Framing Effects

Abstract

A framing effect is a change in preference produced by describing the same options in different but logically equivalent ways, as when the Asian disease problem reverses risk preference between a gain wording and its equivalent loss wording. Such reversals violate description invariance and fall into three empirically distinct kinds — risky-choice, attribute, and goal framing. Prospect theory explains the reversal through a reference-dependent value function, concave for gains and convex for losses; fuzzy-trace theory attributes it to reasoning over a problem's qualitative gist rather than its verbatim numbers; and neuroimaging ties susceptibility to the amygdala and resistance to prefrontal control. This article surveys the phenomenon, its taxonomy, its explanations, its neural basis, and its uses in health and political communication. Three interactive demonstrations let the reader reframe the disease problem, an attribute, and the value curve's reference point.

Keywords: framing effect, decision making, prospect theory, description invariance, message framing

A framing effect occurs when a decision maker's choice shifts because the options are worded differently even though nothing about their substance changes. The framing may cast an outcome as a gain or a loss, describe an attribute in positive or negative terms, or stress the benefits of acting versus the costs of not acting; in each case the information conveyed is the same, and only the surface description differs. Framing effects matter because they cut against the deepest assumption of rational choice: that preferences attach to outcomes, not to the language used to present them (Tversky & Kahneman, 1986). If a mind ranks the very same prospect differently depending on its packaging, then preference is not simply read off a stable internal scale but is in part constructed at the moment of choice, and the words supplied by whoever poses the question become part of what is chosen.

What a Framing Effect Is

The defining case is the Asian disease problem posed by Tversky and Kahneman (1981). Respondents are told that an unusual disease is expected to kill 600 people and asked to choose between two programs. In the gain version, Program A saves 200 people for certain, while Program B has a one-third chance of saving all 600 and a two-thirds chance of saving no one. In the loss version, Program C lets 400 people die for certain, while Program D carries a one-third chance that nobody dies and a two-thirds chance that all 600 die. Program A and Program C describe exactly the same outcome — 200 alive and 400 dead — and so do Programs B and D. Yet when the choice is framed as saving lives, a large majority prefer the sure option; when it is framed as lives lost, a majority prefer the gamble. The preference reverses on wording alone.

Figure 1

One Outcome, Two Frames, Opposite Preferences

The force of the demonstration comes from its logical structure, not from any deception: both descriptions are accurate, and a respondent who saw the two versions side by side would surely call them equivalent. This is why framing is treated as a normative problem rather than a mere misunderstanding. Description invariance is a condition every theory of rational choice imposes, because a preference that flips between equivalent statements cannot be represented by any consistent ranking of outcomes (Tversky & Kahneman, 1986). The interactive demonstration below poses both versions of the disease problem and shows that the underlying outcomes never change as the wording does.

Three Kinds of Framing

Not every framing effect works the same way, and treating them as one phenomenon obscures real differences in mechanism and reliability. Levin, Schneider, and Gaeth (1998) drew the influential distinction among three types, separated by what exactly is being framed and what is being measured. Risky-choice framing, the Asian disease type, presents a choice between a sure option and a gamble of equal expected value, and manipulates whether the outcomes are cast as gains or losses; the measured effect is a shift in risk preference. Attribute framing varies the description of a single characteristic of an object or event — ground beef described as 75% lean rather than 25% fat, a medical procedure as having a 90% survival rate rather than a 10% mortality rate — and measures the resulting change in how favorably the object is evaluated. Goal framing varies whether a message stresses the benefit of performing an action or the cost of failing to perform it, and measures the effect on persuasion.

Table 1
The Three Types of Framing Effect
Note. The typology follows Levin et al. (1998); a meta-analysis across all three types is reported by Piñón and Gambara (2005). The types differ in mechanism: attribute framing is the most consistent, risky-choice framing depends on the equal-expected-value structure, and goal framing is the weakest and least stable.

