Case Study 01: Earthquakes and Bestsellers — The Same Curve
Context: This case study accompanies Chapter 4 (Power Laws and Fat Tails). It traces the identical mathematical pattern — a power law distribution — through two seemingly unrelated domains: seismology and the publishing industry. The goal is not to argue that earthquakes and bestsellers have anything in common in their substance, but that they share a structural signature that reveals deep truths about how extreme events arise in complex systems.
I. The Earth Speaks in Power Laws
Counting Quakes
The Gutenberg-Richter law is one of the most robust empirical relationships in all of science. Since Beno Gutenberg and Charles Richter first published their findings in 1944, the law has been confirmed with ever-larger datasets, across every tectonic region on Earth, and over timescales ranging from months to millennia.
The relationship is starkly simple. If you record every earthquake above a certain minimum magnitude in a given region over a given time period, and then count how many earthquakes occurred at each magnitude level, you find that the frequency drops by a nearly constant factor for each unit increase in magnitude. In most regions, that factor is approximately ten. Ten times as many magnitude-3 earthquakes as magnitude-4. Ten times as many magnitude-4 as magnitude-5. Ten times as many magnitude-5 as magnitude-6. And so on, up to the largest events the tectonic system can produce.
On a log-log plot — with the logarithm of earthquake frequency on the vertical axis and earthquake magnitude on the horizontal axis — this relationship appears as a straight line. The slope of the line is approximately -1 (varying slightly by region), which means the exponent of the power law is approximately 1. This is the Gutenberg-Richter b-value, one of the most measured quantities in seismology.
What the Straight Line Means
The straight line on the log-log plot tells you something profound about the Earth: it does not distinguish between small earthquakes and large ones in any fundamental way. There is no separate mechanism that produces magnitude-3 earthquakes and a different mechanism that produces magnitude-8 earthquakes. They are all produced by the same process — the slow accumulation of strain along a fault, followed by sudden slip. The difference between a small quake and a large one is not the type of failure but the extent of failure: how far the rupture propagates along the fault before it stops.
This has a deeply unsettling implication. It means that every small earthquake was, in a sense, a large earthquake that happened to stop early. The same stress accumulation that produces a magnitude-3 event could, under slightly different conditions — a slightly longer fault, a slightly higher stress concentration, a slightly different arrangement of asperities — have produced a magnitude-7 event. The small earthquake and the catastrophic earthquake are not different species. They are the same species at different scales.
Seismologists describe this by saying that the earthquake process is scale-invariant — it looks the same at every scale. Zoom in on the small earthquakes and you see the same statistical pattern as when you zoom out to include the large ones. This scale invariance is the hallmark of a power law, and it is the reason the Gutenberg-Richter law works: the physics is the same at every magnitude, so the statistics are the same at every magnitude.
Fukushima Revisited
The chapter's opening discussed the 2011 Tohoku earthquake and the Fukushima nuclear disaster. Let us examine this in more detail, because it illustrates the practical consequences of the Gutenberg-Richter law with terrible clarity.
The design basis for the Fukushima Daiichi nuclear plant was determined in the 1960s, when the plant was being planned. Engineers estimated the maximum credible earthquake and tsunami for the region based on the historical record — the catalogue of known past events. The largest known tsunami in the region had been approximately 5.7 meters, so the seawall was built to that standard, with a modest safety margin.
The problem with this approach is that the historical record is short. Japan has detailed seismological records going back roughly a century, and fragmentary records going back several centuries. But the Gutenberg-Richter law applies to timescales much longer than any historical record. A magnitude-9 earthquake in the Japan Trench might occur once every 500 or 1,000 years — well beyond the historical window. The absence of such an event in the historical record does not mean it cannot happen. It means it has not happened yet.
The Gutenberg-Richter law, properly applied, would have told the engineers something different from what the historical record alone suggested. Given the observed frequency of magnitude-6 and magnitude-7 earthquakes in the region, the law predicts the approximate frequency of magnitude-8 and magnitude-9 events. Those predictions would have indicated that a magnitude-9 event — and a corresponding tsunami far larger than 5.7 meters — was not a fantasy but an eventuality, with a meaningful probability of occurring within the plant's operating lifetime.
The engineers were not ignorant of seismology. They were, in a sense, imprisoned by Gaussian thinking — by the implicit assumption that the historical record captures the full range of possibilities, that the past is a reliable guide to the future, and that events far outside the observed range can be dismissed. The Gutenberg-Richter law says the past is not a reliable guide to the extremes, because the extremes are rare enough that any finite historical window is likely to miss them.
This is the difference between Mediocristan thinking and Extremistan thinking, applied to a concrete engineering decision with life-and-death consequences.
II. The Publishing Industry Speaks the Same Language
The Shape of Book Sales
Now let us cross from geology to culture — from the movement of tectonic plates to the movement of books off shelves — and find, improbably, the same curve.
If you plot the sales of all books published in a given year, ranked from most to least, on a log-log scale, you get a straight line. A handful of mega-bestsellers sell millions of copies. A larger group of successful books sell hundreds of thousands. A still larger group sell tens of thousands. A vast ocean of books — the majority of everything published — sell fewer than a thousand copies. Many sell fewer than a hundred.
The specific numbers vary by year and by market, but the shape is consistent: a power law distribution, with an exponent typically between 1.5 and 2.5 depending on how the data is measured and what is included. The top 1 percent of titles account for somewhere between 50 and 80 percent of all sales. The bottom 50 percent account for a tiny fraction.
This is the publishing industry's version of the Gutenberg-Richter law. And just as with earthquakes, the extreme events (mega-bestsellers) are not anomalies or lucky accidents separate from the rest of the distribution. They are the tail of the same curve.
The Mechanism: Preferential Attachment in the Bookstore
But here is where the analogy deepens. The mechanism that generates the power law in book sales is not the same as the mechanism that generates it in earthquakes (the physics of fault rupture), yet it is structurally identical at an abstract level. In both cases, there is a positive feedback loop that amplifies initial advantages.
For books, the feedback operates through multiple channels:
Bestseller lists. When a book appears on the New York Times bestseller list, it gains visibility. Bookstores display it more prominently. Readers who are browsing without a specific title in mind are more likely to encounter it. The increased visibility drives more sales, which keeps the book on the list, which maintains its visibility. This is textbook preferential attachment: popularity breeds more popularity.
Word of mouth. The probability that you hear about a book from a friend is proportional to the number of your friends who have read it — which is proportional to the book's total sales. A book with ten thousand readers has far more word-of-mouth ambassadors than a book with a hundred readers. The larger the readership, the faster it grows. The reinforcing loop again.
Media coverage. Publishers invest marketing dollars in proportion to expected sales, and media outlets cover books they believe their audiences want to hear about — which are, disproportionately, the books that are already selling well. A runaway bestseller generates its own media ecosystem of reviews, interviews, and cultural commentary. This media attention drives further sales.
Algorithmic recommendation. On platforms like Amazon, Goodreads, and social media, algorithms recommend books based on popularity signals: what other people bought, what is trending, what has high ratings from many reviewers. These algorithms are preferential attachment machines — they systematically funnel attention toward books that are already receiving attention.
Each of these mechanisms is a reinforcing feedback loop (Chapter 2). Each amplifies existing advantages. And together, they produce a power law distribution of book sales — the same mathematical shape that the Gutenberg-Richter law produces for earthquake frequencies, through a completely different physical process.
The Aspiring Author's Dilemma
The power law in book sales has practical implications that most aspiring authors do not appreciate.
If book sales followed a Gaussian distribution, then talent and effort would be the primary determinants of success. A good book would sell moderately well; a great book would sell somewhat more; a mediocre book would sell somewhat less. The distribution would be centered on a meaningful "average," and most authors would cluster around it.
But book sales follow a power law. This means that the difference between "successful" and "unsuccessful" is not a matter of degree but of orders of magnitude. The feedback loops that generate the power law amplify small initial differences — differences that may be due to talent, but may equally be due to timing, marketing, luck, social connections, or the unpredictable dynamics of cultural attention.
The sociologist Duncan Watts demonstrated this experimentally. In his "MusicLab" study, he created an online marketplace where participants could listen to and download songs by unknown bands. In some conditions, participants could see how many times each song had been downloaded (a popularity signal). In other conditions, they could not. When the popularity signal was visible, the market exhibited extreme winner-take-all dynamics: a few songs accumulated enormous download counts while most languished in obscurity. When the signal was hidden, the distribution was much more equal, and the "winners" were different in different experimental runs — suggesting that which specific songs became hits was substantially determined by early random fluctuations amplified by preferential attachment, not by intrinsic quality alone.
This does not mean quality is irrelevant. A truly terrible book is unlikely to become a mega-bestseller (though it has happened). But it means that quality is necessary but not sufficient. The power law tells you that the distribution of outcomes is shaped more by the feedback mechanism than by the underlying quality of the inputs. The same lesson applies to music, film, apps, startups, and academic careers.
III. The Same Curve, Different Soil
What the Isomorphism Reveals
The fact that earthquake frequencies and book sales follow the same mathematical distribution is, at first glance, absurd. Tectonic plates and bestseller lists share no substrate, no history, no mechanism in any physical sense. And yet the same curve fits both.
This is the core insight of cross-domain pattern recognition, the organizing principle of this entire textbook. The power law is not a property of rocks or of books. It is a property of processes — of any process where events of varying size are generated, and where there is some mechanism that connects the probability of large events to the probability of small events in a specific, scale-invariant way.
For earthquakes, that mechanism is the physics of fault rupture: the same stress-accumulation-and-release process operates at every scale, and the probability of a rupture propagating to a given size follows a power law.
For book sales, that mechanism is preferential attachment: popularity breeds more popularity through bestseller lists, word of mouth, media coverage, and algorithmic recommendation, and the feedback loop produces a power law distribution of outcomes.
The mechanisms are different in their substance. They are identical in their structure. Both involve a process where the probability of an event of a given size is systematically related to the size itself, in a way that produces fat tails and extreme events.
The Practical Lesson
The practical lesson is the same in both domains: plan for the tail, not the average.
For earthquake engineering, this means designing for events far larger than anything in the historical record. The Gutenberg-Richter law is a tool for extending our risk horizon beyond experience.
For the publishing industry (and, by extension, for any cultural or creative market), this means recognizing that the market is inherently winner-take-all, that a few titles will dominate, and that the fate of individual books is determined as much by the dynamics of the feedback loop as by the quality of the writing.
For readers of this textbook — for anyone learning to see patterns across domains — the lesson is deeper still. The same mathematical curve in earthquakes and bestsellers is not a curiosity. It is evidence that the world is more unified than it appears. The patterns of geology and the patterns of culture are, at a certain level of abstraction, the same pattern. And learning to see that pattern — to recognize the straight line on the log-log plot, wherever it appears — is one of the most powerful intellectual tools this book can give you.
Discussion Questions
The case study argues that the Gutenberg-Richter law could have helped predict the tsunami risk at Fukushima. Why, in your view, was this information not incorporated into the plant's design? What institutional or psychological factors might explain the reliance on historical records over statistical models?
Duncan Watts's MusicLab experiment showed that making popularity visible increased inequality of outcomes. What other systems can you think of where making information about popularity visible (or invisible) might change the shape of the distribution?
The case study claims that quality is "necessary but not sufficient" for success in power law markets. Do you agree? Can you think of examples where a clearly superior product lost to an inferior one that benefited from preferential attachment? What does this imply for how we think about merit?
If you were advising an aspiring author, how would your advice differ based on whether you assumed book sales followed a Gaussian or a power law distribution?