Broadening Access to Data Science Education in High School and Higher Education through Open Source Tools, Infrastructure, and Training

Broadening Access to Data Science Education in High School and Higher Education through Open Source Tools, Infrastructure, and Training Teon L. Brooks*†1, Steven Azeka*1,2,3 1 Gotham Data Clinic; 2 Mozilla; 3 Teachers College, Columbia University * Joint first author † Correspondence: [email protected] Introduction As data science, machine learning (ML), and artificial intelligence (AI) and other data-driven technologies play an increasingly central role in society, the importance of data science education is growing rapidly. From predicting health outcomes to influencing college admissions and criminal sentencing, data systems shape decisions that affect millions. Yet, these technologies are far from neutral. Several cases show how algorithmic bias has disproportionately harmed marginalized communities. For instance, the Correctional Offender Management Profiling for Alternative Sanctions (COMPAS) risk assessment tool has been used in thousands of criminal cases in the U.S. It was found to incorrectly label Black defendants as high-risk more often than white defendants (Angwin et al., 2016). We also see algorithmic bias in Amazon’s AI recruitment system when it was shown to downgrade resumes that mentioned women’s colleges or organizations, reinforcing gender bias in hiring (Dastin, 2022). These cases highlight the urgent need to prepare a new generation of data scientists capable of critically interrogating the systems they build and the models they train. But this raises a foundational question: Who has access to the tools and infrastructure needed to learn, question, and ultimately reshape these technologies? In many U.S. schools, particularly those serving Black, Latinx, Indigenous, and rural students, access to foundational tools for data science education remains limited. This includes not just internet access and computing devices, but also exposure to essential open-source programming languages such as Python and platforms like Jupyter notebooks. These disparities make it difficult to implement robust data science programs and leave many students excluded from academic and career pathways in the field (D’Ignazio & Klein, 2020). Data science education is important beyond future job prospects; data literacy helps us make decisions in our everyday lives, helps us be well informed, and helps us critically evaluate information presented to us. Beyond hardware, deep inequities exist in access to computer science instruction, which often serves as an entry point for data science. Research shows that in the United States, only 47% of Black students and 49% of Hispanic students attend high schools offering a computer science course, compared to 58% of white students. Gender disparities are also striking—in the U.S., just 16% of high school girls report interest in computer science and only 48% feel confident in their abilities, compared to 34% and 65% of boys, respectively (Google Inc. & Gallup Inc., 2016). Addressing these inequities requires not only broadening access to coursework but also building the instructional and technological infrastructure needed for meaningful engagement with data science. During the 2024-2025 school year, over 88% of U.S. public schools provided students with individual computing devices (National Center for Education Statistics, 2025), with Chromebooks capturing around 60% of the education market (Zion Market Research, 2025). While these devices offer an entry point, they are often not capable of running industry-standard tools used in data science. To build true equity, schools must pair hardware access with culturally responsive curricula and hands-on experiences that allow students to use data to analyze and address real-world problems in their communities. We will highlight efforts to broaden participation in data science in high school and higher education since this is where most data science education is currently taking place. The State of Data Science Education Tools Momentum for K–12 data science education is growing. As of 2024, 29 states have adopted data science into their academic standards, signaling widespread recognition of data literacy as a civic, academic, and workforce imperative (Data Science 4 Everyone & Sukol, 2024). States like California, Oregon, and Ohio are exploring ways to integrate data science into high school math pathways, broadening access to meaningful quantitative reasoning. However, access to high-quality tools and instruction remains uneven. Despite the post-pandemic proliferation of school-issued Chromebooks (Zahn & Geho, 2023), Chromebooks are poorly suited for running data science environments such as Python, Jupyter, or RStudio since these devices require additional IT support to enable local development or administrator privileges. Districts face further technical limitations due to restricted administrator rights, bandwidth constraints, and strict firewall settings that complicate software deployment. In many of these instances, teachers