Cheers to Julia for pointing this paper out to me: Planetary-Scale Views on Instant-Messenger Network, Its nothing exotic, but what a data set: They had access to the compete MS IM dataset, modulo individual identifiers. They think they got half of all the conversations during the study. 30 billion conversations, 240 million distinct users, unless you assume a significant number of multiple accounts and robots, which I don’t think was addressed.
They get lots of broken power laws (yawn), and estimate the average degree of separation is 6.6, (double yawn), or that people in the 20s-30s are over represented with respect to the world population (ZZzzz…). There is cool stuff though, like inter-gender conversations last longer on average than single-gender, and there’s some weird off diagonal nodes on reported age correlations or participants. There’s a cool map of users per capita of the world, and you see a significant asymmetry in the US, with more users per capita in the western half as in the east, with about the same density as in Australia. This is clear anticorrelation with population density, but it looks cool. People in Arabic nations seem to have significantly long conversations on IM, why this is is not immediately obvious to me.
There’s some weird extremities of the dataset: The tail of the AddBuddy events distribution shows that some people actually have 600 contacts (the maximum) on IM. Wild. This smells like robots to me.