Risk-Taking

Abstract

Risk-taking is a form of behavior: the voluntary pursuit of an action whose outcomes are uncertain and include the possibility of loss as well as gain. Cognitive psychology studies it on three loosely coupled levels — what a person believes about a hazard, what a person chooses when the odds are known, and what a person does when stakes are live — and these measures of risk correlate only weakly. This article traces the construct from Slovic’s psychometric mapping of perceived risk, through prospect theory’s account of choice under risk, to the behavioral tasks and self-report scales that operationalize it, the dual-systems model that explains its adolescent peak, and recent work recovering a general risk-preference factor stable across the lifespan. A worked example shows how loss aversion can turn a positive-expected-value gamble into one a person rationally declines.

Keywords: risk-taking, risk perception, prospect theory, loss aversion, dual-systems model

Introduction

Risk-taking is the voluntary engagement in behavior whose consequences are uncertain and carry a real chance of loss as well as gain. The word carries three distinct meanings in cognitive psychology, and keeping them apart is the first task of any serious treatment. The first is risk perception — the subjective judgment of how dangerous an activity or technology is, studied as a problem in cognition and affect. The second is decision under risk — the formal choice among options whose probabilities are known or estimable, the territory of expected-utility theory and its descriptive successor, prospect theory. The third is risk behavior — what a person actually does when a hazard is in front of them, measured by behavioral tasks in the laboratory and by self-reports of real-world conduct such as reckless driving, unprotected sex, substance use, and gambling. A person who rates skydiving as extremely dangerous may still jump; a person who accepts a favorable laboratory gamble may never speed. The weak coupling among the three is not a nuisance to be averaged away but one of the field’s durable empirical facts.

The Medical Subject Headings thesaurus places Risk-Taking under Behavior in its classification of psychological phenomena. That placement is an indexing convenience, not a theoretical claim: MeSH organizes the biomedical literature for retrieval, and its single parent edge captures neither the construct’s cognitive machinery nor its affective and developmental dimensions. The sections below use the classification only to locate the topic among its neighbors, and treat the psychology on its own terms.

Types of Risk-Taking

The Medical Subject Headings thesaurus classifies Risk-Taking under its parent descriptor Behavior and records one narrower descriptor beneath it.

The single child is a reminder that the MeSH tree is a retrieval index, not a theory of the construct. Psychologically, risk-taking is better carved by domain than by this one-item hierarchy: the Domain-Specific Risk-Taking framework distinguishes financial, health and safety, recreational, ethical, and social risks, and a person willing in one domain is not reliably willing in another (Weber, Blais, & Betz, 2002; Blais & Weber, 2006). These domains are orthogonal to the MeSH parent–child relation: Gambling is a financial-and-recreational risk that MeSH happens to index as the one named subtype, while the health, ethical, and social domains that dominate the psychological literature have no separate descriptor here. The classification should therefore be read as a pointer into the biomedical index, not as an exhaustive taxonomy of the ways people take risks.

Measuring Risk-Taking

Because the three senses of risk come apart, so do their measures. Risk perception is mapped with the psychometric paradigm, in which respondents rate many hazards on scales such as dread, controllability, and catastrophic potential; factor analysis then locates each hazard in a low-dimensional space whose two main axes — roughly “dread” and “unknown” — predict public demand for regulation far better than expert fatality estimates do (Slovic, 1987). The paradigm’s lasting lesson is that lay risk judgments are not simply mistaken estimates of probability; they are structured, reproducible, and sensitive to qualitative features of a hazard that technical risk assessment ignores.

Risk behavior is measured in two complementary ways. Self-report scales ask directly about the likelihood of engaging in risky activities; the Domain-Specific Risk-Taking (DOSPERT) scale is the most widely used, separating risk-taking propensity, risk perception, and expected benefit within each of its five domains so that an apparently risk-seeking choice can be decomposed into how risky the person thinks the act is and how much they expect to gain from it (Weber, Blais, & Betz, 2002; Blais & Weber, 2006). Behavioral tasks instead elicit risk in real time. The Balloon Analogue Risk Task (BART) asks participants to inflate a virtual balloon one pump at a time, earning money with each pump but losing the balloon’s accumulated value if it bursts; the average number of pumps on unexploded balloons indexes risk-taking and correlates with self-reported real-world risk behaviors (Lejuez et al., 2002). The first interactive demonstration below is a BART, and it makes concrete the task’s central tension: every pump is individually attractive, yet the sequence of pumps is what destroys the reward.

