Encouraging Diversity in AI Test and Evaluation

The use of Artificial Intelligence (AI) is rapidly increasing with applications touching most aspects of daily life across a range of disciplines, including entertainment, news dissemination, healthcare, and national security. Despite the high risk of sociotechnical impacts, especially when deployed in consequential systems, AI is often unreliable and insufficiently tested. Thus, there is a need to build a diverse workforce with expertise in Test and Evaluation (T&E) and AI. The computer science (CS) degree programs training this future workforce exhibit strong gender disparities which are even stronger within the AI subfield (World Economic Forum, 2021), leading to low diversity and too few AI T&E professionals.

According to the 2023 NSF report “Diversity and STEM: Women, Minorities, and Persons with Disabilities,” women earned 21% of bachelor’s degrees in CS as of 2020 down from 29% in 1995 (National Center for Science and Engineering Statistics, 2023). Per the Pew Research Center, workplace inequities are most likely among three groups of women in Science, Technology, Engineering, and Mathematics (STEM) – in settings where men outnumber women, in computer jobs, and advanced degree holders – which all apply to women in CS research (Pew Research Center, 2018).

The Graduate Research in AI Test and Evaluation (GRAITE) Women Workshop is an initiative led by a team of female faculty and staff from the Virginia Tech National Security Institute (VTNSI) designed to encourage women in their junior or senior years of undergraduate study to pursue doctorates in CS and careers in AI T&E. The goals of the workshop were to provide students with a high-level technical background in AI T&E as well as information about careers in AI T&E with the US Department of Defense. Additionally, the workshop provided information on research more broadly as well as guidance on applying to graduate school including how to fund graduate education and gain real-world experience through research assistantships on government projects.

Theories abound to explain gender differences in STEM education and career fields. The increasing skills shortage in STEM fields has been attributed to the gender gap where many female students select careers in non-STEM areas (Hertweck & Lahner, 2025), negative stereotypes of women’s abilities in mathematics and science (Hunt et al., 2022; Master et al., 2021), and the social emphasis on the “masculine framing of computer science” as power and control (Ashlock & Tufekci, 2024, p. 8). The appendix contains a list of selected works on the gender gap in STEM and suggested interventions. Exploring the underlying theories in the literature helps us understand the reasons for fewer women, thus leading to interventions intended to increase representation for more diverse perspectives and capabilities in STEM fields.

Situated Expectancy-Value Theory

Past studies have examined expectancy-value theory, more recently renamed situated expectancy-value theory (SEVT), to understand student expectations for success (Eccles & Wigfield, 2020; Rosenzweig et al., 2022). Relating those expectations to motivations enables the development of interventions that address educational issues such as the gender gap (Rosenzweig et al., 2022). SEVT posits that students must believe they can succeed and that doing so is important, a combination of task value and expectancy (Hulleman et al., 2016). Expectations for success are individuals’ beliefs that they will do well on a task (Rosenzweig et al., 2022). Task value is the student’s reason for doing an activity (Robinson et al., 2022).

SEVT explains the relationship between motivations and academic outcomes such as performance, engagement, and persistence (Chen et al., 2024). Task values predict education and occupational choices (Chen et al., 2024; Robinson et al., 2022). The type of task is important in understanding how students attribute value to a task. Hulleman et al. (2016) specify four task types: intrinsic (enjoyment of an activity), utility (usefulness), attainment (importance), and cost (perceived adverse outcomes), in the SEVT framework. Furthermore, perception of positive ability on tasks drives students’ interest in careers involving those tasks. Skills gained through enrollment in information and communication technology (ICT) and computing courses increases the likelihood that students select STEM majors (Herweck & Lehner, 2025). Similarly, female teens choose STEM majors if they are above average in ICT skills.

