Dunbar's layers, evidence-checked
The 5 / 15 / 50 / 150 circles are the most famous numbers in friendship science — and the most misread. This page checks them against the data they actually came from: which layers replicate across measurement methods, which deserve the word "numerology," and what the innermost ring means for your health. The parent topic sketches the architecture; this is the audit.
What the evidence supports
- A layered network structure (roughly 5 / 15 / 50 / 150) shows up in Christmas-card lists, questionnaire data, phone records, and online social data (Social Networks, 2016).
- The innermost layer — about five people you could call in a crisis — is the ring most consistently tied to emotional well-being.
- Adjacent layers scale in rough multiples of three to four, a pattern first shown by clustering analysis of personal networks (Proceedings of the Royal Society B, 2005).
What remains uncertain
- Whether the layers are cognitive limits or statistical averages — the brain-size argument behind 150 has been seriously critiqued (American Anthropologist, 2011; Biology Letters, 2021).
- Individual networks vary enormously, from under a hundred to well over two hundred, so no layer number works as a personal target.
- Whether position in the layers or time invested in each tie is what actually produces closeness.
Evidence last reviewed: August 15, 2026. Conclusions may change as new research is published.
your social world, layer by layer
Where the Numbers Come From
The story starts with brain anatomy, which is why it matters where the story goes next. In 1992, Robin Dunbar plotted neocortex size against group size across primates and found a tight relationship; extending the line to the human neocortex predicted a group of roughly 150 (Journal of Human Evolution, 1992). That prediction then met three independent kinds of human data. The Christmas-card study asked people to list everyone they'd send a card to — a meaningful-acquaintance census — and the average came to about 154 recipients including household members (Human Nature, 1995). Questionnaire studies of "everyone you know well enough to have a relationship with" landed around the same number across samples (Human Nature, 2003). And in 2005, a clustering analysis of personal-network data found that human networks don't thin smoothly — they organize into discrete layers scaling by roughly three to four times: about 5, 15, 50, and 150 (Proceedings of the Royal Society B, 2005). Later work extended the pattern to phone-call and face-to-face records (Social Networks, 2016) and to Twitter, where users actively maintain stable relationships with 100 to 200 others (PLOS ONE, 2011). Whatever you think of the interpretation, the numbers did not come from a blog post.
The Five Layers, Walked Through
- 🫂 Support clique (~5): the people you would call in the middle of the night — the crisis ring. This is the layer that shows up in health research as "confidants," and it is the one that carries most of the well-being signal.
- 😢 Sympathy group (~15): the people whose death would genuinely devastate you. Big enough to include your best decade's friends, small enough to grieve individually.
- 👥 Friendship circle (~50): the weekend-barbecue tier — people you'd invite, travel with, or call with news, but not in crisis.
- 🌐 Active network (~150): meaningful acquaintances — nameable, contextual, remembered. The famous ceiling; the people you could sit next to at a funeral without it being strange.
- ☁️ Weak-tie cloud (500+): recognized faces and nodding relationships — the barista, the neighbor, the old colleague. Huge, cheap, and mood-relevant, as the weak-ties page shows.
What's Real and What's Numerology
Here is the audit the numbers deserve. What has replicated: the layered pattern itself. Clustering analyses find discrete rings, not a smooth gradient, across card lists, questionnaires, call records, and online data — that part is real, and it is genuinely useful: it says depth is structured, not free-form. What has been contested: the precision and the explanation. Anthropologists re-examining the 1992 brain-size argument found that the 150 prediction rests on assumptions about group-size estimates and scaling that do not hold as cleanly as the story suggests (American Anthropologist, 2011). A 2021 re-analysis in Biology Letters concluded that "Dunbar's number" is best treated as a rough heuristic with enormous variance across populations, not a cognitive constant. The honest synthesis: the layers are a descriptive map of how human networks typically organize — not a quota, and not a law of your brain. A network of 90 or 240 is not broken; variance is the rule, not the exception.
Why the Innermost Layer Carries the Health Signal
When researchers measure which part of the network predicts well-being, the answer keeps landing on the smallest ring. The support clique — the ~5 — is where the buffering happens: it is the layer that responds in a crisis, the one whose responsiveness (not its mere existence) tracks emotional health in the social brain literature (British Journal of Psychology, 2012). The loneliness topic owns the mortality evidence, and the biology of connection topic owns the physiology — but both point at the same structure: the innermost ring is the health-relevant layer, and the outer rings are mostly life-quality and information layers. The practical corollary is blunt: auditing and repairing your inner five matters more than expanding your cloud by five hundred. The parent topic's "quality beats quantity above the floor" claim is, at bottom, a claim about this ring.
