip
 Hide
  • [+] Dice rolls
The Seal of Approval
Austria
Vienna
flag msg tools
badge
Avatar
mbmbmbmbmb
Quote:
If we could somehow get ratings only from the people "like us", that would be more useful.


People "like us", that would be people who rate games similar "to us". Should be possible to find by matching existing game ratings.
1 
 Thumb up
 tip
 Hide
  • [+] Dice rolls
Chris
United States
Ronkonkoma
New York
flag msg tools
badge
Avatar
mbmbmbmbmb
Is there any consideration for the sports simulations?

I'd imagine they're not considered Euros - but, when you realize that, for me, at least, 3 of my top 10 are sports sims (and 2 of my top 5), isn't ANY cluster going to be somewhat off?

Chris
 
 Thumb up
 tip
 Hide
  • [+] Dice rolls
United States
Boise
Idaho
flag msg tools
badge
Avatar
mbmbmb
Ack! Reading this made my temples throb and my eyes bleed... then I got to Grognad's post and the pressure was relieved... ahh, sweet reality.
2 
 Thumb up
 tip
 Hide
  • [+] Dice rolls
Mark McEvoy
Canada
Mountain
Ontario
flag msg tools
badge
Avatar
mbmbmbmb
Lemur wrote:
Is there any consideration for the sports simulations?

I'd imagine they're not considered Euros - but, when you realize that, for me, at least, 3 of my top 10 are sports sims (and 2 of my top 5), isn't ANY cluster going to be somewhat off?

Chris


If his site was currently up, I could plug in those gameIds and see what clusters they correlate to. Hope he gets it back online sometime soon.

But, as I mentioned before, some peculiarities *will* arise. RoboRally matches all three Euro clusters as well as Ameri-1. Tichu matches a Euro cluster. Crokinole matches a Euro cluster.

It seems there are also some games that don't seem to match with any cluster at all. "Mu&Mehr" is the only one of my top 10 that seems to fall nowhere - it's noit in enough top 10's. I expect the detailed sports sims will either fail to register (not on enough lists), or will probably fall in with the wargamers on Ameri-2 (Blood Bowl maybe Ameri-1). But we'll see. I expect some lighter sports fare like StreetSoccer probably hits a match with a Euro list or two (if any).
 
 Thumb up
 tip
 Hide
  • [+] Dice rolls
Chris
United States
Ronkonkoma
New York
flag msg tools
badge
Avatar
mbmbmbmbmb
Oh, I believe they'll fail to register, as well (in fact, I think I used his tool, and they didn't pop up).

I was basically wondering, based on my own top 10 - which includes 7 Euro/Strategy and 3 sports sims, which cluster do I fit in? I know which one, based on the 7 games that probably DO register; I don't know if the sports games impact that at all.

It was a completely selfish question.

Chris
 
 Thumb up
 tip
 Hide
  • [+] Dice rolls
1 , 2 , 3  Next »   | 
Front Page | Welcome | Contact | Privacy Policy | Terms of Service | Advertise | Support BGG | Feeds RSS
Geekdo, BoardGameGeek, the Geekdo logo, and the BoardGameGeek logo are trademarks of BoardGameGeek, LLC.