The Milgram Small World Experiment: Six Degrees of Separation

In 1967, a social psychologist in New York mailed packets to random strangers in the American Midwest and asked them to forward each one to a Boston stockbroker they had never met — but only through people they knew personally. He was testing a hunch: that the United States, with its 200 million inhabitants spread across a continent, might be less socially sprawling than anyone had imagined. What he found became one of the most quoted statistics in social science and one of its most contested.

The Problem Milgram Was Trying to Solve

The small world problem — the question of how socially close any two randomly chosen people actually are — had floated through mathematics and sociology since at least the 1950s. The Hungarian writer Frigyes Karinthy had speculated about it in a 1929 short story, imagining that any person on Earth could be linked to any other through a chain of five acquaintances. It was a compelling idea, but no one had tested it empirically.

Ithiel de Sola Pool and Manfred Kochen spent the 1950s working on a mathematical model of the problem but found it intractable without reliable data on how many acquaintances people actually maintain and how those networks overlap. Their paper circulated in unpublished form for years and was not formally published until 1978, long after Milgram had done the experiment that made the question famous.

Source: Pool, I. de S. & Kochen, M. (1978), "Contacts and influence," Social Networks 1(1), 5–51 — the foundational theoretical paper, written in the 1950s but published only after Milgram's empirical work had already popularized the question.

Milgram came at the problem from a different angle. Rather than building a model, he designed a postal experiment that could trace real social paths through real people, revealing how the American population was wired together.

How the Experiment Worked

Milgram ran two main studies in 1967, published together with Jeffrey Travers in a 1969 paper in Sociometry. The basic design was the same for both: recruit a group of starting participants, give each one a packet, and ask them to get it to a specific target person by forwarding it through personal acquaintances.

The rule was strict. Participants could only forward the packet to someone they knew personally on a first-name basis. They could not send it directly to the target unless they knew the target personally. Each packet contained a roster on which every person who handled it wrote their name, creating a full record of the chain.

Source: Travers, J. & Milgram, S. (1969), "An experimental study of the small world problem," Sociometry 32(4), 425–443 — the primary publication, with full methodology and results.

The target for both studies was the same: a stockbroker living in Sharon, Massachusetts, who worked in Boston. Milgram chose a target who was geographically distant from the starters and professionally distinct from most of them — someone not easily reachable through obvious channels.

The two starter groups were:

• Nebraska random group: 100 people selected at random from Omaha, Nebraska, representing a geographical cross-section of the Midwest.
• Nebraska stockholder group: 96 people in Omaha who were chosen because they worked in finance or investments — people theoretically more likely to know a Boston stockbroker.
• Boston random group: 100 people selected at random from Boston itself, used as a near-distance comparison.

Source: Travers & Milgram (1969), Sociometry — the paper describes all three starter groups and their characteristics.

The Results That Spawned a Cultural Phrase

Starter Group	Packets Sent	Chains Completed	Completion Rate	Mean Chain Length
Nebraska (random)	100	18	18%	5.9 intermediaries
Nebraska (stockholder)	96	24	25%	4.8 intermediaries
Boston (random)	100	22	22%	4.4 intermediaries
Combined Nebraska	196	42	~21%	~5.2 intermediaries

The completed chains averaged between 4.4 and 5.9 intermediaries depending on which starter group you examine. Milgram reported the overall median as approximately 5, but the "six degrees" framing entered popular culture partly because it was a round number and partly because John Guare used it as the title of his 1990 play, which became a 1993 film.

The most striking result was not the average but the minimum. One chain was completed in just two intermediaries. Some packets reached the Boston stockbroker in as few as four days. The target consistently received packets through a handful of specific individuals — close friends or professional contacts who served as what network scientists would later call hubs, people with unusually large and well-connected social networks.

Source: Milgram, S. (1967), "The small world problem," Psychology Today 1(1), 60–67 — the popular article in which Milgram first described the results to a general audience. Note: this was distinct from the full Sociometry paper with Travers.

The Critical Flaw That Almost No One Mentions

The statistic that circled the globe was 6. The fact that almost never circulated alongside it: only about 20 to 30 percent of all packets ever arrived.

