Email updates six degrees theory
Technology Research News
The world has known about the small-world
phenomenon since sociologist Stanley Milgram's 1967 study found that it
took, on average, six exchanges among acquaintances to get a letter from
a random correspondent in Omaha, Nebraska to a Boston recipient identified
only by a brief description.
But the experiment, which started with 96 messages, 18 of which
eventually reached the recipient, has been criticized for not being thorough.
Columbia University researchers have filled in the blanks by carrying
out a larger, more detailed experiment over the Internet. The results
match many of the broad conclusions of Milgram's work, but show that Milgram's
conclusion about the importance of hubs -- people who have many connections
-- may be off, at least in regards to social networks.
The results could contribute to better knowledge bases and peer-to-peer
The researchers' global social search experiment, posted on the
Internet, prompted 24,163 email volunteers to attempt to reach one of
18 target persons in 13 countries by forwarding messages to acquaintances.
"People were given the description of the target individual and asked
to send an email to a contact of theirs who they thought was in some way
closer to the target," said Peter Sheridan Dodds, an associate research
scientist at Columbia University.
Recipients of these messages were instructed to do the same until
the message reached the target individual. The researchers also asked
participants for demographic data and their reasons for choosing the contacts
they did, said Dodds.
The experiment generated a total of 61,168 emails that ran through
166 countries; 384 of the original 24,163 reached their target. The researchers
concluded that individual apathy or disinclination to participate was
the major reason for broken chains. When they questioned senders who did
not forward their messages after one week, only 0.3 percent said they
could not think of an appropriate recipient, according to Dodds.
The experiment confirmed that, in social searches, a message initiated
by a random person reaches its destination in a median of five to seven
steps, depending on the separation of source and target, said Dodds. "People
can find other people regardless of how distant they are," he said. "Search
in large-scale networks is a very difficult problem, and yet people working
collectively are able to do quite well."
With a greater number of successful searches to analyze, however,
they found something surprising: the primary avenues from source to target
are not the highly connected social hubs that Milgram's experiments pointed
to. "Successful chains were... far less likely to use hubs than unsuccessful
chains," said Dodds. "Hubs -- people with many friends -- don't seem to
be so important for this kind of social search."
Participants in successful chains were less likely to send messages
to hubs -- 1.6 percent versus 8.2 percent -- than those in incomplete
The participants' answers also reflected this. They rarely chose
a sender based on the number of friends that person had, according to
Dodds. The main reasons for choosing the next person were geography- and
work-related," he said. These two categories accounted for at least half
of all choices, with geography dominating in the early stages of the chain.
When the researchers compared successful chains to those that
did not reach their target, they found that successful chains involved
more professional ties -- 33.9 percent versus 13.2 percent -- and fewer
familial relationships -- 59.8 percent versus 83.4 percent.
The experiments also showed that the success of a search is highly
dependent on individual incentives, said Dodds. This is because a small
change in the attrition rate -- the probability that people don't send
a message on -- leads to a substantial change in the number of chains
getting through, said Dodds.
One of the researchers' targets, a U.S. professor, received many
more completed chains than any of the other 10 targets reached. This was
probably because the professor appeared reachable, which makes sense because
the participants were largely college-educated and 50 percent lived in
the U.S., said Dodds. "We interpret this as meaning that people's perceptions
greatly affect their chances of success," he said. "In a nutshell, if
you think the world is small, it is."
The study is a confirmation of the six degrees of separation in
social networks, but also debunks some ideas associated with the six degrees
that have entered the popular culture, said Stephen Strogatz, a professor
of applied mathematics at Cornell University. "Milgram didn't really have
enough participants to figure out what kind of methods people were using"
to reach the target, he said.
The new study shows social networks don't depend on super-connected
people, or hubs, said Strogatz. "There was no evidence for hubs in this
study, and yet people were still able to get the messages to the targets
in the successful chains in five to seven steps," he said.
Although there are many similarities between social networks and
virtual networks like the Internet, it makes sense that hubs may not be
as prominent in the real world, said Strogatz. When hubs are virtual,
like on the Internet, "there's no physical or economic cost to having
many people point to your Web page," he said. In the real world, however,
network hubs contain costs -- maintaining a rolodex of 100,000 people,
for instance, takes time, and maintaining a powerplant with many transmission
lines costs money.
In the Milgram study, all successful chains went through one person
-- a well-connected tailor. The Columbia study did not show this funnel
effect, however, said Strogatz. It showed, rather, "that there are a lot
of roads to Rome," he said.
The new research also shows that the key to good social searches
is weaker friends, or more distant acquaintances, said Strogatz. This
makes sense -- closer friends are less useful in this case because people
who know each other well tend to have the same friends, he said.
"This is good work," said Jim Moody, an assistant professor of
sociology at Ohio State University. "By understanding the structure of
email communication networks we might be able to better design tools for
spreading information or stopping virus flow," he said.
The weak-tie findings are consistent with research about how people
use their acquaintances to find information when they're looking for work,
More work on the structure of close ties is needed, Moody said.
"The weak-tie findings... will eventually be quite important for information
flow," he said. "For virus flow through email, the key rests on the structure
of close, trusted email contact -- we won't open attachments from people
we don't know well."
The next step in the research is an experiment currently on the
researchers' Web site that is designed to gain more information about
how the small-world phenomenon works, said Dodds. "People can now send
more than one email regarding any given target," he said. Also, "we've
altered the questions we ask about why people choose the people they do,
and extended the descriptions of the targets."
The current results could be used now to improve databases and
networks, according to Dodds. "This... could be useful in the design of
knowledge bases or in the construction of peer-to-peer networks," he said.
The researchers are also working to model a range of social and
economic problems including the spread of contagion agents like diseases
or fads, the evolution of cooperation, and the structure of modern organizations,
Dodds's research colleagues were Roby Muhamad and Duncan J. Watts.
The work appeared in the August 8, 2003 issue of Science. The research
was funded by the National Science Foundation (NSF), the James F. McDowell
Foundation and the Office of Naval Research (ONR).
Funding: Government, Institution
TRN Categories: Internet; Applied Technology; Computers
Story Type: News
Related Elements: Technical paper, "An Experimental Study
of Search in Global Social Networks," Science, August 8, 2003; Researcher's
book, "Six Degrees, The Science of a Connected Age," by Duncan Watts;
Current small-world phenomenon experiment Web site: smallworld.columbia.edu
August 27/September 3, 2003
Software speeds modeling
DNA plays tic-tac-toe
Email updates six
Cellophane turns LCDs 3D
Detector senses single
makes heat chips
Light makes molecule
Research News Roundup
Research Watch blog
View from the High Ground Q&A
How It Works
News | Blog
Buy an ad link