Information cascade

An information cascade is a pattern of collective decision-making in which people infer what is best from the choices of those who acted before them, sometimes setting aside their own information. The mechanism can help a group outperform any one member, yet it can also turn a few early decisions into a self-reinforcing sequence that no longer gathers what the group knows. Its consequences depend not simply on whether people follow others, but on whether the first signals are sound and whether later participants can still contribute independent judgment.

How a Cascade Forms

The model begins with individuals who receive imperfect information about which of two alternatives is better. In the example offered by economists Sushil Bikhchandani, David Hirshleifer, and Ivo Welch, a group must choose between a new Indian restaurant and a new Thai restaurant; the Indian restaurant is objectively better, but any individual’s information may point the wrong way. People supplement their own signals by observing what others do. If decisions occur one after another, the first few choices can exert disproportionate influence: should the early diners, by chance, receive misleading signals and choose the Thai restaurant, those arriving later may infer from its apparent popularity that it is the better option. They may follow it even when their own information favors the Indian restaurant.

This is not necessarily mindless conformity or a desire to imitate for its own sake. The later diners are trying to learn from other people’s behavior, on the reasonable assumption that those people had useful information. The catch is that an observed choice does not reveal what actually informed it. Once the sequence begins, each person may be responding less to direct knowledge than to an inference about the knowledge of earlier decision-makers. The crowd appears to be accumulating evidence, while its members increasingly copy one another’s conclusions.

That change creates the cascade’s central weakness. Once enough people have made the same choice, it can be rational for the next person to discount a private signal and follow the apparent majority. But if everyone does so, the decisions cease to provide new information: each person assumes previous choices reflect others’ knowledge, though those choices may themselves be based on earlier imitation. Instead of combining independent signals, the group produces a chain of choices that can preserve an initial mistake.

Plank Roads: A Costly Cascade

The plank-road boom in nineteenth-century America illustrates how a cascade can grow from genuine benefits into a bad collective investment. The first road in Salina succeeded, and roads built soon afterward also appeared to work. Plank roads cut travel time between towns in half, could be used in different kinds of weather, and enabled small farmers to reach wider markets. Each successful-looking road offered evidence in favor of the next one, while making alternative solutions seem less promising. George Geddes became a prominent spokesman for the industry; within a decade, New York had 352 plank-road companies and the United States had more than a thousand.

The underlying weakness was delayed: the roads did not last as long as their promoters had predicted. Rather than surviving eight years, as Geddes had promised, their life span was closer to four, making them too costly to maintain. By the late 1850s, the plank road’s limits were becoming clear, and by the end of the Civil War almost all plank roads had been abandoned. A few lasted longer, including a thirteen-mile stretch along what is now Route 27A in Jamaica, Queens, which remained in operation until the 1880s. By then, however, the wider investment had already followed a pattern in which adoption itself seemed to confirm the technology’s promise.

The example distinguishes a cascade from a simple fad. The roads offered real improvements, so the spread was not merely a matter of a persuasive promoter or towns copying their neighbors. But early successes made the benefits seem durable before the crucial evidence about maintenance costs had emerged. As the number of roads grew, the common choice grew more persuasive, while the reasons to investigate alternatives weakened. The eventual failure was therefore not proof that the initial choice had been irrational; it showed how reasonable inferences from visible examples can crowd out less visible, later-arriving information.

When Following Works—and When It Fails

Cascades are not invariably harmful. In classroom experiments, cascading groups chose the better alternative about 80 percent of the time, a higher rate than any individual in the groups achieved. That result suggests that observing others can help aggregate imperfect knowledge. The structural problem is that decisions are made sequentially rather than simultaneously. People may decide at different times because some are cautious, some willing to experiment, and some have more money than others; nevertheless, the order gives early actions special power over what later participants do. Making decisions simultaneously, or nearly so, can help preserve the contribution of independent signals.

A beneficial historical example is William Sellers’s campaign in the 1860s to establish a standardized screw. At the time, American screws were handmade by machinists, a system that restricted mass production and kept customers dependent on the craftsmen who made and repaired their parts. Sellers believed interchangeable parts and mass production were inevitable, and designed a screw intended to be easier, cheaper, and faster to produce than its chief competitor, a British screw. Because the machinist community was tightly connected and had much at stake, he targeted influential users, including the Pennsylvania Railroad and the U.S. Navy. Each new customer made broader adoption seem more likely and helped make it more likely. Within a decade, his design was on its way to becoming a national standard, helping lay the groundwork for modern mass production.

The success came with an unsettling qualification: if machinists adopted the design because Sellers’s influence started a cascade, rather than because each independently judged it superior, the group reached the right answer partly by luck. A good result does not erase the vulnerability of the process. Collective decisions are most likely to be reliable when people with diverse opinions reach independent conclusions using their own information; a cascade can instead give decisive influence to whoever acts first or occupies a persuasive position in a network.

Bubbles and Shared Expectations

The late-1990s telecommunications bubble shows the cascade mechanism operating at vast scale. In the early years of the Internet, traffic grew at a rate of 1,000 percent a year. From about 1996, that growth slowed sharply, but the headline figure remained conventional wisdom. Telecommunications companies invested tens and eventually hundreds of billions of dollars in capacity intended to meet the expected demand. With the prevailing view insisting that traffic would continue to surge, refusing to invest could seem tantamount to business suicide, even to people who had doubts. Once the bubble burst, most companies were bankrupt or close to going out of business, and the assumption that had driven the investments was finally questioned.

The same logic complicates the explanation of financial bubbles. They do not begin from nothing: popular bowling in the 1950s, biotechnology’s transformation of pharmaceuticals, and the Internet’s genuine importance supplied real trends on which investors could initially build reasonable expectations. But as investors increasingly judge stocks by what other people are doing, rather than by independent assessment, each person’s expectations become dependent on everyone else’s. The group’s apparent wisdom then declines, even as the common belief can make the boom feel increasingly normal. During the 1990s stock-market bubble, access to financial news expanded through the Internet and cable television. CNBC’s nonstop market coverage and ticker brought investors constant reports of what other traders were saying and how the market was moving; the network could begin to affect trading simply because its reports were broadcast, regardless of their substantive content. The result was a feedback loop in which shared expectations could become harder to escape, with media amplifying rather than originating the convergence.

The account does not reduce bubbles to collective hysteria. People may follow the crowd because they believe the crowd is informative, just as later diners infer that a busy restaurant is better. That strategy can be useful when the group is drawing on independent knowledge. It becomes dangerous when everyone is piggybacking on judgments that no one is independently testing. Information matters, but so does its structure: a group needs both what its members share and what each holds privately. When expectations converge and private judgment loses its force, a market that can aggregate knowledge can instead begin to behave like a mob.

Limits and Implications

The cascade model does not describe every instance of social influence. Robert Shiller observes that people often make choices independently, without watching what others do; cascades are most plausible where people closely observe earlier actions before deciding. Nor is there just one account of how ideas spread. Malcolm Gladwell’s description of epidemics emphasizes influential people—mavens, connectors, and salesmen—who carry different kinds of information through social ties. In Bikhchandani, Hirshleifer, and Welch’s model, by contrast, early adopters need not possess special influence or expertise: their actions matter because they are early enough to be seen by everyone who follows.

The distinction clarifies the general stakes. A sequence can make useful information visible, but the same sequence can conceal the weak or accidental basis of an early choice. As more people follow, their numbers may look like independent confirmation even when they are not adding new evidence. Whether the result is a failed network of plank roads, a successful national screw standard, or overbuilt telecommunications capacity, the decisive question is whether collective agreement continues to reflect independent knowledge or has become imitation of an inference.