Two sociologists have conducted a study on "unfollow behavior" on Twitter. They have discovered that, over a time period of 6 months, the average Twitter user in their sample lost 39% of his or her followers. They also looked at which factors had an influence on the probability of unfollowing in a given "pair" of Twitter users. Reciprocity (are the users following each other) lowered the probability of an unfollow, as did the overall follow-back ratio of a Twitter user, and the "embeddedness", i.e. how many mutual "friends" the two users had. 

This is all hardly surprising (except maybe for the figure of 39% of unfollows - I had expected a much lower number), since it mirrors real-life behavior: The closer two people are, and tighter they are integrated into the same social network, the lower the probability that relationship is (for whatever reason) terminated. But it might suggest that social media platforms like Twitter and Facebook provide a vast field for sociologists (and other social scientists) to study human behavior - because on these platforms, data on relationships between users is easily available, and scientists don't need to rely on interviews and experiments.