The research, conducted by Marius C. Vollberg of the University of Zurich and University of Geneva and Jörg Gross of the University of Zurich, spans nine preregistered online experiments involving 8,769 adults, predominantly residents of the United States. The central finding is striking in its symmetry: both selfish and prosocial individuals distort the social consequences of their choices, but they do so in predictably different directions. Selfish people shrink the apparent harm of their actions, while prosocial people inflate the apparent benefit of theirs. Crucially, when these distorted interpretations are transmitted to others, only the selfish-serving version reliably spreads, nudging those who encounter it toward more selfish behaviour.

The experimental logic was elegantly simple. Participants, cast as ‘Deciders’, chose between two options with fully known personal payoffs: option A yielded 7 points for themselves, option B yielded 5. Each option also carried an ambiguous consequence for another anonymous participant, the ‘Receiver’. Option A gave the Receiver between 1 and 3 points, option B between 3 and 5. Because the Decider’s own payoff was certain while the Receiver’s was a range, the design captured a fundamental asymmetry in social life: personal outcomes are immediate and tangible, whereas social outcomes are shadowy and hard to verify. After choosing, participants estimated the Receiver’s actual outcome for both options, with real money riding on the accuracy of their guesses.

If participants treated each outcome within a range as equally likely, their estimates should cluster around the midpoint and should not depend on what they chose. That is not what happened. Selfish Deciders, those who took the higher payoff, reported the two options’ social consequences as more similar than prosocial Deciders did, effectively portraying their choice as socially harmless. Statistically, the difference was robust: prosocial Deciders’ estimates exceeded selfish Deciders’ by a coefficient of 0.24, with a 95 percent confidence interval of 0.14 to 0.34. Selfish Deciders exaggerated the Receiver’s outcome under the selfish option while downplaying it under the prosocial option, compressing the apparent moral gap between the two choices until their selfishness looked costless.

Prosocial Deciders distorted too, but differently. They exaggerated the social benefit of the option they chose, and, intriguingly, they also exaggerated the benefit of the selfish option they rejected. Rather than painting their sacrifice as heroic against a dark alternative, they construed a world in which even a modest social benefit justified forgoing personal gain, casting themselves as people who need little prompting to do good. In a neutral condition where the self-maximizing option also maximized the Receiver’s payoff, removing any conflict between self-interest and social interest, estimates drifted closer to the true midpoints, though some downplaying of the unchosen alternative persisted.

Sceptics might wonder whether these deviations reflect genuine beliefs about how the ambiguous outcomes were generated rather than motivated distortions. The team addressed this in a conceptual replication that made the outcome-generation process fully transparent: participants watched an animation showing numbers being sampled into an urn and one drawn at random, making clear that the expected value equalled the range midpoint and that their choice could not influence the process. Distortions persisted. The selfish pattern replicated in full, and the overall difference between Decider types remained strong, indicating that the effects cannot be dismissed as private theories about hidden sampling procedures.

Two further experiments probed the psychology behind the distortions. When participants reported their estimates before choosing, the distortions were already present, though attenuated, ruling out a purely post-hoc justification account. More decisive was a forced-choice design in which participants were randomly assigned to implement either the selfish or the prosocial option. If distortions served to justify specific actions, forcing a choice should have reshaped them accordingly. Instead, distortions tracked participants’ underlying social preferences, measured under free choice, better than the choice they had been forced to make; a Bayesian model comparison strongly favoured the preference-based account. Distortions, in other words, are not excuses manufactured after the fact but preference-consistent interpretations of social reality.

Yet the distortions also revealed a telling double standard. In a separate task, participants faced the very same ambiguous ranges, but this time the outcomes determined their own bonuses. Their earlier estimates about others’ outcomes did not predict how they treated equivalent ambiguity when their own money was at stake. People who had reported that a 3-to-5 range favoured the Receiver did not behave as if they believed it when choosing gambles for themselves. This suggests the distortions are not sincere beliefs about underlying distributions but selective interpretations reserved for the social domain, beliefs held too weakly to guide decisions where self-interest is directly measurable.

Do these distortions actually pay off socially? A fifth experiment asked naive observers to rate the trustworthiness of Deciders whose choices were presented with fully known social outcomes drawn from the distortion patterns observed earlier. Selfish Deciders whose outcomes implied both options were socially equivalent, the aggregate selfish distortion, were rated significantly more trustworthy than selfish Deciders in any other scenario, including the undistorted midpoint case. Prosocial Deciders likewise earned higher trust ratings when outcomes matched their aggregate distortion, implying a generous sacrifice. Distortion, when it goes undetected, appears to buy reputation, offering a concrete social function for what might otherwise look like mere self-deception.

The most consequential finding came last. In a final experiment, participants saw estimates attributed to a previous participant, an ‘Influencer’, before making their own choices and estimates. When the Influencer’s estimates were selfish-serving, portraying the selfish option as no more harmful than the prosocial one, Followers adopted those estimates and became significantly more likely to choose selfishly themselves, an effect that held against every comparison condition. Prosocial-serving distortions, though sometimes echoed in individual estimates, produced no reliable shift in average estimates or choices. Because fewer than 10 percent of Deciders actually produce the selfish aggregate pattern, a small minority may exert a disproportionate influence on collective beliefs. The authors suggest that information lowering the psychological cost of selfishness appeals to undecided actors in a way that glorified sacrifice does not. In domains from climate mitigation to foreign aid, the study offers a psychological foundation for why polarized views about the impact, rather than merely the value, of our actions take hold: when social consequences are open to interpretation, people may predictably form divergent interpretations, and the most self-serving of them can spread.

Subject of Research: How selfish and prosocial individuals distort estimates of the ambiguous social consequences of their decisions, and how those distortions spread socially

Article Title: Selfish and prosocial individuals both distort their actions’ impact, but selfish distortions can be more contagious

Article References: Vollberg, M. C., & Gross, J. (2026). Selfish and prosocial individuals both distort their actions’ impact, but selfish distortions can be more contagious. Nature Human Behaviour. https://doi.org/10.1038/s41562-026-02585-3

Image Credits: AI Generated

DOI: 10.1038/s41562-026-02585-3

Keywords: social preferences, decision making, motivated reasoning, selfishness, prosocial behaviour, ambiguity, trustworthiness, social contagion, behavioral economics, moral psychology, Nature Human Behaviour, preregistered experiments