The Signal and the Noise

The Signal and the Noise Plot Summary

Introduction

Prediction shapes nearly every aspect of our lives, from personal decisions to global policies, yet our ability to forecast the future remains remarkably limited. Despite unprecedented access to information in the digital age, we frequently witness spectacular failures of prediction across domains - from economic crashes to political upheavals, from disease outbreaks to climate events. This paradox forms the central investigation: why does more information often lead to worse predictions rather than better ones? At the heart of this challenge lies the fundamental distinction between signal and noise - between meaningful patterns that reflect underlying truths and random fluctuations that distract and mislead. Mastering prediction requires not just technical expertise but a particular mindset: one that embraces uncertainty rather than denying it, updates beliefs systematically rather than haphazardly, and maintains intellectual humility in the face of complexity. By examining both failures and successes across diverse fields, we discover a unified framework for thinking probabilistically about an uncertain world - an approach that helps us distinguish genuine insights from misleading patterns and make better decisions even when perfect prediction remains beyond our reach.

Chapter 1: The Prediction Paradox: Why More Information Doesn't Equal Better Forecasts

The modern world presents us with a striking paradox: despite unprecedented access to information, our predictions frequently fail spectacularly. The 2008 financial crisis caught most economists by surprise despite warning signs visible years in advance. Political pundits confidently predict election outcomes that prove wildly inaccurate. Intelligence agencies miss critical security threats despite vast surveillance capabilities. This pattern suggests that prediction failures stem not from insufficient information but from how we process that information. The explosion of data in the digital age has paradoxically made prediction more difficult by increasing the amount of noise we must filter. Noise represents random patterns that distract us from meaningful signals - the underlying truths we seek to discover. With more data comes more potential for spurious correlations, misleading patterns, and information overload. Financial analysts examining thousands of economic indicators inevitably find some that appear predictive purely by chance. News media reporting every market fluctuation creates an illusion of meaningful patterns in what is largely random movement. This noise drowns out genuine signals, leading to worse predictions despite more information. Our cognitive architecture compounds this problem through various biases that distort prediction. We naturally seek patterns even in random data, a tendency that served our ancestors well for quick decision-making but undermines careful forecasting. We overvalue information that confirms our existing beliefs while discounting contradictory evidence. We construct narratives to explain past events, creating an illusion of predictability that rarely translates to accurate forecasts. These biases lead us to mistake noise for signal, seeing meaningful patterns where none exist while missing genuine indicators of future developments. Institutional factors further undermine prediction quality. Organizations often reward confidence over accuracy, creating incentives for bold but unreliable forecasts. Experts face professional penalties for acknowledging uncertainty, while those making definitive but wrong predictions rarely suffer lasting consequences. Media outlets prefer simple narratives and confident pronouncements to nuanced probabilistic statements. These structural factors encourage precisely the wrong approach to prediction - overconfidence in the face of inherent uncertainty. The prediction paradox reveals a fundamental truth: better prediction requires not just more information but better thinking. Successful forecasters distinguish themselves not through access to exclusive data but through how they process information. They think probabilistically rather than deterministically, expressing forecasts as ranges of possibilities with associated likelihoods. They update their beliefs systematically as new evidence emerges rather than clinging to initial judgments. Most importantly, they maintain intellectual humility, recognizing the inherent limitations of prediction in complex systems. This paradox ultimately points toward a different approach to prediction - one that embraces uncertainty rather than denying it. Instead of seeking illusory certainty, effective forecasters focus on quantifying uncertainty more precisely. They recognize that being approximately right is better than being precisely wrong, and that acknowledging what we don't know is as important as asserting what we do. This mindset transforms prediction from a binary exercise (right or wrong) into a continuous process of refining our understanding of possibilities.

