James’s Musings

thoughts, photography, and geeky stuff
from an unrelentingly curious Silicon Valley entrepreneur

Age *DOES* Matter: On the Demographics of Social Networks (I)

by James G. Beldock on November 24, 2007

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!

Methodological Notes:

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. 

Be Sociable, Share!
Tracey Lee November 27, 2007 at 12:21 am


Ynema Mangum August 8, 2008 at 5:22 am

Where can I get more up-to-date demographics on this subject?

Vivek Sodera September 17, 2008 at 4:38 pm


Great post and analysis of the data. And thanks for the nice words on the NY Times blog.

Feel free to ping me if you have any questions about our data/studies.


Vivek Sodera

Joe Hunkins March 4, 2009 at 12:08 pm

Excellent post. Hope you’ll update with respect to Twitter and how it’s mixing up the social network equation. I’m wondering if we’ll see regular people mimic “early adopter” behavior and increasingly use Twitter over Facebook because it’s faster, simpler, more mobile friendly.

Comments on this entry are closed.

{ 1 trackback }

Previous post:

Next post: