"The Signal and the Noise" by Nate Silver: Summary and Review


10 min read≈ 2,270 words5 questions · up to 5 points
What this is about
Why can a weather forecaster tell you tomorrow's rain fairly well, while economists and TV experts so often miss? You learn the main ideas of Nate Silver's book in the order of its argument, from foxes and hedgehogs to Bayes's theorem, how it compares with other books on judgment, and which of its claims critics doubt. At the end you know whether to read it, and in which format.
The book in three sentences· skip ahead
The Signal and the Noise argues that most predictions fail because people trust confident stories and take random patterns in data for real ones. Nate Silver visits forecasters in weather, baseball, poker, economics, earthquakes and politics to see who succeeds and why. His answer is that good forecasters think in probabilities, start from what they already know, update step by step as evidence arrives, and stay humble about what cannot be known.
The Signal and the Noise: Why So Many Predictions Fail, but Some Don't
- Author
- Nate Silver
- Published
- 2012
- Publisher
- Penguin Books (paperback with a new preface, 2015)
- Length
- 576 pages
- Audiobook
- 16h 21m, read by Mike Chamberlain
- Best for
- Curious readers who meet forecasts every day, from weather apps to election polls, and want to learn how to judge them without learning statistics first.
On the night of the 2012 American presidential election, the forecasting model of the statistician Nate Silver called the winner correctly in all fifty states, and Silver suddenly became famous as a man who could see the future. He later wrote that he felt like a fluke, because most of that perfect result was luck. The Signal and the Noise had come out six weeks earlier, and its main message was that we predict the world much worse than we think. Amazon named it the best nonfiction book of 2012, and its central question still matters wherever someone states a confident forecast: how do you tell the predictions that deserve your trust from the ones that do not?
What kind of book this is
Silver began as an economic consultant, then built a system for forecasting the performance of baseball players, earned money as an online poker player and in 2008 started the election website FiveThirtyEight. The book uses this unusual career as a tour guide. In thirteen chapters he visits people who make predictions for a living, among them weather forecasters, seismologists, economists, epidemiologists, chess programmers, poker players and investors, and asks why some of them succeed and most of them fail.
The chapters are long pieces of reporting with interviews and charts, and they hardly use formulas. Behind the tour stands one argument, which Silver states plainly in the preface to the 2015 paperback:
Far more often, as this book will explain, we overrate our ability to predict the world around us.
This review follows the logic of that argument rather than the order of the chapters. The ideas on foxes, calibration, overfitting and Bayes carry most of the weight, so they get the most room.
Predictions fail when we mistake noise for signal
The title names Silver’s basic distinction. The signal is the real pattern in the world, such as a true trend in voters’ opinions, and the noise is the random variation around it, which can look like a pattern for a while. A forecaster must separate the two, and Silver argues that our age makes this harder, not easier. More data means more noise, and people are very good at finding patterns that confirm what they already believe.
The first chapter shows what happens when this goes wrong on a large scale. Before the financial crisis of 2008, rating agencies such as Standard & Poor’s gave their top rating to new mortgage securities, based on models that put the chance of default at about one eighth of one percent. According to the figures Silver reports, around 28 percent of these securities later failed, a miss of more than two hundred times. The agencies had no track record for these new products, so their models rested on assumptions, above all the assumption that house prices across the country would not fall together.
Foxes forecast better than hedgehogs
The second chapter asks why political experts so often get it wrong, and it borrows an answer from the psychologist Philip Tetlock. Tetlock collected thousands of predictions by experts over many years and found that most did little better than chance. Some, however, did much better, and he described them with an old image that the philosopher Isaiah Berlin had made famous. Hedgehogs know one big thing and force every event into it. Foxes know many small things, think in probabilities and change their minds when the facts change.
Silver adds his own test. He collected predictions from The McLaughlin Group, an American TV show on which political commentators make forecasts at the end of each episode, and found that they were right about as often as they were wrong. The problem, he argues, is not only lack of knowledge but also incentives. Hedgehogs make better television, because a bold and certain answer is more fun to watch than a careful “probably”, so the media reward exactly the people who forecast worst.
