Choice Is Constructed in Context

Imagine seeing the same meal on two restaurant menus.

On the first, it is among the most expensive options. On the second, it appears beside several meals that cost substantially more. Its ingredients, portion and price remain identical, yet its position within the comparison can make it appear extravagant in one setting and moderate in another.

This does not show that menus control choice or that people have no stable preferences. It shows that options are not always evaluated in isolation. Their meaning can depend partly on the conditions under which they are encountered.

A choice begins with a context.

A decision uses an effective choice set

At any moment, more actions are physically possible than a person can represent and compare.

A restaurant may offer twenty meals, but only five receive serious consideration. Another meal may be technically available but unnoticed, forgotten, misunderstood or dismissed as unsuitable. A person choosing what to do next may possess many theoretical options while experiencing only two as genuinely possible.

The effective choice set consists of the alternatives that enter the decision as candidates for selection. It is not necessarily identical to everything objectively available.

Visibility makes some options easier to notice. Memory makes some alternatives easier to retrieve. Time pressure can narrow consideration. Social conditions can make an otherwise available action feel inappropriate. Current capacity can make one response appear feasible and another merely theoretical.

Choice construction therefore begins before explicit comparison. Context helps determine what participates in the decision.

An observed choice does not reveal where the other possibilities disappeared. An option may have been considered and rejected, or it may never have entered the effective choice set. Behaviour alone cannot establish which occurred.

Construction is not invention

To say that choice is constructed can sound as though every preference is improvised in the moment.

That is not the claim.

People bring histories into decisions. They have learned associations, recurring goals, established commitments, identities, habits and expectations. These can produce considerable stability across situations. Someone who consistently values autonomy may continue to favour it across many job decisions. Someone with a strong aversion to a particular risk may reject it under several descriptions.

Context does not erase these influences. It affects how they are represented and weighted in a particular episode.

Construction describes this interaction. Prior learning constrains the decision, while present conditions determine which parts of that learning become relevant, what information is available and how the alternatives relate to one another.

Choice is neither the pure discovery of a finished internal ranking nor an invention unconstrained by the past.

Frames change what outcomes mean

Consider the classic 600-person disease problem. One version describes how many people a treatment is expected to save. Another describes how many are expected to die. Although the numerical outcomes can be formally equivalent, people often show different levels of risk preference across the two descriptions.

This is a framing effect. The description changes how the decision is represented even when the underlying quantities remain the same.

Risky-choice framing is well established, but it does not produce an inevitable response. A systematic review found that its magnitude varies with problem construction, response format and numerical presentation. Framing is better understood as a family of related effects than as one invariant result (Kühberger and Gradl, 2023).

Some classic gain and loss descriptions are also less perfectly matched than they initially appear. They may emphasise different comparisons or omit different information. When descriptions are made more complete and closely matched, part of the framing effect becomes smaller—but does not necessarily disappear. Large-sample evidence has found a residual effect even with matched, complete descriptions (DeKay and Dou, 2024; DeKay, 2026).

“Framing” names a change in representation. It does not identify a single mechanism. Valenced language, categorical meaning, attention and reference-dependent evaluation may each contribute under different conditions.

Reference points organise evaluation

An outcome does not always acquire meaning from its final quantity alone. It can also be evaluated in relation to a reference point.

Receiving an amount can be represented as a gain relative to a zero or status-quo reference point, a loss relative to what was expected or a disappointing improvement relative to a better possible outcome. The amount remains unchanged while its represented meaning changes.

Reference points can arise from:

• the current situation;
• an expectation;
• a recent outcome;
• an explicitly supplied frame;
• another option in the choice set.

They need not be deliberately selected or clearly verbalised. A person can respond to an outcome as a deterioration without consciously stating the standard against which it was judged.

Reference dependence helps explain why survival and mortality descriptions can produce different choices. Each description can organise evaluation around a different relation to the possible outcome. But reference dependence is not a label for every contextual influence. An added alternative, uncertain outcome and change in physical state need not operate through the same process.

Other options change the comparison

Return to the restaurant menu.

Suppose the diner is choosing between an inexpensive basic meal and a more expensive meal with higher-quality ingredients. The restaurant then adds a third option that is nearly as expensive as the high-quality meal but clearly worse on quality.

Under some conditions, this third option can increase the appeal of the high-quality meal. The new option is rarely selected, but it provides a nearby alternative that the high-quality meal clearly dominates.

