Our research seeks to discover best practices for using avatars to enhance performance, engagement, and STEM identity development for diverse public middle and high school computer science students. As sites of our research we run workshops in which students learn computer science in fun, relevant ways, and develop self-images as computer scientists.

Our workshops involve eliciting student- generated themes, questions, challenges, and goals. This process includes taking an anti- deficit ideological stance on students and their achievement. We have also developed our own custom platform called MazeStar, used in the workshops, that allows students to explore their own ideas by creating customized games while learning about human-computer interaction, web design, privacy , coding, debugging, and
more (we utilize aspects of the nationally recognized Exploring Computer Science (ECS) curriculum). A core component of
MazeStar is our game for learning programming called Mazzy (see Figure 2) in which play requires
learning building blocks of coding. Ultimately, we approach STEM education and access to high quality, relevant learning opportunities as a social justice issue of our time.

Using qualitative, quantitative, and AI/machine learning analysis techniques, we have already formulated a few best practices and guidelines when it comes to avatar use in education. We have also systematically explored the impacts of different avatar types on users, beginning with distinctions between anthropomorphic vs. non-anthropomorphic avatars, user likeness vs. non-likeness avatars, and other conditions informed by insights from the learning sciences and sociology in crowd-sourced studies (with over 10,000 participants).

Taken together, our studies have revealed that avatars can support, or harm, student performance and engagement. A few notable trends are: 1) ‘role model’ avatars (in particular scientist avatars) are positively effective, 2) ‘likeness’ avatars (avatars in a user’s likeness) are not always positively effective, 3) simple ‘abstract’ avatars (such as geometric shapes) are especially positively effective when the player is undergoing failure, e.g., ‘debugging,’ and 4) ‘successful likeness’ avatars that look like the user when doing well and appear ‘abstract’ otherwise are very positively effective.

Research Areas

Impact Areas


Sneha Veeragoudar Harrell

Aziria Rodriguez

Laurel Carney