DataSample Avatar
Description
The final sample consisted of adult participants who were either currently engaged in psychological therapy or had received psychological treatment within the previous six months. This inclusion criterion was adopted to ensure that respondents had recent and relevant experience with therapeutic contexts, thereby increasing the ecological validity of the avatar selection task. Participants were recruited using a convenience sampling strategy and completed an online survey administered via a mobile-optimized platform. After providing informed consent, respondents were exposed to a decision task in which they were asked to select a preferred therapist avatar from a set of AI-generated faces. The avatars varied systematically along two primary dimensions—race (e.g., Black and White) and gender (male and female)—as well as in visual realism (photorealistic vs. stylized representations), depending on the experimental condition. Each participant viewed a subset of avatars presented individually on their device. The faces were explicitly described as illustrative representations only, with no differences in underlying behavior or competence attributed to any avatar. Participants were made aware of this aspect to isolate the effect of visual identity cues on selection decisions. The composition of the avatar set was randomized across respondents, such that each participant encountered a unique combination of faces drawn from a larger pool, thereby minimizing potential image-specific biases.
Files
Steps to reproduce
1.Participant Recruitment Recruit adult participants who are either currently engaged in psychological therapy or have received therapy within the past six months. Ensure informed consent is obtained prior to participation. 2. Survey Setup Implement the study using an online survey platform (e.g., Qualtrics), optimized for both desktop and mobile devices. Structure the survey into three main sections: (a) background and attitudes toward AI in mental health, (b) avatar selection task, and (c) post-choice evaluation. 3.Experimental Design Use a between-subjects design manipulating visual realism (e.g., photorealistic vs. stylized avatars). Within each condition, construct avatar sets that systematically vary by race (e.g., Black, White) and gender (male, female). 4.Avatar Pool Construction Create a database of AI-generated faces for each race × gender category (e.g., at least 10 unique faces per category). Ensure consistency in facial expression, lighting, and background across images. 5.Randomization Procedure For each participant, randomly generate a subset of avatars (e.g., 1–2 images per race × gender category) to form a unique choice set. Randomize both the composition and presentation order of avatars to minimize image-specific and ordering effects. 6.Instructions to Participants Inform participants that all avatars are illustrative representations only and do not differ in behavior, competence, or therapeutic ability. This ensures that selection decisions are based solely on visual and identity-related cues. 7.Avatar Selection Task Present avatars individually on the participant’s device. Ask participants to select the therapist avatar they would prefer for a hypothetical psychological consultation. 8.Outcome Measurement Record the selected avatar and classify it according to the intersection of race and gender (e.g., Black woman, White man). This constitutes the primary dependent variable. 9.Post-Choice Measures After the selection, collect self-reported measures on the perceived drivers of choice, including trust, perceived similarity, and the influence of gender and ethnicity (e.g., Likert or 0–10 scales). 10.Control Variables Collect socio-demographic data (e.g., age, gender, income, education) and prior experience or attitudes toward AI in mental health contexts.