Balloon Analogue Risk Task

Each pump adds $0.05 to this balloon’s bank, but every balloon has a hidden bursting point. Bank the balloon to keep its value; pump once too often and the whole balloon’s value is lost. The sequence of burst points is fixed, so the task is a test of how far you push, not of luck. Your average pumps on the balloons you banked is the standard BART risk index.

A recurring complication is that these measures disagree. Behavioral tasks such as the BART, delay-discounting procedures, and laboratory gambles correlate only modestly with one another and with self-report propensity scales, which is why no single number captures a person’s “risk-taking” (Frey et al., 2017). The disagreement is itself informative: it tells us that the laboratory task, the questionnaire, and the real-world act each tap a partly different process, and that the construct is a family of related tendencies rather than one underlying quantity.

Risk-Taking as Decision Under Risk

The formal study of choice among options with known probabilities is where risk-taking meets decision theory. Classical expected-utility theory holds that a decision maker should weight each outcome’s utility by its probability and choose the option with the highest expected utility; risk attitude in this framework is nothing but the curvature of the utility function. Prospect theory replaced this normative account with a descriptive one built on three systematic departures from it (Kahneman & Tversky, 1979). First, outcomes are evaluated as gains and losses relative to a reference point, not as final states of wealth. Second, the value function is concave for gains and convex for losses, producing risk aversion when choosing among gains and risk seeking when choosing among losses. Third, and most consequentially, the function is steeper for losses than for gains — loss aversion — so that the pain of losing an amount exceeds the pleasure of gaining the same amount, by a factor usually estimated near two.

Loss aversion and the reference point together explain the framing effect: describing an identical option as a gain or as a loss shifts the majority preference between a safe and a risky alternative, because the description moves the reference point and so flips the relevant branch of the value function (Figner & Weber, 2011). The second demonstration lets the reader take one fixed gamble and watch the predicted choice reverse as the framing changes, with the underlying value computation shown alongside.

Prospect value of a gamble

A single gamble: win the upper amount with the stated probability, lose the lower amount otherwise. Expected value weighs the two outcomes linearly; prospect theory passes each through a value function that is concave for gains and steeper for losses by the loss-aversion factor λ. Where the two verdicts disagree, loss aversion — not the arithmetic of the gamble — is deciding.

The practical upshot is that “risk-taking” in the decision-theoretic sense is not a stable appetite for variance. It is the joint product of where the reference point sits, whether the options are coded as gains or losses, and how steeply losses are weighted — three things that the situation, not the person, largely sets. The same individual will take the risky option under a loss frame and refuse it under a gain frame, which is why Figner and Weber (2011) argue that the question is not who takes risks but when and why.

Development, Age, and the Dual-Systems Model

Risk-taking is not constant across the lifespan. Across many cultures and measures it rises sharply in early adolescence, peaks in the mid-to-late teens, and declines through adulthood — a curve that tracks neither the steady improvement of cognitive control nor the monotonic accumulation of knowledge (Steinberg, 2008; Boyer, 2006). The dominant explanation is the dual-systems (or maturational-imbalance) model: a socioemotional system governing reward sensitivity and sensation seeking matures early, driven by pubertal change, while a cognitive-control system supporting impulse regulation and foresight matures slowly into the twenties. The gap between the two — an accelerator that comes online before the brakes — produces the adolescent peak (Steinberg, 2008; Crone & Dahl, 2012).

A complementary account comes from fuzzy-trace theory, which argues that adolescents often take risks not because they underestimate them but because they reason about them too precisely, trading off quantified risks and rewards where experienced adults rely on a protective, categorical gist — “this is dangerous, don’t” (Reyna & Farley, 2006). The two accounts are not rivals so much as descriptions of different levels: one neural and motivational, the other representational. A meta-analysis of age differences in laboratory risk-taking confirms that adolescents take more risks than adults under some conditions, particularly affectively “hot” ones, while the difference shrinks or reverses in “cold” deliberative tasks (Defoe et al., 2015). The third demonstration plots the two systems across adolescence, letting the reader see how an early-rising sensation-seeking curve and a late-rising self-regulation curve combine into the familiar inverted-U of risk behavior.

