The Racialized Basis of Trait Judgments from Faces

人们通常会根据别人的面部表情来推断他们。面部暗示被用来衡量他人的情绪表达(如高兴或愤怒)，并对他人的特征(如可信度或能力)做出推断。这些判决很重要，因为它们会影响就业、医疗保健或刑事审判等领域的决定。面部暗示也被其他人用来推断一个人的种族。这个项目的重点是，感知者有时可能使用完全相同的面部暗示来推断一个人的特征和他们的种族。因此，根据面部暗示来评估另一个人的特征的感知者，实际上可能是在根据种族的差异来评估那个人。所提出的种族-特征重叠模型表明，相同的面部特征被用来推断种族和特征，这有助于解释从面孔进行特征推断时产生偏差的原因和结果。这项研究的最终目的是更好地了解种族偏见是如何在大脑中发挥作用的，并找出减少这种偏见的方法。这个研究项目调查了人们如何使用种族主义的面部暗示(如肤色)来推断他人的特征。一组实验试图为种族-性状重叠假说建立证据。另一种观点表明，这些影响既受个体差异的制约，也受群体差异的制约。第三组研究如何通过将感知者对特征的表示与他们对种族的表示分离来减少这些重叠效应。种族-性状重叠模型产生了多个新颖的、可检验的预测。例如，该模型表明，这些影响存在于感知者的头脑中，这意味着改变感知者的心理表征应该改变这种重叠。该项目提供了几个方法学上的进步，并对如何在商业和工业中使用面部分析具有重要的影响。该模型有助于解释利用面部线索的机器学习算法中出现的社会偏见的新证据，并指明了减少此类偏见的方法。该奖项反映了NSF的法定使命，并通过使用基金会的智力优势和更广泛的影响审查标准进行评估，被认为值得支持。

People commonly make inferences about others based on their facial expressions. Facial cues are used to gauge others’ emotional expressions (such as happiness or anger) and to make inferences about others’ traits (such as trustworthiness or competence). These judgments are important because they affect decisions in such areas as employment, healthcare or criminal trials. Facial cues are also used by others to make inferences about a person’s race. This project focuses on the idea that perceivers may sometimes use the very same facial cues to draw inferences both about a person’s traits and about their race. As a result, perceivers who are evaluating another person’s traits based on facial cues may actually be evaluating that person based on variations in race. The proposed Race-Trait Overlap model suggests that the same facial features are used to infer race and traits, which helps to account for causes and consequences of biases in trait inferences from faces. The ultimate aim of the research is to better understand how racial prejudice works in the mind, and to identify ways of reducing such biases.This program of research investigates how people may employ racialized facial cues (such as skin tone) to infer others’ traits. One set of experiments seeks to establish evidence for the Race-Trait Overlap hypothesis. Another demonstrates that these effects are qualified both by individual differences and group differences. A third set investigates how these overlap effects can be reduced by decoupling perceivers’ representations of traits from their representations of race. The Race-Trait Overlap model yields multiple novel and testable predictions. For example, the model suggests that these effects exist in the minds of perceivers, which means that changing perceivers’ mental representations should change the overlap. The project offers several methodological advances, and carries important implications for how facial analytics are used in business and industry. The model helps to account for emerging evidence of social bias in machine learning algorithms that utilize facial cues, and points the way to reducing such bias.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.