In the introduction we watched a line of people guess whether an urn was mostly-blue or mostly-red, and saw the whole herd freeze onto one answer after just two early guesses. That was the sketch. This lesson is the machine taken apart on the workbench, gear by gear — the exact experiment that economists Sushil Bikhchandani, David Hirshleifer, and Ivo Welch published in 1992 to give information cascades their first rigorous model, worked here with real numbers you can check by hand.
No formulas to memorise. We’ll count evidence the way a careful person actually would: tallying up how many little nudges point toward “blue” versus “red,” and asking which way the balance tips. By the end you’ll be able to point at the precise decider — the third person in the line — where a perfectly rational mind first overrides its own eyes, and see why everyone after them is trapped in the same box.
Before you read — take a guess
Before we build it — take a guess. In the urn line, how many guesses in a row does it take before a rational person should start ignoring their own draw and just copy the crowd?
The setup: two urns and one private peek
Picture two urns sitting on a table, identical from the outside. One is majority-blue: for every three balls inside, two are blue and one is red. The other is majority-red: two red, one blue. A coin flip — genuinely 50/50 — decides which urn gets placed in front of the group, and nobody is told the result. The whole game is one question: which urn is it?
People approach the urn one at a time, in a fixed order. Each person reaches in, draws exactly one ball, looks at its colour privately, puts it back, and then announces out loud a single guess — “blue urn” or “red urn.” Here is the rule that makes the whole thing tick: everyone hears every earlier guess, but nobody ever sees anyone else’s ball. You get your own private peek plus the public running commentary of everyone’s guesses — and that is all.
Before anyone moves, both urns are equally likely, so your belief starts at a clean 50/50. Then you draw. Because the majority-blue urn is two-thirds blue and the majority-red urn is only one-third blue, a blue ball is twice as likely to have come from the blue urn as from the red urn. So a single blue draw, starting from 50/50, pushes your belief to 2/3 blue, 1/3 red. (That’s Bayesian updating — revising a belief when evidence arrives — the prerequisite from the intro. We’ll keep it to counting.) A red draw does the mirror image: 2/3 red, 1/3 blue.
The signal, in one line
Each private draw is a signal: one ball of evidence. A blue draw makes the blue urn twice as likely as the red urn (2/3 vs 1/3), and vice versa. One signal is suggestive — not proof. Keep that “twice as likely” in your pocket; the entire lesson is just stacking these two-to-one nudges against each other.
A trick that keeps the arithmetic painless
Instead of juggling fractions, count in odds. Start at 1:1 (blue-to-red). Every blue signal multiplies the blue side by 2; every red signal multiplies the red side by 2. Two blues and one red? That’s 2×2 : 2 = 4:2 = 2:1 in favour of blue. We’ll use this all lesson — it turns Bayes’ rule into simple bookkeeping.
Person 1 and Person 2: honesty, then a genuine tie
Person 1 has nothing to go on but their own ball. They started at 1:1, drew — say — blue, and now sit at 2:1 for the blue urn. So they guess “blue.” Notice the quiet gift they just handed the room: because Person 1 has no crowd to copy, their guess is a perfectly faithful readout of their draw. Everyone now knows Person 1 drew blue. One signal is public.
Person 2 heard “blue” and then draws their own ball. Two cases:
- They also draw blue. Now two blue signals stack: odds are 2×2 : 1 = 4:1 for blue. Easy — they guess “blue,” and the room can infer a second blue signal.
- They draw red. One blue signal (Person 1’s, inferred from the guess) against one red signal (their own draw). The odds are 2 : 2 = 1:1 — a dead-even tie. Their own eyes and the crowd cancel out exactly.
What do you do on a genuine coin-flip? The experiment adopts a simple tie-break convention: follow your own signal. It’s the natural choice — your private draw is the one piece of evidence nobody else in the room has, so leaning on it is the way to keep contributing information rather than just echoing. So a tied Person 2 guesses the colour they drew. This matters enormously later: as long as people break ties toward their own draw, their guesses keep leaking real signals into the room. The cascade only begins when that leak stops.
Why the tie-break rule isn't arbitrary
Breaking ties toward your own signal is what keeps the line informative for as long as possible. If Person 2 instead copied Person 1 on a tie, they’d be echoing rather than reporting — and the room would learn nothing from them. The convention is chosen precisely so that people reveal their private draw whenever it isn’t clearly outweighed. Hold onto that: the cascade is exactly the moment this stops being true.
Person 1 guesses 'red.' Person 2 then draws a blue ball. Using the odds trick (each signal doubles its side), what should Person 2 rationally guess?
Person 3: the pivotal moment
Now the trap springs. Suppose the first two both guessed “blue” — maybe both drew blue, maybe Person 2 tied and happened to have drawn blue too; either way the room can infer two blue signals are on the board. Along comes Person 3, reaches in, and draws a red ball.
Walk their reasoning explicitly. Person 3 knows:
- Two blue signals from the guesses ahead of them (odds ×2 twice on the blue side).
