Older adults are relatively more susceptible to impulsive social influence than young adults

The study examines how age influences susceptibility to social influence on intertemporal choice, specifically whether exposure to others who are more impulsive or more patient shifts individuals’ discounting preferences. Prior work shows people’s impulsivity or patience can be swayed by social contexts, yet it is unclear whether influence tends to increase impulsivity or patience, and how this varies across the lifespan. Adolescents show heightened susceptibility to social influence, but the effects of aging are less clear, with competing hypotheses: socioemotional selectivity theory predicts greater socioemotional goals with age and potentially greater susceptibility, whereas experience and social reasoning could enable older adults to resist influence. Moreover, older adults’ learning is sometimes impaired for self-relevant outcomes but preserved when learning about others, suggesting potential parity with young adults in learning social information. Baseline age differences in discounting are mixed across studies, with meta-analyses suggesting no robust age effect, though individual studies vary. This study tests these hypotheses using a delegated intertemporal choice task, Bayesian modeling, and a signed KL divergence metric to quantify both magnitude and direction of social influence in young vs. older adults, and explores how socio-affective traits relate to susceptibility.

Background literature indicates that social influence shapes preferences across domains including impulsivity and risk. Adolescents are typically more susceptible to peer influence and risk-taking in social contexts, with susceptibility decreasing across adolescence. The impact of aging on social influence is underexplored; theoretical accounts diverge between increased socioemotional goals (potentially increasing susceptibility) and enhanced social reasoning with age (potentially decreasing susceptibility). Learning abilities in older adults can be reduced for self-related reinforcement but preserved for socially relevant outcomes. Baseline intertemporal preferences across adulthood show inconsistent results (older more impulsive, more patient, or no difference), with recent meta-analyses indicating no robust age effect and possible non-linearities. Prior social influence paradigms in discounting and related domains demonstrate that observing others can shift preferences, but directionality (toward impulsivity vs. patience) and age-dependent mechanisms remain unclear. These literatures motivate testing age differences in susceptibility to patient versus impulsive social influence and the role of individual socio-affective traits.

Design: Cross-sectional study with a delegated intertemporal choice task and computational modeling. Two age groups: young adults (18–36) and older adults (60–80).

Participants: Recruited from university databases, social media, and community. Inclusion: normal/corrected vision; no neurological/psychiatric history; older adults scored above ACE-III cutoff (82). Exclusions: current/previous psychology study, neuropsychiatric diagnosis at testing, ACE-III risk, incomplete task. Final sample: N=154 (young: 76; mean age 23.1, 31 men/45 women; older: 78; mean age 70.0, 37 men/41 women). Questionnaire data missing for one participant per age group in relevant analyses. Some participants lacked data for one simulated agent due to overlapping preferences; their data were excluded from corresponding analyses. Compensation: £10/hour plus a random bonus (£1–£10) paid at testing.

Task: Five blocks of 50 trials (Self1, Other1, Self2, Other2, Self3). On Self blocks, participants chose between a smaller-sooner (SS) immediate reward (£1–£20 today) and larger-later (LL) delayed reward (1–90 days; delays dynamically adapted in Self blocks) according to their own preferences. On Other blocks, participants made decisions on behalf of two named others (gender-matched), instructed that these reflected previous participants’ choices (actually simulated). Feedback was provided to learn others’ preferences. The two others were constructed to be more impulsive or more patient than the participant’s baseline. Order of others’ preferences was counterbalanced.

Simulation of others: Using participant’s baseline k (from Self1, KT model), other agents were simulated as hyperbolic discounters with k shifted by +1 (more impulsive) or −1 (more patient) relative to participant baseline, with softmax choice noise (inverse temperature t=1).

Computational modeling: Temporal discounting modeled with hyperbolic discounting. Two primary model families: (i) Preference-Temperature (KT): single discount rate k and inverse temperature t with softmax choice; (ii) Preference-Uncertainty (KU): discount rate as a normal distribution with mean km (impulsivity) and SD ku (preference uncertainty), choices determined deterministically by sampled k relative to indifference. Variants of KU included additional self- or other-noise parameters. Parameters were fit separately per block.