The typology matters because the three types do not share a single cause and do not behave alike. Attribute framing is the most robust and the easiest to interpret: casting a lone attribute in positive language leaves a positive trace in memory that colors the later evaluation, and the effect appears reliably across many domains. Risky-choice framing depends on the gain-loss structure that prospect theory formalizes and is therefore sensitive to whether the options really are of equal expected value. Goal framing is the weakest and most inconsistent, with the direction of the effect varying with the behavior at issue. A meta-analysis spanning all three confirms that lumping them together understates attribute framing and overstates the reliability of goal framing (Piñón & Gambara, 2005). The attribute-framing demonstration below rewrites one attribute from a positive to a negative frame, and the evaluation moves even though the quantity is fixed.

Why Framing Works: The Reference Point

The leading explanation of risky-choice framing is prospect theory, which replaced utility defined over final wealth with value defined over changes — gains and losses measured from a reference point (Kahneman & Tversky, 1979). Its value function has a distinctive shape: concave above the reference point, so that gains bring diminishing satisfaction, and convex below it, so that losses bring diminishing pain. Concavity over gains produces risk aversion, because the sure gain sits higher on the curve than the average of a larger and a smaller gain; convexity over losses produces risk seeking, because the sure loss sits lower than the average of a larger and a smaller loss. This mirror-image pattern is the reflection effect: risk preferences reverse when a problem about gains is turned into the corresponding problem about losses.

Framing works by moving the reference point. The word saved sets the reference at everyone dead, so the outcomes are read as gains and the concave branch governs the choice, favoring the certain 200 lives. The word die sets the reference at everyone alive, so the same outcomes are read as losses and the convex branch governs, favoring the gamble that might avoid any deaths. The value function never changes; only which part of it is consulted changes, because the wording relocates the psychological zero from which gains and losses are counted. The demonstration below places a reference-point control on the value curve so that sliding it recodes a fixed outcome from a gain into a loss and flips the predicted risk attitude.

Loss aversion — the greater steepness of the value function below the reference point than above it — is often invoked alongside the reflection effect, and it has been the subject of vigorous debate. A prominent critique argued that many findings credited to loss aversion are better explained by inertia or the salience of losses in attention, and that a fixed two-to-one weighting of losses over gains is not robustly supported (Gal & Rucker, 2018). A large reappraisal answered that loss aversion survives once its genuine moderators are taken into account, concluding that reports of its death were greatly exaggerated (Mrkva et al., 2020). The exchange has sharpened the concept rather than discarded it: loss aversion is real but bounded, and its magnitude depends on the stakes, the domain, and the reference point in play.

Gist over Verbatim: The Fuzzy-Trace Account

Prospect theory is not the only explanation, and a serious rival locates framing not in the shape of a value function but in the form of the mental representation people reason from. Fuzzy-trace theory holds that a decision maker encodes a problem at two levels at once: a verbatim representation that preserves the exact numbers and a gist representation that captures their qualitative meaning (Reyna & Brainerd, 1991). People prefer to reason with the simplest gist that supports a decision. In the Asian disease problem, the gain frame reduces to a contrast between saving some people and possibly saving none, so the gist is some lives versus the chance of no lives, which favors the sure option; the loss frame reduces to some deaths versus possibly no deaths, so the gist is some deaths versus the chance of no deaths, which favors the gamble. The numerical expected values, identical across frames, are stripped away in gist processing, and the categorical contrast that remains differs between the two wordings.

The gist account makes predictions that a purely numerical theory does not. Framing effects should grow when the wording invites a coarse categorical reading and shrink when the numbers are made unavoidable, and they should change across development as reliance on gist increases rather than decreases with expertise — a counterintuitive reversal that fuzzy-trace theory predicts. The two explanations are not strictly exclusive; prospect theory describes the valuation of coded outcomes while fuzzy-trace theory describes how the problem is mentally represented before it is valued, and a full account of framing may need both a theory of representation and a theory of value.

Framing in the Brain

If framing reflects the interplay of an intuitive, affect-laden response and a more deliberate one, that structure should be visible in the brain, and it is. In an influential neuroimaging study, De Martino, Kumaran, Seymour, and Dolan (2006) had participants make risky-choice framing decisions while their brains were scanned. Choices that followed the frame — risk-averse under gains, risk-seeking under losses — were accompanied by heightened activity in the amygdala, a structure central to rapid emotional processing. Choices that ran against the frame, resisting its pull, were accompanied by activity in the orbitofrontal and anterior cingulate cortex, regions associated with the integration of emotion into deliberate control. Moreover, participants whose behavior was least susceptible to framing showed the greatest engagement of these prefrontal regions.