This is a catastrophic selection bias. The chains that completed were not a random sample of all possible social paths. They were the paths taken by people who were motivated enough to participate at every step — and motivated people are likely to be more socially engaged, better networked, and more connected to the kinds of people who finish things. The thousands of chains that died quietly in someone's desk drawer represent paths that were likely longer, more broken, or more geographically isolated.

The drop-out problem compounds geometrically. If each person in a chain has an 80% chance of forwarding the packet, a five-step chain has only a 0.8^5 = 33% chance of completing. If the real chains were seven or eight steps long and each person had a 70% forwarding rate, almost nothing would arrive — and the apparent mean would dramatically underestimate the true mean.

Source: Kleinfeld, J.S. (2002), "Could it be a big world after all? The 'six degrees of separation' myth," Society 39(2), 61–66 — the key methodological critique, which examined Milgram's original data and argued that the completion bias makes the six-degrees conclusion unsupportable.

This is not an obscure academic objection. It is a fundamental limitation of the design that Milgram himself acknowledged in passing but did not fully address. The "six degrees" finding is better understood as an upper bound on the length of completable chains, not a reliable estimate of typical social distance.

Why the Experiment Still Mattered — Even Imperfectly

Pointing out that the six-degrees number is suspect should not erase what Milgram actually demonstrated. The experiment showed three things that were genuinely non-obvious in 1967.

First, long-distance social chains exist at all. The intuition of many sociologists at the time was that American society was so fragmented — by class, geography, race, and occupation — that chains connecting random strangers would almost never complete. They completed at 20–30%, which was surprising.

Second, the chains that succeeded did not snake through hundreds of weak acquaintances. They tended to converge on a small number of people who served as key connectors. In Milgram's data, the final step to the stockbroker was made through just three of his close friends in the vast majority of completed chains. This foreshadowed the concept of highly connected nodes that would become central to network science.

Third, geographic proximity was a weaker predictor of connection than professional similarity. The stockholder group in Nebraska, despite being 1,800 km (1,100 miles) from Boston, outperformed the geographically local Boston random group on completion rate, because professional networks bridge distance more efficiently than neighborhood ties.

Source: Travers & Milgram (1969), Sociometry 32(4) — analysis of which individuals served as bridges and how chains converged.

What Network Science Did With the Idea

The small world experiment was largely a cultural artifact for three decades. It re-entered science in 1998 when mathematicians Duncan Watts and Steven Strogatz published a paper in Nature showing that real networks — from power grids to the neural wiring of the roundworm C. elegans — share the property of being simultaneously locally clustered and globally short-pathed. They called this the small-world network property, providing the formal mathematical structure that Milgram's experiment had gestured at.

Source: Watts, D.J. & Strogatz, S.H. (1998), "Collective dynamics of 'small-world' networks," Nature 393, 440–442 — one of the most cited papers in network science.

The following year, Watts and Barabási independently developed models showing that real networks are not random graphs but contain hubs — nodes with far more connections than average — and that these hubs are responsible for the short average path lengths. The network behind Facebook's users, for example, was measured in 2011 by researchers at Facebook and the University of Milan to have an average path length of 4.74 degrees — slightly closer than Milgram's estimate, but in the same general neighborhood.

Source: Backstrom, L. et al. (2012), "Four degrees of separation," ACM Web Science Conference — Facebook-scale network analysis, using 721 million users.

In 2003, Columbia University sociologist Duncan Watts ran a direct digital replication of Milgram's experiment via email, with 60,000 participants in 166 countries trying to reach 18 targets. The email chains that completed averaged between five and seven steps. The completion rate remained low — between 1.5% and 3% of chains — reinforcing the selection-bias concern but also confirming that short chains do exist.

Source: Dodds, P.S., Muhamad, R. & Watts, D.J. (2003), "An experimental study of search in global social networks," Science 301(5634), 827–829 — the large-scale email replication.

The Findings People Get Wrong

"Six degrees" is not a law of nature. It is an observation derived from a methodologically flawed experiment conducted on a specific population (mid-twentieth century Americans) at a specific time. Networks with different structures — smaller clusters, less urbanization, more social stratification — would yield different numbers.

