Chapter 19 Further Reading: Probabilistic Forecasting and Uncertainty

Foundations of Probabilistic Thinking

Silver, Nate. The Signal and the Noise: Why So Many Predictions Fail — But Some Don't. Penguin Press, 2012. The book that brought probabilistic forecasting to a general audience, with strong chapters on calibration, the Bayesian mindset, and why most predictions fail by being overconfident. Silver's discussion of what it means to "be right" probabilistically remains the clearest introduction for general readers.

Taleb, Nassim Nicholas. The Black Swan: The Impact of the Highly Improbable. Random House, 2007. Taleb's provocative argument that rare, high-impact events are systematically underestimated by forecasting models. Particularly relevant to correlated errors and the limitations of models estimated on historical data that may not capture tail risks. Read critically — Taleb's prescriptions are more controversial than his diagnosis.

Tetlock, Philip E., and Dan Gardner. Superforecasting: The Art and Science of Prediction. Crown, 2015. Based on Tetlock's decades of research on expert prediction, this book examines what distinguishes superforecasters — people who make strikingly accurate probabilistic predictions across many domains. The lessons about epistemic humility, calibration, and updating on evidence are directly applicable to election forecasting.

On Election Forecasting

Gelman, Andrew, and Jennifer Hill. Data Analysis Using Regression and Multilevel/Hierarchical Models. Cambridge University Press, 2006. The technical foundation for many election forecasting approaches, including the multilevel regression models used in MRP forecasting. Not light reading, but the definitive reference for analysts who want to build their own models.

Linzer, Drew A. "Dynamic Bayesian Forecasting of Presidential Elections in the States." Journal of the American Statistical Association 108 (2013): 124-134. One of the most rigorous academic treatments of Bayesian probabilistic election forecasting, demonstrating how polling data can be integrated into a dynamic model that updates continuously as Election Day approaches. Technically demanding but extremely relevant.

Lock, Kari, and Andrew Gelman. "Bayesian Combination of State Polls and Election Forecasts." Political Analysis 18 (2010): 337-348. A foundational paper on combining state-level polls into national probabilistic forecasts, addressing the correlated error problem directly.

On Correlated Errors

Shirani-Mehr, Houshmand, David Rothschild, Sharad Goel, and Andrew Gelman. "Disentangling Bias and Variance in Election Polls." Journal of the American Statistical Association 113 (2018): 607-614. The most rigorous empirical analysis of correlated polling errors, finding that bias (systematic error affecting multiple pollsters simultaneously) is often more important than variance (random error). Essential reading for understanding why the 2016 result was less surprising than many models implied.

Kennedy, Courtney, et al. "An Evaluation of the 2016 Election Polls in the United States." Public Opinion Quarterly 82 (2018): 1-33. The American Association for Public Opinion Research's official post-mortem on 2016 polling errors, including analysis of which states were most wrong and why. Identifies the role of non-response bias (better-educated voters were more likely to respond to polls) as a significant contributing factor.

On Calibration

Roulston, Mark S., and Leonard A. Smith. "Evaluating Probabilistic Forecasts Using Information Theory." Monthly Weather Review 130 (2002): 1653-1660. From meteorology, where probabilistic forecasting is most mature, a rigorous treatment of how to evaluate probabilistic forecast quality including both calibration and resolution.

Gneiting, Tilmann, and Adrian E. Raftery. "Strictly Proper Scoring Rules, Prediction, and Estimation." Journal of the American Statistical Association 102 (2007): 359-378. The technical foundations for scoring probabilistic forecasts, explaining why the Brier score and logarithmic scoring rules are "proper" — they incentivize honest probability reporting rather than strategic under- or over-confidence.

On Communicating Uncertainty

Westwood, Sean J., Solomon Messing, and Yphtach Lelkes. "Projecting Confidence: How the Probabilistic Horse Race Confuses and Demobilizes the Public." Journal of Politics 82 (2020): 1530-1544. Experimental evidence that probabilistic win probabilities reduce turnout intention compared to margin-based presentations, raising important questions about how forecasters should communicate probability. The findings are contested and don't obviously translate into a policy prescription, but they're essential context.

Ibrekk, Halvor, and Morgan G. Morgan. "Graphical Communication of Uncertain Quantities to Nontechnical People." Risk Analysis 7 (1987): 519-529. Classic work on how visual representations of uncertainty affect non-expert understanding, from the risk communication literature. The insights transfer directly to election forecast visualization.

Fischhoff, Baruch, and Wändi Bruine de Bruin. "Fifty-Fifty = 50%?" Journal of Behavioral Decision Making 12 (1999): 149-163. Research on how people interpret probabilistic statements, finding systematic misinterpretations. Directly relevant to the challenge of communicating election probabilities to mass audiences.

Accessible Online Resources

Andrew Gelman's blog (statmodeling.stat.columbia.edu) Gelman and collaborators regularly discuss election forecasting, calibration, and uncertainty communication with a level of statistical rigor that is rare in public discourse. The 2016 election-related posts are particularly valuable.

538's methodology documents (fivethirtyeight.com) FiveThirtyEight publishes detailed methodology articles for their election models. Reading these alongside the popular coverage reveals the gap between technical model and public communication — itself a valuable lesson.

The Monkey Cage (Washington Post, monkeycage.org) Political scientists, including Sides, Vavreck, Gelman, and many others, write accessible pieces on election forecasting, polling, and uncertainty. The accumulated archive covering 2016-2024 is a valuable resource.

Probably Overthinking It (blog by Allen Downey) Downey applies probabilistic thinking to elections and many other domains, with readable and technically sound analysis. Particularly strong on Monte Carlo methods and Bayesian updating.