This specific pattern is the attraction effect: increased target choice following the addition of an appropriately positioned, asymmetrically dominated alternative. “Decoy effect” is sometimes used more broadly, but not every added alternative produces attraction.

An intermediate option may instead gain appeal because it appears to offer a compromise between two extremes. An option closely resembling an existing alternative may draw choice share away from it. Depending on the alternatives’ represented relationships, adding an option can produce attraction, compromise, similarity, repulsion or no reliable shift.

A systematic mapping of decoy locations found that different contextual patterns occupy different regions of the attribute space. The decoy’s position can change both the magnitude and direction of its influence (Dumbalska et al., 2020).

The mechanism remains disputed. An added option may change local comparisons, attribute weighting, attention, the subjective range of values or the ease with which a choice can be justified. Different mechanisms can produce similar observed reversals.

Context-sensitive reversals do not necessarily show that reasoning has collapsed. Formal modelling demonstrates that they can emerge when a system attempts to maximise expected value using noisy estimates of values and attribute orderings (Howes et al., 2016). This does not make every reversal adaptive. It shows that changed choice is insufficient, by itself, to diagnose irrationality.

Attention determines what contributes

Information must be represented in some form to contribute to comparison, although that representation need not be conscious or verbally accessible.

Visual prominence and placement can direct attention during the current choice. Information receiving more attention may then contribute more strongly to the developing comparison.

Eye-tracking studies find systematic relationships among fixation, deliberation time and eventual selection. Attentional evidence-accumulation models can reproduce patterns in which currently attended alternatives receive greater influence (Krajbich, Armel and Rangel, 2010; Krajbich et al., 2012). Gaze-dependent models have also accounted for aspects of multi-alternative contextual choice, although strong context effects can be underestimated (Molter et al., 2022).

The relationship runs in both directions. Looking longer at an option can give its attributes more opportunity to influence comparison, while an option that already appears promising may attract more attention. Experimental manipulation provides some evidence that attention can contribute causally, but observational fixation alone cannot establish that direction.

Fixation is therefore not a direct measurement of value. Attention is also broader than vision. A consequence can become mentally prominent because it was recently discussed, emotionally charged or repeatedly imagined.

Context extends into memory

Recent exposure may affect accessibility or the comparison sample available from memory.

Decision by Sampling proposes that attribute values are evaluated through ordinal comparisons with samples drawn from memory and the immediate environment. Instead of consulting a fixed internal scale for whether a salary is high or a delay is long, the system may register how the current value ranks among sampled alternatives.

A salary can therefore seem high after repeated exposure to lower amounts and modest after exposure to higher ones. The salary has not changed; its relative position within the comparison sample has.

Within Decision by Sampling, recent or frequently encountered values may be more likely to enter that sample. The surrounding environment can therefore influence evaluation partly by affecting what gets compared (Stewart, Chater and Brown, 2006).

This is one model of contextual valuation, not the explanation of every context effect. Reference points, attention and option structure may produce related behavioural changes through different processes. The model nevertheless clarifies that the decision context can include traces of earlier environments.

State changes the decision environment

Context also extends into the person’s current state.

The same intended action can appear different when someone is rested than when exhausted. The external task may remain unchanged while its anticipated—and sometimes actual—cost changes with the person’s state. An option that remains preferred in principle may lose immediate priority because it no longer appears feasible.

Hunger, stress, fatigue, affect and arousal are not interchangeable. Each can alter different aspects of valuation, learning and selection, and their effects vary with timing, task and person.

In one set of experimental tasks, increasing hunger was associated on average with greater risk tolerance in monetary decisions. Effects varied across rewards, however, and the evidence does not support the general claim that hunger makes people reckless (Levy, Thavikulwat and Glimcher, 2013).

Stress research is similarly heterogeneous. Acute and chronic stress can affect valuation, learning and risk-taking differently depending on when stress occurs and what the decision requires (Porcelli and Delgado, 2017).

Fatigue also requires care. Exhaustion can change anticipated effort and perceived feasibility. That does not establish that decisions consume a literal internal resource or that previous choices mechanically deplete a general capacity to choose well.

State-dependent choice is not necessarily false choice. If acting has become more costly, the represented value of the options may change with it. Neither the rested nor exhausted choice should be treated as a complete expression of what the person always values.

