The databases listed below are freely available to researchers who intend to conduct non-profit, academic research. Researchers who download the databases should use the stimuli for non-profit research only and should acknowledge the proper sources of the stimuli and any references relevant to the data set. The relevant references are listed in the description of each database.
The first set of databases consists of synthetic faces parametrically manipulated to vary on their perceived value on social dimensions such as trustworthiness and dominance. These faces were generated by data-driven computational models. The models for manipulating face shape were first described by Oosterhof & Todorov (2008), and later extended to face reflectance by Todorov & Oosterhof (2011). Todorov & Oh (2021) provide a detailed description of the computational methods. Research manuscripts using the synthetic face stimuli should cite all of these papers, including papers specific to the validation of the respective database of faces.
300 randomly generated faces used to build the models in Oosterhof & Todorov (2008) and in Todorov & Oosterhof (2011). Ratings of the 300 faces on 9 trait dimensions are also included.
525 faces manipulated on face shape: 25 (face identities) x 3 (trait dimensions: perceived dominance, threat, and trustworthiness) x 7 (parametric face manipulations, ranging from -3 to +3SD with a step of 1SD). These databases were validated by Oosterhof & Todorov (2008).
490 faces manipulated on face shape and orthogonally on perceived trustworthiness and dominance: 10 (face identities) x 7 (parametric face manipulations on perceived dominance, ranging from -3 to +3SD with a step of 1SD) x 7 (parametric face manipulations on perceived trustworthiness, ranging from -3 to +3SD with a step of 1SD). These orthogonal models were created by Oosterhof & Todorov (2008).
3,675 faces manipulated on face shape and reflectance: 25 (face identities) x 7 (trait dimensions: perceived attractiveness, competence, dominance, extroversion, likability, threat, and trustworthiness) x 7 (parametric face manipulations, ranging from -3 to +3SD with a step of 1SD) x 3 (face race: Asian, Black, White). These databases were validated by Todorov, Dotsch, Porter, Oosterhof, & Falvello (2013).
13,125 faces manipulated on face shape and reflectance: 25 (face identities) x 7 (trait dimensions: perceived attractiveness, competence, dominance, extroversion, likability, threat, and trustworthiness) x 25 (parametric face manipulations, ranging from -3 to +3SD with a step of 0.25SD) x 3 (face race: Asian, Black, White). These databases were validated by Todorov, Dotsch, Porter, Oosterhof, & Falvello (2013).
4,000 faces used to build a model of attractiveness by Said & Todorov (2011). Text files, data files, and python and Matlab scripts are also included.
1,400 faces manipulated on face shape and reflectance by gender-specific models built by Oh, Dotsch, Porter, & Todorov (2020): 25 (face identities) x 2 (gender models: for males and females) x 2 (trait dimensions: perceived dominance and trustworthiness) x 7 (parametric face manipulations, ranging from -3 to +3SD with a step of 1SD) x 2 (face gender: male and female).
350 faces manipulated on perceived competence controlling for attractiveness: 25 (face identities) x 7 (parametric face manipulations, ranging from -3 to +3SD with a step of 1SD) x 2 (models: attractiveness-subtracted and attractiveness-orthogonal). The models were built and validated by Oh, Buck, & Todorov (2019).
The second set of databases consists of ratings of real faces. The ratings of faces from the Karolinska institute were used by Oosterhof & Todorov (2008) to derive a dimensional model of social judgments from faces. They were also used by Oh, Dotsch, Porter, & Todorov (2020). Todorov & Oh (2021) provide a detailed analysis of these ratings and show how the dimensional model generalizes across the world. Research manuscripts using these ratings should cite all of these papers, as well as the source of the faces. The ratings of politicians’ faces were used by Todorov, Mandisodza, Goren, & Hall (2005), Ballew & Todorov (2007), and Olivola & Todorov (2010) to predict electoral success. Note that all these ratings simply reflect perceptions and should not be interpreted as indicating actual characteristics of the individuals whose faces were rated. About the inaccuracy of impressions from faces, see Todorov (2017), Todorov, Olivola, Dotsch, & Mende-Siedlecki (2015), Todorov, Said, & Verosky (2011), and Olivola & Todorov (2010).
Ratings of 66 faces from the Karolinska database on 14 trait dimensions.
550 photos of US politicians who competed either in a gubernatorial race (248) or in a house race (302). The database also contains the politicians’ perceived competence from their photos, as measured in a forced choice competence judgement of participants unfamiliar with the politicians. As such, these judgments simply indicate perceptions and are in no way indicative of the actual competence of the politicians.
The third set of databases consists of images of 126 novel 3D objects. These objects were generated by Kurosu & Todorov (2017) and used to study aesthetic judgments.