What is the framing effect?
A disease that will kill 600 people, two programmes, and a switch of majorities caused by a change of verb. Kahneman's Thinking, Fast and Slow reprints the evidence. Later replications show an effect that is real, smaller than the original, and still disputed in its medical version.
Answer to the question
What is the framing effect?
The framing effect is the finding that logically equivalent descriptions of the same options lead people to choose differently. In Tversky and Kahneman's 1981 Asian disease problem, a sure option won when outcomes were described as lives saved and a gamble won when they were described as lives lost.
Key takeaways
- Tversky and Kahneman (1981) found that a sure option was preferred when outcomes were described as lives saved and a gamble when they were described as lives lost. Kahneman's Appendix B gives 72 per cent of 152 and 78 per cent of 155.
- McNeil, Pauker, Sox and Tversky (1982) asked 238 patients, 491 graduate students and 424 physicians to compare surgery with radiation. Surgery looked relatively more attractive in all three groups when outcomes were stated as the probability of living.
- Levin, Schneider and Gaeth (1998) argued that three different effects share the name: risky choice, attribute and goal framing.
- Klein and colleagues (2014) ran the Asian disease problem in 36 samples with 6,271 participants. No sample reversed, and the typical effect was about half the original size.
- Macchi and Zulato (2021) report that the medical reversal appears with cumulative figures and disappears with a linear layout, where 73 and 71 per cent chose surgery in the two frames.
Thirty-six samples and 6,271 people later, the best-known framing result still comes out in the original direction, and in none of those samples did it reverse. It is also about half the size the original reported, and one of its medical versions is under dispute. The framing effect is the finding that equivalent descriptions of the same options lead to different choices.
The classic demonstration is a disease expected to kill 600 people. One programme saves 200 for certain. The other gives a one-in-three chance of saving everyone and a two-in-three chance that nobody is saved. Describe the same two programmes by deaths instead of survivors and most people switch sides.
What Does the Asian Disease Problem Show?
Amos Tversky and Daniel Kahneman published the problem in Science in 1981. Kahneman's Thinking, Fast and Slow reprints their later 1984 paper as Appendix B, and that appendix is where the book prints its numbers. The chapter on frames gives the result in words only.
When the programmes were described by lives saved, 152 people chose between the sure option and the gamble, and 72 per cent took the sure option. When they were described by lives lost, so that the sure option read as 400 deaths, a separate 155 people chose, and 78 per cent took the gamble. The two descriptions fix the same outcomes and the same probabilities. Only the wording moves, and the majority reverses with it.
The pattern fits what the two authors had been building in prospect theory, their account of choice under risk. Most respondents who read about gains avoided the gamble, and most who read about losses took it. The framing effect in this problem is that asymmetry, triggered by nothing more than a change of verb. The loss side of that asymmetry has its own article, Loss Aversion and the Bet You Turn Down.
The appendix also reports a result that quoted summaries often leave out. Shown two options side by side, one plainly better than the other, all 86 respondents picked the better one. When the same pair arrived as two separate decisions to be made together, 73 per cent of 150 respondents chose the worse combination, and 3 per cent chose the better one. Each decision looked reasonable in its own frame. Joined, they produced a choice that no one made when the options could be seen whole.
Does Framing Work Beyond the Disease Problem?
Barbara McNeil, Stephen Pauker, Harold Sox and Tversky carried the idea into medicine in the New England Journal of Medicine in 1982. They asked 238 patients with chronic conditions, 491 graduate students and 424 physicians to compare surgery with radiation therapy for lung cancer. In all three groups, surgery looked relatively more attractive when outcomes were stated as the probability of living than as the probability of dying. The physicians were not exempt.
The paper also varied whether the treatments were named and whether results came as life expectancy or as cumulative probability, and surgery looked more attractive in all three groups when they were named and when life expectancy was given. The study is about how treatments are described, and it makes no claim about which treatment is better.
A supermarket version had appeared in 1988. Irwin Levin and Gary Gaeth had consumers rate ground beef labelled 75 per cent lean or 25 per cent fat. The lean label drew the more favourable ratings, and the gap was smaller after the consumers tasted the beef. The authors explain this with an averaging model, in which a real experience of the product dilutes the effect of the label.
By the end of the 1990s the results under the framing label looked inconsistent. Levin, Sandra Schneider and Gaeth argued in 1998 that studies defined framing in different ways and so measured different processes. They proposed a typology of three.
The first type, risky-choice framing, changes willingness to take a gamble, and the Asian disease problem is the standard case. The second, attribute framing, changes how one feature of a thing is judged, as with beef labelled 75 per cent lean or 25 per cent fat. The third, goal framing, changes how persuasive a message is, depending on whether it is built on what acting gains or on what not acting loses.
On that reading, the framing effect is a family of effects with a shared trigger, and a result for one member says little about the others. A charity appeal built on what is lost if nobody gives is a goal frame, for instance. The three are tested with different stimuli, and they behave differently in the pooled data below. Price presentation is one applied case, covered in The £3.95 You Spent to Save £3.95, and narrow framing, a separate idea about deciding one choice at a time, has its own article, What is narrow framing?
How Large Is the Effect?
Two large projects have since put numbers on it. Adelson Piñon and Hilda Gambara pooled 51 studies, 151 effect sizes and nearly 13,500 participants, drawn from papers published between 1997 and 2003. Risky choice and goal framing had similar average effects, attribute framing was the weakest of the three, and results varied considerably from study to study.
