Uncertainty, Not Just Thrills: How Sensation Seekers Weigh Risk Differently

What looks like a hunger for excitement may actually be a different relationship with uncertainty. A new two-part study published in Communications Psychology reports that sensation-seeking is not simply the pursuit of stimulation itself, nor a general tolerance for unpredictable outcomes, but a specific interaction between the two: how much a person values expected stimulation changes whether they tolerate or avoid uncertainty about that stimulation. Crucially, the same research shows that impulsivity — a trait often lumped together with sensation-seeking — shapes choice through a different and partly opposite computational route.

Sensation-seeking has long been defined as the pursuit of varied, novel, complex and intense experiences, even when those experiences carry potential costs. It is linked to a wide range of behaviours, from alcohol and substance use to antisocial conduct, but also to positive outcomes such as greater psychological well-being and constructive risk-taking. This duality has long puzzled researchers: the same trait can support healthy exploration in one setting and costly risk-taking in another. The new work suggests a possible reason — the consequences may depend on how stimulation value and uncertainty are jointly weighted in a given decision environment.

The study, led by Ern Wong, Tobias U. Hauser, Pietro Pietrini and Charley M. Wu, tackled a stubborn confound. In real life and in most laboratory tasks, options that promise stimulation are also uncertain: novel, exciting or poorly understood things tend to have unpredictable outcomes. Someone who repeatedly chooses a stimulating option might be attracted to the stimulation itself, or might simply be more willing to accept uncertainty, or both. Because these two quantities are usually correlated, standard designs cannot tell them apart.

The first experiment reanalysed previously published data from an operant sensation-seeking task in which participants chose between abstract stimuli associated with monetary rewards and probabilistic, non-painful electric stimulation to the hand. The researchers fit hierarchical Bayesian learning models that tracked, trial by trial, each participant’s beliefs about the probability of stimulation and the uncertainty around that probability. Because the original design made expected stimulation and uncertainty highly correlated, the team compared two single-process models: one assigning a bonus to expected stimulation and one assigning a bonus to uncertainty. Model comparison using leave-one-out cross-validation favoured the uncertainty model decisively (LOOIC 6712 versus 7673 for the expected-stimulation model). In that winning model, higher sensation-seeking scores were associated with less negative uncertainty weights — meaning that whereas most participants avoided uncertain stimulation, high sensation-seekers were more willing to sample it, even though the uncertainty carried no useful information about reward.

This reanalysis was exploratory and could not decisively separate the mechanisms. The second experiment was therefore preregistered on the Open Science Framework and designed to break the collinearity. A total of 619 participants were recruited through Prolific; after exclusions for failed attention checks and below-chance learning, the final sample comprised 527 adults aged 18 to 80. In a web-based adaptation of the task, participants completed four rounds, each split into a reward-learning phase with full feedback and a test phase in which they could self-administer a brief coloured spinning animation. Six objects per round were assigned reward levels and stimulation probabilities of 0.15, 0.50 or 0.85, with the two factors independent by design. Participants were explicitly told that stimulation did not influence rewards, so uncertainty about stimulation was non-instrumental — sampling it could not help them earn more money.

Behaviourally, sensation-seeking and impulsivity were only modestly correlated (r = 0.17), and they showed clearly different signatures. Higher sensation-seeking was associated with broader, more diffuse sampling, measured as higher Shannon entropy over choices (r = 0.12, p = 0.004), and with lower overall reward performance (r = −0.12, p = 0.005). This performance gap was stable rather than widening across the session, and it persisted even after statistically controlling for trial omissions, suggesting it was not simply fatigue or disengagement. Reaction-time analyses also ruled out a speed-accuracy trade-off. Impulsivity, by contrast, showed no reliable association with either performance or choice entropy.

