All Aboard the Hype Train
What makes the board game great?
Being an avid boardgamer for over ten years I have always wondered what does make a game successful? After starting my education at Lambda School in their data science program, I finally had the necessary tools to answer a decade-old question. So, I cracked my knuckles and started my first solo foray into the jungle of practical data analysis. The findings were… Interesting. Here’s what I’ve found.
The very first step on this journey was to figure out what exactly I’m looking for! My hypothesis was simple: game ratings depend on what game mechanics are, to which genre it belongs and who created it.
After all, it made perfect sense. Traditional game mechanics, most familiar to non-gamer are dice rolling, card drafting, bidding and blocking. These are all featured in the games from the olden days i.e. Risk or Monopoly. But today’s board games are significantly more complicated, and they can be built on a myriad of game mechanics which are too long to list here. In order to learn a new game, you must learn a new mechanic, therefore I proposed that people will tend to fall back to what’s already familiar to them.
Similarly, I thought that the genre of the game might be important as people gravitate towards the familiar setting, whether that was an economic game or sci-fi saga. Finally, in my view game designers should be like movie directors — everyone knows that whatever Michael Bay touches becomes pure gold.
My dataset featured the top 5000 ranked games from bgg.com (the ultimate board gaming database website). The baseline of the ratings was simply a mean rating of all the games in this set. This equaled 6 out of 10 or as the geeks would say — It’s ok, will play if in the mood.
The second step was to narrow down what mechanics and genres will we use in our analysis? According to the BGG there are 52 unique mechanics and 84 individual genres! After I ranked them, I got excited! Top 3 ranked mechanics showed up as Dice Rolling, Hand Management and Variable Player Powers; while the top 4 genres were Card Game, Wargame, Fantasy, and Economic. So far so good, my suspicions that popular mechanics and genres will be most heavily represented in top games appeared correct.
On top of these features, my dataset also had information which you would normally see printed on the back cover of a board game: recommended number of players, age and expected time to complete the game. The only piece of information which was added in the mix was how many people voted on the game ratings at bgg.com. It was finally time to run our predictive model.
The results were in and they were majestic! Our model was showing an R² score of 0.89! In simple English, this meant that our selected game characteristics explained the game ranking almost perfectly (max R² = 1). It was time to see which of these features was most important in building our model!
The number of votes was dominating! This sole feature contributed to over 70% of the game rating. Something wasn’t right! Normally when one feature dominates this much it means that our data is leaking (we are using the information to predict something, which we would not know before our prediction). But how can that be the case? After all, we only knew that people voted on the game, and not necessarily on how they voted.
To see how the mechanics and the genre contributed to the ratings I created a correlation matrix between them and the ratings. And these were the results:
Look at all that pale real estate! The findings were abysmal. Correlations ranged from -0.15 to 0.18 in both categories. Since correlation ranges between -1 to 1, this effectively meant that our star features barely had any impact on the game rating! Game designers also had a correlation of only 0.08. This simply meant that none of these features can be used to reliably predict the game ratings.
Running our model once more without the number of votes yielded an R² of 0.24. This meant that only a quarter of the games in this set can be explained accurately by using features from our starting hypothesis. This also implied that strongest indicator of game success will be the number of votes.
Anecdotal evidence supports the revised hypothesis as games that are normally popular on Kickstarter (a platform that allows for crowdfunding before the product is even made) end up being popular after publishing. It is in our innate psychology to try to justify our liking of something that we physically own. Does this mean that the skeleton of these games (mechanics), their skin (genre) and game designer don’t matter? Not necessarily. It just seems that hype surrounding the game will tend to propel it to the higher ranking. Marketing works and in board game industry you either board the hype train or you get run over.
All the math used in this post can be found at the following link:
