Cues driving trait impressions in naturalistic contexts are sparse

Trait impressions are ubiquitous and consequential. Understanding how people form trait impressions is at the core of social cognition. Prior literature primarily relied on constrained designs due to methodological challenges, which may have precluded an ecologically valid understanding. Here, we overcome longstanding methodological challenges by leveraging a range of computational tools to investigate trait impressions in naturalistic context. Using these tools, we quantified comprehensive cues identified from prior theories (facial, bodily, clothing, environmental cues) in naturalistic images, modeled their unique and shared predictions of trait impressions, and manipulated cues realistically in naturalistic images to test causal effects. Across two large-scale, pre-registered studies (N1 = 2,435 representative of U.S. population; N2 = 569), we found that with rich information available, the cues that predicted trait impressions in out-of-sample data were sparse. Predictive cues were distinct for traits that represented different psychological dimensions of social cognition. We confirmed for a subset of cues that these predictions were causal. Variance partition and interaction analyses revealed why predictive cues were sparse: co-occurring cues in naturalistic contexts tended to convey consistent information; predictive cues carried unique information beyond shared consistent information; unpredictive cues played a role by modifying the utilization of predictive cues through interactions. Together, our findings suggest that the mind may have evolved to utilize the naturalistic relations between cues in the real world to simplify what information to attend to when forming trait impressions.