Social networking websites would seem to be the perfect venue for conducting a “Six Degrees of Separation” experiment over a large population. The advantages are that the networks involved are very explicit, and that as long as a large, popular network was used, there would be tremendous amounts of data accessible to the experimenters.

Early on in Facebook’s career, there was a section on each person’s profile that indicated how many friend links it took to get from you to them. (Sadly, it appears to have been removed from the more recent versions). From what I remember, this count was never more than three or four for people in the cornell community (the only profiles I could look at back when Facebook was closed), which would indicate (although by no means prove) that Cornell has a lower separation degree internally than the oft quoted six. This makes sense due to the small size of the campus, and the relatively close quarters in which all of us live and work.

Even now on Facebook, after a search for groups involving Six Degrees of Separation, there are a number that propose to research this phenomenon. However, these groups vary in size. The only sizable group is here , and has more than 500,000 members at the moment. The person who started the group has made one trial run where he tried to connect himself to several randomly selected members of the group, and succeeded with average separation degrees of 4.67 the first time and 4 the second time (taken at group sizes of 350k and 400k respectively). However, he only tested himself connecting to 6 people each time, so the data is perhaps somewhat suspect.

All the other Facebook groups are significantly smaller (none over 1000 members), despite some of them having better experimental procedure .For instance, one group aims to collect the friend data of all the users who join and then to analyze that offline, which would be much more effective at checking large numbers of chains. However, the most telling experiment would be if Facebook itself conducted the experiment, using all of the data stored on their servers. That would at least give an estimate of the separation over a very widely geographically distributed population (albeit on connected by a common interest). Because of this common interest, I would expect that such a survey would result in degrees of separation much lower than the original Milgram study (perhaps in the neighbor hood of 4 or 5, rather than the 6-7 that Milgram found).

Sadly, none of these steps really addresses the question of whether this is true in the world as a whole. For that, we are likely to wait a long, long time.