Calling Bullshit: The Art of Skepticism in a Data-Driven World
Bergstrom and West explain how to identify misleading claims that borrow authority from numbers, graphs, and scientific language.
Carl Bergstrom and Jevin West examine misleading communication that gains credibility by looking quantitative, technical, or scientifically sophisticated to its audience.
A product advertised as twice as effective might improve an outcome from one case in a hundred to two. The relative comparison is real, but the absolute numbers change what the promise means for a buyer.
They argue that you often need careful questions about inputs and conclusions more than specialist knowledge of every calculation in between.
Numbers can be accurate while the message remains deceptive. Selective comparisons, distorted axes, missing denominators, and unrepresentative samples can all steer interpretation.
Correlation is a recurring concern. Two measurements moving together do not by themselves establish what caused what, or whether another factor shaped both.
The authors also explore selection effects: the process that puts people or events into a dataset can generate patterns that resemble substantive discoveries.
An algorithm predicting employee success from previous hires inherits the hiring process that produced its data. If some groups rarely had a chance to be hired, their absence is not evidence of lower ability.
Machine learning and large datasets do not escape these problems. A complicated model can amplify flawed labels, unsuitable measurements, or assumptions its audience cannot easily inspect.
Scientific publishing brings additional pressures, including incentives for striking results. A single study deserves attention to methods and uncertainty rather than automatic treatment as settled knowledge.
The book's skepticism is not hostility toward expertise. It aims to protect credible inquiry by distinguishing responsible evidence from impressive-looking persuasion that avoids responsibility.
Calling out a misleading claim should be accurate and proportionate. Explain the problem, consider innocent error, and avoid replacing someone else's exaggeration with your own.
The practical lesson is to examine how a number became a claim: who collected it, what it measures, what comparison matters, and what conclusion it can genuinely support.
A single study deserves attention to methods and uncertainty rather than automatic treatment as settled knowledge.
What standard of evidence would let you challenge a persuasive numerical claim without slipping into indiscriminate distrust of research?