Social Identity and Gender-role Identity

Identity formation occurs early in childhood, where an individual develops multiple identities, or self-concepts, throughout their lifetime (Stets & Burke, 2000). Identity influences human behavior and forms through repeated positive interactions. Multiple identities affect motivation, interest, and learning (Ashlock & Tufekci, 2024) and include personal, social, gender-role, and other forms of identity (Chen et al., 2024). Social identity forms with membership in a group engendering a sense of belonging (Ashlock & Tufekci, 2024). Further, public perceptions, social beliefs, and support structures in computer science are associated with developing a sense of belonging and the ability to identify with roles in the STEM profession.

Gender-role identity incorporates socially constructed attitudes, behaviors, and norms associated with a gender in its formation (Chen et al., 2024). Gender stereotypes influence gender-role and social identities as early as elementary school and bias self-assessments of skills and competencies (Chen et al., 2024). For example, boys perceive math and science tasks as aligned with their identity more than girls do. Gender differences in choices are explained by variations in gender motivations towards STEM tasks and influenced by social environments. Differences highlight the importance of identity-based motivations and interventions to address gender gaps in STEM fields (Chen et al., 2024).

Relating identity with SEVT, attainment value is associated with gender-role identity predicting academic choices (Chen et al., 2024). In addition, the acquisition of attainment task values during social identity development partially explains gender differences. However, research involving attainment value is scarce. When students perceive engagement with a task as consistent with their personal or social identity, the likelihood of choosing the task increases (Chen et al., 2024).

Prevalent in education and social environments, the masculine framing of computer science emphasizes programming and technical aspects associated with men, rather than including concepts of collaboration, anticipating user needs, socialization, and creativity that are more often associated with women (Ashlock & Tufekci, 2024; Sällvin et al., 2024). This framing leads to a lower sense of belonging for women (Hunt et al., 2022) and devalues social computing uses (Ashlock & Tufekci, 2024).

Self-efficacy

Self-efficacy is another theoretical framework for understanding the gender gap in STEM fields. Bandura (2001) explained self-efficacy as the belief in one’s ability to execute the skills necessary to produce desired results. It develops from previous experience, verbal encouragement, physiological reactions to situations or tasks, and learning through observation (Hulleman et al., 2016). Self-efficacy is measured at a task or subject level, such as mathematics or computer science. Increased self-efficacy is associated with excitement and teacher encouragement (Hulleman et al., 2016). Lower self-efficacy is associated with anxiety, cultural and gender stereotypes, and a feeling of not belonging, which decreases persistence in computer science (Hunt et al., 2022).

Studies show that women entering computer science have lower self-efficacy than men and are more likely to revise self-efficacy positively or negatively based on course feedback (Hunt et al., 2022). Low self-efficacy is associated with lower retention rates in a major and impacts career choice. Furthermore, women tend to hold themselves to higher standards and are likelier to self-assess their skills as lower than men, even when grades are the same (Chen et al., 2024; Hunt et al., 2022). Self-assessment of one’s capabilities influences goal aspirations and career choice (Bandura, 2001). Individuals who perceive high skill levels, task success, and support are likelier to experience high expectancy (Hulleman et al., 2016).

Interventions

Understanding the factors that contribute to the gender gap enables the development of interventions that increase student motivations to persist in STEM fields. Literature supports numerous intervention examples related to expectancy values, self-efficacy, and identity. Research has targeted one specific factor for intervention or multiple (Hulleman et al., 2016; Rosenzweig et al., 2022). This section is not meant to be inclusive of all interventions but rather focuses on those more closely related to and supportive of the GRAITE initiative.

Interventions associated with high expectancy include student perceptions of high skill or ability level, successful experiences, support in completing tasks, feedback that effort matters and skills can be increased, high teacher expectations of students, and student perceptions of low task difficulty (Hulleman et al., 2016). Instructors can support student motivation and achievement by assigning challenging work, providing encouragement, promoting autonomy in completing assignments, connecting course content with student interests (Robinson et al., 2022), and specialized courses and mentoring programs (Hertweck & Lehner, 2025). These actions increase self-efficacy and success expectancy. Similarly, clear expectations and observed peer success increase success expectancy (Hulleman et al., 2016).

