Information Cascades
Once enough people in a sequence have made the same choice, it can become rational for the people who follow to set aside their own weaker private information and follow the visible pattern instead, which produces consensus that looks stronger than the evidence actually behind it.
What Is It?
Economists studying herd behavior in the early 1990s formalized a pattern that shows up anywhere people decide one after another rather than all at once: each person has some private information or judgment, but also observes what the people before them decided. Early on, people mostly act on their own private information. But once a few people in a row have made the same choice, it becomes rational for the next person to follow that visible pattern rather than their own weaker private signal, reasoning that so many people probably wouldn’t have chosen this unless it were right. Once that happens, something stranger sets in. Each additional visible choice makes the consensus look stronger without actually adding any new information to it. The people responding to the cascade aren’t revealing their own private signals anymore, they’re just responding to the same early decisions everyone after them is also responding to, so a long chain of apparent agreement can rest on far fewer independent judgments than its length suggests, sometimes only the first one or two. The unsettling part is that every individual decision in the cascade is locally rational. Nobody has to be foolish or lazy for it to happen. It’s simply more rational, at each step, to trust the accumulated visible pattern of everyone who went before than to trust your own single data point against it. That’s also what makes a cascade fragile in a specific way: because the visible consensus may rest on very little actual independent evidence, a single sufficiently strong new signal can overturn it entirely, and can even start a cascade running the opposite direction. A decision that looks unanimous can be both large and shallow at the same time.
Why Does It Matter?
Sequential approval processes, a proposal that goes past one reviewer, then a committee, then leadership, create ideal conditions for cascades. Each reviewer sees that the people before them approved it, and that visible approval becomes evidence in its own right, often stronger evidence than whatever private doubt the current reviewer actually has. A proposal can sail through five approvals not because five people independently found it sound, but because the first one or two approvals created a visible signal that made dissent at every subsequent stage feel like second-guessing a settled matter rather than raising new information. This is a different failure than groupshift or shared information bias, both of which happen inside a single room, at a single moment. A cascade is sequential: it’s specifically about what happens when decisions are made one after another, each one visible to the next, rather than all at once. And it means the order people are asked matters as much as who’s asked, since an early skeptic placed later in the sequence faces much stronger pressure to defer than the same skeptic asked first.
What Changes Once You See It?
You start collecting judgments independently before revealing anyone else’s answer, so five approvals actually represent five pieces of information rather than one judgment echoed four times. You start treating a long chain of approvals as weaker evidence than it looks, since a cascade can produce five confident approvals from one real judgment and four acts of reasonable deference. You become more cautious about letting anyone go first at all when independent judgment genuinely matters, using blind or simultaneous assessment where possible so the organization captures everyone’s private information before social information has a chance to overwrite it.
Common Misunderstandings
- It isn’t a claim that deferring to others is always irrational. Following the visible pattern of people who went before you is often the individually sensible choice, which is exactly what makes cascades so persistent and hard to interrupt.
- It isn’t the same as Groupshift or Shared Information Bias, which describe distortions inside a single group discussion happening at the same time. A cascade specifically requires a sequence, one decision visible to the next, made at different times.
- It doesn’t mean the cascade’s outcome is necessarily wrong. A cascade can land on the correct answer, the problem is that the apparent strength of the consensus exaggerates how much independent information actually backs it.
- It isn’t fixed by simply asking people to think independently. The incentive to defer to the visible pattern is real and rational at the individual level, structural changes to the order and visibility of decisions are what actually interrupt it.
Diagnostic Question
How many of these apparent approvals represent genuinely independent judgments, and how many were made after the emerging consensus was already visible?
Explore Further
Field Notes
None yet.
Related Field Guide
Origin
Formalized independently by Sushil Bikhchandani, David Hirshleifer, and Ivo Welch, “A Theory of Fads, Fashion, Custom, and Cultural Change as Informational Cascades” (1992), and by Abhijit Banerjee, “A Simple Model of Herd Behavior” (1992).