The framing effect
The evidence is real and the argument about it is still running: how strong it is, how far it travels, or whether it repeats. Trust the direction, and hold the numbers loosely.
Amos Tversky and Daniel Kahneman published the demonstration in Science in 1981, in a paper called The Framing of Decisions and the Psychology of Choice. Participants read that a country was preparing for an outbreak of an unusual disease expected to kill six hundred people, and that two programmes had been proposed.
One group of 152 people saw the programmes described in terms of lives saved. If programme A is adopted, two hundred people will be saved. If programme B is adopted, there is a one third probability that six hundred people will be saved and a two thirds probability that nobody will be saved. Seventy-two percent chose A.
A different group of 155 people saw the same two programmes described in terms of lives lost. If programme C is adopted, four hundred people will die. If programme D is adopted, there is a one third probability that nobody will die and a two thirds probability that six hundred people will die. Seventy-eight percent chose D.
The two pairs are the same pair. Two hundred saved of six hundred is four hundred dead, and the gambles are identical. Each person saw one version only, so this is a comparison between two groups rather than a reversal inside anyone's head. Presented as gains, most people avoided the risk. Presented as losses, most people sought it. Tversky and Kahneman read this as a violation of the requirement that a choice should not depend on how the options are described, and it became the canonical example of what prospect theory predicts: people are risk averse when they see gains and risk seeking when they see losses, relative to whatever reference point the wording sets.
The effect is one of the more thoroughly checked findings in the field. Kühberger's 1998 meta-analysis pooled 136 studies and found a mean effect around d 0.31, moderate and consistent. The Many Labs 2 project replicated the disease problem across dozens of sites and found it held. Later work distinguishes three kinds of framing: risky choice framing, which is this problem; attribute framing, where 95 percent lean and 5 percent fat describe the same meat; and goal framing, where the same action is described by what you gain from doing it or lose from not.
The live argument is about what the result means. Since 2014 David Mandel and others have pointed out that the certain options are incompletely described. Two hundred will be saved does not say that four hundred will not be, and people may reasonably read it as at least two hundred, which would make the safe option genuinely better in the gain frame and genuinely worse in the loss frame, so that the switch is sensible rather than irrational. It is a serious objection. Data Colada tested it directly by adding the word exactly to both descriptions and found the effect largely survived. Later work finds the direction and size depend on whether the options are described completely, and that a residual effect remains even when they are. So the effect is real, and whether it counts as an error is still being argued.
Every number you put in front of someone is already framed, and the only question is who chose the frame.
A retention rate of ninety percent and a churn rate of ten percent describe the same customers. A proposal that will save the company two hundred thousand a year and a proposal that will stop it losing two hundred thousand a year describe the same proposal. A hire that succeeds seven times in ten and a hire that fails three times in ten is one hire. Each pair lands differently, and whoever wrote the slide picked one, whether they noticed picking or not.
The useful discipline runs in two directions. When you are the one presenting, decide the frame deliberately rather than by habit, and be honest with yourself about why. A gain frame makes people cautious. A loss frame makes them willing to take the risk. If you find yourself reaching for the loss frame every time you want approval for something bold, that is worth knowing about your own persuasion. When you are the one being presented to, do the arithmetic that turns one frame into the other before deciding. Ask what the same number looks like the other way round, and notice whether your preference survives the translation. If it does not, the frame was doing the choosing.
The mechanism sits close to two others in this library. Anchoring is about the first number in a conversation setting the range for everything after it. Loss aversion is the underlying asymmetry, that losses weigh more than equivalent gains, which is what makes the loss frame push people toward the gamble in the first place. Framing is the presentation layer that switches which of those asymmetries is in play.
Natural frequencies say most people, including doctors, cannot reason about a problem posed in percentages and can reason about the identical problem posed as counts. Framing says the same choice described as two hundred saved or four hundred dead flips the majority from the sure thing to the gamble. They look like the same finding and they are not, and the proof is that the framing study was already in counts. Asking for the count fixes the reasoning error: it exposes the denominator and what sits under the rate. It does not fix the frame, because a count can be a gain or a loss as easily as a percentage can. So the discipline is two questions in order. Out of how many, how many. Then, what does that same count look like the other way round.
Source: Tversky and Kahneman, The Framing of Decisions and the Psychology of Choice, Science, volume 211, 1981, pages 453 to 458, DOI 10.1126/science.7455683. The meta-analysis is Kühberger, The Influence of Framing on Risky Decisions: A Meta-analysis, Organizational Behavior and Human Decision Processes, volume 75, 1998, pages 23 to 55. The linguistic objection is Mandel, Do Framing Effects Reveal Irrational Choice?, Journal of Experimental Psychology: General, 2014, and the direct test of it is Data Colada on whether precise wording removes the effect.
Rory Sutherland, 2019
An advertising man arguing that the way something is presented is part of what it is worth, rather than decoration sitting on top of the real value. The same claim the disease problem makes with numbers, made across products, prices and cereal boxes.
Draw your own card. It does not take long, and it rewards taking your time.