Suggested interventions that address the masculine framing of computer science include incorporating creative and social aspects of computer science in the undergraduate curriculum to increase retention rates for women. Examples include real-world projects and cases, ethical debates, moral and social justice, peer support networks, and computer design for social improvement (Ashlock & Tufekci, 2024; Sällvin et al., 2024). Other interventions include increased female role models, mentorships, and collaborative learning such as pair programming (Sällvin et al., 2024).

Note that gender issues and disparities extend to underrepresented students, resulting in a lack of ethnic and racial diversity. Barretto et al. (2021) relate many of the same gender gap issues in computer science to AI and machine learning (ML). Their study of diversity in AI and ML found that underrepresented minorities were six times less likely to take traditional introductory courses on this subject. However, an AI and ML course that examines its use for social good aligns with student interests and attracts a more diverse student enrollment. Interestingly, AI and ML research increased student interest.

The GRAITE Women Workshop was sponsored by Google’s exploreCSR program 2023 cycle. The exploreCSR program is dedicated to fostering an inclusive future for computing research (Google Research). The goal of exploreCSR is to support institutions that create opportunities for historically marginalized groups to pursue computing research. The unrestricted gift covered venue facility rental and catering costs of hosting the workshop as well as travel expenses for bringing teams from their universities to attend the workshop in person at Virginia Tech Research Center in Arlington, Virginia.

The VTNSI organizers leveraged involvement in the Department of Defense Senior Military College (SMC) program. As the primary goal of the workshop was to communicate opportunities for students to continue their education in CS PhD programs, VTNSI issued a call for participation from SMC schools that do not have PhD programs – the Citadel, Norwich University, Virginia Military Institute, and University of North Georgia. Norwich University and the University of North Georgia (UNG) responded to the invitation, each bringing a team of five students to the GRAITE Women workshop, as well as a faculty mentor to attend the workshop and to work with the team during the Spring academic semester.

The GRAITE Women Workshop was a one-day event held on January 12th, 2024. The agenda began in the morning with presentations on career paths in AI T&E by Dr. Kristen Alexander, chief learning and artificial intelligence officer in the Office of the Director, Operational Test and Evaluation (The Office of the Director, Operational Test and Evaluation, n.d.), and Dr. Anna Rubinstein, chief of responsible artificial intelligence for the National Geospatial-Intelligence Agency (National Geospatial Intelligence Agency, n.d.). Next, VTNSI faculty gave an overview of AI T&E including background on AI, the use of T&E for responsible AI with concerns like explainability and fairness, performance metrics for AI T&E, and testing for bias. VTNSI faculty also provided information on how to conduct and present research. After this introduction, the teams participated in a walkthrough of a code notebook to explore how the composition of training and test datasets impact research results. Students were given time to explore the data provided and generate and test a hypothesis based upon the data. Each team then presented their initial findings and started formulating their research question for their semester project. The day concluded with an overview of graduate school opportunities presented by faculty from Virginia Tech’s CS department. Students were encouraged to network with each other and with presenters during meals and breaks.

After the workshop, each team identified an AI T&E research question to pursue during the Spring academic semester. This enabled students to apply what they learned at the workshop on a new problem and as a low-stakes, small scale exploration of AI T&E research as a potential career path. The teams presented their semester projects at VTNSI’s 11th Annual Hume Center Colloquium in Blacksburg, VA on April 9th, 2024.

In addition to the in-person engagement at the GRAITE Women Workshop and Hume Center Colloquium, the organizing team created a listserv for the teams to build a cohort of women. This list was used after the workshop to share various opportunities such as conferences and scholarships with the participants. It was also used to contact the VTNSI team with questions regarding their semester projects.