The number is not universal. Milgram's experiment excluded people without sufficient literacy or motivation to participate, and his sample was skewed toward white, middle-class Americans. Social distances within isolated rural communities, across caste boundaries in South Asia, or between economically segregated urban populations might be dramatically longer.

Knowing a path exists is not the same as being able to navigate it. Even in Watts's email study, only 1.5–3% of chains completed, suggesting that even if short paths exist between any two people in principle, in practice most people cannot identify or traverse them. The topology of the network and the information available to walkers navigating it are entirely different things.

Source: Watts, D.J. (2004), "The 'new' science of networks," Annual Review of Sociology 30, 243–270 — a thorough review distinguishing network structure from navigability.

The popular phrasing inverts the logic. The experiment did not prove that any two people on Earth are six handshakes apart. It showed that a subset of chains between specific pairs of Americans, conducted through motivated participants, completed in a median of about five to six steps. Generalizing from 64 completed chains to "the world" is a substantial extrapolation.

From Letters to Facebook: Measuring Six Degrees in the Digital Age

Milgram's postal chains closed in 1967 with 64 completed journeys and a sample size that would barely qualify as a pilot study by modern standards. The obvious next move — run the same experiment at internet scale — took nearly four decades, and when researchers finally did it, the results were both validating and humbling.

In 2003, Duncan Watts, Peter Dodds, and Roby Muhamad at Columbia University recruited 60,000 volunteers across 166 countries and asked them to reach one of 18 target individuals — a professor in the United States, an archival inspector in Estonia, a veterinarian in Norway, and others chosen to span geography and profession — using only email forwarding through personal acquaintances. The study was a deliberate digital mirror of Milgram's design, substituting email for physical packets but preserving the "first-name basis" rule: you could only forward to someone you knew personally.

Source: Dodds, P.S., Muhamad, R. & Watts, D.J. (2003), "An experimental study of search in global social networks," Science 301(5634), 827–829 — the first large-scale cross-national replication of Milgram's design, covering 166 countries and 18 targets.

The completed chains averaged between five and seven steps — within the range Milgram had reported, and consistent across remarkably diverse country pairings. But the completion rate told a more sobering story: only 1.5% to 3% of chains ever finished. That is worse than Milgram's already-low 20–30%, which means the selection bias problem did not disappear when the medium changed — it got more severe. Email chains die faster than postal ones because the friction of forwarding is lower, which sounds paradoxical, until you realize that low friction also means people feel less committed to following through. Deleting an email requires almost no effort; throwing away a physical packet requires a small but non-trivial act of decision.

Navigability — the ability of participants to identify the correct next person in a chain, not merely the existence of a short path — emerged from the Dodds study as the binding constraint. Participants who reported high confidence in their choice of recipient produced chains that completed at higher rates, and professional similarity to the target turned out to be a better navigation signal than geographic proximity. This echoed what Milgram's stockholder group had already suggested: people can bridge large distances quickly when they share an occupational frame of reference.

Eight years after the email study, Facebook's own data team produced what remains the most statistically powerful test of the small-world idea. Lars Backstrom and colleagues at Facebook, working with researchers at the University of Milan, analyzed the friendship graph of 721 million active users — roughly 10% of the global population at the time — and computed the distribution of shortest paths between all pairs of users.

Source: Backstrom, L., Boldi, P., Rosa, M., Ugander, J. & Vigna, S. (2012), "Four degrees of separation," Proceedings of the ACM Web Science Conference — analysis of the Facebook social graph using 721 million users and 69 billion friendship links.

The mean shortest path across all user pairs was 4.74 degrees. By 2016, as the platform grew to 1.6 billion users, the same team updated the figure to 4.57 degrees — connectivity had tightened as the network expanded, because each new user added more bridging edges than they added path length. The number is close to Milgram's estimate, but the mechanism behind the closeness is different from what Milgram could observe.

Think of it like the difference between knowing a road exists on a map versus being able to navigate to your destination without a GPS. The Facebook study measured the map — the actual shortest-path topology of the network. Milgram measured navigation: how far could motivated humans travel using only local information and social intuition? The Facebook result shows the road genuinely exists. The email and postal studies show that humans, left to their own devices, find it roughly half the time at best.