Uncertainty changes what can be selected

Not all uncertainty concerns the same object.

A person may be uncertain about:

• which outcome will occur;
• how probable each outcome is;
• how valuable an option will be;
• which option they prefer;
• whether the selected option is better.

Risk, ambiguity, outcome uncertainty, value uncertainty and choice confidence are related but distinct.

Uncertainty about outcomes may increase information search. Ambiguity may change whether an option is treated as credible. Uncertainty about value may cause a person to revisit particular attributes. Low choice confidence may delay selection or make it easier to reopen.

Value certainty and choice confidence do not simply average certainty across every attribute. Attributes carrying more weight in the decision can also contribute more strongly to confidence. Deliberation can refine both attribute estimates and the certainty attached to them (Lee and Hare, 2023).

Greater certainty in an attribute is not the same as assigning that attribute greater value. Certainty concerns the precision of the representation; value concerns its direction and influence within the choice.

Preference can consequently become more definite through comparison without uncertainty always favouring the safest, most familiar or most immediate response.

Social context changes meaning

An option can change when other people are present.

A private choice may carry different anticipated consequences when it becomes observable. The presence of others can introduce empirical expectations about what people usually do, normative expectations about what they approve, reputational consequences and identity concerns.

Experiments manipulating empirical and normative expectations show that both can influence norm-relevant decisions, although their effects depend on how the expectations relate to one another (Bicchieri and Xiao, 2009).

A social change in behaviour does not automatically establish conformity or a change in private preference. Other people may supply information, alter the consequences of acting, activate an existing commitment or change what an action communicates. Someone may also make one choice publicly while retaining another private evaluation.

Social context matters because it changes what the alternatives can mean, not because it supplies one universal social mechanism.

Stable preferences and constructed choices coexist

Context-sensitive choice does not imply that stable preferences are absent.

Learning, goals, commitments and identity can produce durable constraints. But those constraints do not fully specify every decision. Their influence depends on which attributes enter attention, how alternatives are framed and what current conditions make possible.

Deliberation can also refine preference. A person may begin with uncertain evaluations, examine the attributes and develop a more definite ranking. Choice can therefore report an existing preference while also contributing to its precision.

Post-choice rating studies initially appeared to show that selecting an option causes people to value it more and devalue the rejected alternative. Genuine updating may occur, but the traditional design contains a measurement problem. Initial ratings are noisy, and the later choice supplies additional information about the underlying preference. Part of the apparent change can therefore arise without choice creating a new preference (Chen and Risen, 2010).

It is necessary to distinguish:

• revealing a preference that was imperfectly measured;
• refining an uncertain preference during deliberation;
• updating evaluation after selection;
• constructing a preference without prior constraint.

These processes can produce similar observations while making different claims about what changed.

Context sensitivity is not a verdict

Some context effects violate formal principles of consistent choice.

If two descriptions are genuinely equivalent, description invariance says that they should produce the same preference. Within formal choice models governed by regularity, adding an option constrains how selection probabilities for existing options should change. Certain framing and attraction effects depart from these standards.

That departure can be significant without becoming a verdict on the person.

A changed choice may reflect a different interpretation, comparison, inferred signal, perceived constraint or noisy representation. Some reversals may emerge from adaptive responses to uncertainty. Others may produce outcomes the person would reject under fuller information or different conditions.

Context sensitivity is neither proof that every reversal is rational nor proof of global irrationality.

It shows that observed choice depends on how the decision is represented.

Choice is constructed at the point of comparison

People do not enter every decision with a complete ranking of every possible action.

Context helps determine which options become available for consideration. Frames and reference points shape what outcomes mean. Other alternatives change the comparison field. Attention determines which information contributes. Memory supplies possible comparison samples. State alters anticipated costs and feasibility. Uncertainty changes information search and confidence. Social conditions change what actions represent.

History, learning, identity, goals and prior commitments constrain how choice is constructed in context. The same setting does not produce the same choice in everyone, and many preferences remain stable across substantial changes in circumstance.

But stable influence is not the same as a finished ranking.

The option that gains priority expresses an interaction between the person and the conditions of decision. Even then, selection does not guarantee execution. The context in which action must begin may support, obstruct or transform the selected response.

Choice is neither the pure discovery of a fixed internal preference nor the product of context alone.

It is constructed where a person’s history, current state and available world meet.