Good forecasts are honest probabilities
If political experts are the warning, weather forecasters are the model, and chapter 4 is the most hopeful part of the book. Silver shows how much they have improved. Twenty-five years before the book, a three-day forecast of where a hurricane would reach land missed by about 350 miles on average, and by the time he wrote the miss was about 100 miles. The reason is a mix of better computer models, which have to deal with the chaotic behaviour of the atmosphere, and experienced humans who correct the models where they know them to be weak.
The chapter also explains the central quality of a good forecast. A forecaster is calibrated when the probabilities they state match how often things happen: on all the days with a forecast of 30 percent chance of rain, it should rain on about three in ten. Silver found that the US National Weather Service comes close to this.
Commercial forecasters often do not, because they have a “wet bias”: they forecast rain more often than it comes, since viewers forgive an unneeded umbrella more easily than a ruined picnic. The lesson reaches far beyond weather. A forecast that says “70 percent” is not wrong when the other 30 percent happens; it is wrong only if such events happen much more or much less often than it said.
Complex models often fit the noise (“overfitting”)
The chapters on earthquakes and economics show the opposite of the weather story: fields in which forecasts have hardly improved. Seismologists can say how often large earthquakes happen in a region, but not when the next one will come. A simple rule called the Gutenberg-Richter law says that each step up in magnitude makes a quake about ten times rarer, and Silver argues that the many attempts to beat this rule with more detailed models have failed.
His explanation is overfitting, which Slate’s reviewer quotes him describing as “the act of mistaking noise for signal”. A model overfits when it bends itself to every accident in past data, so that it looks perfect on the past and predicts the future badly. Silver argues that the models behind the safety standards of the Fukushima nuclear plant in Japan bent in this way and made a quake as large as the one of 2011 look almost impossible. For the mechanism itself, read What Is Overfitting?, which works through it with one clear example.
Economists face a related problem. The economy is a system in which causes are tangled, data are revised for years afterwards, and almost every variable is connected to every other. Silver reports that the ranges economists give for future growth missed the real outcome far more often than their stated confidence allowed, which means that their forecasts were badly calibrated, in the sense of the weather chapter.
Bayes’s theorem shows how to become “less and less wrong”
Chapter 8, “Less and Less and Less Wrong”, is the centre of the book. Silver presents Bayes’s theorem, a rule from the eighteenth century for updating a belief when new evidence arrives, as the method that ties his good forecasters together. You start with a prior, an honest estimate of how likely something is before the new evidence. Then you ask how likely the evidence would be if your idea were true, and how likely it would be if it were false. The theorem combines these numbers into a new estimate, which becomes the prior for the next piece of evidence.
Silver’s best-known example is domestic. Imagine you find a stranger’s underwear in your partner’s drawer. If you think the chance of an affair was low before, Silver works out that even this alarming discovery raises it only to about 29 percent, because there are innocent explanations too. The example shows why a strong-looking clue can leave the question open, and why your starting point matters so much. Bayes’ Theorem, Explained walks through the same rule with a medical test.
The chapter also attacks the statistical method that most sciences use. Silver blames the approach of the statistician Ronald Fisher, which tests whether a result is “significant” without asking how plausible it was in the first place, for the many published findings that later cannot be repeated.
Feedback and incentives decide who improves
Several chapters show that the fields where forecasts get better share two features. Forecasters there get quick, clear feedback, and they are rewarded for being right rather than for being interesting. Weather forecasters learn the next day whether they were right. Baseball analysts can check their player forecasts every season, and Silver argues that the best results came from combining statistics with the judgment of scouts, not from replacing scouts. Poker, his own former trade, teaches the same lesson in a hard way: a good player can lose for months, so you must judge your decisions by their quality and not only by their results.
Where these conditions are missing, forecasts stay poor. Fund managers are judged against each other, so following the crowd is safer for their careers than being right alone, and this herding helps bubbles grow. The chess chapter shows how people misread machines. In the 1997 match between IBM’s computer Deep Blue and the world champion Garry Kasparov, a strange move that Kasparov took as a sign of deep intelligence was, Silver reports, caused by a bug.
