Generative neural networks for experimental manipulation: Examining dominance-trustworthiness face impressions with...

Area of Science:

• Psychology
• Cognitive Science
• Computer Vision

Background:

• Face impressions are crucial in social interactions, with dominance and trustworthiness being key judged traits.
• Recent research questions the independence of these trait judgments, proposing competing theoretical accounts.
• Existing methods for manipulating face impressions are often data-intensive and less experimentally flexible.

Purpose of the Study:

• To develop a data-efficient deep model for manipulating face impressions.
• To test competing predictions regarding the independence of dominance and trustworthiness judgments.
• To propose an alternative account for the functional roles of these impressions.

Main Methods:

• Developed a novel deep model for face impression generation, requiring significantly less data per dimension.
• Conducted two studies (N=128, N=234) to build and validate the impression models.
• Precisely manipulated dominance and trustworthiness impressions in face photographs.

Main Results:

• Demonstrated the ability to build accurate impression models with 15 times less data than previously required.
• Observed distinct patterns in how cues for one trait influence impressions of the other, challenging existing accounts.
• Achieved larger effect sizes using a broader feature set compared to previous manipulation methods.

Conclusions:

• Proposed an altered account where functional roles necessitate impressions of both dominance and trustworthiness to be based on cues of both traits.
• The developed method offers a versatile tool for researchers to manipulate face stimuli efficiently.
• Facilitates new investigations in face perception with reduced data collection requirements.