- One red signal from their own eyes (odds ×2 on the red side).
Stack them: 2 × 2 : 2 = 4 : 2 = 2:1 in favour of the blue urn. Two blue signals outweigh one red signal by exactly two to one. So the rational guess — the one that maximises Person 3’s own chance of being right — is “blue.”
Read that again, because it is the whole lesson: Person 3 holds a red ball in their hand and correctly says “blue.” Not out of cowardice, not to fit in — because the math genuinely says the blue urn is twice as likely, and their single red draw isn’t enough to overturn two earlier blue signals. Their private evidence is real, but it is outvoted by better evidence.
Here is the running tally, written as inferred signals:
| Person | Own draw | Signals the room can infer so far | Odds (blue : red) | Rational guess | Does the guess reveal their draw? |
|---|---|---|---|---|---|
| 1 | blue | 1 blue | 2 : 1 | blue | Yes — pure readout of the draw |
| 2 | blue | 2 blue | 4 : 1 | blue | Yes — draw still shows through |
| 3 | red | 2 blue (their red is hidden) | 4 : 2 = 2 : 1 | blue | No — they’d say blue either way |
| 4 | anything | still just 2 blue | 2 : 1 (before own draw) | blue | No |
Look at the last column flipping from Yes to No at Person 3. That flip is the birth of the cascade — the instant a guess stops carrying private information.
The counterintuitive core
A rational Person 3 ignores the evidence in their own hand. Not because they’re weak or conformist — because two independent blue signals genuinely outweigh one red signal, two to one. The unsettling truth of this whole model is that following the crowd here is the correct individual decision. The disaster is collective, and we’re about to see why.
Person 3 draws red but there are two blue guesses ahead of them. Which statement about their 'blue' guess is exactly right?
Drive the line yourself
Before we prove the cascade locks, get your hands on it. Below is a line of rational deciders you can steer. Each shows a faint private-signal letter (their draw) and a coloured public choice box. Turn the accuracy slider down toward 55% and watch wrong cascades bloom — the line confidently marches onto the incorrect urn on the strength of a couple of unlucky early draws. Notice the readout: the number of signals that actually informed the line barely moves off two or three, no matter how long the line is. Then flip the public signal on and watch it reroute the entire line at once.
Information cascades — the line of deciders
Drive the line yourself
Each person in the line has a private signal about which option is right, and decides seeing only the earlier choices. Lower the signal accuracy and watch a couple of early picks stampede everyone into copying the herd — even against their own evidence. Then drop a public signal and watch it reroute the whole line.
Only 4 of 14 private signals actually informed the line — 4 people overrode their own evidence to copy the herd. Cascade locked at position 5. The line landed on the CORRECT option — but almost by luck: it rode a handful of early signals, not the crowd’s combined knowledge.
Three experiments worth running:
- Sweep the accuracy slider down toward ~60%. As you drag it across the low range, each setting redraws the line’s private signals — and you’ll routinely catch it locking onto the wrong urn. Proof that a cascade doesn’t need bad reasoners, just a couple of unlucky early draws.
- Watch “signals used.” However long the line, the number of private draws that genuinely shaped the outcome stays tiny — usually two. Everyone after the lock is an echo. This is the point of Lesson 3: the crowd pools almost no information.
- Toggle the public signal. One shared piece of evidence, visible to all, can start a cascade instantly — or, if it contradicts the herd, break one. Hold that thought; Lesson 4 is built on it.
What to actually watch
Not whether the line agrees — it almost always does. Watch how little it took to make them agree, and how often that agreement is wrong when signals are noisy. A confident, unanimous line resting on two draws is the whole danger in one picture.
The cascade locks: why Person 4 is trapped too
Here’s the part that turns a single odd decision into an unbreakable pattern. Person 3 guessed “blue” regardless of their draw — we just showed they’d say blue whether they pulled blue or red. So their guess reveals nothing about what they drew. Everyone behind them can deduce this. Person 3’s guess is informationally empty: it tells you about the situation Person 3 was in, not about the ball Person 3 held.
Now step into Person 4’s shoes. What do they actually know?
- The two genuine blue signals from Persons 1 and 2. (Still just two.)
- Person 3’s guess — which they correctly discard as empty.
- Their own fresh draw.
That’s identical to the situation Person 3 faced: two blue signals in the public pool, plus one private draw. So Person 4 does the same arithmetic — 2:1 for blue before their draw — and if they pull red, they too land at 2:1 and guess “blue,” hiding their red. And so their guess is also empty. Person 5 inherits the exact same frozen pool. And Person 6. And everyone after.
The public tally is frozen at the first two signals. No later draw — however many reds pile up in people’s hands — ever enters the record, because from Person 3 onward every guess is a foregone conclusion. Fifty people could each privately draw red, all guess blue, and the line would look like an avalanche of unanimous, confident evidence for the blue urn while secretly sitting on a pile of contradicting red draws nobody will ever hear about.