The result gave framing a mechanistic reading: the effect arises from an emotional system that responds to the gain or loss wording, and it is modulated, though rarely abolished, by a control system that can override the immediate response. This maps naturally onto dual-process descriptions of judgment, in which a fast, automatic route and a slow, effortful route jointly determine the choice. Kahneman's two-system synthesis makes the mapping explicit: framing effects are a paradigm case in which an intuitive System 1 response, keyed to the surface wording, is only sometimes corrected by a deliberate System 2 that consults the underlying outcomes (Kahneman, 2003). It also reframes individual differences in susceptibility as differences in the balance between these systems rather than as differences in intelligence, since resistance to framing tracks prefrontal engagement rather than general ability.

Framing Beyond the Laboratory

Framing is not confined to hypothetical gambles; because the same information can almost always be worded more than one way, framing is a permanent feature of real communication, and two applied literatures have taken it up in depth. In health communication, message framing asks whether to stress the gains of a healthy behavior or the losses of failing to adopt it. The influential proposal of Rothman and Salovey (1997) held that gain-framed appeals are more effective for prevention behaviors, which are low in perceived risk, whereas loss-framed appeals are more effective for detection behaviors, which feel risky because they can reveal a problem. An early demonstration found that a loss-framed pamphlet urging breast self-examination produced stronger intentions and behavior than a gain-framed one (Meyerowitz & Chaiken, 1987). A later meta-analysis tempered the picture, finding that framing effects on health behavior are real but small and moderated by the behavior in question, so that the prevention-detection rule holds only weakly across the full body of studies (Gallagher & Updegraff, 2012).

In political communication, framing is the emphasis a speaker places on some considerations over others when describing an issue. Chong and Druckman (2007) synthesized this literature, distinguishing equivalence framing, in which logically identical information is reworded as in the disease problem, from emphasis framing, in which different true aspects of an issue are foregrounded to shift opinion. Emphasis framing is the workhorse of political persuasion, and its effects depend on the strength and credibility of the frame, its repetition, and whether competing frames are present to cancel it. Across both applied literatures the same lesson recurs: framing effects are genuine and consequential but bounded, subject to moderators that determine when they appear and how large they are. Meta-analyses of risky-choice framing put the average effect at a modest but reliable size and show that it grows or shrinks with features such as the response mode, the size of the stakes, and whether respondents see one frame or both (Kühberger, 1998; Steiger & Kühberger, 2018).

Worked Example

The Asian disease problem turns the framing reversal into arithmetic that makes the equivalence unmistakable. A disease threatens 600 people. In the gain frame, Program A saves 200 for certain, and Program B saves all 600 with probability one-third and no one with probability two-thirds. The gamble's expected number saved is (1/3 × 600) + (2/3 × 0) = 200, exactly equal to the sure 200 of Program A. A risk-neutral chooser is therefore indifferent, yet a clear majority pick the certain option, revealing risk aversion for gains.

Now recode the identical outcomes as losses. Program C lets 400 die for certain, and Program D has a one-third chance that no one dies and a two-thirds chance that all 600 die, for an expected number of deaths of (1/3 × 0) + (2/3 × 600) = 400. Program C leaves 600 − 400 = 200 survivors, the same as Program A, and Program D leaves an expected 600 − 400 = 200 survivors, the same as Program B. Every program specifies an expected 200 alive and 400 dead; the four options collapse to two distinct prospects, each presented twice. The only thing that changes between the gain and loss versions is the word attached to the certain outcome — saved or die — and that single word moves the reference point from which the value function is read, flipping the majority from the sure option to the gamble. The demonstrations above compute these identical outcomes for any population size, so that no arithmetic difference can be hiding behind the reversal.

Discussion

Framing effects occupy a special place in the study of judgment because they isolate the contribution of description itself. Other well-known departures from rational choice can be read as responses to genuine features of a problem; framing cannot, because the feature it responds to — the wording — carries no information about the outcomes. This is what makes framing so direct a challenge to the assumption that preferences are defined over outcomes. The challenge is not that people are irrational in some global sense but that the act of choosing is more constructive than the classical picture allows: the frame supplied by the environment becomes an ingredient of the preference rather than a neutral window onto a fixed one (Tversky & Kahneman, 1986).

