How to Kill 400 People and Save Them at the Same Time

Amos Tversky and Daniel Kahneman published the framing effect study in 1981, in a landmark paper called "The Framing of Decisions and the Psychology of Choice" in Science. By that point, the two had already produced a decade of research dismantling the rational-agent model of human judgment, but the framing paper struck even their colleagues as startling.

The rational-agent model holds that people have consistent preferences and that logically equivalent descriptions of a problem should produce the same choices. If you prefer saving 200 people to a one-third gamble on saving 600, you should make that same choice regardless of whether the options are described in terms of lives saved or deaths prevented. The numbers are the same; the outcomes are the same; only the presentation differs.

In their study, 72 percent of participants chose Program A (the certain option) when the problem was framed in terms of lives saved. When the identical problem was framed in terms of deaths, only 22 percent chose Program A. The majority switched to the gamble. Changing the label from "saved" to "die" reversed the majority preference.

Tversky and Kahneman ran the experiment on groups of physicians and statisticians, people who work with probabilities professionally. The framing effect did not disappear. It was slightly smaller in magnitude, but the majority still reversed their preferences when the frame changed. Knowing about decision theory does not protect you from the decision theory. Neither does being a doctor or a mathematician.

The Asian disease problem is the most cited demonstration of the framing effect, but it is far from the only one. Tversky and Kahneman showed that the effect generalizes across an astonishing range of domains.

Medical consent is one of the most consequential. A 1982 study by Barbara McNeil and colleagues presented patients, physicians, and graduate students with a choice between surgery and radiation therapy for lung cancer. When outcomes were described in terms of survival rates, more participants chose surgery. When the identical outcomes were described in terms of mortality rates, more participants chose radiation. The numbers were mathematically equivalent. The preference was not.

Consumer behavior is another. The same ground beef labeled "80% lean" is rated more favorably than ground beef labeled "20% fat." Bank accounts described as having a "monthly fee unless you maintain a minimum balance" attract fewer customers than accounts described as offering "a monthly bonus for maintaining a minimum balance," even when the financial terms are identical.

The framing effect works because human judgment is not absolute. We evaluate outcomes relative to a reference point, and whether something looks like a gain or a loss depends on where that reference point is set. "200 saved" is a gain relative to zero survivors. "400 dead" is a loss relative to 600 survivors. Losses feel more aversive than equivalent gains feel attractive, so frames that trigger loss thinking tend to push people toward risk-seeking behavior, and frames that trigger gain thinking push people toward risk-aversion.

What makes the framing effect particularly unsettling is that people typically do not notice it. In Tversky and Kahneman's studies, participants who were shown both versions of the problem often acknowledged that the programs were equivalent and felt embarrassed by their inconsistency, but reported being unable to do anything about it. The preference shift felt compelling even to people who could see, with the two frames in front of them, that it was irrational.

This is different from ordinary persuasion, where you are at least aware that someone is trying to influence you. With framing, you are unaware that the presentation is doing any work. The decision feels like a considered judgment about the options. It is partly a judgment about the words.

The practical implication is that whoever controls the frame controls, to a significant degree, the decision. Insurance sold as protection against catastrophic loss attracts different customers than insurance sold as a financial planning tool, even if the policy is identical. Salary negotiations framed as "what you would be giving up by not taking this offer" produce different outcomes than negotiations framed as "what you would be gaining." Public health messages framed around deaths prevented often land differently than the same messages framed around lives saved.

None of this is conscious manipulation on anyone's part, most of the time. Frames emerge naturally from how people talk and write. But they are not neutral.