DATABASE
Here are some materials created and data collected for my work. They are available for strictly non-commercial, research purposes. Detailed explanations can be found in the papers under “Citations”. For any inquiries, email me directly.
Person Images – Clothing Manipulation

This image set entails 600+ male person images and visual mask images as well as 74 clothing-only images. These images vary in the upper-body clothing type, rated as appearing either “expensive” or “cheap” by a group of human raters (that is, status impressions have been verified by human ratings to be successfully manipulated via clothing as intended). Sample images below do not show faces for privacy. Actual stimuli entail the faces. The stimuli come with ratings of the person images on competence impressions.
Citation:
Oh, D., Shafir, E., & Todorov, A. (2020). Economic status cues from clothes affect perceived competence from faces. Nature Human Behaviour, 4, 287–293. doi:10.1038/s41562-019-0782-4
Person Images – Face Manipulation

This image set entails 1,400 synthetic (25 females, 25 males) and 1,400 real-life images of individuals (25 females, 25 males) that were made to appear “trustworthy” (or “untrustworthy”), or “dominant” (or “submissive”) to various degrees. Impressions have been verified by human ratings to be successfully manipulated. Both male and female face images are available for both synthetic and real-life images. Synthetic images consist of White faces. Real-life images consist of racially diverse individuals (self-identified White, Black, Asian (broadly defined and not confined to the East Asian category), and Latinx). The face manipulation was done using data-driven, statistical models of gender-specific trustworthiness impressions and gender-specific dominance impressions. The stimuli come with ratings of the person images.
Person images varying on competence impressions without attractiveness confound – https://osf.io/86kfq/
Citation:
Oh, D., Dotsch, R., Porter, J., & Todorov, A. (2020). Gender biases in impressions from faces: Empirical studies and computational models. Journal of Experimental Psychology: General, 149(2), 323-342. doi:10.1037/xge0000638
Person images varying on competence impressions without attractiveness confound – https://osf.io/86kfq/

