i

Description

Delicate Arch, arche de grès rouge qui encadre des montagnes enneigées au loin

Source

National Park Service · original page

Licence

PD Public domain

Original photo, unretouched. The background is a version of it stylised by the site.

The framing effect

Saying “90% success” or “10% failure” describes the same reality, but not the same decision.

In brief

• What: the choice changes depending on how two equivalent options are presented (gain or loss, survival or mortality, percentage or number).
• Why: we evaluate options against a reference point, and the wording sets that reference point. A loss weighs more than an equivalent gain.
• Detect: a decision changes when the same data is reworded; a report highlights only one angle, either favorable or alarming.
• Avoid: restate the information in the other frame before deciding, ask for the raw figures and the reference point, present both angles.

On this page

1. Definition
2. The mechanism
3. Where you encounter it
   1. In project management
   2. In management
4. Famous cases
   1. Tversky and Kahneman's “Asian disease” 1981
   2. Surgery or radiation therapy: survival or mortality 1982
5. Warning signs
6. Avoidance reflexes
7. With AI: detect, remedy, mitigate
   1. Detect
   2. Fix
   3. Mitigate
   4. Prompts to build into your routines
   5. How to build them in
   6. Precautions
8. Limits
9. Sources and verification
10. See also
    1. Combines with, or reinforces
    2. Related biases
    3. Counterweights and related reflexes

Definition#

Fictitious illustration

Illustration inspired by the person

The framing effect is the change in judgment or choice caused by the presentation of information whose content is logically the same. It was demonstrated by Amos Tversky and Daniel Kahneman (1981) and is part of their prospect theory: decision makers do not compare end states, but gains and losses measured from a reference point.

There are several forms, which are best told apart (Levin, Schneider and Gaeth, 1998): risky choice framing (a choice between a sure option and a risky one), attribute framing (“75% lean meat” or “25% fat”) and goal framing (a message stressing what you gain by acting or what you lose by not acting).

The mechanism#

Three drivers add up:

• Reference point: the wording designates what is “normal.” “200 people saved” takes the catastrophe as its baseline; “400 people die” takes the current situation as its baseline.
• Diminishing sensitivity: between 0 and 100, a difference of 10 counts for more than between 900 and 1,000. On gains, this makes us cautious; on losses, it pushes us to take a risk to avoid them.
• Loss aversion: a loss is felt more strongly than a gain of the same size (see loss aversion). A “loss” frame therefore mobilizes people more.

The result is a preference reversal: cautious in the “gain” frame, risk-seeking in the “loss” frame, for exactly the same options.

Where you encounter it#

In project management#

• Progress reports: “80% of the features are delivered” and “20% remains to be done, i.e. six weeks” are about the same thing, but the first reassures and the second alerts. The choice of frame steers the committee's decision.
• Dashboards: a “green, 92% availability” light does not trigger the same reaction as “8% downtime, i.e. more than two days a month.”
• Go / no-go decision: “invest another €200,000 to save the project” (possible gain) and “give up and lose the €600,000 already spent” (certain loss) do not lead to the same choice; the second frame pushes toward continuing and joins up with sunk costs.
• Communication with the client: “we can meet the date if you give up this work package” or “we lose this work package to meet the date”; the client reacts to the loss.
• Risk management: “95% chance that the migration succeeds” or “a one-in-twenty chance of losing a day of data.”

In management#

• Feedback: “9 of your 10 deliverables were good” or “one deliverable in ten was rejected”; the same finding can motivate or demoralize.
• Bonus policy: a bonus “paid if the target is met” (gain) and a bonus “withdrawn if the target is not met” (loss) have different effects, the loss frame often being felt as more threatening.
• Reorganization announcements: “90% of jobs are preserved” and “10% of jobs are cut” reassure or worry teams that have nonetheless been given the same figures.
• Job interview: a candidate who “succeeded in 70% of their projects” is perceived differently from one who “failed 30% of them.”

Famous cases#

Warning signs#

• The same data is always presented from the same angle (the reassuring percentage, never what remains to be done).
• A committee's decision changes when the information is reworded as a gain or a loss, as a percentage or an absolute number.
• Loaded words accompany the figures: “only,” “already,” “up to,” “lose,” “save.”
• The reference point is not stated: gain or loss relative to what?
• Two reports on the same state of progress give opposite impressions.

