A power law is a mathematical relationship where one quantity changes as a fixed power of another. The formula is simple: f(x) = Axp, where A is a constant and p is the exponent. What makes power laws fascinating isn’t the formula itself but what they reveal: the same pattern shows up in word frequencies, earthquake sizes, city populations, biological metabolism, and wealth distribution. When you understand power laws, you start seeing them everywhere.
The Basic Pattern
In a power law relationship, small values are extremely common and large values are extremely rare, but those rare large values are far bigger than you’d expect from a normal bell curve. Think of city sizes: thousands of tiny towns, hundreds of mid-size cities, and a handful of massive metropolises. There’s no “typical” city size the way there’s a typical human height. The distribution stretches out dramatically toward the high end.
This stretched-out tail is what statisticians call a “fat tail.” In a bell curve (the familiar normal distribution), values cluster tightly around an average, and extreme outliers are vanishingly unlikely. In a power law distribution, the average can be mathematically infinite, because the extreme values are so large they pull the mean upward without limit. That single property is why power laws matter so much for understanding risk, wealth, natural disasters, and the structure of networks.
Scale Invariance: The Defining Feature
The signature property of a power law is scale invariance. This means the pattern looks the same no matter how much you zoom in or out. If you examine earthquakes between magnitude 3 and 5, the ratio of smaller to larger quakes is the same as the ratio between magnitude 5 and 7. There’s no characteristic scale where the behavior changes. Cities, wealth distributions, and earthquake frequencies all share this self-similar quality, which is why power laws are sometimes called “scale-free.”
This is deeply counterintuitive if you’re used to thinking in bell curves. Height has a characteristic scale (most adults are within a foot of the average). But wealth has no characteristic scale. The gap between someone with $1 million and $10 million is structurally identical to the gap between $10 million and $100 million. The same proportional pattern repeats at every level.
The 80/20 Rule Is a Power Law
The Italian economist Vilfredo Pareto observed that 80% of society’s wealth was controlled by 20% of its population. This “Pareto Principle” or “80/20 Rule” is one of the most widely cited power laws in everyday life. The Pareto distribution that describes it is a power law probability distribution, and the specific 80/20 split corresponds to an exponent of about 1.16.
The principle extends well beyond wealth. Roughly 80% of a company’s sales often come from 20% of its products. A small fraction of bugs cause most software crashes. A handful of customers generate most complaints. These aren’t coincidences or loose metaphors. They’re all manifestations of the same underlying mathematical structure, where outcomes are concentrated in a small number of cases rather than spread evenly.
Word Frequencies and Zipf’s Law
One of the cleanest examples of a power law appears in language. In any large body of text, the most frequent word appears roughly twice as often as the second most frequent word, three times as often as the third, and so on. This pattern, known as Zipf’s law, has held up across languages and text types for over 70 years of study.
In English, “the” dominates. The second-ranked word has a frequency proportional to 1/2, the third proportional to 1/3, and onward down the list. The vast majority of words in any language are used rarely, while a tiny core of words does most of the heavy lifting. This rank-frequency relationship follows a power law with an exponent close to 1, and it emerges naturally in every sufficiently large corpus of natural language.
Earthquakes and Natural Disasters
The Gutenberg-Richter law describes how earthquake frequency relates to magnitude: log(N) = a – bM, where N is the number of earthquakes at or above magnitude M. In plain terms, for every step up in magnitude, earthquakes become roughly ten times less frequent. Magnitude 5 quakes happen about ten times more often than magnitude 6 quakes, which happen about ten times more often than magnitude 7 quakes.
This is a power law in logarithmic form, and it has practical consequences. It means truly catastrophic earthquakes are rare but not impossible, and their probability can be estimated from the frequency of smaller quakes. The same type of relationship appears in wildfire sizes, flood levels, and volcanic eruptions. Nature generates extreme events more often than a bell curve would predict.
Networks and the Rich-Get-Richer Effect
Power laws also describe how connections are distributed in networks, from social media to scientific collaboration to the structure of the internet. Most nodes have few connections, but a small number of hubs have an enormous number. This pattern arises through a process called preferential attachment: new nodes tend to connect with nodes that are already well connected. The probability of gaining a new connection is proportional to how many connections you already have.
This is essentially a “rich get richer” mechanism, sometimes called the Matthew effect. A website with many inbound links is more visible, so it attracts more links. A scientist with many collaborators gets invited to more projects, generating still more collaborations. The result is a power law distribution of connections, where the gap between ordinary nodes and supernodes is vast. Social scientists have confirmed that preferential attachment acts as the primary driver in the evolution of co-authorship networks and technology industry collaboration networks.
Biological Scaling
One of the most striking power laws in biology is Kleiber’s law, first uncovered in 1930. It states that an animal’s metabolic rate scales with its body mass raised to the power of roughly 3/4. A cow doesn’t burn energy ten times faster than a cat just because it’s ten times heavier. Instead, metabolic rate increases more slowly than mass, following a consistent exponent across organisms spanning many orders of magnitude, from tiny insects to whales. The same 3/4 exponent appears in plants as well, suggesting a deep structural constraint on how living things use energy.
Why Power Laws Matter for Risk
If you assume the world follows bell curves, you’ll dramatically underestimate the probability of extreme events. Financial markets, for instance, experience crashes far more often than a normal distribution would predict. These rare, high-impact events, sometimes called “black swans,” live in the fat tails of power law distributions. Economists use the concept of fat-tailed distributions to describe phenomena ranging from market crashes to climate change impacts.
The practical difference is enormous. Under a bell curve, a stock market drop of 20 standard deviations from the mean is essentially impossible. Under a power law, it’s unlikely on any given day but inevitable over a long enough time horizon. This is why power laws have reshaped how risk analysts, insurers, and financial regulators think about extreme events. Ignoring fat tails means being perpetually surprised by disasters that were, statistically, entirely predictable.
How Scientists Verify Power Laws
Not every dataset that looks like a power law on a graph actually is one. Many distributions, including lognormal distributions, can mimic the shape of a power law over certain ranges. The gold standard for testing involves combining maximum-likelihood fitting methods with statistical tests that compare how well a power law fits the data versus alternative explanations. A 2009 framework published in SIAM Review by Clauset, Shalizi, and Newman established the widely used approach, and their analysis found that several datasets commonly claimed to follow power laws didn’t hold up under rigorous testing.
This matters because the implications of a true power law (infinite possible variance, scale-free behavior, predictable ratios of small to large events) are dramatically different from those of a distribution that merely looks similar. A lognormal distribution, for example, always has a finite mean and variance. Mistaking one for the other can lead to badly wrong predictions, especially about extreme events.