Chapter 2: Distinguishing Signal from Noise: The Core Challenge of Prediction

The fundamental challenge in prediction lies in distinguishing meaningful signals from random noise. Signals represent genuine patterns that help us understand underlying reality, while noise consists of random fluctuations that create the illusion of patterns where none truly exist. This distinction proves extraordinarily difficult in practice, as human cognition naturally seeks patterns even in random data, leading us to detect "signals" that have no predictive value. The signal-to-noise ratio varies dramatically across different domains, influencing prediction difficulty. Weather forecasting benefits from strong physical signals governed by atmospheric laws, allowing relatively accurate short-term predictions despite complexity. Financial markets, by contrast, contain weak signals buried within enormous noise, as prices reflect countless factors including random fluctuations and self-reinforcing behaviors. Political prediction occupies a middle ground, with genuine signals from demographic trends and economic conditions obscured by media narratives and short-term events. Understanding the inherent signal-to-noise ratio in different domains helps calibrate appropriate confidence in predictions. Scale plays a crucial role in signal detection. Patterns that appear meaningful at one scale often disappear at another. Daily stock market movements consist almost entirely of noise, while long-term market trends reflect more meaningful economic signals. Similarly, local weather fluctuations may seem random, but global climate patterns reveal clear signals when viewed over decades. Choosing the appropriate time scale for analysis proves essential for distinguishing signal from noise. Many prediction failures stem from applying short-term patterns to long-term forecasts or vice versa. The scientific method provides our most powerful framework for signal detection, using controlled experiments, statistical analysis, and theoretical understanding to separate genuine patterns from random fluctuations. However, even science faces significant challenges in noisy domains where controlled experiments prove impossible. Climate science, economics, and epidemiology all confront this difficulty, relying on natural experiments, statistical controls, and theoretical models to extract signals from observational data. These fields demonstrate that signal detection remains possible even without perfect experimental conditions, though it requires greater caution and methodological sophistication. Technological advances have dramatically improved our signal detection capabilities while simultaneously increasing noise. Satellite observations provide unprecedented data about weather patterns, enabling better forecasts. Genetic sequencing reveals signals about disease risk previously invisible to medicine. Yet these same technologies generate vast quantities of data that must be filtered and interpreted, creating new opportunities for mistaking noise for signal. The most successful prediction systems combine technological tools with human judgment, using algorithms to process data while relying on human expertise to interpret results within broader contexts. Ultimately, distinguishing signal from noise requires both statistical rigor and domain expertise. Statistical methods help identify patterns that exceed random variation, while domain knowledge provides context for interpreting these patterns. Neither approach alone suffices. Statisticians without domain expertise may identify correlations without understanding causation, while experts without statistical training may see patterns in random data. The most successful forecasters combine both perspectives, using statistical tools to test hypotheses derived from theoretical understanding while remaining vigilant against the human tendency to detect false patterns.

Chapter 3: Bayesian Thinking: A Superior Framework for Handling Uncertainty

Bayesian thinking provides a powerful framework for prediction under uncertainty, fundamentally different from conventional statistical approaches. Named after Thomas Bayes, an 18th-century mathematician and minister, this approach treats probability not as a fixed frequency of outcomes but as a degree of belief that evolves with new evidence. This perspective transforms prediction from a static judgment into a dynamic process of continuous learning and refinement. The core of Bayesian reasoning lies in updating prior beliefs with new evidence to form posterior beliefs. We begin with initial probabilities based on existing knowledge, then systematically revise these probabilities as new information emerges. This process mirrors how humans naturally learn, but Bayes's theorem provides a mathematical framework that quantifies this intuitive process. The formula calculates exactly how much to adjust our beliefs given the strength of new evidence and our prior confidence. This approach prevents both excessive conservatism (sticking too firmly to initial beliefs) and excessive reactivity (overweighting new evidence). Medical diagnosis illustrates Bayesian thinking in action. A doctor begins with base rates of disease in the population (prior probability), then updates this assessment based on test results (new evidence). If a test for a rare disease returns positive, the Bayesian approach prevents overreaction by considering both the test's accuracy and the disease's rarity. This prevents the common error of ignoring base rates, which leads to unnecessary treatments and anxiety when screening for uncommon conditions. Similar principles apply in security screening, financial risk assessment, and other domains where false positives create significant costs. Bayesian thinking excels in environments characterized by limited data and evolving knowledge. Traditional statistical methods often require large, representative samples and fixed hypotheses - conditions rarely met in real-world prediction challenges. Bayesian methods, by contrast, incorporate prior knowledge when data is sparse and continuously update as information accumulates. This makes them particularly valuable in dynamic situations like pandemic response, where initial data is limited but rapidly evolving. The approach allows forecasters to start with imperfect information and systematically improve predictions as evidence accumulates. The Bayesian framework also explicitly acknowledges subjective elements in prediction. While traditional statistics often claims complete objectivity, Bayesian methods recognize that prior beliefs inevitably influence interpretation of evidence. Rather than pretending these subjective elements don't exist, Bayesian thinking brings them into the open where they can be examined and refined. This transparency helps prevent hidden biases from distorting conclusions. By making assumptions explicit, Bayesian methods allow for more honest evaluation of prediction quality. Perhaps most importantly, Bayesian thinking embraces uncertainty rather than hiding from it. Instead of making binary predictions (will happen/won't happen), it expresses forecasts as probability distributions that reflect confidence levels. This nuanced approach prevents overconfidence in predictions and helps decision-makers understand the range of possible outcomes. By acknowledging what we don't know alongside what we do, Bayesian methods produce more honest and ultimately more useful forecasts. This humility about prediction limitations paradoxically leads to more reliable results over time.

