Understanding the viral invitation process and successful virality of Facebook social games is a subject that warrants deep investigation, and a formal paper on just such a topic has been released by a group of researchers and industry veteran Manu Rekhi. In “Diffusion Dynamics of Games on Online Social Networks”, the group investigates the invitation process of users within social games, specifically focusing on “predicting invitation efficiency and understanding the group and social dynamics of invitation networks”.
Written by Xiao Wei, Jiang Yang from the University of Michigan, Ricardo Matsumura de AraÃºjo from the Federal University of Pelotas, Brazil and Manu Rekhi, VP of strategy, marketing, business and corporate development for Lolapps, the investigation begins by defining a smart terminology for the analysis of the viral spread of an application. That is, who is inviting who, who is accepting, and how/why are these processes occurring. Below are some of the key findings and statistics, but I suggest you view the entire paper available here, because I only analyze a small section of their very complete findings.
Firstly, the paper defines the process of transmitting invitations between each other as ‘diffusion’, and adopts models of disease transmission to analyze the diffusion. To get data for the analysis, the group uses the extremely detailed data that is available for two of Lolapps popular games, Yakuza Lords (YL) and Diva Life (DL). As a testament to the level of analysis in the paper, here’s a quote about the information they know about their userbase:
As of February 2010, YL had reached one million active users, with over 85% being male. Although launched 2 months later than YL, DL gained over 2 million monthly active users in the same period. As a game targeted at women, DL has more than 96% female users. The age distributions are similar and range from teens to 70s, with the majority being in the 18 to 38 years old range.
One of the first striking elements of the paper is their breakdown of user behavior. Looking at this graph below, entitled “Figure 2: Distribution of Different Actions in YL/DL”, we can see that for both games, a great percentage of users begin by looking at the items and character screens, but as they last longer in the game and become more expert, most of their time is spent on the mission and battle areas. This corroborates well with common sense, but to see that the transition area is around 5 to 10 actions is the precise kind of tool that social game designers need to iterate their games and reduce the number of players that leave.
Furthermore, they go on to say that since each user has to explicitly share their information with the application, they can understand a bit more about their sharing habits: “around 90% users share their locale information, 40% users share their friend list but only 1% share their relationship status”. This is interesting to note, as Facebook’s pre-application screen is not just clicked through for many users, a great percentage choose what they will share with the applications they play.
The paper goes on to analyze some key factors of inviters and invitees. Specifically, they look at the various factors that cause an inviter’s invite to be successful, and then examine what we can learn about invitees based on their choices to accept invites. I’ve included some short form points from the paper that I found interesting, and strongly suggest you read the entire paper here.
- Out of all players who downloaded the two games analyzed here, more than 37% (for YL) and 25% (for DL) received invitations from their friends before starting to play the game.
- Compared to players who have never been invited, invited users remain in the game longer. Around 80% of non-invited players leave the game within the first day
and almost none keep playing longer than 80 days. But among invited players, over 50% kept on playing for more than a day and 20% of all invited users were still playing 80 days later.
- On average, each inviter has invited 26 friends while the median number is 10.
- For both games, there is a quick expansion in the first 6 generations of invitations. We see in the graph below the invitation patterns of a typical inviter. This shows that on average, the first few cascades generate an increasing number of ‘invited people, where a cascade refers to the total number of invites sent by people who have been just invited to the game. This total number of invites increase until we get to the fifth level of invites, where we see that the fifth group if inviters generates, on average, the peak number of total invites into the system. The paper discusses this diffusion of invites in more depth, and finds some fascinating results.
- Just 10% of users account for 50% of successful invites.
- The paper breaks down the invitation strategies of users to determine which invites will be successful. They break the “efficiency” of an invite, meaning the success rate, into several interesting factors and determines their affect on the efficiency.
- volume (# of friends invited): the more friends invited, the less success per invite
- pacing (time between invites):Â invitations that are more spread out in time are more likely to succeed
- repetition (# of invites to a friend): higher number of invites represent a higher chance of success
- selectivity (number of invites per invite session): uusers who invite friends individually tend to have a higher yield, possibly because they target their invitations to those who are more likely to accept.
- The paper also attempts to determine the demographics of the best type of inviter:
- demographics: no correlation: influential inviters are distributed randomly across various demographics with the exception of age: we observe that being older does confer a bit more authority and influence.
- ego-network profile (structure of friends):high friend count means weaker connections and lower overall success rate
- “It is puzzling that the shape and density of one’s FB network has little predictive power. A possible explanation is that these networks can be largely static and do not reflect the level of interaction between friends.”
- gamer activity: the longer they play, the more invites they send
- Overall, they find that invitation strategy is more important than demographics in determining invitation success rate.
- The paper goes on to cover the details of the invitee and which factors determine whether someone will accept an invite.