Abstract
People readily infer personality traits from physical appearance. These judgments can be consequential in various settings, ranging from job applications to judicial settings. Trait inferences from the face have been studied primarily. Here we investigated whether people infer personalities from body shapes. We synthesized 140 (70 female; 70 male) three-dimensional bodies, using the Skinned Multi-Person Linear Model (Loper et al., 2015). In this model, each body is represented as a set of coefficients in a PCA space derived from laser scans of over 1700 real bodies. Participants rated each body on 30 social traits. These traits consisted of positive and negative items from the Big Five Factors of Personality (e.g., conscientiousness: self-disciplined +, lazy -) (Gosling, Rentfrow, & Swann, 2003). First, we visualized the structure of the multivariate body-trait space separately for male and female bodies. The main axes captured the valence (good/bad) and agency (active/passive) of the traits. Positive and negative traits within each Big Five domain were contrasted along the main diagonals and main axes of the space. Second, we predicted trait ratings from body shape coefficients using multiple linear regression with cross-validation. Global trait profiles were predicted accurately from body coefficients (p < .0001), as was a subset of individual traits (16 for males, 15 for females; Bonferroni corrected α = .002). Predictions were most accurate for traits related to extraversion (e.g., enthusiastic, dominant, quiet), conscientiousness (e.g., self-disciplined, disorganized, lazy), and to a lesser extent, openness (e.g., curious, intelligent). Third, we visualized bodies that typified individual traits. This visualization indicated that body weight relates to trait valence and gender-specific shaping (e.g., female, pear-shaped/rectangular; male, wide-shoulder/rectangular) relates to trait agency. The present study takes a first step towards understanding the range, diversity, and reliability of personality inferences made from body shapes.
Meeting abstract presented at VSS 2018