Recently, I’ve become intrigued by the demographics of social network websites. By now, of course, they’re nothing new. The early networks (Classmates.com as early as 1995, SixDegrees.net in 1997) pale in their relative success to the second generation of social network sites: Friendster started in 2002, then came LinkedIn and MySpace, and of course Facebook which now has boasts a shocking 55 million members and was founded by a Harvard undergraduate in 2004. All of these sites predicate their success to some extent on the Network Effect, a term coined by Ethernet inventor Robert Metcalfe, which describes the phenomenon that a network becomes incrementally more valuable to its users each time an additional member joins the network. (Specifically, Metcalfe’s Law states that the value of a network with n nodes is proportional to n2.)
For various reasons, Facebook has become my social network of choice. In the gross network size sumo match, it wins handily (55 million users at last count), and Facebook keeps growing in my consciousness, so I got curious: Just who is using these sites? Is the demographic changing?
For starters, I was convinced I could predict the likelihood of a person being a Facebook member based on their age. Could I? You betcha! It took some digging, but after a couple of days of running searches and compiling statistics, I had solid data supporting the hypothesis. (See below for methodological notes.) Here, for example, are the gross number of Facebook members who voluntarily associated themselves with specific graduating classes at four top US universities:
So the number of Facebook members increases as college graduation year increases. Not terribly surprising, but I was somewhat surprised by the “spread” between Harvard and Princeton. According to this first cut, it appeared that Princeton graduates are less likely to adopt Facebook if they are older, whereas Harvard graduates are somewhat more likely to do so. Once I corrected for the size of the undergraduate populations of the respective schools, the spread shrank significantly:
There is still a difference between the universities, but the trend is considerably tighter. (For those wondering why the lines stop earlier for Harvard than for Princeton, and why they don’t reach as high, see the methodological notes.) Those who like to look closely at charts will notice that the trend is really surprisingly monotonic: there just aren’t that many later years which have lower membership than earlier years. Taking the average makes the trend even more obvious:
Now it gets interesting. These data pretty much prove that Facebook usership is seeing exponential demographic growth: the exponential regression shown above (dotted white line) is by far the most accurate regression for these data (r2>0.99; nothing else comes close).
As an aside, note that members counted in the statistics above graduated college before Facebook, or any of the social networking sites with any real following, launched. This is particularly interesting because it points to some further areas of research: what is causing this adoption? Is this a “sideways” look at Metcalfe’s Law in action?
So age does matter, if you want to find someone on Facebook!
Next question: what can we make of the differences between adoption rates at different universities? It turns out there are some trends there as well. For starters, I broadened the focus beyond the admittedly minuscule sample of the four top US universities above. Arbitrarily, I chose to expand the list to include a few state universities of much larger size, as well as another top private university (MIT). Here are the total number of members of Facebook who have voluntarily associated themselves with their university networks, along with the respective universities’ undergraduate and total student populations, for comparison:
This chart tells us two important things: one, on a gross numbers basis, graduates and students at the University of Michigan would seem to be experiencing the greatest network effect value on Facebook, followed by Harvard and UCLA. This isn’t altogether surprising, however, given the relative sizes of the student populations at Michigan and UCLA. But correct (normalize) these data for size of student population, and an entirely different trend jumps out:
What do these charts say? If you correct for size of undergraduate population, Stanford University students and graduates are far more likely to be members of Facebook than are the students of these few other universities (59.3% more likely than nearest “competitor” Princeton, and 9.5 times more likely than UCLA students). Perhaps this is to be expected from a university in the heart of Silicon Valley, a mere few minutes’ drive from the venture capitalists who fund so much of the social networking industry. But the trend is even more interesting when one corrects for the overall student body size: then Princeton University is the hands-down winner, outstripping nearest competitor Stanford by 42.2% and besting UCLA by 2.7 times.
At first blush, the answers to this second question would seem to fly in the face of the first: how can Princeton students and graduates be so much more likely to be users of Facebook while still lagging behind the other schools in “older” graduate adoption? As you have no doubt surmised, the difference lies in the activity of the current student population. What Princeton loses in “older” graduate adoption of social networking technology it more than makes up in adoption by its current student body. Or to put it another way, if you meet someone on Facebook, and she is from Princeton, she’s likely to be younger than that fellow you met from Harvard!
All of the data in this posting were collected either by using Facebook’s Friend Finder or reviewing the “Network” home pages for each of the various universities. Due to result size limits imposed by Facebook, presumably for performance reasons, queries by class year and university which return more than 500 listings report “of over 500 found.” By paging through the results, it is possible to get beyond 500 and reliably spot results up to about 550 or so hits, but in no case will Facebook display the 551st hit. Therefore, in all the datasets, any value reported above 500 was treated as an “overflow” and was not calculated into averages.
There are several potential problems with the above techniques used to quantify class membership. First, not all members of an undergraduate class are of the same age. Second, Facebook does not reliably distinguish between undergraduate and graduate class membership when searching (it does, to some extent, when reviewing an individual’s profile). Third, Facebook does not distinguish between students, faculty and university staff when calculating the size of its Networks. Especially for the larger universities—with larger staffs—there is likely to be some impact from this particular source of error. Finally, affiliation with class years and universities is strictly optional, so there is undoubtedly selection bias present in these figures. Of particular note is that there are undoubtedly some people who affiliate with a University network but intentionally choose not to divulge their graduation year.