Subterfuge by the Numbers
A data dive into the best iPhone game I’ve ever played
The best way I can describe Subterfuge is Risk that moves in real-time instead of turns and with private chats. But this game can take up to a week or more to finish. The teaser here is the quickest way to learn how it works.
After the first day of playing, I was hooked. Eventually, I found the forums where players discussed strategy and their love of the game and landed on the leaderboard page. After finding where I stood among all players, I thought this would be a perfect example to learn how to scrape data from the web, analyze it and visualize it, all in R.
I had a few questions that I hoped to answer with my analysis and visualizations.
- What can we tell about the success of Subterfuge as a game from this information?
- Does experience, either days since joining or number of games played, matter?
- Does the rate of games played per week matter?
Was Subterfuge a Success?
I wanted to see how many people have found and played this wonderful game. The leaderboard provides a few metrics to review for this.
- Left: I placed a red line on October 15, 2015 when the app was publicly released. There area some users who were beta-testers and joined to the left of that line. There was a huge amount of users who joined in the immediate days after release, but that rate was not sustained. There seems to be around 10 users joining per day in the past year.
- Middle: Once users join, they also tend to play a few games then drop off. I have played in 34 games (note that 80 users have played in more than 45 games but I removed them for the sake of the visualization).
- Right: This graph tells the same story as the middle graph; users tend to play a game, and assuming 1 week per game, not play soon after finishing the first one. See later on for more analysis on the rate of games played per week.
Unfortunately, it looks like the developers were right not to spend more resources on Subterfuge. After the initial boom of downloads, there hasn’t been another sharp rise and users tend to not play multiple games (which is a shame, in my opinion).
Subterfuge Experience
My first assumption was that as you play more, you get better. You learn how to combine specialists to attack well and how to defend against other specialist combos. And most importantly, you learn how to chat and make allies.
First, I needed to look into the ranting system. The developers decided to use an ELO scale, meaning your rating will increase or decrease if you win or lose, respectively. Everyone starts with an ELO of 1200. The magnitude of change to your rating also depends on the rating of your opponents after a game. On the left is a distribution of every rating for every player. It is a fairly standard bell curve, with a majority of players to the left of 1200. Perhaps players lose or resign ni their first game, then abandon the app. It should be noted that all of the observations are for Registered Users, meaning they paid $9.99 to access the rated games.
But let’s look to me. As of February 19, 2017, my current rating is 1329 and that places me at 905 of 7712 total players. Look me up, my username is Baer.
Above are two identical graphs showing the relationship between the total number of Rated Games Entered and the Rating of that individual player. The purple horizontal line is at 1200, the starting ELO.
The red line shows a linear model and the blue shows a general additive model. There is a huge amount of variance as the amount of experience grows. With the blue general additive model, 25 games seem to be the sweet spot for rating level. After 25 games, there isn’t much left to learn. The wiggle in the line is reacting to the amount of variance. it is likely a plateau.
On the right, you can see me, Baer, sitting just above both models. I beat the curve!
Rate of Games played per Week
My rate of games played per week is .51 games per week, or once every two weeks. I have taken a break from playing for the past month due to international travel and a broken phone.
My rate seems to be the perfect amount to stay with the highest rating. That makes sense to me. Playing in more than one game at a time (which I have done a few times) is difficult. Subterfuge doesn’t command 100% of your time, but it does require your attention throughout a day. I have sent messages to the wrong people before.
But again, the linear model shows a positive correlation, showing that playing at a more frequent rate does increase a player’s rating.
Conclusions
Subterfuge is a rich game that has provided me hours of entertainment. The graphs didn’t show anything too surprising, but it was good to confirm my theory: the more you play, the higher the rating (until a point).
However, there are some more data that I wish I had and analysis left to be done:
- Change over time. All of the analyses above were for the leaderboards at one specific moment (Jan. 31, 2017). I would love to see my rating count on a timeline with its peaks and valleys. If I was dedicated enough, I could run my web scraper script every day and create a database…
- In-game data. The leaderboard is a table of results of every game. But how do choices in the game lead to wins and losses? Which specialists are chosen most? Who builds the first mine and when? How much time has each user logged with the app open?
- Player categories. Can I determine player groupings based on the amount and type of medals earned? I’ve noticed that some players go for gold all the time at the risk of being eliminated by betraying allies in the end game (aggressive type) while some players are happy with a silver or bronze if their ally gets the gold (support type).
Leave me a comment if you want to see where you compare on these graphs or if you think you can help take this to the next level.
All code to scrape, clean and visualize this data can be found on my GitHub, found here.