Chapter 4: Why Predictions Fail: Cognitive Biases and Institutional Blindspots

Prediction failures rarely stem from simple ignorance or random error. Instead, they typically result from systematic biases that distort our judgment in predictable ways. These biases operate at both individual and institutional levels, creating persistent blindspots that undermine forecasting across domains from economics to politics, from weather prediction to intelligence analysis. Overconfidence represents perhaps the most pervasive bias in prediction. Study after study demonstrates that experts express far more certainty than their track records justify. Economic forecasters routinely provide precise GDP projections without acknowledging the enormous uncertainty involved. Political analysts confidently predict election outcomes based on minimal evidence. This excessive certainty stems partly from incentive structures that reward bold predictions over accurate ones. Media outlets prefer definitive statements to nuanced probabilities, and clients prefer consultants who project confidence rather than acknowledge uncertainty. This creates a prediction environment where being wrong but confident often proves less professionally damaging than being uncertain but accurate. Confirmation bias further undermines prediction by causing us to seek and interpret evidence selectively. Once we form a hypothesis - whether about market trends, political outcomes, or scientific theories - we naturally gravitate toward information that supports it while discounting contradictory evidence. This tendency creates self-reinforcing feedback loops where initial beliefs become increasingly entrenched regardless of their accuracy. Financial analysts who predicted market growth before 2008 continued finding "positive indicators" even as warning signs multiplied. Intelligence analysts similarly interpreted ambiguous information about Iraqi weapons programs through the lens of preexisting beliefs, missing contradictory signals. The narrative fallacy leads us to construct coherent stories that explain past events, creating an illusion of predictability. These retrospective narratives oversimplify complex phenomena by emphasizing certain factors while ignoring others, particularly random variation. When these simplified models are applied to future prediction, they inevitably fail to capture the full complexity of real-world systems. The financial crisis exemplifies this problem - post-crisis narratives identified "obvious" warning signs that seemed clear only in hindsight. This fallacy creates an illusion of understanding that masks the inherent unpredictability of complex systems. Institutional structures further distort prediction through specialized departments, bureaucratic boundaries, and organizational incentives that impede holistic analysis. Intelligence about potential terrorist attacks before 9/11 was fragmented across agencies, with each possessing partial information but none seeing the complete picture. Financial risk assessment similarly suffered from departmental silos, with mortgage originators, rating agencies, and investment banks each seeing only part of the systemic risk building in housing markets. These structural blindspots prevent organizations from integrating diverse information sources into coherent predictions. Group dynamics exacerbate prediction failures through phenomena like groupthink and herding behavior. Experts within fields develop shared assumptions that rarely face serious challenge. Dissenting views face social and professional penalties, creating powerful incentives for conformity. This dynamic explains why entire professional communities - from intelligence agencies to financial institutions - can simultaneously miss critical developments that outsiders sometimes identify. The collective intelligence that should improve prediction instead produces collective blindness when group dynamics suppress diverse perspectives and critical thinking. Addressing these biases requires both individual awareness and institutional reform. At the individual level, forecasters must recognize their own cognitive limitations and adopt practices like explicit probability estimates, consideration of alternative hypotheses, and systematic tracking of prediction accuracy. At the institutional level, organizations need structures that reward accuracy over confidence, encourage diverse perspectives, and facilitate information sharing across traditional boundaries. Without addressing both levels simultaneously, prediction failures will continue despite advances in data collection and analytical methods.

