Preferential Attachment

Imagine you’ve just moved to a new city and are looking for a sports club. You probably won’t pick the smallest club, but the one everyone’s talking about. That’s exactly the behavior Preferential Attachment describes: whoever joins new preferentially attaches to whoever already has many followers, links, or customers. The advantage of the big players thus keeps growing on its own, without anyone having to plan it. Experts also call this the rich-get-richer effect. It’s one of the few simple rules that can describe the growth of remarkably different systems astonishingly well.

Why some platforms simply can’t be caught up anymore

When this principle takes effect, no balanced field emerges, but an extreme imbalance. A handful of providers collect almost all the connections, while thousands of others receive barely any attention. This distribution is called the long tail: a few giants stand at the front, followed by a long tail of tiny players. This explains why there are many social networks, but only three or four that really count.

This is highly relevant for the financial world. A company that gained the most users early on becomes difficult to attack later. A better product alone is often not enough, because new users still go where everyone already is. Investors therefore pay high valuations for companies that have achieved this lead. This is referred to as a network effect, which is further amplified by Preferential Attachment.

There is also an unpleasant side to this. When attention always accumulates with the same players, good new ideas have a hard time. In science, frequently cited papers get cited even more often, regardless of whether newer works are better. Quality and visibility then diverge.

The rule behind the growth

Such systems are described as a network. A network consists of nodes, i.e. the individual members, and edges, i.e. the connections between them. In the case of websites, the nodes are the pages and the edges are the links. The number of connections a node has is called its degree.

The best-known model for this comes from physicists Albert-László Barabási and Réka Albert, dating from 1999. It works in two steps that keep repeating. First: in every time step, a new node is added. Second: this node does not choose its connections randomly with equal probability. The probability of docking onto a particular node is proportional to that node’s degree.

A node with a hundred connections is thus chosen ten times as often as one with ten. As a result, every lead reinforces itself over time. An important difference from a purely random network: there, almost every node would have roughly the same number of connections. A model, then, doesn’t need to be complicated to produce realistic structures. A common misconception, by the way, is that timing doesn’t matter. In fact, early nodes have a systematic advantage, simply because they’ve had longer to accumulate connections.

From citation index to recommendation algorithm

The effect is most visible in social media. An account with many followers appears more often in recommendations and thereby gains even more followers. Recommendation systems actively reinforce this because they preferentially surface popular content. Some platforms therefore deliberately mix in new or small accounts to mitigate the imbalance.

The pattern also shows up in the AI world. Developers preferentially build their applications on models for which there is already a lot of documentation and many examples. This attracts further developers and makes the ecosystem even more attractive. This distribution is also very clearly visible with code libraries on platforms like GitHub.

In news articles you rarely encounter the term literally. Instead, there’s talk of market concentration, winner-takes-most, or the Matthew Effect. Anyone familiar with the principle understands such reports better. It provides an explanation for why markets tip over, even though no one planned a monopoly.