The competing explanations are best read as accounts of different stages of one process. Prospect theory specifies how outcomes, once coded as gains or losses, are valued, and it predicts the direction of the risky-choice reversal from the curvature of the value function (Kahneman & Tversky, 1979). Fuzzy-trace theory specifies how a problem is represented before it is valued, and it predicts when framing will be strong from the gist that the wording invites (Reyna & Brainerd, 1991). The neural evidence adds the implementation, tying the frame-consistent response to emotional circuitry and resistance to prefrontal control (De Martino et al., 2006). Meanwhile the applied and meta-analytic literatures supply the crucial correction to any tidy story: framing effects are real but bounded, and their size is governed by moderators that a mature science of framing must specify rather than ignore (Levin et al., 1998; Kühberger, 1998).

Current Directions

The most consequential recent work has been the turn to large-scale replication and moderation. A meta-analytic re-appraisal of the framing effect confirmed a reliable overall effect while documenting how strongly it depends on design features and how much heterogeneity the literature contains, sharpening the estimate that older narrative reviews had left loose (Steiger & Kühberger, 2018). On a larger canvas, a 19-country replication of the choice patterns predicted by prospect theory found that the reflection and framing signatures reproduce robustly across cultures, establishing that the phenomena underlying framing are not artifacts of a narrow set of Western samples (Ruggeri et al., 2020).

A second strand continues to test the boundaries of loss aversion, the asymmetry that underwrites much of risky-choice framing, adjudicating when the loss-over-gain weighting appears and when it does not, and concluding that the effect is genuine but carries identifiable moderators rather than a universal constant (Mrkva et al., 2020). Together these directions mark a shift in emphasis from establishing that framing exists, long since settled, to specifying precisely how large it is, for whom, and under what conditions — the questions that determine whether a frame will move a real decision in health, policy, or the marketplace.

Glossary

Attribute framing.

A framing effect in which describing a single characteristic in positive versus negative terms changes how favorably the object is evaluated; the most robust of the three types.

Description invariance.

The requirement of rational choice that logically equivalent descriptions of the same options yield the same preference; framing effects are violations of it.

Dual-process account.

The view that a fast, automatic response to a frame and a slow, deliberate one jointly determine the choice, with resistance to framing tied to deliberate control.

Emphasis framing.

In political communication, foregrounding some true aspects of an issue over others to shift opinion, as distinct from rewording logically identical information.

Equivalence framing.

Framing in which the alternative descriptions are logically identical, as in the gain and loss versions of the Asian disease problem.

Framing effect.

A change in preference produced solely by describing the same options in different but logically equivalent ways.

Fuzzy-trace theory.

The theory that people encode a problem as both verbatim numbers and qualitative gist and reason from the simplest gist that supports a decision, producing framing effects.

Gain frame.

A description that casts an outcome as a gain relative to a reference point, such as lives saved; it tends to elicit risk aversion.

Gist representation.

In fuzzy-trace theory, the qualitative, meaning-based encoding of a problem that strips away exact numbers and drives many framing effects.

Goal framing.

A framing effect in which stressing the benefit of acting versus the cost of not acting changes an appeal's persuasiveness; the weakest and least stable type.

Loss aversion.

The tendency for a loss to weigh more heavily than an equal gain, reflected in a value function steeper below the reference point than above it.

Loss frame.

A description that casts an outcome as a loss relative to a reference point, such as lives lost; it tends to elicit risk seeking.

Message framing.

The applied use of gain versus loss wording in persuasive communication, especially health messages promoting prevention or detection behaviors.

Prospect theory.

The descriptive theory of choice under risk in which value is defined over gains and losses from a reference point, supplying the standard account of risky-choice framing.

Reference point.

The baseline against which outcomes are coded as gains or losses; framing effects arise because the wording relocates it.

Reflection effect.

The reversal of risk preference between gains and losses: risk aversion among gains becomes risk seeking among the corresponding losses.

Risky-choice framing.

A framing effect in which casting a choice between a sure option and an equal-expected-value gamble as gains or losses shifts risk preference; the Asian disease type.

Verbatim representation.

