Retention and progression: Seven months in World of Warcraft

Knowing which players will stop playing is important for online game companies. Players generate massive amounts of behavior data every day. Can we mine this data to predict if they will churn? Is churn related to to the pace of progression through game content? We collected demographic and motivational data from a survey of 1350 players from China and North-America, and matched it with raiding and player-versus-player data collected from the game between December 2011 and June 2012. We find that the ratio of active player base who raids remains constant around 50% across all seven months. However, the active player base of June raids half as much as its December counterpart. Our results also indicate that Chinese players are more focused, while North-Americans more adverse to difficulty. While 10% of the player base churn every month, 5% come back, thus netting a 5% active player loss per month. A simple regression model predicting player churn the following month using only three in-game features achieves 0.90 recall but only 0.28 precision, suggesting that churn remains challenging to predict.