This National Science Foundation Improving Undergraduate STEM Education funded project integrated data science into six undergraduate STEM courses at a public university, a private university, and a Historically Black College and University. A research–practice partnership involving faculty, graduate students, and an evaluation team developed 12 discipline-specific data science modules that were implemented multiple times and reached over 1,000 students. The modules used real-world datasets, including environmental and traffic data, to embed data science concepts into existing STEM courses and better prepare students for data-driven careers.

Using a mixed methods approach, the team collected instructor interviews, student surveys, pre and post data, and assessment results. Instructors valued the flexibility to tailor content, emphasizing hands-on learning, data visualization, and statistical analysis. Challenges included varied student backgrounds and adjustments during COVID-19.

Students reported increased interest, confidence, and skills in data science after completing the modules. They appreciated real-world applications but requested more guidance and time for complex tools and topics. Overall, the project demonstrates that embedding discipline-specific data science into existing STEM courses is an effective and sustainable way to strengthen data literacy.

Open-ended learning environments (OELEs) are educational settings that encourage students to explore, experiment, and construct their own understanding of STEM concepts. While OELEs can increase engagement and deepen learning, they can be challenging for novice learners who may lack the self-regulated learning (SRL) skills needed to manage their own learning effectively. Recent research highlights the importance of students’ emotions (affective states) and thinking processes (cognitive processes) in determining performance in these environments, but the connection between these factors has not been fully explored.

In this study, we examined how students’ emotional states relate to the consistency of their cognitive strategies while they worked on building causal models of scientific processes in the XYZ OELE. Our findings show that high-performing students use more coherent cognitive strategies, and their emotional states differ significantly from those of low-performing students, reflecting the impact of cognitive strategy coherence on affect. This work provides new empirical insights into the interplay between thinking and emotions in OELEs, highlighting how students’ understanding of their own learning processes shapes their emotional experiences in STEM learning.

Data science skills are increasingly important for undergraduate students in engineering and science, but it can be challenging to integrate these skills effectively into existing courses. This study examined the impact of embedding discipline-specific data science modules into undergraduate STEM courses at three universities in the United States, using a collaborative research-practice approach. Researchers analyzed survey responses from 877 students, instructor grades, and interviews across six courses to understand changes in student perceptions of data science across different demographics, academic levels, and disciplines, as well as how students’ self-assessments compared with instructor evaluations.

The results showed that students’ confidence and perception of their data science abilities improved significantly after completing one or more modules, regardless of the course or university. Student self-assessments closely matched instructor evaluations, indicating consistent recognition of skill development. Students reported benefits such as connecting data science to real-world applications and career relevance, though they also noted challenges with learning new data analysis tools and navigating varying prior experience levels. These findings provide valuable guidance for educators seeking to integrate data science into STEM curricula in ways that are meaningful, engaging, and practical.

Figure 1.ĚýResults of the mixed-effects models indicating the effect of taking one or more modules on constructs. Adj. Diff indicates the adjusted difference (i.e. difference after controlling for course and institutions) between pre- and post-survey scores, Eff. Size the effect size calculated using Westfall, Kenny, and Judd () method,ĚýZĚýthe test statistic,ĚýPĚýtheĚýp-value, andĚýNĚýthe sample size. Blue and orange dots represent pre- and post-survey mean adjusted scores, respectively, with horizontal red lines indicating 95% confidence intervals.

Testing is an essential part of software engineering, and many tools have been created to help students learn how to test effectively. Prior research has shown that a simple testing checklist can improve learning, but it does not fully address the challenge students face when writing test code that reflects their actual design or intentions. At the same time, generative AI tools like ChatGPT are emerging as promising new forms of software assistance.

In this study, we examined how different support tools—a checklist, ChatGPT, or both—affect students’ performance in unit testing. We worked with 42 students and found that whether used individually or together, these tools produced similar results in terms of testing performance. Students generally preferred the checklist but recognized ChatGPT’s value in speeding up task completion and helping with programming language difficulties.

However, while ChatGPT showed potential benefits for testing education, it did not fully solve the challenges identified in earlier work. In addition, students often engaged only superficially with ChatGPT’s responses, which could limit their deeper understanding of new concepts and reduce opportunities for critical thinking.

Based on these findings, we provide recommendations for both students and instructors on how to adapt learning and teaching strategies in the AI era, as well as insights into the evolving role of AI in education.

Reading assessments are important for improving students’ understanding, but many educational technology tools focus mostly on final scores, offering little insight into how students actually read and think. This study explores using multiple types of data—including eye-tracking, test results, assessment content, and teaching standards—to gain deeper insights into reading behavior. We use unsupervised learning techniques to identify distinct reading patterns, and then a large language model (LLM) summarizes this information into easy-to-read reports for teachers, simplifying the interpretation process. Both LLM experts and human educators evaluated these reports for clarity, accuracy, relevance, and usefulness in teaching. Our results show that LLMs can effectively act as educational analysts, turning complex data into insights that teachers find helpful. While automated reports are promising, human oversight is still necessary to ensure the results are reliable and fair. This work moves human-centered AI in education forward by connecting data-driven analysis with practical classroom applications.