Chapter 5: Learning from Success: Case Studies in Effective Forecasting

While prediction failures receive considerable attention, examining successful forecasts provides equally valuable insights. These success stories reveal common principles that transcend specific domains, offering a blueprint for more effective prediction across fields. Though perfect prediction remains impossible in complex systems, these cases demonstrate that significant improvement is achievable through specific approaches and attitudes. Weather forecasting represents one of prediction's greatest success stories. Fifty years ago, a five-day forecast was less accurate than today's ten-day forecast. This remarkable improvement stems from three factors: massive data collection through satellites and sensors; sophisticated computer models that simulate atmospheric physics; and continuous verification against actual outcomes. Weather services meticulously track forecast accuracy, using this feedback to refine their models and methods. This commitment to verification distinguishes weather forecasting from domains where predictions face little systematic evaluation. The field also excels at communicating uncertainty through probability estimates and visualization tools like the "cone of uncertainty" in hurricane forecasting. Certain political forecasters consistently outperform their peers by adopting what psychologist Philip Tetlock calls a "foxlike" approach - drawing on multiple information sources and theoretical perspectives rather than relying on a single framework. These successful forecasters express predictions probabilistically, continuously update their assessments as new information emerges, and maintain intellectual flexibility. They avoid the overconfidence that plagues "hedgehog" forecasters who view events through a single ideological lens. This approach proves particularly valuable in domains characterized by limited data and complex causality. The Good Judgment Project demonstrated that ordinary people trained in probabilistic thinking and cognitive debiasing could outperform intelligence professionals in geopolitical forecasting. In epidemiology, successful disease outbreak predictions typically combine statistical models with deep domain knowledge. During the H1N1 influenza pandemic, the most accurate forecasts came from teams that integrated mathematical modeling with understanding of transmission mechanisms, population behavior, and historical patterns from previous outbreaks. These forecasters recognized both the value and limitations of their models, supplementing quantitative projections with qualitative insights about factors models couldn't capture. They also updated predictions frequently as new data emerged, treating forecasting as a continuous process rather than a one-time judgment. Financial prediction presents perhaps the greatest challenge, yet even here, certain approaches yield better results. The most successful investors typically avoid short-term market prediction altogether, focusing instead on fundamental value assessment. They maintain emotional discipline during market volatility, resist herding behavior, and exploit the predictable irrationality of other market participants. Rather than attempting to predict precise market movements, they identify situations where risk-reward ratios favor particular positions. This approach acknowledges the inherent unpredictability of markets while still extracting valuable signals from the noise. Across these diverse domains, successful prediction shares common characteristics. First, effective forecasters maintain appropriate humility about prediction limitations. They express forecasts as probability distributions rather than point estimates, acknowledging inherent uncertainty. Second, they systematically track prediction accuracy, learning from both successes and failures. Third, they combine quantitative methods with qualitative judgment, recognizing that neither approach alone suffices for complex prediction challenges. Finally, they view prediction as an iterative process rather than a one-time judgment, continuously updating forecasts as new information emerges.

Chapter 6: The Unknown Unknowns: Navigating Fundamental Uncertainty

The concept of "unknown unknowns" - factors we don't even know we should consider - represents one of prediction's most profound challenges. Former Defense Secretary Donald Rumsfeld popularized this term, distinguishing between "known unknowns" (recognized uncertainties) and "unknown unknowns" (uncertainties we fail to anticipate). This distinction helps explain why even sophisticated prediction methods frequently fail when confronting novel situations or complex systems. Unknown unknowns emerge from system complexity and interconnectedness. In financial markets, for instance, new instruments like collateralized debt obligations created unforeseen risk correlations that traditional models couldn't capture. These instruments transformed housing market fluctuations into systemic financial instability through mechanisms that experts failed to anticipate. The problem wasn't merely underestimating known risks but failing to recognize entire risk categories - a fundamental blind spot in prediction methodology. Similar dynamics appear in climate systems, where feedback loops and tipping points create potential for sudden, nonlinear changes that models struggle to anticipate. Cognitive limitations contribute significantly to unknown unknowns. Human imagination naturally draws from past experience, making truly novel threats difficult to conceptualize. Before September 11, 2001, intelligence agencies collected numerous signals suggesting potential terrorist attacks, but lacked a conceptual framework for interpreting these signals coherently. The possibility of using commercial aircraft as weapons existed in their data but not in their mental models, illustrating how cognitive frameworks determine which possibilities we consider. This limitation affects all prediction domains, from technological innovation to geopolitical developments, where genuinely novel events defy imagination until they occur. Historical analysis reveals that unknown unknowns often emerge from system transitions - periods when established patterns break down and new dynamics emerge. The 2008 financial crisis occurred partly because models based on historical housing price data couldn't account for structural changes in mortgage markets. Climate change similarly creates prediction challenges by pushing environmental systems beyond historical parameters. These transition periods render historical data less relevant for future prediction, undermining the foundation of most forecasting methods. The accelerating pace of technological and social change makes such transitions increasingly common, expanding the territory of unknown unknowns. Addressing unknown unknowns requires fundamental changes in prediction approaches. First, prediction methods must incorporate scenario planning that explicitly considers novel possibilities outside historical experience. This approach doesn't predict specific outcomes but prepares for diverse contingencies, reducing vulnerability to surprises. Second, organizations need processes that challenge conventional wisdom and surface alternative perspectives. Red teams tasked with identifying vulnerabilities, prediction tournaments that aggregate diverse viewpoints, and cross-disciplinary collaboration all help identify blind spots in prediction frameworks. These methods won't eliminate unknown unknowns but can expand awareness of possibilities beyond conventional thinking. Perhaps most importantly, effective prediction requires intellectual humility - recognizing the limits of our knowledge and the inevitability of surprise. This doesn't mean abandoning prediction entirely, but rather approaching it with appropriate caution. Expressing predictions probabilistically, maintaining flexibility when new information emerges, and designing systems resilient to unexpected developments all help manage the inevitable surprises that unknown unknowns create. This humility paradoxically improves prediction quality by preventing overconfidence and encouraging continuous refinement of forecasting methods.