University of North Georgia: Predicting Severity of Aviation Accidents

In response to the invitation for the University of North Georgia (UNG) to participate in VTNSI’s GRAITE Women Workshop, five undergraduate students were selected by faculty to participate. These students met the defined criteria of being in a technology major, having a GPA of 3.0 or higher, and being in the Junior or Senior academic levels with research interest. Participants included students majoring in CS and cybersecurity.

After attending the GRAITE workshop in January 2024, the UNG GRAITE team met weekly to develop a topic and conduct their research. The team selected the research topic of using machine learning to predict the severity of general aviation crashes.

The National Transportation Safety Board (n.d.) aviation accident database, Avall, provided source data for general aviation crashes and served as the training and testing dataset from January 1, 2008, through December 31, 2023. Key variables influencing accident outcomes were investigated to increase understanding of accident severity prediction in general aviation. The resulting research objectives were:

• Develop predictive models using machine learning and neural networks to determine accident injury levels.
• Identify key variables influencing accident severity for improved prediction accuracy.

The UNG GRAITE team’s approach to predictive modeling began with data pre-processing, including data coding, variable selection, and finalizing a valid dataset for testing. Secondly, model construction involved the development of a neural network model using TensorFlow’s Keras API. The neural network model contained one input layer, two hidden layers, and one output layer. Third, the model was trained using an Adam optimizer, a binary cross-entropy loss function, and an accuracy metric. Computational efficiency was optimized with a batch size of 100 while adjusting internal parameters to minimize loss. Finally, model evaluation compared Tanh and ReLU activation functions for predictive modeling (Jadon, 2018). The Sigmoid activation function performed binary classification tasks for the output layer to aid in predicting accident severity. Visualizations were created utilizing MatPlotLib to plot training and validation loss and accuracy over epochs.

The results indicated that after optimization and training, the data consistently achieved 85% accuracy with loss values below 0.06. Although there was no significant performance difference between Tanh and ReLU activation functions, ReLU achieved peak accuracy significantly faster (under 10 epochs) than Tanh (greater than 40 epochs). Predictive modeling in the severity of general aviation accidents is a promising research area.

Figure 1: Poster created by the UNG GRAITE team and presented at the Hume Colloquium

The GRAITE research project enabled the participating UNG students to gain exposure to developing and conducting a research project for testing and evaluating artificial intelligence. The students created a topic, conducted a literature review, developed research methods and models, and analyzed machine learning algorithms. This project greatly enhanced the students’ knowledge of research methods, analytical techniques, and the testing and evaluation of machine learning algorithms, thus, increasing self-efficacy and expectancy. Through attendance at the January workshop and the April Colloquium, UNG students were exposed to various military and governmental careers, student research presentations, and graduate programs. The students gained knowledge of the research process and ideas for future research and connected with fellow student researchers.

Norwich University: AI and the Creation of False Citations

For their semester research project, the team from Norwich University chose the topic Artificial Intelligence and the creation of false citations: A study on national intelligence within generative chatbots. The team consisted of five undergraduate students in the cyber and CS and criminal justice and criminology disciplines. They examined the citation accuracy of five generative AI chatbots; ChatGPT, Gemini, CoPilot, Rytr, and Perplexity; focusing on their reliability in producing credible citations on national intelligence.

The chatbots represented various AI methodologies, providing a comprehensive sample for analysis. The research team specifically explored the incidence of false or altered citations, an issue with significant implications for research integrity, particularly in fields like national intelligence where accurate information is critical. As AI-generated content increasingly supports academic research, understanding its potential for citation inaccuracy is essential. This research highlights the need for robust validation mechanisms within these models to ensure reliable information, especially in complex, sensitive areas where misinformation can have serious consequences.

Using Python’s JupyterHub, the students prompted each chatbot with similar requests for citations on national intelligence. Each bot generated citations, targeting 24 per bot. Manual verification revealed that about half of the sources were only partially accurate, with ChatGPT showing the highest accuracy (despite lacking URL links) and RYTR the lowest, often providing incorrect or outdated links. Citation inaccuracies were frequent across the models, with common issues including altered titles and modified data; this illustrates a tendency within these models to prioritize prompt coherence over factual accuracy. Of the 122 citations generated, only 24 were accurate, 72 partially correct, and 26 entirely incorrect. While most chatbots used authentic sources, they often made unsupported adjustments to citation details.