The Many Labs project, led by Richard Klein, ran the Asian disease problem in 36 samples, 27 in laboratories and 9 online, with 6,271 participants in total. Twenty-five samples came from the United States and 11 from other countries. The framing effect appeared in the original direction in the pooled data and was significant in 86 per cent of samples. None reversed. The typical size was about half that of the 1981 original.
| Project | Studies or samples | People | Finding |
|---|---|---|---|
| Piñon and Gambara pooled analysis | 51 studies, 151 effect sizes | nearly 13,500 | Attribute framing weakest; results varied considerably |
| Many Labs, Asian disease problem | 36 samples | 6,271 | Original direction, significant in 86 per cent of samples, about half the original size |
Variation between samples was negligible, and neither country nor setting made a difference. The replication also dropped the word "Asian" and named each participant's own country in its place, so the result did not depend on the original wording.
Why Is the Medical Version Disputed?
Laura Macchi and Edoardo Zulato argued in 2021 that the reversal in the surgery problem comes from how the numbers were laid out. The original gave cumulative figures, and many people, they suggest, read them as if each figure had to be added to the last.
Their first study, with 150 students across medicine, statistics and lay subjects, reproduced the original pattern, and many people in the mortality frame said the numbers did not add up. In a second study with 180 participants, the cumulative layout produced a framing effect, but a linear layout did not: 73 and 71 per cent chose surgery in the two frames.
A third study, with 430 participants, compared three layouts in the mortality frame. With the cumulative layout, 83 per cent chose radiation. With a linear layout, 68.5 per cent chose surgery, and with an interval layout, 59.1 per cent did. About 52 per cent of those given the cumulative layout judged the data wrong, against about 11 per cent in the other layouts. The authors conclude that the medical reversal reflects misleading communication more than a cognitive bias.
The authors do not dispute that wording moves choices. They dispute why it moved them in the surgery problem, and their own first study reproduced the original surgery pattern. The medical reversal may be a separate artefact of number layout. The Asian disease result stands either way.
Common questions
What is the framing effect?
The framing effect is the finding that logically equivalent descriptions of the same options lead people to choose differently. Tversky and Kahneman demonstrated it in 1981: a sure option was preferred when the outcomes were described as lives saved, and a gamble was preferred when the identical outcomes were described as lives lost.
Does the framing effect replicate?
Largely, in the best-known test, though at a smaller size and with a dispute over the medical version. The Many Labs project ran the Asian disease problem in 36 samples and found the framing effect in the original direction in the pooled data, significant in 86 per cent of samples, though at about half the original size. A pooled analysis of 51 later studies, published between 1997 and 2003, found it weakest for attribute framing and similar for risky-choice and goal framing.
Who discovered the framing effect?
Amos Tversky and Daniel Kahneman, in a 1981 paper in Science, with Tversky as first author. Kahneman's Thinking, Fast and Slow reprints their 1984 paper as Appendix B, where the exact figures for the Asian disease problem, the best-known demonstration, appear.
Sources
Kahneman, D. (2011). Thinking, Fast and Slow. New York: Farrar, Straus and Giroux. Includes a reprint of Kahneman and Tversky (1984); the chapter on frames gives the result in words only. Seed book; vault hub Books/Thinking Fast and Slow.
Tversky, A., & Kahneman, D. (1981). “The framing of decisions and the psychology of choice.” Science, 211(4481), 453-458. doi:10.1126/science.7455683 https://doi.org/10.1126/science.7455683
Kahneman, D., & Tversky, A. (1984). “Choices, values, and frames.” American Psychologist, 39(4), 341-350. doi:10.1037/0003-066X.39.4.341 https://doi.org/10.1037/0003-066X.39.4.341
McNeil, B. J., Pauker, S. G., Sox, H. C., & Tversky, A. (1982). “On the elicitation of preferences for alternative therapies.” New England Journal of Medicine, 306(21), 1259-1262. doi:10.1056/NEJM198205273062103 https://doi.org/10.1056/NEJM198205273062103
Levin, I. P., & Gaeth, G. J. (1988). “How consumers are affected by the framing of attribute information before and after consuming the product.” Journal of Consumer Research, 15(3), 374-378. doi:10.1086/209174 https://doi.org/10.1086/209174
Levin, I. P., Schneider, S. L., & Gaeth, G. J. (1998). “All frames are not created equal: A typology and critical analysis of framing effects.” Organizational Behavior and Human Decision Processes, 76(2), 149-188. doi:10.1006/obhd.1998.2804 https://doi.org/10.1006/obhd.1998.2804
Piñon, A., & Gambara, H. (2005). “A meta-analytic review of framing effect: Risky, attribute and goal framing.” Psicothema, 17(2), 325-331. Average effect sizes as reported in the abstract: risky 0.437, attribute 0.260, goal 0.444. https://www.psicothema.com/pi?pii=3107
Klein, R. A., Ratliff, K. A., Vianello, M., Adams, R. B., Jr., Bahník, Š., et al. (2014). “Investigating variation in replicability: A 'Many Labs' replication project.” Social Psychology, 45(3), 142-152. doi:10.1027/1864-9335/a000178 https://doi.org/10.1027/1864-9335/a000178 (gain versus loss framing: median effect size 0.58 against 1.13 in the original).
Macchi, L., & Zulato, E. (2021). “Numbers do not add up! The pragmatic approach to the framing of medical treatments.” Judgment and Decision Making, 16(3), 596-613. doi:10.1017/S1930297500007750 https://doi.org/10.1017/S1930297500007750