Do selfish and prosocial people distort their actions' social impact differently?

Testing whether ambiguity becomes self-serving

People often choose between options without knowing exactly how much good or harm each one does to others. The paper asks whether that ambiguity becomes a tool: do people interpret uncertain social consequences in ways that flatter their own choices, and do selfish and prosocial people do this differently?

The design turns an abstract decision into a measurable estimate. Participants acted as Deciders choosing between two options whose consequences for a Receiver fell within ranges rather than fixed values. Deciders then estimated the Receiver's outcome for each option. Because each range has a midpoint, the deviation of an estimate from that midpoint gives a concrete measure of distortion: positive deviations inflate an option's social benefit, negative ones downplay it.

The authors classify Deciders as selfish or prosocial by the payoff structure of the option they chose, not by their motives. That distinction matters for reading every result: keeping more money is treated as a selfish choice, and the paper explicitly notes this need not reflect uncharitable intent.

Selfish deciders shrink the difference between options

The headline contrast is about the gap between the two options. Selfish Deciders inflated the social benefit of their chosen option while downplaying the alternative, reporting the two options as more similar than prosocial Deciders did. In experiment 1a this difference was b = 0.24, 95% confidence interval 0.14 to 0.34.

Prosocial Deciders moved the other way: they portrayed their choice as comparatively more socially beneficial. They also inflated their implied relative sacrifice, meaning the personal cost they appeared to give up, although this pattern was less consistent across experiments.

The paper stresses that these are aggregate tendencies, not universal behavior. Individual Deciders varied substantially in whether and how they distorted, and the aggregate selfish pattern of reporting both options as 3 and 3 was produced by fewer than 10% of Deciders. The finding is a systematic directional tendency, not a claim that everyone distorts.

Distortion survives before and after the choice

A simple explanation would be post-hoc justification: people pick an option, then bend the numbers to defend it. The experiments test this by measuring estimates before a choice is made and by forcing choices.

Distortions appeared even before people chose and when choices were forced. In experiment 3b, a model using participants' free-choice preferences predicted social outcome estimates better than a model using the forced allocation (BIC 3,832 versus 3,840; Bayes factor 68 in favor of the preference model). That pattern fits distortions tied to underlying social preferences rather than to defending a specific action.

The authors are careful about what this supports. They describe the evidence as more consistent with a social-preference account than a justification account, not as proof that justification plays no role.

The distortion stops at one's own money

If people genuinely believed their distorted estimates, the same beliefs should guide them when their own payoffs are at stake. Experiment 4 tested this by offering Deciders choices between a fixed payoff and an uncertain range with the same structure as the social outcomes.

The link was absent. Selfish Deciders showed no evidence of choosing the low range over its midpoint more often than prosocial Deciders (b = −0.32, 95% CI −0.65 to 0.02, P = 0.064) or neutral Deciders (b = −0.001, 95% CI −0.34 to 0.33, P = 0.994). Prosocial Deciders likewise showed no evidence of preferring the high range over its midpoint relative to selfish Deciders (b = 0.05, 95% CI −0.28 to 0.38, P = 0.757). Bayesian comparisons favored the absence of an overall link (BF01 > 1,000).

The authors read this as a double standard: distortions apply to social consequences but not to equivalent ambiguity about one's own outcomes. That undercuts the idea that these reports are sincere beliefs about the underlying distributions.

Distorted reports can buy trustworthiness

Experiment 5 asked whether these distortions pay off socially. A separate group of observers saw choice scenarios in which both the Decider's and Receiver's outcomes were presented as fully known, with Receiver outcomes drawn from the distortions observed earlier. Each observer rated one Decider's trustworthiness, so every rating is a naive judgment of a single scenario.

Selfish Deciders were rated less trustworthy than prosocial Deciders in the no-distortion scenario with Receiver outcomes of 2 and 4 (b = 33.45, 95% CI 27.09 to 39.81, t = 10.31, P < 0.001). But selfish Deciders gained when outcomes were 3 and 3, the aggregate selfish distortion: relative to no distortion, b = −14.20, 95% CI −20.93 to −7.46, t = −4.13, P < 0.001, meaning the distorted scenario was rated higher.

Prosocial Deciders also gained from their aggregate distortion of 3 and 5 relative to all other outcome scenarios (b = −8.71, 95% CI −13.74 to −3.68, t = −3.39, P < 0.001), but showed no evidence of a difference from no distortion (b = −6.38, 95% CI −13.20 to 0.45, P = 0.067). The trust benefit is clearer for the selfish pattern than for the prosocial one.

Only selfish distortions spread to others

Experiment 6 tested contagion. Followers saw an Influencer's distortion and then reported their own Receiver outcome estimates and made their own choice. The conditions were no distortion, a control, a prosocial-serving distortion, and a selfish-serving distortion.

Followers in the selfish-serving condition overestimated the Receiver outcome of the selfish option relative to both the range midpoint (b = 0.28, 95% CI 0.20 to 0.36, t = 7.06, P < 0.001) and the no-distortion condition (b = −0.19, 95% CI −0.30 to −0.08, t = −3.48, P < 0.001). They also chose the selfish option more often than in the no-distortion condition (b = −0.49, 95% CI −0.88 to −0.10, z = −2.48, P = 0.013), the control (b = −0.80, 95% CI −1.19 to −0.40, z = −3.99, P < 0.001), and the prosocial-serving condition (b = −0.56, 95% CI −0.95 to −0.18, z = −2.85, P = 0.004).

Prosocial-serving distortions did not show the same effect. Selfish choice frequency did not differ between the prosocial-serving condition and the control (b = −0.23, 95% CI −0.62 to 0.15, P = 0.236) or the no-distortion condition (b = 0.07, 95% CI −0.31 to 0.46, P = 0.716). The authors connect this asymmetry to the small share of Deciders producing the selfish aggregate pattern, suggesting a minority may exert disproportionate influence. Why selfish distortions spread more remains an open question in the paper.

What the evidence covers and where it stops

The distortions were measured against range midpoints, which count as distortions only if participants treat the underlying distributions as uniform. The authors note a conceptual replication with explicit instructions about how outcomes were generated, but they cannot fully exclude that some participants assumed non-uniform distributions. The neutral condition, where the self-maximizing option was also the other-maximizing option, is a useful reference but not directly comparable to the main conditions.

The decision contexts were deliberately abstract, designed to capture the structure of ambiguous social consequences rather than any specific real-world setting. That supports a clean dissection of the mechanism but limits direct extrapolation to applied contexts. Participants were English-speaking, predominantly US residents, so generalization to other cultural and linguistic settings is untested. The social judgment measure was limited to trustworthiness; fairness, likability, and moral character were not assessed.

The most consequential asymmetry is the one the paper emphasizes: selfish-serving distortions changed both others' estimates and their choices, while prosocial-serving distortions did not shift average estimates or choices. For readers weighing how ambiguous social impact claims circulate, the paper's priority is that direction of distortion matters more than its mere presence, and the selfish direction carries the demonstrated contagion risk.

Sources used in this answer