These findings underscore the importance of validation processes within AI models, especially for fields like national security. The inaccuracies suggest that without rigorous source verification, generative AI tools risk introducing significant misinformation. The research findings also seem to suggest that AI chatbots combine data from a variety of documents to generate new, inaccurate sources.
Participation in the GRAITE Women Workshop and this project provided interdisciplinary students from Norwich University a direct opportunity to engage in AI research.

Figure 2: Poster created by the Norwich University GRAITE team and presented at the Hume Colloquium

This initiative incorporated many interventions specifically targeted to encourage undergraduate women to pursue PhDs in computer science and careers in AI T&E. The selected interventions were based on increasing expectancy-value, self-efficacy, and social identity in computer science. Specifically, we increased interest in AI and research, connected research topics with student interests, and increased student perceptions of skills in problem-solving, AI T&E, and research. The initiative provided successful experiences with research into AI T&E methods. A collaborative learning environment was established through teamwork and a peer support network. Faculty advisors were female role models that set high expectations of participants, encouraged skills development, fostered challenging work, and supported the completion of tasks.

The participants’ feedback supports the initiative’s success in meeting the stated targets and encourages continued efforts. As one participant said,

“The VT GRAITE workshop provided me with an opportunity to learn about research in the technology field and gain a better understanding of what graduate school will look like. As someone planning to do a master’s program in a niche field, this has been an invaluable opportunity to learn how to work in a research team and to start thinking in a more research-oriented way.”

The collaborative learning environment, social connections, and peer support network were valuable to one participant who noted,

“Being able to be a part of the GRAITE program has not only pushed me to test my skills but also has strengthened my connections with other institutions in different states. I loved being able to meet other students and collaborate with them during workshops. I am very thankful to have been a part of something that stretched beyond my university.”

One participant was so inspired that she is subsequently applying to PhD programs. Her feedback highlights the program’s positive impact on her abilities and confidence.

“When I was selected to be one of the first UNG students to be a part of the GRAITE program, I was both overjoyed and nervous. As a college student, adding more projects or other responsibilities to your plate can be daunting. However, this research was unlike any other. While I was a part of the GRAITE program, we focused on graduate research in AI test and evaluation, and it was something I had never done before, despite being a student with a major in cybersecurity and a minor in computer science. We had the opportunity to fly out to the Virginia Tech Research Center, meet with another senior military college, Norwich University, working alongside us, and learned how to begin the project. The workshop encompassed everything from research to project examples, and we even got to play around with a master’s student’s personal project to get a feel for how to create our own!

Once we planned, drafted, and created our research from January to April, it was time to present at Virginia Tech’s main campus at their Annual Research Conference. We could watch dozens of presentations, view everyone’s posterboards, and network with everyone who attended. This opportunity not only broke me out of my shell for researching the unknowns but also gave me the confidence to present my research far beyond the walls of UNG.”

This same participant further endorsed the program and recommended it to students with similar interests, stating,

“I highly recommend this opportunity to anyone who wants to tackle a challenge beyond their studies and grow their knowledge in emerging technical developments.”

Lessons learned

Timing was the primary obstacle as the initiative was intended to be two-phased with a workshop component followed by a research semester culminating in the end of Spring Colloquium. The exploreCSR program applications closed Summer 2023 with notification and receipt of award late Fall, leaving little time for student recruitment and travel planning before the beginning of Winter break and forcing the workshop into late Winter. As a result, some SMC schools did not have the opportunity to participate. Planning challenges for the January workshop were offset by the convenience and accessibility of Arlington, a Washington DC suburb, by air travel. This was not the case for the April Colloquium in Blacksburg. Flights to the small regional airport as well as transportation to VT’s campus were considerably more limited and expensive, requiring budget adjustments.

