A new study on the behavior of Facebook users claims that users influence each other when it comes to installing applications, but only once an application has exceeded a certain level of popularity. Under that level, the effect of social influence is negligible.
The study, led by Oxford University, is based on data collected over two months in 2007 that covered all 2720 Facebook applications available at the time and the entire population of 50 million users. There has been a lot of previous research into the diffusion of innovations among populations, but typically such studies focus on the spread of a single successful technology across a subset of all its potential users. This study looked at all possible technologies (in this case all Facebook applications) and all possible users.
The researchers used a technique called fluctuation scaling to determine the extent to which the behavior of users was correlated, i.e. how much users influence each other when making the decision to install an application. In 2007, users got notifications every time a friend installed a new application, so a user with many Facebook friends had the potential to influence the behavior of a large number of other users.
Two distinct scenarios emerged. When installing applications under a certain threshold of popularity (55 installs per day in this data set), users did not seem to influence each other's behavior at all. When the number of installs exceeded 55 a day, social influence became extremely important in fueling further popularity. There was a huge range in popularity between apps. The top application had 12 million users, and the 100th most popular had 180,000, while the 1,000th had a mere 1,300 users.
The researchers concluded that this threshold of social influence is an inherent property of the way information about installations was disseminated in Facebook. Therefore, above that threshold, the popularity of a particular application was due more to social influence than to the inherent properties of the application. The researchers also believe the findings can be generalized to other online systems such as Amazon or Netflix, both of which allow their users to rate products and, consequently, influence their future popularity.