Cascades in the Wild

The earlier lessons built the machine in a laboratory: a line of people, an urn, one private signal each, and the exact moment when copying the crowd becomes the rational move — the moment a cascade forms and information stops aggregating. That’s the clean version. This lesson takes the machine outside and points it at the messy world, because the whole reason cascades matter is that they are everywhere, wearing disguises.

Here’s the one sentence to carry through every case below. A cascade forms when other people’s visible actions outweigh your own private signal, so you copy them — and once you copy, your action stops carrying your information and starts carrying theirs. Every example in this lesson is that sentence in a costume. The skill you’re building isn’t spotting a crowd; crowds are easy. It’s asking, each time, how much of this crowd is real pooled information, and how much is just people copying the person in front?

Restaurant queues: the line is the menu

Start where lesson one started — hungry, in a strange city, between a packed restaurant and an empty one. You pick the packed one, and we already established that’s rational: the queue is a stack of other people’s choices, and each choice is a scrap of evidence.

But now look at what the queue does. The person behind you sees a line one longer because of you — and you joined partly on the strength of the people ahead of you, not on anything you actually know about the kitchen. The visible action (waiting) gets copied; the copying makes the action more visible; the loop feeds itself. This is self-fulfilling popularity: the place is busy because it’s busy, and that can be completely detached from whether the food is any good.

The naive read is “the busy place must be better.” Sometimes true! But the queue by itself can’t tell you, because a queue built by copying looks identical to a queue built by quality. That indistinguishability — real signal and pure echo wearing the same face — is the theme of this entire lesson.

Bestseller lists, charts & app-store rankings

Now industrialise the queue. A bestseller list, a music chart, an app-store “Top 10,” a “trending” tab — each is a queue rendered as a number, broadcast to millions. And here’s the sharp part: being near the top causes more sales. Rank is a public action people copy. You browse the chart, see a book at #2, and buy it partly because it’s at #2 — which nudges it toward #1, which sells more copies, which cements the rank.

Economists have a name for this shape — cumulative advantage, or more bluntly, “the rich get richer.” A tiny early lead in sales earns a higher rank, the higher rank earns more sales, and the gap widens far beyond any gap in quality. Two books that would sell identically in a world with no visible rankings can end up an order of magnitude apart once the ranking is public and copyable.

Viral products, songs & trends

Why do two nearly identical songs — same genre, same polish, released the same week — end up worlds apart, one streamed a hundred million times and the other forgotten? The honest answer is uncomfortable: often, not much. A small early lead, handed out by luck or timing or a single well-placed share, snowballs.

The classic demonstration is the Music Lab experiment (Salganik, Dodds & Watts): thousands of people were given unknown songs to listen to and download. Some saw how many times each song had already been downloaded; others saw nothing. In the independent world (no download counts), quality had a modest, steady effect and outcomes were predictable. In the social world (counts visible), the same songs produced wildly different hits in different parallel copies of the experiment — and the results were far more unequal and far less predictable. Same songs, same people. The only new ingredient was seeing what others chose, and it manufactured runaway winners that reshuffled from one run to the next.

Financial bubbles & fashions

Point the same lens at markets and it gets expensive. You buy an asset because the price is rising; the price is rising because people are buying; and the price itself becomes the visible action everyone copies. Notice the swap: in a healthy market, price is supposed to summarise information. In a bubble, price replaces information — people stop asking “what is this worth?” and start asking “what are others about to pay?”

This is a cascade in its purest financial form. The crowd looks maximally confident — soaring prices, giddy headlines, your neighbour bragging about gains — yet it is pooling almost no independent information about value. Everyone is reading everyone else. And because the cascade rests on so little real information (exactly the fragility from the earlier lessons), it is primed to shatter. The crash isn’t a separate disaster bolted onto the boom; the crash is the fragility becoming visible. One credible signal — a big holder sells, a default hits the news — and the copied action reverses direction. Now the visible action is selling, and it cascades down just as fast as it climbed up.

