The motive cocktail in altruistic behaviors

The study investigates what motivates altruistic third-party punishment (3PP) and third-party helping (3PH), behaviors where an unaffected observer sacrifices personal resources to punish a transgressor or help a victim. Prior accounts emphasize either strategic, future-oriented benefits (signaling trustworthiness or deterring harm) or strong reciprocity, where individuals sanction norm violations without expected personal gain. A normative framework of utility maximization that quantifies inequality (for example, Fehr–Schmidt) can unify diverse prosocial behaviors, and neuroimaging supports human sensitivity to inequality. However, most studies examine only one or two motives at a time, impeding unified prediction across contexts and individuals and making motives hard to disentangle due to overlapping predictions. This work extends the normative framework by proposing that altruistic behavior is jointly driven by multiple socioeconomic motives and tests whether a comprehensive ‘motive cocktail’ can explain detailed behavioral patterns across many conditions, including interactions among inequality, costs, and intervention impact.

Two theoretical traditions have explained third-party interventions: (1) strategic accounts highlighting long-term benefits via signaling or deterrence, and (2) strong reciprocity accounts focusing on norm enforcement absent personal gain. A prominent normative approach quantifies inequality as losses within utility (Fehr–Schmidt), explaining altruistic punishment and helping, with neural evidence for inequality coding. Yet prior empirical work typically contrasts a few motives, assumes exclusivity, and employs designs where motives yield similar predictions, limiting identifiability and synthesis. The present work aggregates established motives—self-interest (SI), self-centered inequality aversion (SCI; disadvantageous and advantageous), victim-centered inequality aversion (VCI), efficiency concern (EC), and reversal preference (RP)—and proposes new compound motives, inequality discounting (ID), to capture interactions with cost, thereby addressing gaps in identification and integration across prior studies.

Two experiments implemented a third-party intervene-or-watch task that factorially varied key variables to dissociate motives. Participants always held 50 tokens per trial and observed a dictator-game outcome (transgressor vs victim: 50:50, 60:40, 70:30, 80:20, 90:10 with ±2 jitter). They then received an intervention offer with specified cost (10, 20, 30, 40, 50) and impact ratio (1.5 or 3.0), in either a punishment scenario (reducing the transgressor’s payoff by ratio×cost) or a helping scenario (increasing the victim’s payoff by ratio×cost). They chose to accept (intervene) or reject (no intervention). Each participant completed 300 trials: 5 inequalities × 5 costs × 2 ratios × 2 scenarios × 3 repetitions, arranged in six blocks (three punishment, three helping). Decision windows were self-paced; visual feedback followed responses. Experiment 1 (laboratory): N=157 university students (mean age 21.24±2.56), compensated ¥60–120. The design included attention to task comprehension (instruction quizzes, practice) and post-task personality questionnaires (Social Value Orientation, MACH-IV, Interpersonal Reactivity Index). Generalized linear mixed models (GLMMs; binomial link) quantified main and interaction effects of scenario, inequality, cost, and ratio on intervention probability P(yes). Linear mixed models examined decision times. Behavioral modeling: A series of models incrementally incorporated motives into utility for yes/no options, mapped to choices with a softmax. Models included: (1) baseline (coin flip), (2) SI, (3) SI+SCI (α disadvantageous, β advantageous), (4) SI+SCI+VCI (γ), (5) +EC (ω), (6) +RP (κ; rank reversal preference), and (7) +ID (η_no inaction ID, η_yes action ID) forming the full seven-motive cocktail. Model fitting used maximum likelihood with extensive multi-start optimization and comparison via AICc and protected exceedance probability (PEP). Model recovery and parameter recovery validated identifiability. K-means clustering on behavioral patterns identified participant subtypes; associations with personality measures were examined via (partial) correlations with FDR correction. Experiment 2 (online preregistered): N=1,258 university students (from >60 countries), on Prolific with the same task coded in PsychoPy/PsychoJS. Preregistration: OSF (https://osf.io/gcsqp). Attention checks (12 questions across blocks) ensured engagement; 107 participants below 75% accuracy were excluded from model-based analyses but included in model-free summaries as specified. Sample size determination used model-based power analysis targeting the inequality×cost×ratio interaction. Additional models included a lapse-augmented cocktail (Pmin, Pmax) and a simple-response logistic model for participants showing low-dimensional or random response patterns. Group-level modeling and clustering paralleled Experiment 1, with comparisons of clusters and replication of GLMM effects.