Choice set optimization: In Self blocks, 25 trials used a generative grid near indifference points spanning k in [−4,0]; 25 trials used adaptive Bayesian design, updating a prior (Normal mean −2, SD 1; t=0.3) after each trial to probe the current indifference point. Other blocks used generative method exclusively.

Social influence metric: Signed Kullback–Leibler divergence between posterior distributions of k at consecutive Self blocks quantified both magnitude and direction of shift toward or away from the observed other’s preference. Positive signed D_KL indicates a shift toward the other; negative indicates a shift away.

Model fitting and comparison: Hierarchical Bayesian modeling in Stan (HMC). Four chains, 2000 warm-up + 2000 samples (8000 total posterior draws). Weakly-informative priors; constraints: k and km negative, ku positive, noise parameters within bounds. Separate hierarchical fits for young and older groups; priors reset each block. Model comparison via Leave-One-Out Information Criterion (LOO-IC); winning model: KU without noise. Parameter recovery performed by simulating 160 synthetic participants, fitting the winning model, and correlating simulated vs. recovered parameters (Spearman’s rho > 0.85 for km and ku).

Questionnaires: ACE-III (older adults only), WTAR (IQ), Autism Quotient (AQ), Apathy Motivation Index (AMI), Toronto Alexithymia Scale (TAS), Self-Report Psychopathy (SRP-IV-SF), Questionnaire of Cognitive and Affective Empathy (QCAE), plus task-specific ratings of learning confidence and perceived similarity (0–10 scales). Exploratory factor analysis (maximum likelihood, oblimin rotation) on questionnaire subscales yielded a 3-factor solution: (1) Autistic & alexithymic traits; (2) Psychopathic traits; (3) Affective empathy & emotional motivation.

Statistical analysis: Linear mixed-effects models tested effects of age group, other’s preference, and their interaction on learning accuracy, signed D_KL, and task ratings, with random subject intercepts. Additional models included baseline km and WTAR as covariates and interactions where appropriate. Nonparametric tests (Wilcoxon) for simple/post hoc comparisons; binomial tests vs. chance for learning accuracy; Spearman correlations for associations with factor scores; Z-tests for correlation differences; FDR correction for multiple comparisons; Bayes factors to assess evidence for null effects. Visualizations via ggplot2.

