Over the last two decades, investments in digital infrastructure for statewide longitudinal data systems mean researchers now have access to longitudinal postsecondary transcript data. However, because this data was created for administrative use rather than designed for researchers, these transcripts lack standardized measures of course content that make it challenging to compare across institutions and systems.
We address this measurement challenge through a text as data approach to classify college courses into a standardized hierarchical taxonomy developed by the National Center for Education Statistics, called the College Course Map. Drawing on nationally-representative transcript studies, we fine-tune and release open-source ModernBERT models placing courses into this taxonomy. Our training data, methods, and performance are described here. The models will be the engine behind a new software tool that practitioners and researchers can use to classify their own course data into this standardized scheme.
Our research team is using the tool for several applications, including to identify "curricular deserts," bottleneck courses, and to describe long-term trends in course-taking.
We are currently looking for researchers and practitioners that would be interested in piloting an early release version of the CCM software tool. See more information on our Getting Involved page.
This project is primarily funded via grant R305D240029 from the Institute of Education Sciences, U.S. Department of Education. Seed funding was provided by the Michigan Institute for Data and AI in Society (MIDAS) Propelling Original Data Science (PODS) Grant program. Paulson is also grateful for support from PR/Award R305B200011 from the Institute of Education Sciences, U.S. Department of Education, and from a 2024 National Academy of Education/Spencer Foundation Dissertation Fellowship. We are grateful for data partnerships with several state agencies on several of the research applications.
