Model torturing means repeatedly or selectively modifying a statistical model until it produces a preferred result. It is a form of statistical abuse: the final analysis may look statistically validated while hiding biased choices. section 1.4
When the answer is chosen first
A researcher may be pressured to use a favorable dataset to impress investors. Similar pressures appear when leaders want research supporting a cause or campaign, or when financial ties encourage reassuring claims about sugary drinks. These incentives can steer model choices toward a desired conclusion rather than a trustworthy one. section 1.4
The practical defense is critical thinking: look beyond statistical validation and ask whether confirmation bias, data choices, or model changes may have shaped the outcome. section 1.4