Avoidance reflexes#

1. Restate in the other frame. Turn “90% success” into “10% failure,” “200 saved” into “400 lost,” and see whether your choice changes.
2. Ask for the raw figures. Absolute number, base, period: a percentage without a base says nothing.
3. Name the reference point. “Relative to what is this a gain or a loss?”: the initial plan, the current state, the competition?
4. Present both angles by convention. In a report template, systematically require “done” and “remaining,” “availability” and “downtime.”
5. Compare end states. Reason about the resulting situation (this budget, this deadline, this quality) rather than about the gain or loss relative to a benchmark.
6. Separate presentation from decision. Leave the decision to someone who did not write the report, or have the report reviewed by someone who must propose the opposite frame.
7. Frame honestly. When you are the one presenting, choose the frame that best represents the situation, not the one that gets the answer you want; a misleading frame destroys trust.

With AI: detect, remedy, mitigate#

Prompts to build into your routines#

1. Reword in both frames. Before deciding on the basis of a figure or a sentence from a report.

Here is a piece of information as it was presented to me: [text or figure].
Rewrite it in three equivalent ways: 1) in a “gain” frame; 2) in a “loss” frame; 3) in absolute numbers and frequencies (“out of 100…”), without adjectives.
State the implicit reference point of the original version and what it leaves in the shadows. Do not add any figure that is not in my text.

2. Neutralize a progress note. To review a report intended for a committee.

Here is a project progress note: [text].
1. Identify the wording that steers the reader (loaded words, percentages without a base, a single angle).
2. Propose a neutral version that presents both what is done and what remains, with units and dates.
3. List the three questions a critical reader should ask about this note.
Do not change any figure; if a piece of data is missing, flag it.

3. Decision stability test. When a decision depends on how an option is described.

Here are two options: [A] and [B], described as follows: [description].
Evaluate them three times, in three different presentations (gain, loss, raw figures), without seeing your previous answers. Then say whether your opinion changed from one presentation to another, and what explains it. Finish with an opinion based solely on the raw-figures version.

How to build them in#

• Before a committee: prompt 2 reviews the progress note; prompt 1 is applied to the two figures that matter most.
• In a report template: the two-column layout is fixed once and for all, so that the frame no longer depends on the author.
• Difficult decision: prompt 3 serves as a check; if the AI's opinion changes with the frame, the frame carries weight.

Precautions#

• Do not describe your preferred option to the AI in qualifying terms: you frame it before it even begins.
• Check that the rewordings change no figure; the AI may round or miscalculate.
• A “neutral” version produced by the AI is still a version: review it, because the choice of words is still a choice.
• Do not send confidential reports or personal data to an unauthorized tool.

Limits#

• The effect is not universal: a meta-analysis (Kühberger, 1998) concludes that the effect is real but of moderate size and varies with the type of framing and the wording of the problems.
• Part of the effect comes from the wording conveying information: saying “half of projects succeed” rather than “half fail” sometimes reflects what the speaker actually thinks (“informational” framing).
• The figures of the Asian disease problem are results of an experiment on students, not rates generalizable to every steering committee.
• For the McNeil et al. (1982) case, only the trend is reported here; the exact proportions per group are to be read in the article.

Sources and verification#

• Tversky & Kahneman (1981), “The framing of decisions and the psychology of choice,” Science — the “Asian disease.”
• McNeil, Pauker, Sox & Tversky (1982), “On the elicitation of preferences for alternative therapies,” New England Journal of Medicine.
• Tversky & Kahneman (1986), “Rational choice and the framing of decisions,” Journal of Business.
• Levin, Schneider & Gaeth (1998), “All frames are not created equal: A typology and critical analysis of framing effects,” Organizational Behavior and Human Decision Processes.
• Kühberger (1998), “The influence of framing on risky decisions: A meta-analysis,” Organizational Behavior and Human Decision Processes.
• Kahneman & Tversky (1979), “Prospect Theory: An Analysis of Decision under Risk,” Econometrica.
• Echterhoff et al. (2024), “Cognitive Bias in Decision-Making with LLMs” — framing in language models.

See also#