Research & citation bandwagons

Cascades don’t spare the people whose whole job is being careful. Scientists cite prior work, and a paper that is already heavily cited is more likely to be cited again — it’s what turns up in searches, what reviewers expect to see, what feels safe to build on. That’s a queue made of footnotes. A fashionable early result attracts a cascade of follow-up papers, reviews, and citations, and the field can arrive at an impressive-looking near-unanimity that rests, when you trace it back, on one early study rather than many independent confirmations.

The distinction that matters here is between citation and replication.

This is one engine behind the “replication crisis”: a striking early finding gets cited thousands of times, becomes textbook fact, and then — when someone finally tries to reproduce it — evaporates. The unanimity was a citation cascade all along, information-poor and fragile, waiting for the one independent draw to break it. The lesson isn’t that scientists are foolish; it’s that even careful people, acting rationally, can build a towering consensus on a single unrepeated signal.

Reputational herding: safety in being wrong together

So far every cascade has been informational: you copy because others’ actions look like better evidence than your one signal. But there’s a second, sneakier engine that produces identical-looking herds for a completely different reason — and once you see it, you can’t unsee it in finance, politics, and management.

Reputational herding (the model is Scharfstein & Stein, 1990) is copying the pack not because you believe they’re right, but to avoid being the only one who’s wrong. The asymmetry is everything: being wrong alone is career-ending — “how did you miss what everyone else saw?” — while being wrong along with everyone is forgivable — “nobody saw it coming.” So the rational move, if you care about your reputation, is to huddle in the middle of the pack even when your private signal says the pack is mistaken.

Notice how this differs from a pure informational cascade. In the urn line, you copy because you genuinely think the crowd is probably right. In reputational herding, you might privately think the crowd is wrong and copy anyway — your motive is protecting your reputation, not updating your belief. Different mechanism, same footprint: your private signal gets suppressed and the herd loses information.

Standing ovations, bank-run links & coordination

Last stop, and it’s a bridge to the neighbouring course. Watch a standing ovation begin. A few people rise; others see them and rise; the rising crowd is itself a public action that recruits more rising, until the whole hall is on its feet — sometimes for a merely-good performance. You stood partly because they stood, and your standing pulled up the row behind you. A queue, a chart, and an ovation are the same object seen from three angles: a visible, copyable action snowballing.

But an ovation hints at something beyond information. When you stand, you’re not only reading a signal — you’re coordinating: nobody wants to be the last one seated (or the first, lonely one standing). That’s the seam where cascades meet common knowledge and critical mass, the models from the sister course. A bank run is the sharp-edged version: it’s a coordination failure — everyone would be fine if everyone stayed calm — that a herding signal can trigger. You see others rushing to withdraw (a visible action), and whether or not the bank is sound, the rational move becomes to rush too.

Sorting the wild from the merely crowded

Time to make the core skill explicit and drill it. The hardest part of applying this model is resisting it — not every crowd is a cascade. When people reach a shared answer through independent judgement, agreement is good news: it’s real pooled information. The tell is whether each person’s decision was made seeing others’ choices (cascade risk) or sealed off from them (independent, trustworthy).

Recap

Every case in this lesson was the same sentence in a different costume: a visible action gets copied, which makes the action more visible, which freezes each copier’s private information out of the crowd. The crowd grows confident while staying information-poor — and therefore fragile, one credible contrary signal away from reversing. Your job as a clear thinker is never to be impressed by a crowd’s size or confidence, but to ask what’s underneath it: independent draws, or echoes?

You can now spot cascades in the wild and, crucially, tell them apart from crowds that genuinely know something — the difference between a queue and a jury, a chart and a replication. The next lesson, Cascades, Common Knowledge & Critical Mass, walks straight onto the boundary we just touched: how a cascade relates to the models underneath it, and how a single public signal can start a cascade — or break one.