The biggest risks are the ones we fail to imagine
The last two chapters, on climate change and terrorism, deal with uncertainty that cannot be reduced to a neat probability. Silver argues that climate science is convincing above all because its basic physics is well understood, while precise forecasts of temperatures and sea levels decades ahead carry large uncertainty. The terrorism chapter compares the attacks of 11 September 2001 with Pearl Harbor. In both cases, Silver argues, the signals were there, but an attack of that kind seemed so unfamiliar that nobody gave it a serious probability. Terrorist attacks also follow a power law, a pattern in which rare, very large events cause most of the damage, so planning for the typical attack is not enough.
How it differs from other books on prediction
The Signal and the Noise sits between psychology books about error and statistics books about method. Three neighbours show its place.
- Superforecasting by Philip Tetlock and Dan Gardner (2015) comes from Tetlock’s later research, which followed the study Silver draws on in chapter 2. It is shorter, closer to the evidence and more practical, because it describes a forecasting tournament and the habits of its best participants. Silver’s book is wider: it covers fields Tetlock never studied and shows how the problems differ between them.
- The Black Swan by Nassim Nicholas Taleb (2007) argues that rare, extreme events matter most and that forecasting them is mostly hopeless. Silver takes these events seriously in his last chapters but stays more hopeful, because weather and baseball show that some forecasting really does improve.
- Thinking, Fast and Slow by Daniel Kahneman (2011) explains the mental shortcuts behind overconfidence. Silver describes the same overconfidence from the outside, through track records of real forecasts. Our review of Kahneman’s book covers the psychological side.
What makes Silver’s book different is that he writes as a practitioner who has built forecasting models himself and has been judged in public on their results. The Bayes and overfitting ideas also appear in Algorithms to Live By, but there they serve everyday decisions, while Silver applies them to public forecasts.
Where it falls short
The book has three weaknesses. The first is that it oversells Bayes. In a widely read piece in The New Yorker in January 2013, the cognitive scientist Gary Marcus and the computer scientist Ernest Davis argued that Silver’s picture of other statistical methods was one-sided. They also argued that he gave Bayesian reasoning credit for successes that may not have used Bayesian methods at all. Silver often uses “Bayesian” to mean a general attitude of updating, which is sound advice, but not proof that one school of statistics is right.
The second weakness is uneven chapters. The climate chapter drew sharp criticism from the climate scientist Michael Mann, who argued that Silver used the same statistical approach for climate trends as for voters’ opinions, although the two problems are very different. The chapter on terrorism, as the political scientist Andrew Hacker noted, says less about how agencies should handle far more warnings than they can follow up.
The third weakness is length and structure. Reviewers such as Matthew Yglesias at Slate found that the main argument sometimes gets lost in the reporting, and many examples come from the years before 2012, so the material on elections and on “Big Data” now feels dated. The core ideas, calibration, overfitting and updating, have aged much better than the examples.
Which format to choose
The paperback of 2015 is the best choice, because it adds a new preface in which Silver explains how much luck lay behind his perfect 2012 forecast. Charts carry a real part of the argument in the chapters on weather, earthquakes and economics, so print or an ebook on a large screen works better than a phone. The unabridged audiobook, read by Mike Chamberlain, lasts more than sixteen hours. It suits the reported chapters on poker, baseball and weather, but the chapters on Bayes and overfitting are easier to follow on the page. If time is short, read the introduction and chapters 2, 4, 5 and 8, which together hold the core of the argument.
Quiz
5 questions · up to 5 points
Open a question and pick one answer. Only your first answer counts: +1 if it is right, −1 if it is wrong.
What does Silver mean when he says that more information can make predictions worse?
Show answer
More data contains more random patterns, so it becomes easier to take noise for a real signal and to find support for what you already believe. Silver's 'signal' is the real pattern and 'noise' is everything random around it. More data adds both, and people are good at seeing patterns that are not there.
In the book's terms, why do 'foxes' make better forecasts than 'hedgehogs'?