Trial-level mixed-effects logistic regressions then revealed a striking opposition. Participants as a whole reliably tracked reward and avoided uncertainty: larger differences in expected reward increased the odds of choosing an option (β = 3.38), while larger differences in stimulation uncertainty decreased them (β = −1.70). But the traits moderated uncertainty in opposite directions. Higher sensation-seeking weakened avoidance of uncertain stimulation (β = 0.090), whereas higher impulsivity strengthened it (β = −0.099). A direct statistical contrast between the two interaction coefficients confirmed they differed significantly (Δβ = 0.189, p = 0.004). A three-way interaction further showed that the strongest uncertainty aversion occurred in people low in sensation-seeking and high in impulsivity, while for those high on both traits, uncertainty no longer systematically deterred choice at all.

The hierarchical Bayesian model then decomposed choices into parameters for expected reward, expected-stimulation weighting (θ), uncertainty weighting (ω), motor stickiness and softmax temperature (τ), which indexes choice stochasticity. Higher sensation-seeking was linked to greater choice randomness (a positive partial correlation with log temperature, r = 0.137, p = 0.002), suggesting that broader sampling partly reflects a less deterministic choice policy rather than pure information-seeking. Yet stochasticity alone did not capture the full picture. In exploratory follow-up regressions, sensation-seeking showed a negative θ × ω interaction (β = −0.187, p = 0.003): higher sensation-seeking was associated with greater tolerance of uncertainty when expected-stimulation weighting was low, but with greater avoidance of uncertainty when expected-stimulation weighting was high. Impulsivity showed the opposite sign of interaction (β = 0.147, p = 0.024), and a joint path model confirmed that this θ × ω association differed significantly between the two traits (Δβ = −0.284, p < 0.001).

The authors also probed whether subjective feelings about the stimulation explained these patterns. Participants rated the visual stimulation as pleasant and moderately arousing, and sensation-seeking was positively associated with arousal ratings while impulsivity was associated with lower valence and lower arousal. But neither valence nor arousal predicted the computational parameters, and including them as covariates left the critical trait dissociation intact. In other words, the distinct profiles of sensation-seeking and impulsivity were not explained by how pleasant or arousing the stimulation felt.

The findings carry important caveats. The reanalysis was exploratory and its design correlated the two key quantities; the preregistered experiment used a simplified artificial stimulus, prioritising computational identifiability over ecological richness, so it remains unknown whether the same mechanisms govern clinically salient forms of sensation-seeking such as gambling, substance use or social risk-taking. Temperature-like parameters can also absorb multiple sources of unexplained variability, which is why the authors interpret the stochasticity effect as evidence for a broader, less deterministic choice policy rather than pure random exploration. Both experiments used adult samples, so no conclusions about developmental change can be drawn. Still, the framework generates testable predictions: environments such as gambling venues, where salient audiovisual cues raise the subjective value of stimulation, might bias high sensation-seekers toward persistent engagement with predictable but costly options — a speculative interpretation the authors flag for future experimental work. What the study establishes now is more fundamental: sensation-seeking and impulsivity do not merely differ in how much risk people take, but in how the value of stimulation and the weight of uncertainty are combined in the computations that drive every choice.

Subject of Research: Computational mechanisms separating sensation-seeking from impulsivity in reward-guided choice under stimulation uncertainty

Article Title: Stimulation uncertainty reveals distinct mechanisms of sensation seeking and impulsivity

Article References: Wong, E., Hauser, T. U., Chandrasekaran, A., Pietrini, P., & Wu, C. M. (2026). Stimulation uncertainty reveals distinct mechanisms of sensation seeking and impulsivity. Communications Psychology, 4(1), Article 141. https://doi.org/10.1038/s44271-026-00525-5

Image Credits: AI Generated

DOI: 10.1038/s44271-026-00525-5

Keywords: sensation-seeking, impulsivity, uncertainty, decision-making, computational modelling, reinforcement learning, exploration, risk-taking, Bayesian modelling, behavioral economics, personality traits, Stimulation