The dual-systems model across adolescence

In Steinberg’s account, a socioemotional system driving sensation seeking matures early and overshoots in mid-adolescence, while the cognitive-control system that reins it in matures slowly into the mid-20s. Risk behavior tracks the gap between an aroused reward drive and a still-immature capacity to regulate it — which is why it peaks in adolescence rather than in childhood, when neither system is strong.

Individual Differences and the Structure of Risk Preference

Beyond age, people differ stably in their appetite for risk. The oldest personality construct here is sensation seeking — the pursuit of varied, novel, and intense experiences and the willingness to take risks for their sake — which Zuckerman linked to a common biosocial substrate shared with impulsivity and certain temperamental traits (Zuckerman & Kuhlman, 2000). Demographic differences are also robust: a meta-analysis of more than 150 studies found that men take more risks than women on average, though the size of the gap varies widely by domain and has narrowed in some over time (Byrnes, Miller, & Schafer, 1999).

The deepest recent question is whether these differences cohere into a trait at all. Using a large battery of propensity measures, frequency reports, and behavioral tasks, Frey and colleagues (2017) found that a general risk-preference factor R emerges across measures and is as temporally stable as major personality traits — but that this general factor is captured almost entirely by self-report propensity measures, while behavioral tasks load weakly and disagree with one another. The reconciliation is that there is a real, stable disposition toward risk, but the behavioral tasks that psychologists prize are noisy and domain-bound indicators of it (Mata et al., 2018). This reframes the measurement disagreement of the earlier section: the tasks are not wrong, they are simply weak windows onto a propensity that self-report, for once, measures better.

Worked Example

Consider a single gamble: a 50–50 chance to win \$150 or lose \$100. Its expected value is clearly positive — a rational wealth-maximizer should take it every time.

Expected value. The arithmetic is immediate: (0.5 × \$150) + (0.5 × −\$100) = \$75 − \$50 = +\$25. On expected value alone, the gamble is attractive.

Prospect-theory value. Prospect theory evaluates the same gamble through a value function anchored at the current wealth as reference point. Using the canonical form v(x) = x^0.88 for gains and v(x) = −λ · |x|^0.88 for losses, the two outcomes are valued at v(150) = 150^0.88 ≈ 82.22 and v(100) = 100^0.88 ≈ 57.54. The gamble’s prospect value is then PV = (0.5 × 82.22) − (0.5 × λ × 57.54) = 41.11 − 28.77λ, where λ is the loss-aversion coefficient.

The decision depends on λ, not on expected value. Table 2 evaluates PV at four values of λ. The gamble is attractive only to a decision maker whose loss aversion is mild; at the textbook estimate of λ ≈ 2, and still more at Tversky and Kahneman’s median estimate of λ ≈ 2.25, the positive-expected-value gamble is declined.

The breakeven coefficient is λ = 82.22 / 57.54 ≈ 1.429: below it the gamble is accepted, above it declined. The example makes the section’s point exact. A choice that looks like timid “risk aversion” need not reflect any distaste for variance at all; it can fall directly out of a single parameter, loss aversion, applied to a reference-dependent value function. “Risk-taking” measured by whether a person accepts favorable gambles is, to a large degree, a measure of their λ.

Discussion

The through-line of a century of research is that risk-taking resists reduction to a single disposition. Perceived risk is organized by dread and the unknown rather than by fatality rates; choice under risk is governed by reference points and loss aversion rather than by a fixed appetite for variance; behavioral risk-taking peaks in adolescence for reasons of neural timing rather than ignorance; and the stable individual differences that do exist are measured better by asking people than by watching them gamble. Each level has its own currency, and translating between them — from a BART pump count to a DOSPERT score to a decision whether to merge into traffic — is exactly where the weak correlations live.

This matters practically because interventions pitched at the wrong level fail. Telling adolescents that an activity is dangerous addresses risk perception, but the dual-systems and fuzzy-trace accounts both predict that information will not touch the affective, reward-driven process that actually produces the behavior. Likewise, a policy that assumes people are consistent risk-takers or risk-avoiders — and so targets “risk-prone individuals” — will misfire against the evidence that the same person’s choice flips with the frame. The construct’s fragmentation is not a failure of the science; it is the finding.