• Learning accuracy: Both age groups learned others’ preferences above chance for impulsive (young mean 83%, older 82%; both P<0.001) and patient others (young 86%, older 85%; both P<0.001). Participants were more accurate for patient than impulsive others (b=0.03, 95% CI [0.01, 0.05], Z=2.71, P=0.007). No age-group difference in learning accuracy (b=−0.01, 95% CI [−0.04, 0.01], Z=−1.22, P=0.22, BF01=1.56). Older adults reported lower confidence in learning (b=−0.59, 95% CI [−1.00, −0.18], Z=−2.82, P=0.005).
• Baseline discounting: Winning model was KU without noise (lowest LOO-IC). No credible age differences in baseline km (young mean [SE] = −4.79 [0.22], older = −5.16 [0.25]; W=3243, Z=−1.01, r=0.08, P=0.314, BF01=3.47) or ku (young 1.37 [0.06], older 1.47 [0.06]; W=2481, Z=−1.74, r=0.14, P=0.081, BF01=2.31).
• Susceptibility to social influence (signed D_KL): Significant interaction of age group × other’s preference (b=−0.56, 95% CI [−0.93, −0.20], Z=−3.03, P=0.002). Older adults were more influenced by impulsive others than young adults (W=1861, Z=−2.67, r(140)=0.22 [0.06, 0.38], P=0.008). No age-group difference in susceptibility to patient influence (W=2723, Z=−1.15, r(138)=0.10 [0.01, 0.25], P=0.252, BF01=3.30). Samples: young impulsive N=68, young patient N=72, older impulsive N=74, older patient N=68.
• Within-group patterns: Older adults showed equal susceptibility to impulsive vs. patient others (V=886, Z=−1.03, r(62)=0.13, P=0.305, BF01=5.49). Young adults were more influenced by patient than impulsive others (V=469, Z=−3.82, r(62)=0.48 [0.27, 0.66], P<0.001). Learning accuracy did not correlate with magnitude of shift (all |r|<0.14, all P>0.27).
• Overall influence: Across groups, signed D_KL differed from zero for both impulsive (grand median 0.12; W=6832, Z=−3.57, r(152)=0.30 [0.15, 0.45], P<0.001) and patient others (grand median 0.37; W=8624, Z=−7.67, r(152)=0.65 [0.53, 0.75], P<0.001). Participants reported feeling more similar to patient than impulsive others (b=1.20, 95% CI [0.53, 1.78], Z=3.62, P<0.001).
• Trait moderators (factor analysis): In older adults, susceptibility to impulsive influence positively correlated with the ‘Affective empathy & emotional motivation’ factor (Spearman rs(71)=0.29 [0.06, 0.48], P=0.014; FDR-corrected P=0.043); not significant in young (rs(66)=−0.13, P=0.30, BF01=5.33). This correlation was significantly stronger in older than young adults (Z=2.45, P=0.014). No correlation of this factor with patient influence in either group. In older adults, susceptibility to patient influence positively correlated with ‘Autistic & alexithymic traits’ (rs(66)=0.34 [0.11, 0.54], P=0.004; FDR P=0.013), not in young; the between-group difference was significant (Z=−2.27, P=0.023). Psychopathic traits did not explain susceptibility in either group (all BF01 > 5.20).
• Controls and robustness: Results unchanged when controlling for baseline km, WTAR IQ, ACE attention/memory (older only), order of others’ preferences, and after outlier checks. Model-free analyses replicated core behavioural patterns.

The research shows that aging is associated with a selective increase in susceptibility to impulsive social influence in intertemporal choice. Despite equivalent learning accuracy for others’ preferences and no detectable age differences in baseline discounting parameters, older adults shifted their preferences more toward impulsive others than young adults, while both groups were similarly influenced by patient others. This suggests that older adults’ vulnerability is not due to deficits in learning social information or baseline impulsivity, but rather to differences in how social cues are integrated into value representations. The findings align with accounts proposing age-related changes in socioemotional motivations and sensitivity to social rewards. Importantly, within older adults, higher affective empathy and emotional motivation were associated with greater susceptibility to impulsive influence, indicating that socio-affective traits can heighten risk for socially driven impulsive decisions. These insights have implications for understanding age-related susceptibility to misinformation and social persuasion, and underscore the importance of considering the content and valence of social influence (impulsive vs. patient) when assessing conformity across the lifespan.

This study demonstrates that older adults are relatively more susceptible than young adults to impulsive social influence on intertemporal preferences, while showing comparable susceptibility to patient influence. The effect occurs despite similar learning accuracy and no differences in baseline temporal impulsivity, and is linked in older adults to higher affective empathy and emotional motivation. The findings advance understanding of how age and socio-affective traits shape social conformity in value-based decision-making, with potential implications for interventions to mitigate harmful social influence. Future research should: (1) examine broader forms of social influence beyond economic preferences; (2) employ longitudinal designs including mid-life samples to capture non-linear lifespan trajectories; (3) incorporate social-cognitive mechanisms (e.g., theory of mind, metacognition) into computational models; and (4) test more ecologically valid, real-world contexts of social influence.

• Domain specificity: The task captures a specific form of social influence in economic intertemporal choice; generalization to other domains may be limited.
• Cross-sectional design: Limits causal inference about developmental trajectories; longitudinal studies including mid-life are needed.
• Modeling scope: Computational models did not include explicit social-cognitive constructs (e.g., theory of mind, metacognition) that may explain why and how influence occurs.
• Ecological validity: The abstract laboratory design may not capture complexities of real-world social influence scenarios.