Which facial features are central in impression formation?

Abstract

Which facial characteristics do people rely on when forming personality impressions from faces? Previous research has uncovered an array of facial features that influence people’s impressions. Even though some (classes of) features, such as facial width-to-height ratio or resemblances to emotional expressions, play a central role in theories of social perception, their relative importance in impression formation remains unclear. Here, we model faces along a wide range of theoretically important dimensions. We use machine learning to test how well 31 features predict impressions of trustworthiness and dominance in a diverse set of 597 faces. In line with overgeneralization theory, emotion resemblances were most predictive of both traits. Other features that have received a lot of attention in the literature, such as facial width-to-height ratio, were relatively uninformative. Our results highlight the importance of modeling faces along a wide range of dimensions to elucidate their relative importance in impression formation.

When faces and voices come together: Face width‐to‐height ratio and voice pitch contribute independently to social perception

“…In contrast, these individuals are seen as less intelligent than the average person (Deska et al, 2018b). However, recent findings suggest that the impact of fWHR on dispositional inferences might be smaller and less reliable than previously thought, particularly when considering male faces (see Durkee and Ayers, 2021;Jaeger and Jones, 2020). Thus, the contribution of fWHR to person perception is still a matter of debate.…”

Mapping physical characteristics in face images to social judgements

“…Indeed, several recent studies have demonstrated that physical characteristics that have shown clear and robust effects on social judgements of faces in studies using this method are relatively poor predictors of social judgements when the physical characteristics and ratings are taken from natural (i.e. unmanipulated) face images that vary simultaneously on multiple dimensions (Holzleitner et al, 2019;Jaeger & Jones, 2022;Lee et al, 2021;Said & Todorov, 2011).…”

“…The mostly commonly used approach is bootstrapped regressions, in which only those components that predict the outcome variable consistently across iterations are retained (Holzleitner et al, 2019;Said & Todorov, 2011). An alternative but less widely used approach is to employ regularized regressions (Jaeger & Jones, 2022), which shrink uninformative predictors to small values, retaining informative ones. These are most naturally approached from a Bayesian perspective (Jones & Kramer, 2021), within which prior distributions put very low probability on large estimates, which can be overcome only with strong data.…”

“…However, it is not well suited for assessing the predictive accuracy of data-driven models, since it is highly sensitive to the number of predictors included in the model. Rather than focusing on explained variance, the root-mean-squared error (RMSE) directly assesses the accuracy of predictions made by data-driven approaches and is directly comparable across models (Holzleitner et al, 2019;Jaeger & Jones, 2022). Thus, we recommend using RMSE, rather than the proportion of variance explained, when evaluating data-driven models.…”

Summarized Attribute Preferences Have Unique Antecedents and Consequences

“…Note that the question of learning how much one likes the attribute Reditry is not the same as learning what facial features cause a face to be high or low in Reditry. Our study is focused on the former evaluative process; the latter attributional process is fascinating, too (and better studied; seeJaeger & Jones, 2020;Zebrowitz & Montepare, 2008), but it was not the focus of the current investigation.…”