This week's model: Information Cascades
An information cascade happens when people stop relying on their own judgment and instead copy the decisions of the people who went before them, because the sheer weight of those earlier choices feels more convincing than their own private information.
The strange part is that each person's choice to follow the crowd can be perfectly rational in the moment. The crowd's outcome can still be wrong, because once a cascade starts, nobody after the first few people is actually contributing new information. Everyone is just copying copies.
This map places Information Cascades among the models it is most often confused with. The horizontal axis asks whether a decision mostly comes from your own independent read of the situation or from copying what others around you are already doing. The vertical axis asks whether the mechanism is a calculated inference or a more automatic, emotional pull.
Imagine a row of people deciding, one after another, whether a new restaurant is good or bad, purely by watching who walks in. Each person also has a small private clue of their own: maybe they read one review, or noticed the smell of the kitchen. No one can see anyone else's clue, only their final choice: walk in, or walk away.
The first person has nothing to go on but their own clue, so they act on it directly. The second person now has two sources: their own clue, and the fact that the first person either walked in or didn't. If those two sources agree, the second person feels even more confident. If they disagree, the second person has a real decision to make.
This is the mechanism economists Sushil Bikhchandani, David Hirshleifer, and Ivo Welch formalized in 1992, building on related work by Abhijit Banerjee the same year. Once enough earlier people have made the same choice, that accumulated choice can carry more weight than any one person's private clue, even a reasonably good one.
At that point, it becomes individually rational to ignore your own clue and copy the crowd. Not because you're lazy or scared, but because, from your seat in line, the crowd looks like stronger evidence than your own small sample of one.
Here is the uncomfortable part. Once a cascade starts, the people who follow it are no longer adding their own information to the public record. The restaurant might actually be mediocre, but if the first two people happened to get unlucky, misleading clues, everyone after them can pile on the wrong conclusion while feeling completely rational the entire time.
This is why a cascade is often described as fragile. It can look unanimous and still be built on a thin foundation: often just one or two original, independent signals that got amplified by imitation, not multiplied by new evidence.
Cascades are fragile in a useful way too: they can be reversed. A single piece of strong, visible, new information, like a scathing public review, a credible whistleblower, or one respected early adopter visibly changing course, can be enough to make the next observer question the crowd and restart independent thinking.
This is also why anonymous or simultaneous decisions (like a sealed-bid auction or a secret ballot) resist cascades: nobody can see anyone else's choice as it's being made, so there's no sequence for a cascade to form along.
Information cascades explain a lot of patterns that otherwise look mysterious: why a mediocre product can rocket to the top of a sales chart once a few visible customers adopt it, why a venture round can fill up fast once one respected investor signs on, why a bad policy can survive for years once enough officials have publicly endorsed it, and why financial bubbles can keep inflating even when individual traders privately suspect prices are too high.
The practical takeaway is not "never follow others." It's knowing when you're adding independent judgment to a decision, and when you're just one more domino falling the same way as the last one, for reasons that have little to do with the actual quality of the choice.
Click a card to see how the same mechanism plays out across very different situations.
This simulator runs the classic sequential-decision model behind information cascades. A hidden "true state" (the restaurant is secretly Good or secretly Bad) is set at random. Each simulated person gets one private signal pointing toward the correct answer, at the accuracy level you choose, and decides in sequence while watching everyone before them.
Describe a real decision where you can feel the pull of the crowd, and this tool will sketch out how to think through it.
This is a portable, plain-language description of the model, written so you can hand it to an AI assistant or a teammate and get consistent, well-grounded help applying it.
Describe a specific situation, and this will generate a ready-to-use prompt for applying Information Cascades to it.