looking to teach data science rely on unplugged activities involving no-tech to low-tech options such as using the whiteboard or Post-It Notes to visualize data (National Academies of Sciences, Engineering, and Medicine, 2023). This infrastructure gap is further compounded by limited teacher preparation. Robertson et al., (2025) found that teachers’ initial understanding of data literacy was narrow and traditional—often confined to graphing and spreadsheets—with low confidence and readiness to teach the subject. Across high schools in the U.S., data science course offerings have a relatively small uptake, with only about 3% of U.S. students enrolled in a dedicated data science course or a course that included a data science module during the 2022–2023 school year (Data Science 4 Everyone & Sukol, 2024). This has resulted in a fragmented implementation landscape in which high-quality data science instruction is concentrated in well-resourced schools and largely absent from those serving historically marginalized populations. In the U.S., the University of California, Berkeley’s “Data 8” course provides a potential model for scaling a more inclusive data science instruction. Designed for all majors, the course has catalyzed a statewide initiative—the California Alliance for Data Science Education (CADSE)—which develops open-source materials and provides training for educators across high schools and colleges. These models demonstrate the power of pairing curriculum with infrastructure and professional learning to achieve scale. To bridge the resource gap, educators often turn to beginner-friendly platforms like TinkerPlots or the Common Online Data Analysis Platform (CODAP). These tools provide drag-and-drop functionality that lowers technical barriers, enabling students to explore real datasets, visualize relationships, and reason about variability without needing programming skills. For example, in Germany, the ProDaBi project (Project Data Science and Big Data at School) used CODAP as an entry point for secondary students. In this context, CODAP served as a stepping stone toward more advanced environments, helping students build confidence before moving into Python and Jupyter (Frischemeier et al., 2021). JupyterHub reduces technical barriers while providing a scalable way for teachers to manage assignments, share datasets, and support collaborative work. JupyterHub allows users to rapidly prototype their code and data analyses because they can run code in small chunks, see results, and iterate. JupyterHub deployments have been shown to improve equity of access, support large-scale courses, and reduce the technical burdens typically associated with coding instruction (Suen et al., 2018; Zonca & Sinkovits, 2018). By enabling students to transition from entry-level tools like CODAP into authentic programming environments like Jupyter, these tools reduce the barrier of entry to gain real-world data science skills. Emerging Data Science Tools and Technologies To close access gaps, a new generation of open-source tools is transforming what’s possible for data science instruction, particularly in resource-constrained environments. These tools leverage modern web technologies to bypass installation barriers and offer fully-fledged, authentic coding experiences through nothing more than a browser. Recent large-scale education reforms have highlighted gaps between policy and implementation for delivering computing education at scale. In 2018, France passed a reform that mandated that all high school students must be taught Python3 as part of their course curriculum (Ministère de l’Éducation nationale et de la Jeunesse, n.d.; Programmes et ressources en numérique et sciences informatiques - voie G, n.d.). The reform did not specify how to implement the curriculum. Running servers for tens of thousands of students can easily become cost prohibitive. The Ministry of Education adopted a unified strategy supporting Python instruction in secondary schools through a hybrid cloud-based platform called Capytale where files are saved to the cloud, but computations are handled in the browser greatly reducing maintenance cost. Built on new open-source technologies, Capytale allows students to run code directly in the browser with a single click, no installation setup and no compute-based server cost. It is currently deployed in over 700 schools serving more than 27,000 students (PyData, 2021). These tools change the operational paradigm for most computational notebook services. Usually, notebooks rely on a client-server architecture where the browser provides a user interface to input, interact, and display content as the client but sends these interactions off to be processed by a server. The server can be hosted locally on the device as with Jupyter Notebooks, but can be hosted on the cloud as with Google Colab or Posit Cloud. In 2017, a new web technology called WebAssembly (Wasm; Haas et al., 2017) was released and brought the ability to take native code like C and make it compatible with modern web browsers. This has opened up the possibilities for new