The digital studies also refined what "degree" means. A Facebook friendship is not the same social relationship as a first-name acquaintance. The 4.74-degree figure includes dormant connections, followed public figures, and casual online contacts that would never function as information-forwarding links in a postal chain. If the Facebook graph were filtered to strong ties — people who exchange messages regularly — the average path length would likely increase substantially, bringing it back toward Milgram's territory. The two numbers are measuring adjacent but distinct quantities, which is why the apparent confirmation is real but should be held loosely.

What the digital era collectively established is that Milgram's central intuition was correct: large human social networks contain short paths between most node pairs, maintained by a relatively small number of high-degree connectors. His estimate of six was reasonable. The methodology that produced it was not.

The Phrase Outran the Data

"Six degrees of separation" entered the culture with a velocity that the underlying data could never justify, and that gap tells us something interesting. The idea that we are all close — that the world is small, that community transcends distance — is a comforting one. People wanted it to be true. The phrase spread through parties, films, and eventually tech companies not because the methodology was airtight but because it confirmed something humans were primed to believe.

This is not a failure of science. It is a demonstration that scientific findings, when they articulate something emotionally resonant, will propagate far beyond the conditions under which they were established. Milgram gave the world a number. The world turned the number into a metaphor. The metaphor changed how engineers designed social networks.

The honest version of the finding is less tidy but more interesting: some human social networks contain short paths between distant nodes, those paths are navigable by certain kinds of people under certain conditions, and the nodes with the most connections play an outsized role in maintaining whatever connectivity the network has. That is a meaningful result. It just doesn't fit neatly on a bumper sticker.

The Takeaway

Stanley Milgram's small world experiment produced a number — six degrees — that became a cultural shorthand for human connectedness, but the number itself rests on shaky foundations: a 20% completion rate, a selected sample, and a rounding decision made for a popular magazine. What the experiment reliably showed is that social networks contain structural shortcuts, that a small number of highly connected individuals serve as bridges across otherwise distant communities, and that the distance between strangers is shorter than intuition suggests. Those findings are real, important, and confirmed by later work at far larger scales. The phrase that came with them just happens to be simultaneously famous, useful, and slightly wrong.

---

Read next:

• The Milgram Obedience Experiment
• The Robbers Cave Experiment
• The Asch Conformity Experiment

---

Q: What exactly did Milgram send in the packets?
A: Each packet contained a document describing the target person (name, address, occupation), a roster on which each handler recorded their name before forwarding, a set of instructions explaining the rules, and a set of "tracer cards" that participants mailed back to Milgram's lab at each step, allowing him to reconstruct the chain even if the packet never arrived.

Q: Why did so many chains fail to complete?
A: Participants dropped out at every step for a variety of reasons: lack of interest, inability to identify a useful next link, privacy concerns, or simple inertia. Since each additional step multiplied the probability of drop-out, longer chains failed at much higher rates, which systematically removed long paths from the completed-chain data and biased the mean length downward.

Q: Did Milgram study race and social distance?
A: Yes, in one variant of the experiment, Milgram used a Black target in New York. Chains directed toward the Black target from white starters completed far less often and were longer when they did complete, suggesting that racial segregation created real structural barriers in the American social network that the six-degrees framing obscured.

Q: How does the Facebook network compare?
A: A 2011 study of 721 million Facebook users found an average path length of 4.74 degrees between any two users — slightly shorter than Milgram's estimate, but this measures friendship links on a platform rather than real-world acquaintance chains, and Facebook's graph is heavily biased toward people who are already somewhat connected.

Q: Was "six degrees of separation" Milgram's own phrase?
A: No. Milgram used the phrase "small world problem" and described his findings in terms of the number of intermediaries. The phrase "six degrees of separation" was coined by playwright John Guare for his 1990 Broadway play of the same name, which dramatized the concept. Milgram had been dead for five years by the time the play opened.

Q: Has the experiment ever been cleanly replicated?
A: The closest attempt was Dodds, Muhamad, and Watts's 2003 email study with 60,000 participants in 166 countries. It found similar path lengths (five to seven steps) in completed chains but the same low completion rate problem. No study has fully solved the drop-out bias, because the only way to do so would be to track every chain regardless of whether participants cooperate — which is not feasible with voluntary subjects.