Framing

Why can two entirely accurate headlines about the same event leave opposite impressions?

Two headlines. "Ninety per cent of patients survive the operation." "One in ten patients dies during the operation." Neither is inaccurate, neither omits anything the other includes, and a moment's arithmetic shows they are the same sentence. Yet readers of the first consent to surgery at higher rates than readers of the second, and this holds for doctors as well as patients. If the information is identical, what exactly is it that differs — and where does the difference get in?

Begin by separating two things that get called framing and behave quite differently.

Sometimes the two descriptions really are logically equivalent: 90% survival and 10% mortality, a discount for cash and a surcharge for card, glass half full. Nothing has been added or withheld. Call this equivalence framing. Sometimes, instead, the two accounts select different true facts to foreground: a protest reported through its crowd size against the same protest reported through its arrests. Call this emphasis framing. Most headlines are the second kind. Most laboratory demonstrations are the first, and it is the first that is genuinely startling, because it appears to break a rule we would have signed up to without hesitation — that a preference should not depend on the wording of a description that contains the same content.

Take the case that made this concrete. Tversky and Kahneman put an unusual disease to their subjects: 600 people are at risk, and two programmes are on offer. Programme A saves 200 people for certain. Programme B has a one-in-three chance of saving all 600 and a two-in-three chance of saving nobody. Seventy-two per cent chose the certain option. Now the same choice, restated: programme C means 400 people die; programme D carries a one-in-three chance that nobody dies and a two-in-three chance that all 600 die. Only twenty-two per cent chose the certain option. The arithmetic is untouched. The majority reverses.

What could produce that? Ask what a person actually evaluates. Not final states of the world — if we did, the two versions would be indistinguishable. We evaluate changes, relative to a reference point, and the wording supplies the reference point. Talk of lives saved sets the reference at 600 dead, so everything reads as a gain; talk of lives lost sets it at 600 alive, so everything reads as a loss. And people treat those two regions asymmetrically: for gains we prefer the sure thing, while for losses we will take a gamble to avoid the certain one. Losses also loom larger than gains of the same size. Put those together and the reversal is not a mystery but a prediction.

Notice what this does to the idea of a neutral description. A choice has to be described from some reference point; there is no view from nowhere. Editors who believe they are merely reporting are still, unavoidably, choosing where to put the zero.

Now the caveat that keeps this honest. The Asian disease result is robust and has been reproduced many times, but framing effects in general vary a lot. They shrink when the stakes are large and real rather than hypothetical, when respondents are experienced in the domain, when they are asked to justify their answer, and — most tellingly — when both framings are put in front of the same person at once, at which point most people notice the equivalence and stop reversing. The effect is a fact about how a description is processed on its own terms, not a fixed constant of human nature.

The tempting misreading is that framing effects show people are irrational, or that they are being tricked. Neither is quite right. Emphasis framing supplies genuinely different information, so responding to it differently is not an error at all — it is what you should do when told different things. And equivalence framing exploits something ordinarily sensible: we are built to evaluate changes rather than absolute states, because gains and losses are what we can act on, and a reference point is the only thing that makes a change definable. The failure is narrow and specific — the reference point is supposed to come from the situation, and instead it leaks in from the sentence.

It also matters that the two kinds have different remedies. For equivalence framing, the fix is mechanical: state both formulations, since seeing them together dissolves most of the effect. For emphasis framing there is no such trick, because the frames are not equivalent and choosing among them is an editorial judgement that cannot be automated away. Being aware that you are inside a frame does not put you outside one.

• How prospect theory's value function generates the gain-loss asymmetry that framing exploits.
• Why default options, which change nothing about the alternatives, move behaviour as strongly as framing does.

Every term the collection defines is gathered in the glossary.