The lead you need is only TWO
A cascade doesn’t require a landslide. The instant two matching signals sit ahead of you and your own single draw can’t break the tie, you rationally join — and your join adds nothing, so the next person is stuck in the identical box. Two is the magic number. Small, early, and permanent.
Could a cascade ever fail to start?
Yes — if the early draws disagree. If Person 1 says blue and Person 2 (drawing red) breaks the tie to “red,” the pool is 1 blue : 1 red and Person 3’s own draw actually decides things — no lock yet. The cascade needs an early lead of two in the same direction. But because a two-signal lead is so easy to get by chance, cascades form almost immediately in practice. Fragile to start, frozen once started — that tension is the engine of Lesson 3.
Pin the locking mechanism in one sentence.
Pick the right option for each blank, then check.
Once two matching signals sit ahead of you, a rational person guesses that colour even against their own draw, so their guess reveals about their private signal — which means the public tally stays at the first signals, and everyone after just the herd.
Which guesses actually carry information?
The heart of the model is a sorting question: some public guesses genuinely leak a person’s private draw into the room, and some are hollow echoes. Telling them apart is the whole skill. Sort these.
For each guess, decide whether it reveals the speaker's private draw or reveals nothing (because they'd have said it regardless).
Place each item in the right group.
- Person 12 guesses 'blue' after a long blue cascade, having drawn red twice in a row earlier for practice.
- Person 1 guesses 'blue' with no one ahead of them.
- Person 3 guesses 'blue' after two 'blue' guesses, while holding a red ball.
- Person 7 guesses 'blue' in a line that has said 'blue' six times running.
- Person 2 draws blue after one 'blue' and guesses 'blue' at 4:1 odds.
- Person 2, hearing one 'blue', draws red and breaks the tie to guess 'red'.
Why this is rational, not foolish
It’s tempting to sneer at the frozen line — a mob of sheep, each too timid to trust their own eyes. That reading is wrong, and getting it wrong is the single biggest misunderstanding of this model. Every person in the line is behaving optimally given what they can see. Person 3, holding a red ball, maximises their personal probability of naming the right urn by saying “blue,” because two blue signals really do make blue twice as likely. Punish them for it and you’d be punishing correct reasoning.
The failure is not in any individual — it’s structural. The problem is that the mechanism the group relies on to pool information (people announcing guesses) quietly stops carrying information after the second person. Each rational decision to copy the crowd is locally correct and globally corrosive: it’s the right move for the decider and a small act of vandalism against everyone downstream, because it removes one more private draw from the shared record. Nobody intends this. It’s an emergent property of individually smart choices.
The clean summary
A cascade is a collective failure built entirely from individually rational choices. Each person trades away their own weak signal for the crowd’s stronger-looking one — a good trade for them — but in doing so they stop the crowd from ever learning what they knew. Smart people, sensible decisions, dumb aggregate. Nobody is the fool; the structure is.
A colleague scoffs: 'The urn cascade just proves people are irrational herd animals.' What's the precise correction?
In a 40-person blue cascade where 25 people secretly drew red, how much of that red evidence made it into the public record?
When to reach for this model
Pull out the urn game whenever you see a crowd converging fast and confidently on a choice where each person could privately check only a little. The diagnostic question is never “what does the crowd know?” but “how many independent signals actually made it into this consensus before people started copying?” If the honest answer is “two, and then everyone echoed,” you’re looking at a cascade, not a pooled verdict — and the crowd’s confidence is telling you nothing about whether it’s right.
That reframing travels everywhere: a startup that “everyone” is investing in, a research result the field piles onto, a product review section where the first two five-star ratings shaped all the rest, a queue outside a restaurant. In every case the model tells you to mentally discount the echoes and hunt for the handful of genuinely independent signals underneath. Usually there are far fewer than the size of the crowd suggests.
Recap
Big picture
The urn game
- The urn cascade, worked by hand
- Setup
- Two urns: 2:1 blue vs 2:1 red, 50/50 which is in play
- Each person draws one ball privately, then guesses aloud
- You hear all guesses, never see a draw
- One blue draw → blue urn 2:1 likely
- The arithmetic (count in odds)
- Start 1:1; each signal doubles its side
- Two blue vs one red = 4:2 = 2:1 for blue
- Tie-break rule: follow your own draw
- Person 3: the pivot
- Draws red, but 2 blue signals outvote it 2:1
- Rationally guesses blue against own eyes
- Guess now reveals nothing about their draw
- The lock
- Empty guesses freeze the public tally at 2 signals
- Everyone after inherits the same trap and echoes
- A lead of just TWO starts it
- Why it matters
- Individually rational, collectively broken
- Information stops aggregating
- Confident ≠ correct
- Setup
You’ve now built the cascade from the ground up and seen the exact decider — Person 3 — where rational copying begins and information stops flowing. The unsettling takeaway: this whole edifice of unanimous confidence rests on just two early draws. Next, in “Rational Yet Fragile,” we push on that fact — measuring just how little information a cascade actually pools, and why that emptiness makes even a long, confident herd shatter and flip at the first credible nudge.