This image set entails 525 synthetic (25 faces, but some female-appearing after manipulation) and 160 real-life images of individuals (10 female, 10 male, all Caucasian) that were made to appear competent (or incompetent) to various degrees. Impressions have been verified by human ratings to be successfully manipulated. Importantly, the set consists of three subsets of images. Faces of multiple identities were varied by either of 3 face models:
Model 1. Original Competence Impression Model
Model 2. Difference Competence Model (Competence – Attractiveness)
Model 3. Orthogonal Competence Model (Competence ⟂ Attractiveness)
Importantly, Model no. 2 and 3 vary person images on competence impressions while not relying on changes in facial attractiveness. The stimuli come with ratings of the individuals on competence, attractiveness, confidence, and masculinity.
This image set can be useful for example when one tests the effect of face-based competence impressions on behavior while controlling for attractiveness’s impact on competence impressions.
Person images of various levels of trustworthiness impressions without attractiveness confound – https://osf.io/h93cu/
This image set entails 525 synthetic images of individuals (25 male, but some female-appearing after manipulation) that were made to appear trustworthy (or untrustworthy) to various degrees. Impressions have been verified by human ratings to be successfully manipulated. Faces of multiple identities were varied by either of 3 face models:
Model 1. Original Trustworthiness Impression Model
Model 2. Subtraction Trustworthiness Model (Trustworthiness – Attractiveness)
Model 3. Orthogonal Trustworthiness Model (Trustworthiness ⟂ Attractiveness)
Importantly, Model no. 2 and 3 vary person images on trustworthiness impressions while not relying on changes in facial attractiveness. The stimuli come with ratings of the individuals on trustworthiness, attractiveness, approachability, and emotional expression (angry–happy).
This image set can be useful for example when one tests the effect of face-based trustworthiness impressions on behavior while controlling for attractiveness’s impact on trustworthiness impressions.
Person images varying on various facial attractiveness cues– https://osf.io/rqgys/
This image set entails 76 synthetic male face images that vary on the attractiveness cues (1) only on the face shape, (2) only on face reflectance, or (3) both.
300 random Asian faces – https://figshare.com/articles/dataset/300_Random_Asian_Faces/7361270
Model 1. Original Competence Impression Model
Model 2. Difference Competence Model (Competence – Attractiveness)
Model 3. Orthogonal Competence Model (Competence ⟂ Attractiveness)
Importantly, Model no. 2 and 3 vary person images on competence impressions while not relying on changes in facial attractiveness. The stimuli come with ratings of the individuals on competence, attractiveness, confidence, and masculinity.
This image set can be useful for example when one tests the effect of face-based competence impressions on behavior while controlling for attractiveness’s impact on competence impressions.
Citation:
Oh, D., Buck, E. A., & Todorov, A. (2019). Revealing hidden gender biases in competence impressions of faces. Psychological Science, 30(1), 65-79. doi:10.1177/0956797618813092
Person images of various levels of trustworthiness impressions without attractiveness confound – https://osf.io/h93cu/
This image set entails 525 synthetic images of individuals (25 male, but some female-appearing after manipulation) that were made to appear trustworthy (or untrustworthy) to various degrees. Impressions have been verified by human ratings to be successfully manipulated. Faces of multiple identities were varied by either of 3 face models:
Model 1. Original Trustworthiness Impression Model
Model 2. Subtraction Trustworthiness Model (Trustworthiness – Attractiveness)
Model 3. Orthogonal Trustworthiness Model (Trustworthiness ⟂ Attractiveness)
Importantly, Model no. 2 and 3 vary person images on trustworthiness impressions while not relying on changes in facial attractiveness. The stimuli come with ratings of the individuals on trustworthiness, attractiveness, approachability, and emotional expression (angry–happy).
This image set can be useful for example when one tests the effect of face-based trustworthiness impressions on behavior while controlling for attractiveness’s impact on trustworthiness impressions.
Citation:
Oh, D., Wedel, N., Labbree, B., & Todorov, A. (in preparation). Data-driven models of face-based trustworthiness judgments unconfounded by attractiveness.
Person images varying on various facial attractiveness cues– https://osf.io/rqgys/
This image set entails 76 synthetic male face images that vary on the attractiveness cues (1) only on the face shape, (2) only on face reflectance, or (3) both.
Citation:
Oh, D., Grant-Villegas, N., & Todorov, A. (2020). The eye wants what the heart wants: Females’ preference in male faces are related to partner personality preference. Journal of Experimental Psychology: Human Perception and Performance, 46(11), 1328–1343. doi:10.1037/xhp0000858
300 random Asian faces – https://figshare.com/articles/dataset/300_Random_Asian_Faces/7361270

This image set entails 300 synthetic East Asian face images. The 300 images are East Asian "twins" of 300 Caucasian faces in Oosterhof & Todorov (2008). How the original 300 Caucasian images were generated is described in the 2008 paper. In this image set, the FaceGen parameter coordinates of the 300 random Caucasian FaceGen faces are shifted so that the coordinates of the face images center around the average coordinates of the East Asian face in the FaceGen model. The center of the East Asian Faces was calculated from the actual East Asian faces, measured via 3d laser scan. This shift in the coordinates essentially makes the faces on average appear more East Asian.
This image set can be useful for example when one builds a data-driven face evaluation model of Asian faces (e.g., a face model of trustworthiness judgement). See Oosterhof & Todorov (2008) and Todorov & Oosterhof (2011) for details.
This image set can be useful for example when one builds a data-driven face evaluation model of Asian faces (e.g., a face model of trustworthiness judgement). See Oosterhof & Todorov (2008) and Todorov & Oosterhof (2011) for details.
Citation:
Oh, D., Wang, S., & Todorov, A. (2018). 300 random Asian faces. from https://figshare.com/articles/dataset/300_Random_Asian_Faces/7361270.
Oosterhof, N. N. and A. Todorov (2008). The functional basis of face evaluation. Proceedings of the National Academy of Sciences of the USA, 105(32): 11087-11092.
Todorov, A. and N. Oosterhof (2011). Modeling social perception of faces. IEEE Signal Processing Magazine, 28(2): 117-122.
HOME