Chapter 7: Probabilistic Forecasting: Building Models That Acknowledge Uncertainty

Effective prediction requires moving beyond deterministic forecasts toward probabilistic models that express uncertainty explicitly. Traditional forecasting often produces single-point estimates - a specific GDP growth rate, election outcome, or hurricane path - that create an illusion of precision. Probabilistic forecasting, by contrast, generates ranges of possible outcomes with associated likelihoods, providing a more honest and ultimately more useful picture of future possibilities. The foundation of probabilistic forecasting lies in quantifying uncertainty sources. Some uncertainty stems from incomplete information about current conditions - we lack perfect knowledge about present economic indicators, disease prevalence, or atmospheric states. Additional uncertainty emerges from model limitations - even perfect information about present conditions wouldn't yield perfect predictions because our models simplify complex reality. Finally, fundamental randomness in many systems creates irreducible uncertainty that no model can eliminate. Effective probabilistic forecasts account for all three uncertainty sources, distinguishing between what might be known with better information and what remains inherently unpredictable. Monte Carlo simulation provides a powerful technique for building probabilistic forecasts. Rather than calculating a single outcome, this approach runs thousands of simulations with slightly varied inputs and model parameters, generating a distribution of possible outcomes. This distribution reveals not just the most likely result but the full range of possibilities and their relative probabilities. Financial risk models, climate projections, and epidemiological forecasts increasingly rely on this approach to express uncertainty meaningfully. The resulting probability distributions help decision-makers understand both central tendencies and tail risks - the low-probability but high-impact events that often drive system behavior. Ensemble methods further enhance probabilistic forecasting by combining multiple models rather than relying on a single approach. Weather forecasting exemplifies this technique, running various models with different assumptions and physics representations, then aggregating their predictions. This approach outperforms individual models because different models capture different aspects of reality, and their collective wisdom exceeds any single perspective. Similar ensemble approaches improve economic forecasting, election prediction, and disease outbreak modeling. The diversity of models helps capture different uncertainty sources and reduces the impact of individual model biases. Communicating probabilistic forecasts effectively presents significant challenges. Human minds naturally seek certainty, making probabilistic information cognitively demanding. Visualization techniques help overcome this limitation by representing uncertainty graphically through confidence intervals, fan charts, or probability distributions. The "cone of uncertainty" in hurricane forecasting exemplifies effective uncertainty visualization, showing the range of possible storm paths while emphasizing increasing uncertainty over time. Similar approaches help communicate economic forecasts, climate projections, and other probabilistic predictions in ways that maintain nuance while remaining accessible to non-specialists. Decision-making under probabilistic forecasts requires shifting from "what will happen?" to "what actions perform well across multiple scenarios?" Rather than optimizing for a single predicted outcome, robust decision-making identifies strategies that perform adequately across the full range of possibilities. This approach acknowledges forecast limitations while still extracting actionable insights from probabilistic information. Climate adaptation planning, financial risk management, and pandemic response all benefit from this robust approach to decision-making under uncertainty, focusing on resilience rather than optimization for a single scenario. The ultimate value of probabilistic forecasting lies not in eliminating uncertainty but in managing it effectively. By quantifying what we know and don't know, these methods prevent both overconfidence in favorable outcomes and paralysis from fear of worst-case scenarios. They transform prediction from a binary exercise (right or wrong) into a nuanced assessment of possibilities that supports better decision-making under inevitable uncertainty. This approach aligns prediction methods with the fundamental nature of complex systems, where perfect foresight remains impossible but improved understanding of possibilities and probabilities enables more effective navigation of an uncertain future.