• Main effects (Experiment 1, GLMM): Participants preferred helping over punishment (mean P(yes): help 0.25 vs punish 0.18; scenario coefficient b = −1.22, 95% CI [−1.64, −0.80], P < 0.001). Intervention increased with more extreme transgressor–victim inequality (b = 1.61, 95% CI [1.40, 1.81], P < 0.001) and higher impact-to-cost ratio (b = 0.82, 95% CI [0.62, 1.01], P < 0.001), and decreased with higher cost (b = −2.12, 95% CI [−2.37, −1.86], P < 0.001).
• Interaction effects (Experiment 1, GLMM): scenario×ratio (b = −0.39, 95% CI [−0.47, −0.30], P < 0.001), inequality×ratio (b = −0.08, 95% CI [−0.14, −0.02], P = 0.017), cost×ratio (b = −0.08, 95% CI [−0.14, −0.02], P = 0.015), and a three-way inequality×cost×ratio (b = −0.21, 95% CI [−0.27, −0.15], P < 0.001), indicating nonlinear utility interactions beyond simple additive motives.
• Model comparison: The full seven-motive cocktail model (SI+SCI+VCI+EC+RP+ID) best fit choices (lowest AICc; PEP > 99.99% among seven models). Model recovery showed no misidentification of alternative generative models as the full model. Adding each motive class improved fit (ΔAICc reductions). The full model captured main and interaction effects across 100 conditions.
• New compound motives: Inequality discounting (ID) emerged in two forms—inaction ID (η_no), acting as if down-weighting others’ inequality when not intervening as cost rises, and action ID (η_yes), acting as if residual inequality after intervention is down-weighted at higher cost—explaining inequality×cost×ratio interactions.
• Individual differences (Experiment 1): K-means clustering revealed three behavioral types: justice warriors (35%) intervened frequently, sensitive to inequality and low cost; pragmatic helpers (18%) preferred helping, relatively insensitive to inequality/cost; rational moralists (47%) intervened mostly at minimal cost. Parameter differences across clusters: η_yes, κ (RP), and η_no differed significantly (Kruskal–Wallis; H(2)=22.18, P<0.001; H(2)=15.57, P<0.001; H(2)=9.71, P=0.008). Correlations linked higher η_yes to lower ratio sensitivity and higher overall P(yes); higher κ to lower overall P(yes); higher η_no to reduced inequality sensitivity.
• Replication (Experiment 2): All main and interaction effects replicated in N=1,258. The full model again outperformed alternatives (PEP > 99.9%). Six clusters emerged; three mirrored justice warriors (16.60%), pragmatic helpers (17.30%), and rational moralists (27.00%). Three additional clusters showed simple-response patterns (scenario-only, cost-only) or random responding and were better captured by a simple-response model; these participants had lower attention-check accuracy.
• Out-of-sample predictions: Using parameters from the intervene-or-watch task, the model reproduced published findings: (1) 2PP > 3PP and monotonic decrease of punishment with decreasing inequality when allocator-favored (Fehr & Fischbacher), and (2) stronger punishment than helping in robbery-like scenarios (Stallen et al.) when EC is attenuated. Justice-warrior parameters best matched prior datasets.
• Personality associations: Greater self-centered disadvantageous inequality aversion (α) or inaction ID (η_no) associated with more selfishness (SVO) across experiments; partial correlations showed θ (interpreted as VCI/related) remained associated with lower selfishness. Action ID (γ, as labeled in text for empathy association) positively associated with empathy, whereas inaction ID (η_no) showed opposite trends (significant mainly in Experiment 2). In dictator allocations collected pre-task, rational moralists gave least; across clusters, rational moralists were most selfish and justice warriors showed highest empathy.

Findings support that altruistic third-party interventions are guided by a mixture of motives rather than any single dominant motive. The seven-motive cocktail, including two inequality discounting components, explains not only standard main effects (inequality sensitivity, cost sensitivity, helping preference) but also higher-order interactions, indicating nonlinear interplay between self-interest and inequality concerns as costs and impacts vary. The inclusion of efficiency concern (EC) accounts for the robust preference for helping over punishment, while reductions in EC for victims or in severe norm violations explain contexts where punishment exceeds helping and why 2PP > 3PP. The generalized reversal preference (RP) indicates a tendency to reverse unjust status differences in intentional transgressions, contrasting with previously reported rank-reversal aversion under luck-based inequalities. By varying cost exogenously, the design uncovered a three-way interaction previously hard to detect, highlighting that perceived inequality is context-sensitive and modulated by intervention costs. The identification of three behavioral types and their parameter profiles advances understanding of individual differences, linking them to prosocial traits and providing a basis for targeted interventions (e.g., lowering barriers to intervention to reduce inaction ID).

The study introduces and validates a comprehensive motive cocktail model that unifies multiple socioeconomic motives—SI, SCI, VCI, EC, RP, and two forms of inequality discounting—to explain altruistic third-party punishment and helping across 100 conditions. The model best predicts behavior, recovers complex interaction effects, and identifies three robust behavioral types with distinct motive profiles. It generalizes to explain second- versus third-party differences and context-dependent preferences for punishment versus helping, and relates motive parameters to prosocial personality traits. Future work should extend the cocktail to incorporate additional real-world motives (reputation, deterrence, reciprocity, broader social norms), test developmental and neurobiological bases of motives, and bridge laboratory tasks with real-world applications to inform policies that foster prosocial behavior.

• The one-shot, anonymous design minimized reputation and reciprocity, excluding motives that can be influential in real-world repeated interactions.
• The simplified laboratory paradigm may not capture the full richness of social contexts; some interventions involved deception regarding the source of dictator offers (though payoff implementation was real for selected trials).
• Online data included a substantial subset with low engagement exhibiting simple or random response patterns, necessitating separate modeling.
• The model does not include all potentially relevant motives (e.g., explicit reputation concerns, broader norm structures, reciprocity, deterrence dynamics) and assumes specific forms (e.g., sigmoid discounting) for ID.
• Samples were student populations, which may limit generalizability; cultural differences were exploratory.