Feedback from the student teams further emphasized an opportunity for improving the timing of the events with a larger gap between the introductory workshop and research results presentation, preferably occurring in different semesters. Hosting the introductory workshop virtually would reduce costs for the program overall as well as limiting the amount of time students are absent from class; however, in-person attendance enables easier networking and allows students to experience geographic areas boasting many AI T&E jobs, such as Arlington. Last, timing logistics limited the workshop cohort size. Larger team sizes would enable more diversity, but might result in less individual engagement. Increasing the number of teams fielded by a school would require recruitment of more students and additional faculty advisors.

Last, just as both AI and T&E benefit from diversity of thought, diversity led to the success of the initiative. The academic faculty and staff utilized backgrounds in business administration, education, computer science, cybersecurity, criminal justice, AI, and T&E to address organizational logistics challenges, create technical material, and mentor student projects. Additionally, the DoD guest speakers both described non-linear career paths to lead research in AI T&E with STEM PhDs outside of CS.

Future Workshops

Feedback from the inaugural workshop provided evidence that initiatives like the GRAITE Women Workshop are impactful for encouraging diversity in CS and workforce development in AI T&E by creating spaces that promote representation of women and other historically underrepresented groups.

After continued collaboration between VTNSI, UNG, and Norwich University, UNG secured funding from the SMC Cyber Institute for a virtual workshop in March 2025 followed by a September 2025 research symposium at UNG. The Research in AI T&E (RAITE) workshop will leverage the lessons learned from the inaugural workshop while maintaining the theme from GRAITE Women of encouraging historically underrepresented groups in computing to pursue research in the field of AI T&E. All SMCs are invited to bring teams to participate.

We hope that continuing to work towards the goals of the GRAITE Women initiative will result in an increase in women completing PhDs in CS and filling the need for increased diversity in the STEM workforce.

The authors thank the students who participated in the GRAITE Women Workshop initiative. The team from the University of North Georgia included: Jennifer Maaskant, Emily Northcutt, Erika Flores, Destani Fountain, and Riley Freeman. The team from Norwich University included: Jessica Bechtold, Bella Belldina Ray, Emma Gizzi, Elaina Latino, and Isabella Ross.

The authors would also like to give a special thank you to Dr. Anna Rubinstein and Dr. Kristen Alexander for taking time away from their busy schedules to speak to our workshop participants.

This work was partially funded by an unrestricted gift from Google.

Ashlock, J. M., & Tufekci, Z. (2025). Gender differences in computing interest: the role of social constructs in early paths. Computer Science Education, 1–33. https://doi.org/10.1080/08993408.2024.2433354

Bandura, A., Barbaranelli, C., Caprara, G. V., & Pastorelli, C. (2001). Self‐efficacy beliefs as shapers of children’s aspirations and career trajectories. Child Development, 72(1), 187-206.

Barretto, D., LaChance, J., Burton, E., & Liao, S. N. (2021). Exploring why underrepresented students are less likely to study machine learning and artificial intelligence. In Proceedings of the 26th ACM Conference on Innovation and Technology in Computer Science Education, 1, 457-463.

Chen, J., Perez-Felkner, L., Nhien, C., Hu, S., Erichsen, K., & Li, Y. (2024). Gender Differences in Motivational and Curricular Pathways Towards Postsecondary Computing Majors. Research in Higher Education, 65(8), 2013–2036. https://doi.org/10.1007/s11162-023-09751-w

Christensen, M. A. (2023). Tracing the Gender Confidence Gap in Computing: A Cross-National Meta-Analysis of Gender Differences in Self-Assessed Technological Ability. Social Science Research, 111. https://doi.org/10.1016/j.ssresearch.2023.102853

Eccles, J. S., & Wigfield, A. (2020). From expectancy-value theory to situated expectancy-value theory: A developmental, social cognitive, and sociocultural perspective on motivation. Contemporary Educational Psychology, 61, 101859

Hertweck, F., & Lehner, J. (2025). The gender gap in STEM: (Female) teenagers’ ICT skills and subsequent career paths. PLoS ONE, 20(1), 1–18. https://doi.org/10.1371/journal.pone.0308074

Hulleman, C. S., Barron, K. E., Kosovich, J. J., & Lazowski, R. A. (2016). Student motivation: Current theories, constructs, and interventions within an expectancy-value framework. In Psychosocial skills and school systems in the 21st century: Theory, research, and practice, 241-278. Cham: Springer International Publishing.