Current Directions

Three strands dominate current work. The first is the recovery of a general risk-preference factor from large multi-method batteries. Frey and colleagues (2017) showed that such a factor is as reliable and temporally stable as the Big Five personality dimensions, and that it is best indexed by propensity self-reports rather than by the behavioral tasks the field had favored — a result that has reframed how risk preference is measured and that Mata and colleagues (2018) have developed into a broader case for integrating the psychological and economic views of risk preference.

The second strand is lifespan stability and change. Using a long panel, Josef and colleagues (2016) found that risk-taking propensity declines with age on average but that individuals maintain their rank order over time — risk preference behaves like a trait with a developmental trajectory, not a fixed constant and not pure noise. This work connects the adolescent-peak literature to the adult decline and treats the whole curve as one phenomenon.

The third strand is cross-cultural developmental research. Large multinational studies of adolescence find that the rise in sensation seeking and the lag in self-regulation appear across very different societies, supporting the dual-systems model’s claim to describe a species-typical maturational pattern rather than a feature of any one culture (Duell et al., 2018; Steinberg et al., 2018). Open questions remain about how much the expression of that pattern — which risks, in which settings — is shaped by local opportunity and norms, a question the behavioral measures are only beginning to address.

Common Misconceptions

“Risk-taking is a single personality trait.”

Perceived risk, choice under risk, and risky behavior correlate only weakly, and risk-taking is strongly domain-specific — a person willing in finance may be cautious in health or recreation (Weber, Blais, & Betz, 2002; Frey et al., 2017).

“Declining a favorable bet shows irrational timidity.”

Loss aversion applied to a reference-dependent value function can make a positive-expected-value gamble genuinely unattractive, as the worked example shows; the behavior is systematic, not a lapse of reason (Kahneman & Tversky, 1979).

“Adolescents take risks because they don’t understand the dangers.”

Adolescents often estimate risks as accurately as adults, sometimes more precisely; the peak reflects an imbalance between an early-maturing reward system and a slower-maturing control system, not an information deficit (Steinberg, 2008; Reyna & Farley, 2006).

“Perceived risk just tracks the objective probability of harm.”

Public risk judgments are organized by qualitative features such as dread and unfamiliarity, and can diverge sharply from expert fatality estimates without being simply mistaken (Slovic, 1987).

Glossary

Balloon Analogue Risk Task (BART).

A behavioral measure in which participants inflate a virtual balloon for sequential rewards, losing the accumulated value if it bursts; mean pumps on unexploded balloons indexes risk-taking.

Decision under risk.

Choice among options whose outcome probabilities are known or estimable, the domain of expected-utility and prospect theory.

DOSPERT.

The Domain-Specific Risk-Taking scale, which measures risk-taking propensity, risk perception, and expected benefit separately across financial, health/safety, recreational, ethical, and social domains.

Dual-systems model.

A developmental account attributing the adolescent peak in risk-taking to an early-maturing socioemotional/reward system outpacing a slower-maturing cognitive-control system.

Expected value.

The probability-weighted average of an option’s monetary outcomes; a normative benchmark that prospect theory systematically departs from.

Framing effect.

The reversal of preference between a safe and a risky option when an identical choice is described in terms of gains versus losses.

Fuzzy-trace theory.

A theory holding that mature decision makers rely on categorical “gist” representations that discourage risk, whereas adolescents reason more literally about quantified risks and rewards.

General risk-preference factor (R).

A stable, cross-measure disposition toward risk recovered from large batteries, best indexed by self-report propensity rather than behavioral tasks.

Loss aversion.

The tendency to weigh a loss more heavily than an equal gain, captured by a coefficient λ usually estimated near 2.

Prospect theory.

A descriptive theory of choice under risk built on reference-dependent evaluation, diminishing sensitivity, and loss aversion.

Psychometric paradigm.

Slovic’s method of mapping perceived risk by having people rate hazards on qualitative scales and factor-analyzing the ratings into a low-dimensional risk space.

Reference point.

The baseline, usually the status quo, against which outcomes are coded as gains or losses in prospect theory.

Risk perception.

The subjective judgment of how hazardous an activity or technology is, structured by affect and qualitative features rather than fatality rates alone.

Sensation seeking.

A personality trait denoting the pursuit of varied, novel, and intense experiences and a willingness to take risks for their sake.

Key Researchers

Renato Frey

is a decision scientist at the University of Basel and the University of Zurich whose work established that risk preference shares the psychometric structure of major personality traits, recovering a stable general risk-preference factor across a large multi-method battery.