In fuzzy-trace theory, the encoding that preserves the exact surface numbers of a problem, distinct from the gist that people usually reason from.

Key Researchers

James N. Druckman (contemporary). Martin Brewer Anderson Professor of Political Science at the University of Rochester; with Chong he synthesized framing theory in political communication, separating equivalence from emphasis framing and cataloguing the moderators of framing effects. Faculty Page - ORCID

Daniel Kahneman (1934-2024). Nobel laureate and Eugene Higgins Professor of Psychology, Emeritus, at Princeton University; with Tversky he developed prospect theory and ran the framing experiments that made the effect a canonical result. Wikipedia - Google Scholar

Anton Kühberger (contemporary). Decision researcher at the University of Salzburg; his meta-analyses quantified the size and moderators of the framing effect, tempering earlier claims about its reliability and reach. Faculty Page - ORCID

Benedetto De Martino (contemporary). Cognitive neuroscientist at University College London; his 2006 imaging study showed that susceptibility to the framing effect tracks amygdala activity while resistance tracks prefrontal control, giving framing a mechanistic reading. Homepage - ORCID

Valerie F. Reyna (contemporary). Lois and Melvin Tukman Professor at Cornell University; with Brainerd she developed fuzzy-trace theory, which explains framing as gist-based reasoning over the qualitative meaning of a problem rather than its exact numbers. Faculty Page - ORCID

Peter Salovey (b. 1958). Chris Argyris Professor of Psychology at Yale University; with Rothman he framed the theory of gain- versus loss-framed health messages, matching frame to prevention or detection behavior. Faculty Page - ORCID

Amos Tversky (1937-1996). Professor of Psychology at Stanford University; with Kahneman he defined prospect theory, posed the Asian disease problem, and formalized description invariance as the axiom framing violates. Wikipedia

Frequently Asked Questions

What is a framing effect?
A framing effect is a change in preference caused solely by how logically equivalent options are described. Casting the same outcome as a gain or a loss, or an attribute in positive or negative terms, can reverse a choice even though nothing about the substance changes (Tversky & Kahneman, 1981).

What is the Asian disease problem?
It is the classic demonstration of risky-choice framing. A public-health choice between a sure option and a gamble of equal expected value is posed once in terms of lives saved and once in terms of lives lost; the gain wording elicits a risk-averse majority and the loss wording a risk-seeking one, though the outcomes are identical (Tversky & Kahneman, 1981).

What are the three types of framing effect?
Levin and colleagues distinguish risky-choice framing, which shifts risk preference; attribute framing, which changes the evaluation of a single characteristic; and goal framing, which changes an appeal's persuasiveness. They differ in mechanism and reliability, with attribute framing the most robust (Levin et al., 1998).

Why do framing effects happen?
Prospect theory explains the classic case: value is defined over gains and losses from a reference point, and the value function is concave for gains but convex for losses. The wording sets the reference point, so the same outcome is read as a gain or a loss and recruits opposite risk attitudes (Kahneman & Tversky, 1979).

Are framing effects a failure of rationality?
They violate description invariance, the axiom that a rational ranking cannot depend on which of two equivalent descriptions is used. This does not mean people are globally irrational, but it shows that preference is partly constructed at the moment of choice rather than simply read off a fixed scale (Tversky & Kahneman, 1986).

What is gain-loss framing in health messages?
Message framing asks whether to stress the benefits of a healthy behavior or the costs of avoiding it. Gain-framed appeals tend to suit prevention behaviors and loss-framed appeals detection behaviors, though meta-analysis shows the average effect on health behavior is real but small (Rothman & Salovey, 1997; Gallagher & Updegraff, 2012).

How large and reliable are framing effects?
Meta-analyses put the average risky-choice framing effect at a modest but reliable size that grows or shrinks with design features such as response mode and stakes. A 19-country study confirmed that the underlying choice patterns replicate across cultures (Kühberger, 1998; Ruggeri et al., 2020).

What does fuzzy-trace theory say about framing?
It holds that people encode a problem as both exact numbers and qualitative gist and reason from the simplest gist that supports a decision. The gain and loss wordings reduce to different categorical contrasts, so framing arises from the mental representation of the problem rather than from the value of its outcomes (Reyna & Brainerd, 1991).

References