experiences for the web. Using a toolchain called Emscripten (Zakai, 2011), to make compiling C and C++ native code to Wasm easier, open-source developers have been working to make the browser a new execution environment for Python. This has enabled Python to run in the browser natively, removes the need for external servers to run reasonable workloads, which greatly reduces the cost for running services. Bringing Python and R to the Web: Pyodide and webR Python and R are widely regarded as the primary programming languages in data science. They both traditionally require either a desktop environment setup. Unfortunately, this setup is not compatible with most Chromebooks or iPads. The alternative is to have a hosted server instance where all the computations happen on the cloud, but this can become price prohibitive at scale. To address the frustration of sharing data science artifacts to the web, Pyodide (Droettboom et al., 2021) was created to port the Python’s reference implementation, CPython, and its standard library to the web using Wasm and Emscripten. Pyodide has created its own toolchain and distribution to support the scientific Python stack. It has ported the core scientific Python libraries: NumPy (Harris et al., 2020), SciPy (Virtanen et al., 2020), Matplotlib (Hunter, 2007), and pandas (McKinney, 2010; The pandas development team, 2025), as well as a variety of packages that rely on native code. Pyodide can import any pure Python package on the Python Packaging Index (PyPI) using micropip. It provides an interactive computing environment that runs entirely in the browser. All the Pyodide-related resources are downloaded the same way other web content is downloaded, which makes the installation process seamless. Parallel advances have been made in the R ecosystem. Inspired by the work of Pyodide, the webR project (Stagg et al., 2023) takes the base R ecosystem and compiles it to Wasm. This toolchain is now used to power kernels in Jupyterlite to provide a notebook environment in R, native to the web. Bringing Data Science Workflow to the Web: JupyterLite The use of computational notebooks is a huge driver of many data science workflows. Jupyter Notebooks (Kluyver et al., 2016) have been a standard in the data science community for sharing work. These notebooks are also used to teach data science as it allows the mixture of text blocks and code blocks for explaining and exploring, respectively (Rule et al., 2018). The Jupyter team has extended the functionality of Jupyter Notebooks to create JupyterLab, a full Integrated Development Environment (IDE) experience for computational workflows. Building on the success of both Pyodide and webR, a new Jupyter subproject was formed to explore a new notebook paradigm that can run Jupyter Notebooks and JupyterLab without the computational server backend. JupyterLite offers a lightweight, browser-based version of JupyterLab. Students can write and execute Python code, analyze datasets, and visualize results—entirely within their browser. This platform is particularly valuable in schools with limited IT support or restricted networks, as it eliminates complex installations and leverages the existing student devices. Piloting Courses and Building Tools with Pyodide Several pilot courses using Pyodide have been launched in secondary schools and universities over the past five years across the globe. In the United States, a team at Stanford University developed PyodideU (Jefferson et al., 2024), a JupyterLab-like IDE powered by Pyodide, that has been used as the primary programming environment across five different classes, taught in-person and online, domestically and internationally, reaching 10,000 students across 150 countries, at minimal cost. In Belgium, the Flemish government has enacted new learning objectives in secondary schools to cover computational thinking and computer science. The eTeacher platform was developed to address these learning objectives and has been piloted in twenty schools (Hoobergs et al., 2023). A computer science course in Verona, Italy (Audrito et al., 2025) and one in Lausanne, Switzerland (Farah et al., 2020) have also been designed and built with Pyodide. At UC San Diego, instructors of Principles of Data Science course have developed the tool PandasTutor (Lau et al., 2023) to help explain data transformation in Pandas with visual cues. With this tool, a learner can input their Pandas code and it will provide a step-by-step diagram of what’s happening to the data. Scalable and Public-Access Models Binder (Jupyter et al., 2018) is another powerful tool for democratizing access. Binder allows instructors to launch shareable, server-hosted notebook environments from GitHub repositories using JupyterHub. Used in courses like Berkeley’s Foundations of Data Science, Binder’s public instance (mybinder.org) supports thousands of users. However, its maintenance cost—estimated at $66,000 per year—raises sustainability challenges for widespread deployment (Jupyter et al., 2018). In contrast, Capytale (Académie de Paris, n.d.; JupyterCon, 2023) demonstrates a scalable, government-supported