Noise: A Flaw in Human Judgment

Kahneman, Sibony, and Sunstein define bias as predictable systematic deviation from a target—for example, a judge who is consistently harsher than average. Noise, by contrast, is unwanted variability in judgments that should be identical, such as two judges giving different sentences for the same crime, or the same judge giving different sentences on different days. The book argues that noise has been neglected because it can only be detected statistically, across sets of judgments, whereas bias can be identified in individual cases. This statistical invisibility makes noise counterintuitive and easy to ignore, even though its contribution to overall error can equal or exceed bias. The authors compare the two to the distinction between accuracy (the opposite of bias) and precision (the opposite of noise).

The book distinguishes three principal forms. Level noise refers to differences in average severity or leniency between judges—some judges are consistently harsh, others consistently lenient. Pattern noise refers to idiosyncratic responses to particular cases, where a judge's reaction depends on features of the case that are irrelevant to the standard of judgment. Occasion noise refers to variability within the same person over time, driven by mood, fatigue, hunger, or other transient factors. The authors use the example of criminal sentencing to illustrate level and pattern noise: defendants receive different sentences partly because of which judge they draw (level) and partly because of how that judge responds to specific case features (pattern). Occasion noise is illustrated by the finding that parole decisions vary with the time of day.

Noise extends the research program Kahneman began with Amos Tversky, but shifts emphasis from bias to variability. The earlier work documented systematic cognitive biases—representativeness, availability, anchoring—that affect judgment in predictable directions. Noise focuses on a different component of error: variability in judgments that should be uniform. The authors explicitly position the book as a complement to the bias literature, arguing that reducing error requires attending to both components. In an interview, Kahneman noted that while bias is a psychological process detectable in individual judgment, noise is a statistical concept that only becomes visible when examining patterns across many judgments. The book thus applies the probabilistic, statistical mindset of the earlier work to organizational and institutional decision-making.

Decision hygiene is the book's prescriptive framework for reducing noise in judgment. Its core principle is that intuition should be disciplined and delayed, not eliminated. The recommended practices include: obtaining independent judgments from multiple people before any discussion, so that opinions are not contaminated by social influence; aggregating those independent judgments, since averaging uncorrelated errors reduces variance; structuring decision processes with clear criteria and standardized scales; and using simple rules or algorithms where judgment adds no value over mechanical combination. The authors also recommend conducting 'noise audits' to measure variability in existing systems. They emphasize that these remedies need not resolve every disagreement—they aim to reduce unwanted variability while preserving legitimate diversity of opinion where it is desirable.

Statisticians Andrew Gelman and colleagues published a detailed critique arguing that Noise contains a fundamental incoherence: it argues that judgments are noisy and unpredictable while simultaneously citing very small studies (some with as few as seven or eight participants) as if they yielded universal truths . They point to specific examples, including a study of software developers' time estimates and research on hypoglycemia and judicial decisions. The critique contends that if judgments are genuinely noisy, drawing general conclusions from tiny samples is methodologically inconsistent. The authors have not published a formal rebuttal to this specific critique, but the book's broader argument acknowledges that noise can be measured and reduced only through systematic, aggregated data rather than individual case analysis.

The book was a New York Times bestseller and received substantial coverage in both academic and popular venues . A British Psychological Society review described it as a 'seminal work' that highlights the ubiquity of noise while noting that the early statistical chapters are 'unlikely to grip the lay reader' . New York Times reviewer Steve Brill praised it as a 'welcome handbook for making life's lottery a lot more coherent,' connecting its message to contemporary concerns about institutional trust and polarization . The LSE Review of Books called it a 'novel book' that would help readers understand decision-making processes better, though it also observed that the book's length and detail may test general readers' patience . Overall reception has been positive but more mixed than for Thinking, Fast and Slow.