Show answer
Foxes combine many small ideas and change their minds when the evidence changes, while hedgehogs force every event into one big theory. Silver takes the distinction from the psychologist Philip Tetlock, whose long study of expert predictions found that the flexible, cautious experts did better.
A weather service says '30 percent chance of rain' on many days. What would make it a well-calibrated forecaster in Silver's sense?
Show answer
On the days with that forecast, it rains on about 30 percent of them. Calibration means that stated probabilities match how often things really happen. Silver praises the US National Weather Service for coming close to this.
According to the chapter 'Less and Less and Less Wrong', how should you change your view after a surprising piece of evidence?
Show answer
Start from how likely the idea was before, then ask how much more likely the evidence is if the idea is true than if it is false, and adjust step by step. This is Bayes's theorem as Silver presents it: a prior, updated by new evidence, so that you become 'less and less wrong' over time.
What does this review say about how to read the book?
Show answer
Read it for its cases and its way of thinking, but treat its strong claims for Bayesian statistics with some caution. The review praises the reported chapters and the main habits of thought, and it reports that statisticians found the picture of other methods one-sided.
Your points are saved only in this browser for now. Accounts that keep them for good are on the way.
Continue learning: Bayes' Theorem, Explained →The takeaway
The Signal and the Noise by Nate Silver argues that predictions fail when people mistake 'noise', random patterns in data, for the 'signal', the real pattern underneath. Its other main ideas are that 'foxes' with many small ideas forecast better than 'hedgehogs' with one big idea, that forecasts should be stated as honest probabilities, and that Bayes's theorem shows how to update them. Its weakness is that it sells Bayes too hard and wanders between topics. When you hear a confident prediction, ask how sure the forecaster really is and how their past forecasts turned out.
Key quotes
“A 90 percent chance is not quite a sure thing.”
“Coming to a better understanding of data and statistics is essential to help us navigate our lives.”
Should you read it?
Read it if
- You want to understand Bayes's theorem, overfitting and probabilistic forecasting through real cases rather than formulas.
- You often hear confident predictions in the news or at work and want a way to tell the careful ones from the empty ones.
- You enjoy long reported chapters about different worlds, such as weather centres, baseball scouting and professional poker.
Skip it if
- You already know statistics well; the method is basic, and the chapters on Bayes give a one-sided picture of other schools.
- You want a short, tightly argued book; at almost 600 pages it moves slowly and wanders between topics.
- You want a practical training course in forecasting; Superforecasting by Philip Tetlock and Dan Gardner covers that better.
As an Amazon Associate, mentua.org earns from qualifying purchases. It costs you nothing and never shapes the verdict.
Ideas from this book
- Bayes' theoremBayes' theorem is a rule for updating a belief when new evidence arrives. You start with how common the thing was before the evidence. Then you weigh how much more often the evidence appears when the thing is true than when it is false. A rare condition stays fairly unlikely even after a good test is positive.
- OverfittingOverfitting happens when a model or a rule is fitted so closely to past cases that it copies their accidents along with the real pattern. It then looks excellent on the cases it was built from and predicts new cases badly. The risk is greatest when there are few cases and many ways to adjust the rule.
- Wisdom of the crowdThe wisdom of the crowd is the finding that the average of many independent guesses is often more accurate than most of the single guesses. It works because individual errors point in different directions and cancel out. It fails when people share the same bias or copy each other, because then their errors add up instead.
References
- The Signal and the Noise. Penguin Random House
The publisher's page with editions, page counts and the audiobook.
- Mostly I was getting credit for having pointed out the obvious. Nate Silver, Salon, 2015
An excerpt from the preface to the 2015 paperback, in which Silver explains why his perfect 2012 election forecast was partly luck.
- The Signal and the Noise, reviewed. Matthew Yglesias, Slate, 2012
A friendly review that explains the Bayesian and overfitting chapters and finds the main thesis partly lost in the stories.
- How He Got It Right. Andrew Hacker, The New York Review of Books, 2013
A political scientist reads Silver as a modest forecaster and questions the chapters on polls and terrorism.