model that pairs open-source tools with national curriculum initiatives. Built on top of Pyodide and Basthon, its custom notebook offering, Capytale respects data privacy with its client-side execution and allows seamless classroom integration, offering a blueprint for U.S. school systems seeking cost-effective solutions. It is integrated into 7,000 schools with 280,000 active users (10% teacher, 90% students). Expanding the Ecosystem: Julia, Pyret, and Visual Programming Tools While Python and R dominate, other languages are emerging in the educational data science space. Julia, known for high-performance numerical computing, is gaining traction, particularly in the higher ed context (JuliaDataScience.io). Though its ecosystem is still maturing, Julia holds promise for future web-based implementations. Projects like Pyret (developed by Bootstrap) and Racket provide programming language alternatives designed with pedagogy in mind. They are intentionally limited in functionality to help focus and facilitate learning as opposed to the more general purpose nature of language like Python. Pyret has been taught at Brown University as part of their Introductory Programming courses. However, both of these projects have limited browser support, which hinders their adoption in K–12 settings. Meanwhile, visual programming tools offer compelling entry points for middle and high school students by combining accessibility with authentic data transformation practices. For younger or novice learners, Tidyblocks (Wilson, 2019/2025) is a proof of concept visual programming language that offers a block-based interface inspired by Scratch but grounded in the logic of R’s Tidyverse. Students can manipulate datasets visually using drag-and-drop blocks, while the underlying R or Python code is simultaneously displayed. This dual-mode interface creates a natural onramp from visual programming to textual coding and supports conceptual understanding of data analysis workflows. JupyterLab Blockly (QuantStack/Jupyterlab-Blockly, 2022/2025) and Edublocks (EduBlocks, n.d.) are other compelling entry points for early-stage learners to introduce programming in a more accessible way while offering the full affordances of the Python programming language. Toward an Ideal Model for Equitable Data Science Education in High School and Higher Ed While open-source tools and policy reforms offer important strides toward equity in data science education, a more coordinated, systemic model is needed to ensure that all students—not just those in well-resourced districts—can access, engage with, and benefit from data science learning. An ideal solution must be built on a foundation of technological accessibility, instructional equity, curricular relevance, and long-term sustainability. 1. Web-Native, Open-Source Infrastructure At the core of this model is a robust, browser-based ecosystem that eliminates the need for local installations or expensive hardware. Tools like Pyodide, webR, Jupyterlite, and JupyterHub already demonstrate that professional-grade computing environments can be delivered entirely through the web. An ideal model would integrate these into a unified, school-friendly platform with single sign-on, privacy compliance, and support for both Pythonand R-based workflows. To ensure accessibility, the platform should be device-agnostic, running seamlessly on any web-capable device, including Chromebooks and tablets, while also supporting offline access and data persistence in contexts where internet connectivity is unstable. It should provide a visual programming layer, such as Tidyblocks, for novice users, alongside a scripting interface that allows students to progress toward industry-standard tools. To ensure long-term sustainability, this infrastructure should be open source and maintained through a public–private consortium involving government agencies, academic institutions, nonprofit organizations, and industry partners. 2. Comprehensive Educator Support Teacher readiness is a critical dimension of equity in data science education. An effective professional development framework should immerse educators in experiential, hands-on engagement with authentic data science practices, supported by collaborative and reflective learning environments that facilitate the integration of these competencies into classroom instruction. Research emphasizes that teachers must first have opportunities to engage meaningfully with data—formulating questions, analyzing complex or “messy” datasets, and applying appropriate analytical tools—before they can design comparable learning experiences for students (National Academies of Sciences, Engineering, and Medicine, 2023). To prevent further inequities within the field, professional development must be incentivized and equitably distributed, particularly to educators in Title I schools and rural districts. To ensure coherence and sustainability, these initiatives should align with state-level data science standards. States and higher education institutions should also establish clear certification pathways for both pre-service and in-service teachers, building a consistent pipeline of qualified data science educators. A key element of this effort will be mentorship from experienced data science practitioners. Finally, to reduce barriers and strengthen implementation, districts can leverage existing data science curricula supplemented with practical guidance and localized support. 