ORCID

Ralph Hertwig

directs the Center for Adaptive Rationality at the Max Planck Institute for Human Development in Berlin, where his research on experience-based risky choice and the description–experience gap has reshaped how psychologists understand decisions from experience rather than from stated probabilities.

ORCID - Wikipedia

Daniel Kahneman

(1934–2024) was a psychologist whose prospect theory, developed with Amos Tversky, provided the descriptive account of choice under risk that underlies the modern field; he received the 2002 Nobel Memorial Prize in Economic Sciences for the work.

Wikipedia

Valerie F. Reyna

is the Lois and Melvin Tukman Professor of Human Development at Cornell University and the originator, with Charles Brainerd, of fuzzy-trace theory, whose application to adolescent risk-taking reframed risky choice as a matter of gist versus verbatim representation.

ORCID - Faculty page

Paul Slovic

is a professor of psychology at the University of Oregon and president of Decision Research whose psychometric paradigm mapped the structure of perceived risk and showed that lay and expert judgments diverge in systematic, interpretable ways.

ORCID - Wikipedia

Laurence Steinberg

is a Distinguished University Professor of Psychology at Temple University and a leading proponent of the dual-systems model of adolescent risk-taking, whose cross-national research established the adolescent peak as a species-typical developmental pattern.

ORCID - Wikipedia

Elke U. Weber

is the Gerhard R. Andlinger Professor in Energy and the Environment at Princeton University and a co-developer of the Domain-Specific Risk-Taking (DOSPERT) scale, whose work decomposed apparent risk attitudes into perceived risk and expected benefit within distinct domains.

ORCID

Marvin Zuckerman

(1928–2018) was a professor of psychology at the University of Delaware who introduced the construct of sensation seeking and its measurement, and traced its common biosocial substrate with impulsivity and related traits.

Wikipedia

Frequently Asked Questions

What is risk-taking in psychology?

Risk-taking is the voluntary pursuit of behavior whose outcomes are uncertain and include the possibility of loss or harm as well as gain. Psychologists study it at three levels — how dangerous a hazard is perceived to be, what a person chooses when probabilities are specified, and what a person actually does — which correlate only weakly with one another.

Why do the different measures of risk-taking disagree?

Because each taps a partly different process. Risk-perception ratings, laboratory gambles, behavioral tasks such as the BART, and self-report propensity scales load onto a common factor only modestly, so no single number captures a person’s overall riskiness. Recent work finds that self-report propensity measures index the stable underlying disposition better than any behavioral task does.

What is loss aversion, and how does it affect risk-taking?

Loss aversion is the tendency to weigh a loss more heavily than an equal-sized gain, by a factor usually estimated near two. It can make a gamble with positive expected value unattractive: in the worked example, a 50–50 chance to win \$150 or lose \$100 has an expected value of +\$25 yet is declined by anyone whose loss-aversion coefficient exceeds about 1.43.

What is the framing effect?

The framing effect is the reversal of preference between a safe and a risky option when the same choice is described in terms of gains rather than losses. Because prospect theory’s value function is steeper for losses, moving the reference point by rewording the problem flips the majority choice, even though the options are objectively identical.

Why does risk-taking peak in adolescence?

The dual-systems model attributes the peak to a maturational imbalance: a reward and sensation-seeking system matures early under pubertal influence, while the cognitive-control system that regulates impulses matures slowly into the twenties. The gap between an early accelerator and late brakes produces the mid-to-late-teen peak, which appears across many cultures.

Do adolescents take risks because they underestimate danger?

Generally no. Adolescents often estimate risks as accurately as adults, and fuzzy-trace theory argues they sometimes reason about them too literally, weighing quantified risks and rewards where adults apply a protective categorical gist. The peak reflects motivational and developmental timing, not an information deficit.

Is there a general trait of risk preference?

Yes, with a qualification. Large multi-method studies recover a general risk-preference factor that is as temporally stable as major personality traits, but it is captured almost entirely by self-report propensity measures; behavioral tasks are noisy, domain-bound indicators of it.

Are men bigger risk-takers than women?

On average men take more risks than women, but the effect is domain-dependent and varies in size: a meta-analysis of over 150 studies found a consistent but moderate male tendency that is large in some domains and negligible in others, with some gaps narrowing over time.

References