Fig.1. Proposed Pipeline for LLM-Driven Assessment Report Generation: By instructing LLMs to role-play as an educational analyst and providing assessment context and data, we construct a prompt that is used to generate a teacher-oriented assessment report

Understanding students’ emotions is important for creating learning environments that adapt to their needs. Advanced computer vision models like HSEmotion and EMONET can detect emotions in real time, but their effectiveness in real classrooms is not well understood. These models are usually trained on adult faces in controlled settings, which makes them less accurate when faced with different camera angles, lighting, image quality, or skin tones. This study examined how these factors—camera angle, lighting, resolution, and skin tone—affect the accuracy and fairness of emotion detection in three different learning environments. Statistical analysis shows that these variables significantly influence how accurately the models estimate students’ emotional responses.

Fig.4 Valence comparison for different camera angles

Previous research has explored how the structure of white matter in the brain relates to academic skills like reading and math. Some studies have suggested that white matter—specifically, how it allows signals to travel through the brain—can predict a child’s academic abilities, while others have found no connection. However, studies that follow children over time (called longitudinal studies) have found that changes in white matter within the same child may be linked to learning progress.

This study aimed to replicate and expand on earlier findings by looking at how changes in a specific white matter pathway in the brain, called the left arcuate fasciculus, relate to reading development. The researchers followed 340 students from first through fourth grade, using diffusion MRI scans to measure white matter and tracking their reading and math scores over time. The results showed that year-to-year improvements in reading—but not math—were connected to changes in the left arcuate fasciculus. These findings offer more evidence that the brain’s white matter can change along with learning, and they underscore the value of long-term studies in understanding how children develop academic skills.

Fig 1

(A) Average estimated tract profiles for MD in the left arcuate fasciculus generated by the GAMM for four different quartiles of reading score change (reading state). Each color represents the magnitude of change relative to the average individual reading score. Shaded areas represent the standard errors of the predictions. (B) The estimated smoothing effect of time elapsed since the first study observation on average MD in the left arcuate. (C) Relationship between overall mean Woodcock–Johnson reading scores and MD in the left arcuate at each time point in the study.

Ěý

The “eye-mind link” idea suggests that where and how we move our eyes can directly show what we are thinking. However, it’s been hard to connect specific eye movements to exact thoughts in real-life situations. One reason is that eye movement measures like how long we look at something (fixation duration), how big our pupils get, and how far our eyes jump (saccade amplitude) are often averaged over time periods that include many different types of events, making it hard to interpret.

To tackle this, researchers tested whether looking at eye movements during specific events, chosen by the participants themselves, could better show how focused someone is and how well they perform a task. Nursing students wore special head-mounted eye trackers while doing simulation exercises. Later, they watched videos of their simulations and divided them into meaningful events, describing what they were doing, how well the task went, and what they were thinking during each event.

The study found that when students spent more time looking at things and had larger pupil sizes, they rated their teamwork as better. Larger pupil size also predicted better communication. On the other hand, when students made bigger eye jumps (larger saccades), they felt more confident about their abilities. These patterns were consistent across different types of events, and eye movement measures didn’t change much during an event. However, eye fixations were longer during the first five seconds of a new event compared to the last five seconds of the previous one, suggesting people focus more when starting a new activity.

In conclusion, looking at eye movements during specific events is a good way to understand how focused people are and how they explore their environment while learning naturally, and this works across different kinds of tasks.

Figure 4

The relationship between mean fixation duration and team quality

Researchers in learning sciences have suggested that there is a strong connection between learning STEM subjects (science, technology, engineering, and math) and developing computational thinking (CT) skills. Studies have shown that CT and STEM learning support each other, but not enough work has been done to clearly define the core CT knowledge and skills that should be taught in K-12 schools to improve both STEM and CT learning. As a result, many important CT concepts are often missing or underrepresented in classrooms for children from kindergarten through 12th grade.

We believe that CT ideas and practices are not only essential for computer science education but also help students build important modeling and problem-solving skills that apply to STEM subjects. In this chapter, we build on our previous framework to promote combined learning of science content and CT skills in middle school classrooms. We explain the main STEM and CT ideas and practices that we have introduced into science classes through our specially designed lessons. We also describe CTSiM, a computer-based learning tool created in our lab, which helps students learn these skills.

We share results from studies in middle school classrooms using CTSiM, showing that students made strong progress in both science and computational thinking concepts. Our assessments also help us understand how students learn and practice these skills, revealing important connections between their learning behaviors and their growth in STEM and CT knowledge.

Fig.Ěý12.1

The STEM + CT framework

Computational models (CMs) give pre-college students a chance to connect science, technology, engineering, and math (STEM) with computational thinking (CT) in ways that reflect how STEM is used in the real world. But not all students and teachers are ready to take on the challenge of learning or teaching programming. To make computing more accessible and reduce the difficulty of learning to code, this study explores simpler versions of computational modeling that involve little or no programming. Instead of writing code, students focus on understanding and evaluating existing code and computer simulations. The paper shares results from a small pilot study that looked at how students engaged with CT practices and what challenges they faced across three types of these easier-to-use CM activities.Ěý