This is the whole course in one sitting. Hold the through-line in your head: a few gargantuan winners take almost everything while a long, thinning tail gets scraps — and that lopsided shape has a name, a signature, an engine, and a set of lies. The shape is the power law (scale-free — the same lopsidedness at every zoom), whose tell is a straight line on log-log axes, and whose defining feature is that there’s no typical case: extremes dominate the total and the #1-to-#1000 ratio is enormous. The engine that manufactures it is cumulative advantage — the reinforcing loop that goes by three names (preferential attachment, the Matthew effect, “the rich get richer”), where more begets more, amplifying a tiny, near-random early lead into a runaway winner. Its friendly face is the 80/20 principle — the vital few and the trivial many, self-similar enough to nest inside itself. Its market form is winner-take-all: scalability plus network effects turn a razor-thin quality edge into a hundred-to-one payoff. Its danger is the tyranny of the tail: the average is unstable, one event can outweigh all history, and luck compounded beats merit proportional. And the model lies if you forget its limits — a straight-ish log-log line proves little, log-normal look-alikes fake the shape, survivorship and the Matthew effect inflate the winner’s apparent merit, the exponent is estimated from scarce tail data, and “power law, therefore this inequality is natural” is a smuggled value judgment. Take a breath. There is no going back once you commit.
How this exam works
This is a real exam, not a practice quiz. Questions appear one at a time. Once you submit an answer it is locked for good — there is no going back, no retry, and no restart. Your score stays hidden until the very end. A few questions ask you to select all that apply (read those carefully — partial credit is not a thing here). You need 70% to pass. Ready when you are.
What does it mean to call a power-law distribution SCALE-FREE?
Select an answer to continue.
Big picture
Cumulative advantage and power laws, in one picture
- Cumulative advantage and power laws
- What a power law is
- A scale-free distribution - the same lopsided shape at every zoom, no typical case, extremes dominate the total, and an enormous #1-to-#1000 ratio; the tell is a straight line on log-log axes, unlike the bell curve where a big middle clusters near the mean - think Zipf words, Pareto wealth, Gutenberg-Richter quakes, city sizes
- The engine - preferential attachment
- One reinforcing loop with three names - preferential attachment, the Matthew effect, the rich get richer - where more begets more; Merton saw it credit famous scientists, Barabasi-Albert and Yule formalised it in networks; it amplifies a tiny near-random early lead into a runaway winner, so a big win is NOT proof the winner is better
- The 80/20 principle
- Pareto vital few and trivial many - the friendly face of a power law; the two numbers need not sum to 100, a steeper exponent pushes 80/20 toward 90/10, and it nests self-similarly - top 4% holds about 0.8 x 0.8 = 64%; use it to find and double down on the vital few, but do not blindly discard the long tail or force the split
- Winner-take-all markets
- Rosen superstars need scalability plus imperfect substitution; Frank-Cook tournaments reward relative rank, not absolute output; network effects amplify the leader so a razor-thin quality edge pays a hundred to one, not ten percent more - it does not fit non-scalable local trades, and where it bites it fuels wasteful arms races
- The tyranny of the tail
- The average is unstable and may never converge, one record event can outweigh all history combined, and outcomes are luck-compounded far more than merit-proportional - skill is only a filter, then luck plus the loop pick the winner; so never plan for the average, build a margin of safety against the tail
- Where the model lies
- A straight-ish log-log line is not proof, log-normals from multiplicative growth mimic the shape so shape hides mechanism, survivorship plus the Matthew effect inflate the winner apparent merit, the exponent is estimated from scarce tail data and is shaky - and beware the normative trap of sliding from is a power law to therefore the inequality is justified
- What a power law is
Key takeaways
You now hold the whole model. A power law is a scale-free distribution — the same lopsided shape at every zoom, no “typical” case, extremes that dominate the total, and a #1-to-#1000 ratio in the hundreds or thousands — whose visual tell is a straight line on log-log axes, in sharp contrast to the bell curve’s comfortable, symmetric middle. The engine that manufactures it is cumulative advantage: one reinforcing feedback loop wearing three names — preferential attachment, the Matthew effect, “the rich get richer” — where more begets more, so a tiny, near-random early lead compounds into a runaway winner (which is emphatically not proof the winner is better). Its friendly face is the 80/20 principle — vital few, trivial many, numbers that needn’t sum to 100, steepening toward 90/10 with the exponent, and self-similar, nesting inside itself (top 4% holds about 0.8 times 0.8 = 64%). Its market form is winner-take-all: Rosen’s superstars (scalability plus imperfect substitution), Frank and Cook’s tournaments (relative rank, not absolute output), and network effects that amplify the leader, so a razor-thin edge pays a hundred to one — while non-scalable local trades stay proportional and the model’s real cost is arms-race waste. Its danger is the tyranny of the tail: the average is unstable (it may never converge), one event can outweigh all history, and outcomes are luck-compounded far more than merit-proportional — skill is a filter, then luck plus the loop crown the winner — so you plan for the tail with a margin of safety, never for the average. And the model lies if you forget its limits: a straight-ish log-log line is not proof, log-normal look-alikes fake the shape so shape never reveals mechanism, survivorship plus the Matthew effect inflate the winner’s apparent merit, the exponent is estimated from scarce tail data and is shaky, and the normative trap — sliding from “this is a power law” to “therefore this inequality is natural and justified” — smuggles a value judgment past you disguised as maths. Never ask “isn’t the winner just that much better?” Ask: how strong is the loop, how much was luck, and what is the shape trying to make me believe?