applets.

Small World

← All apps

About Small World

Put people in a ring where each one knows only the few people nearest them, and a message has a long way to go: in a ring of 100 where everyone knows the 2 on each side, the average chain between two people is about 13 links. Now let a handful of people each know one person at random somewhere across the ring. Five such shortcuts usually cut the average to between 7 and 9 links, while almost everyone's circle of friends stays as close-knit as before. That is the small-world effect: a few long links are enough to make a big world small.

THE RING: the people are lamps round a telephone switchboard; the short links to neighbours are the plain wiring, and every shortcut is a patch cord across the board. ADD A SHORTCUT plugs in one more cord between two people chosen at random, and the ticket shows the average separation falling. ClickTap two lamps to see the shortest chain between them. The Rewired at random slider builds the network the way Watts and Strogatz did: each link is moved to a random far end with that chance. UNPLUG takes every cord out again, and NEW DRAW rolls a new set of random choices.

THE CURVE builds their famous chart point by point: for rewiring chances from 1 in 10,000 to 1, the average separation and the clustering, each as a share of the plain ring's. SPREAD passes a rumour from friend to friend, once a day, on a plain ring and on the same ring with a few shortcuts, side by side. REAL WORLD sets out what the real experiments found, and shows why letters that get lost make the measured chains look shorter than they are. COPY LINK copies the settings.

THE STORY

The idea that everyone is linked by a short chain of acquaintances appears in a 1929 short story, Chains, by the Hungarian writer Frigyes Karinthy, whose characters bet they can reach anyone on Earth through at most five people.

In the 1960s the American social psychologist Stanley Milgram tested it with letters. He described the first studies, started in Wichita, Kansas and Omaha, Nebraska, in Psychology Today in 1967. In the best known, written up with Jeffrey Travers in Sociometry in 1969, 296 starters (196 in Nebraska and 100 in the Boston area) were each asked to get a folder to one stockbroker who worked in Boston and lived in Sharon, Massachusetts, by sending it only to someone they knew on first-name terms who might be closer. 64 chains arrived, and those had 5.2 people in between on average. The phrase "six degrees of separation" came later, from John Guare's 1990 play of that name. The psychologist Judith Kleinfeld later pointed out that most chains never arrived and that some starters were picked for being well connected, so the evidence was thinner than the legend.

In 1998 Duncan Watts and Steven Strogatz, at Cornell, published "Collective dynamics of 'small-world' networks" in Nature. They showed that a ring that is rewired only slightly has short paths like a random network but stays clustered like the ring, and found the same pattern in the network of film actors, the US western power grid and the nervous system of the worm C. elegans. In 2003 Peter Dodds, Roby Muhamad and Watts repeated Milgram's test by email: 24,163 chains aimed at 18 people in 13 countries; only 384 arrived, in 4.05 steps on average, and allowing for the lost chains they estimated a median of five to seven steps. On Facebook the average separation was measured at 4.74 links in 2011 (721 million users) and 4.57 links in 2016 (1.59 billion users), which Facebook reported as 3.57 people in between.

Two games use the same idea. The Six Degrees of Kevin Bacon, made up by three students at Albright College, Pennsylvania in 1994, links actors through films they made together. Mathematicians have their Erdős number: 1 if you wrote a paper with Paul Erdős, 2 if you wrote one with someone who did, and so on; Casper Goffman described it in print in 1969.

HOW IT WORKS

Separations are counted in links: a friend is 1, a friend of a friend 2. Milgram and Facebook counted the people in between, which is one fewer. The average separation is worked out exactly: a breadth-first search from every person finds the shortest chain to everyone else, and the page averages over every pair. Rewiring can, rarely, cut someone off; such pairs are left out of the average and the ticket says how many. Clustering is Watts and Strogatz's measure: for each person, the share of pairs of their friends who are friends with each other, averaged over everyone. A plain ring with k friends each has clustering 3(k − 2) ÷ 4(k − 1), which is 2/3 for 10 friends.

The rewiring follows the 1998 paper: go round the ring once for the nearest neighbours, then again for the next nearest, and with chance p move the far end of each link to someone chosen at random, never yourself and never someone already linked. The dashed Expected line on THE CURVE is Barrat and Weigt's approximation (2000) for the clustering, C(0)(1 − p)³. A fully random network with k friends each has an average separation of about ln n ÷ ln k, which is why a huge network can still be only a few links across. ADD A SHORTCUT adds a link instead of moving one, so the ring itself stays whole.

The model is a picture of an idea, not of a real town: real friendships don't sit on a circle, some people have far more friends than others, and real letters are passed on by people who don't know the shortest route. The lost-letters chart in REAL WORLD assumes the starter always sends the letter, each person after that forwards it with the same chance, and letters go along shortest chains; the Dodds study allowed for losses in a similar but more careful way.

FURTHER WATCHING AND READING

PEOPLE
FRIENDS EACH
none
TOLL TICKET100 people

AVERAGE SEPARATION
CLUSTERING
LONGEST CHAIN
CORDS