Summary

The art of prediction ultimately rests on distinguishing meaningful signals from random noise - a challenge that requires both technical skill and intellectual humility. Across domains from weather forecasting to financial markets, from political elections to disease outbreaks, successful prediction depends not on eliminating uncertainty but on quantifying it accurately. The most reliable forecasters understand that certainty is rarely achievable and instead focus on expressing degrees of belief that evolve with new evidence. They think probabilistically, update their views systematically through Bayesian reasoning, and maintain appropriate humility about the limits of prediction in complex systems. This approach transforms prediction from a static judgment into a dynamic process of continuous learning and refinement. Rather than claiming perfect foresight, effective forecasters acknowledge what they don't know alongside what they do, express predictions as probability distributions rather than point estimates, and systematically track their accuracy to improve over time. They combine quantitative models with qualitative judgment, recognizing that neither approach alone suffices for complex prediction challenges. Most importantly, they understand that prediction serves decision-making - that the goal isn't perfect foresight but better choices under inevitable uncertainty. By embracing this perspective, we can navigate an unpredictable world more effectively, making better decisions even when the future refuses to reveal itself completely.

About Author

Loading...

Nate Silver

Nathaniel Read “Nate” Silver (born January 13, 1978) is an American statistician and writer who analyzes baseball and elections. He is currently the editor-in-chief of ESPN’s FiveThirtyEight blog and a Special Correspondent for ABC News. Silver first gained public recognition for developing PECOTA, a system for forecasting the performance and career development of Major League Baseball players, which he sold to and then managed for Baseball Prospectus from 2003 to 2009.In 2007, writing under the pseudonym “Poblano”, Silver began to publish analyses and predictions related to the 2008 United States presidential election. At first this work appeared on the political blog Daily Kos, but in March 2008 Silver established his own website, FiveThirtyEight.com. By summer of that year, after he revealed his identity to his readers, he began to appear as an electoral and political analyst in national print, online, and cable news media.The accuracy of his November 2008 presidential election predictions—he correctly predicted the winner of 49 of the 50 states—won Silver further attention and commendation. The only state he missed was Indiana, which went for Barack Obama by one percentage point. He correctly predicted the winner of all 35 U.S. Senate races that year.In April 2009, he was named one of The World’s 100 Most Influential People by Time.In 2010, Silver’s FiveThirtyEight. blog was licensed for publication by The New York Times. The newly renamed blog, FiveThirtyEight: Nate Silver’s Political Calculus, first appeared in The Times on August 25, 2010. In 2012 and 2013, FiveThirtyEight won Webby Awards as the “Best Political Blog” from the International Academy of Digital Arts and Sciences.Silver’s book, The Signal and the Noise , was published in September 2012. It subsequently reached The New York Times best seller list for nonfiction, and was named by Amazon.com as the #1 best nonfiction book of 2012. The Signal and the Noise won the 2013 Phi Beta Kappa Award in Science. The book has been published in eight languages.In the 2012 United States presidential election between Barack Obama and Mitt Romney, he correctly predicted the winner of all 50 states and the District of Columbia. That same year, Silver’s predictions of U.S. Senate races were correct in 31 of 33 states; he predicted Republican victory in North Dakota and Montana, where Democrats won.In July 2013, it was revealed that Silver and his FiveThirtyEight blog would depart The New York Times and join ESPN. In his new role at ESPN, Silver would become editor-in-chief of the FiveThirtyEight site. ESPN would own the FiveThirtyEight site and the brand. The ESPN-owned FiveThirtyEight launched on March 17, 2014. Silver’s lead article explained that the site would focus on a broad range of subjects under the general rubric of “data journalism”.

Read more