Hunt, C., Yoder, S., Comment, T., Price, T., Akram, B., Battestilli, L., … & Fisk, S. (2022). Gender, Self-Assessment, and Persistence in Computing: How gender differences in self-assessed ability reduce women’s persistence in computer science. In Proceedings of the 2022 ACM Conference on International Computing Education Research, 1, 73-83.

Jadon, S. (2018). Introduction to Different Activation Functions for Deep Learning. Medium. https://medium.com/@shrutijadon/survey-on-activation-functions-for-deep-learning-9689331ba092

Master, A., Meltzoff, A. N., & Cheryan, S. (2021). Gender Stereotypes about Interests Start Early and Cause Gender Disparities in Computer Science and Engineering. Grantee Submission, 118(48), P. https://doi.org/10.1073/pnas.2100030118

National Center for Science and Engineering Statistics (NCSES). (2023). Diversity and STEM: Women, Minorities, and Persons with Disabilities 2023. Special Report NSF 23-315. Alexandria, VA: National Science Foundation. Available at https://ncses.nsf.gov/wmpd.

National Geospatial Intelligence Agency. (n.d.). Homepage. https://www.nga.mil/

Osunde, O. J., Bacon, L., & Mackinnon, L. (2023). Motivationally Appealing Computer Science e-Learning Games: An Inclusive Design Approach. Electronic Journal of e-Learning, 21(4), 314-327.

Pew Research Center. (2018). Women and Men in STEM Often at Odds Over Workplace Equity https://www.pewresearch.org/social-trends/2018/01/09/women-and-men-in-stem-often-at-odds-over-workplace-equity/

Ren, K., & Olechowski, A. (2020). Gendered professional role confidence and persistence of artificial intelligence and machine learning students. In 2020 ASEE Virtual Annual Conference Content Access.

Robinson, K. A., Lira, A. K., Walton, S. P., Briedis, D., & Linnenbrink-Garcia, L. (2022). Instructional Supports for Motivation Trajectories in Introductory College Engineering. AERA Open, 8(1).

Rosenzweig, E. Q., Wigfield, A., & Eccles, J. S. (2022). Beyond Utility Value Interventions: The Why, When, and How for Next Steps in Expectancy-Value Intervention Research. Educational Psychologist, 57(1), 11–30. https://doi.org/10.1080/00461520.2021.1984242

Sällvin, L., Öberg, L.-M., & Mozelius, P. (2024). Essential Aspects of Gender-inclusive Computer Science Education. In Proceedings of the International Conference on Gender Research (ICGR), 339–346.

Samuel, Y., George, J., & Samuel, J. (2020). Beyond stem, how can women engage big data, analytics, robotics and artificial intelligence? an exploratory analysis of confidence and educational factors in the emerging technology waves influencing the role of, and impact upon, women. arXiv preprint arXiv:2003.11746.

Stets, J. E., & Burke, P. J. (2000). Identity Theory and Social Identity Theory. Social Psychology Quarterly, 63(3), 224–237.

The Office of the Director, Operational Test and Evaluation. (n.d.). Homepage. https://www.dote.osd.mil/

The National Transportation Safety Board. (n.d.). Aviation Accident Database (Avall). https://data.ntsb.gov/avdata

World Economic Forum. (2021). The gender gap in science and technology, in numbers https://www.weforum.org/stories/2021/07/science-technology-gender-gap/

Selected Works