How to Use the Layers Without Worshipping Them
The layers earn their keep as a diagnostic, not a scorecard. The exercise: write down who would answer a 3am call (the inner five), who you'd invite to a milestone birthday (the 50), and who would recognize you in a grocery store (the cloud). Then look at the gaps. An empty inner ring is a health risk worth treating seriously — that is the loneliness topic's territory. An empty cloud is mostly a mood and opportunity cost — annoying, not dangerous. And an overstuffed middle with no inner ring is the classic activity-busy, intimacy-poor pattern the male-friendship page dissects. The layers also predict cost: each step inward roughly triples the maintenance time per person, which is exactly what the maintenance math page quantifies. The construction method — how to actually fill a layer that is empty — belongs to Building Your Tribe.
| Layer | What the data actually shows | Evidence |
|---|---|---|
| 🫂 Support clique (~5) | The crisis-call ring — the layer most consistently linked to emotional well-being | Strong |
| 😢 Sympathy group (~15) | Grief-tier — appears consistently in personal-network datasets | Good |
| 👥 Friendship circle (~50) | Weekend tier — visible in clustering analyses, softer boundaries | Moderate |
| 🌐 Active network (~150) | The famous ceiling — replicable across methods but with wide variance | Moderate |
| ☁️ Weak-tie cloud (500+) | Faces and context, no obligation — size barely constrained by anything | Moderate |
🧮 Averages, not quotas
If your network is 100 people or 250, nothing is broken — the layer numbers describe the typical shape of thousands of networks, and individual variance is the rule. The one finding worth taking personally is the asymmetry: the innermost ring carries the health signal, so its quality deserves your first hour, not your last. The stress pillar explains the physiology that ring buffers; the audit question is simply whether it is stocked.
Questions, Answered Briefly
- 📏 Is 150 a limit I'm stuck with? It's an average ceiling for maintainable meaningful contacts — people land all over the distribution. The inner ring, not the ceiling, is where health relevance lives.
- 🔢 Are the layers really multiples of three? The three-to-four-fold scaling is a statistical pattern found in clustering analyses — a useful description, not a law your friendships must obey.
- 🫂 How do I know if my support clique is healthy? The 3am test: two or three people who would answer and come. Responsiveness matters more than headcount — the buffering happens in the answering, not the listing.
- 🧅 Does family count toward the layers? Yes — and family runs on a maintenance discount: kin reach the same closeness with less contact time than friends, which the maintenance math page covers.
The Bottom Line
- The layers are real as description. A structured 5 / 15 / 50 / 150 pattern replicates across card lists, questionnaires, phone records, and online data.
- The numbers are averages, not quotas. Individual networks vary widely; the brain-size explanation behind 150 has been legitimately critiqued.
- The innermost ring carries the health signal. The support clique of about five does the crisis buffering — audit it before expanding the cloud.
- Use the layers as a diagnostic. Map your own rings, find the gaps, and route the repair work to Building Your Tribe and the maintenance math.
Related Topics
- Dunbar, "Neocortex size as a constraint on group size in primates," Journal of Human Evolution (1992)
- Dunbar & Spoors, "Social networks, support cliques, and kinship," Human Nature (1995)
- Hill & Dunbar, "Social network size in humans," Human Nature (2003)
- Zhou, Sornette, Hill & Dunbar, "Discrete hierarchical organization of social group sizes," Proceedings of the Royal Society B (2005)
- Sutcliffe et al., "Relationships and the social brain: integrating psychological and evolutionary perspectives," British Journal of Psychology (2012)
- Mac Carron, Kaski & Dunbar, "Calling Dunbar's numbers," Social Networks (2016)
- Gonçalves, Perra & Vespignani, "Modeling users' activity on Twitter networks: validation of Dunbar's number," PLOS ONE (2011)
- De Ruiter, Weston & Lyon, "Dunbar's number: group size and brain physiology in humans reexamined," American Anthropologist (2011)
- Lindenfors, Wartel & Lind, "'Dunbar's number' deconstructed," Biology Letters (2021)