Networks

Who you know shapes what happens

Same people. Different wiring. Different world.
Networks aren't about the people. They're about the connections between them.

A network is a bunch of dots (call them people) and lines connecting them (call them friendships). Sounds trivial. It isn't. Change the pattern of lines — leaving every person otherwise identical — and rumors spread faster, ideas travel further, crises cascade or fizzle, and the whole system either shrugs off failures or collapses at the first attack. The wiring is the model.

Below we'll build up to three surprising facts. Along the way we'll play with three of the canonical network shapes: ring lattices where you only know your neighbours, small-world networks where a few random shortcuts change everything, and scale-free networks where a handful of super-connected hubs dominate.

Fact 1: A few shortcuts turn a village into a small world

Below is a village of 40 people. Everyone knows their 4 nearest neighbours — just like a real village where you know the folks next door but not the family across town. To get a message across, it has to be passed from neighbour to neighbour to neighbour… that takes many steps.

Now drag the rewiring slider. Each notch takes a random neighbour edge and replaces it with a random long-distance one — a shortcut. Watch what happens to the "typical steps between two people" number. Then watch what happens to how clustered the village stays.

The rewiring switch · drag to add shortcuts

Rewiring probability p0.00

Typical steps between two people

—

the "six degrees" number

Clustering

—

how many of your friends know each other

Shortcuts added

rewired long-distance edges

The surprise. You only need a tiny amount of rewiring (try p = 0.05) for the "typical steps" number to crash — the network becomes a small world. But clustering barely moves. That's the Watts–Strogatz result: real friendship networks combine "everyone knows their neighbours" with "six degrees of separation" because they only need a handful of long-distance connections to shrink.

Fact 2: Your friends have more friends than you do

Below is a different network — one where a few people are extremely popular hubs, and most people know only a couple of others. (This is what real Twitter follower graphs, citation networks, and airline route maps look like.)

Let's play a game. There are two ways to sample a person:

🎲 Random person. Pick anyone at random.
👥 Random friend. Pick a random person, then pick one of their friends at random. That's the friend you sampled.

Click a few times on each button and watch the two averages fight it out. Common sense says they should be about the same. Common sense is spectacularly wrong.

The friendship paradox · sample a person, or sample a friend

🎲 Random-person average friends

—

click "Random person" to sample

👥 Random-friend average friends

—

click "Random friend" to sample

Why it works. Popular people show up in many friend lists — so when you sample "a friend of a random person", you're much more likely to land on a hub than if you sampled uniformly. This isn't sociology. It's plain combinatorics, and it holds for every non-uniform network ever built. Practical use: to catch an outbreak two weeks earlier, monitor friends of random people — you'll hit the central nodes disproportionately often. Christakis & Fowler (2010) used exactly this trick to spot a Harvard flu wave ahead of population surveillance.

Fact 3: Hubs are both a superpower and a soft spot

Same network family. Two ways it can lose people. On the left, random failures — a laptop crashes, someone loses their phone, an airport shuts for a snowstorm. On the right, a deliberate hub-first attack — an adversary who knows the topology and always removes the most-connected node next.

Watch both destroy the same starting network at the same rate. The colored bars below each panel track the size of the largest surviving cluster. The chart at the bottom overlays the two curves.

Random failures vs targeted attacks · same starting network

🎲 Random failures

nodes chosen uniformly at random

🎯 Targeted attack

highest-degree node next, every time

Nodes removed

0 / 0

out of the original network

🎲 Largest cluster (random)

—

% of original

🎯 Largest cluster (targeted)

—

% of original

The design lesson. The internet, airline route maps, and power grids are all scale-free — dominated by hubs. That gives you robustness for free against random failures (most breaks hit unimportant peripheral nodes) and fragility against a smart adversary (a handful of hub failures cascade instantly). The same wiring; two opposite security stories.

Where networks matter

Every "how does this spread?" question in this book is really a network question underneath. Contagion (Ch 11), diffusion of ideas, financial contagion in the 2008 crisis, the spread of gossip, terrorist and crime networks, the friendship paradox in disease surveillance, the small-world magic behind six degrees of separation. Change the wiring, change the story.

Related models

🦠 SIR epidemic — contagion lives on these networks. The same R₀ spreads very differently on a village vs a scale-free hub graph.
📈 Power laws — Barabási–Albert graphs produce heavy-tailed degree distributions; the two models are the same mechanism told two ways.
🟦🟧 Schelling segregation — the "who is next to whom" question in Schelling is literally a network question underneath.

← Power laws SIR epidemic →