3. Culturally Responsive and Community-Connected Curriculum The curriculum embedded in this model should go beyond technical skills to address data ethics, justice, and civic engagement. Rooted in principles of culturally responsive pedagogy (Ladson-Billings, 1995), such a curriculum would incorporate datasets and case studies that reflect students’ lived experiences and community concerns. It would also encourage critical thinking about data systems, including issues of privacy, surveillance, algorithmic bias, and bias in training datasets, while creating space for students to design and conduct their own investigations on topics that matter to them. Research underscores that connecting data science to social realities can foster both agency and equity by positioning students as critical consumers and producers of data (National Academies of Sciences, Engineering, and Medicine, 2023). This emphasis helps students view data science not only as a career skill but also as a tool for social inquiry, justice, and empowerment. 4. Open Access to Shared Data and Resources To support inquiry-driven learning, the model should include a curated repository of open datasets designed specifically for students. These datasets should be pre-cleaned and annotated with student-friendly documentation to reduce technical barriers, while covering diverse domains such as environmental science, public health, housing, and education. Each dataset should be paired with exploratory questions and scaffolded analysis tasks that guide students in developing statistical reasoning and data science practices. In addition, curricular resources—including projects, rubrics, and exemplars—should be shared openly through a centralized portal maintained by a coalition of educators and researchers. Such an infrastructure would ensure consistent access to high-quality, developmentally appropriate resources, while also enabling adaptation to local contexts and community priorities. 5. Sustainable Funding and Governance Equity in data science education cannot rely on temporary grants or volunteer labor; it requires sustainable investment and coordinated leadership. What is needed is support by dedicated state and federal funding for platform development, teacher training, and curriculum dissemination. Long-term stability also depends on public–private partnerships that engage technology companies, universities, and philanthropic foundations in resource sharing and innovation. To ensure accountability and responsiveness, a governing board should include educators, technologists, students, and community leaders. International and domestic precedents illustrate the power of such coordination: France’s Capytale initiative shows how national strategy can scale access to coding and data tools in high schools, while California’s Alliance for Data Science Education (CADSE) demonstrates how higher education leadership can catalyze statewide investment and curriculum alignment. Conclusion The growth of data science education in high school and higher ed settings presents both a challenge and an opportunity. While momentum is strong, persistent inequities in access to tools, infrastructure, and trained educators threaten to leave many students behind. Open-source technologies—especially those that are web-native—offer a promising path forward. By leveraging tools like Pyodide and JupyterLite, and by integrating block-based platforms like Tidyblocks and Jupyterlab-Blockly, schools can provide authentic data science experiences even in the face of resource constraints. But technology alone is not enough. Achieving equity in data science education requires intentional investment in teacher preparation, curriculum development, and community-centered design. When students from all backgrounds are given access to real tools, culturally relevant content, and supportive instruction, they are not just learning to code—they are learning to question, critique, and shape the data systems that influence their lives. Acknowledgments We would like to thank Pablo Ripollés and Eric Chen for their review and feedback on this paper. Appendix: Resources Projects Links Basthon https://basthon.fr/ Binder https://mybinder.org/ Capytale https://capytale2.ac-paris.fr/ CODAP https://codap.concord.org/ Edublocks https://edublocks.org/ eTeacher https://www.eteacher.be/ Jupyter https://jupyter.org/ JupyterHub https://jupyter.org/hub Jupyterlab-Blockly https://jupyterlab-blockly.rtfd.io JupyterLite https://jupyterlite.rtfd.io/ ProDaBi https://www.prodabi.de/en/ Pyodide https://pyodide.org PyodideU https://ide.stanford.edu/public Pyret https://pyret.org/ Racket https://racket-lang.org/ Tidyblocks https://github.com/gvwilson/tidyblocks/ TinkerPlot https://www.tinkerplots.com/