Building Comprehensive, Longitudinal College and Career Readiness Predictive Analytics for the CORE Districts

• In Event: 66.039 - Developing Indicators to Evaluate College and Career Readiness

Abstract

In order to develop measurable, meaningful, and actionable indicators of students’ college and career readiness (CCR), it is crucial to have a robust system for collecting and analyzing data, to have end users that can make meaning of those data to inform district- and school-level strategic initiatives, and to have wide availability of data to draw upon to build the indicators that are most predictive of the outcomes stakeholders care about. These three pieces of the puzzle have recently been assembled by the CORE Districts—a network of eight of California's largest school districts collaboratively working to innovate, implement, and scale new strategies to improve student outcomes and narrow achievement gaps.
Recently, CORE launched an innovative data system for more than 50 collaborating school districts (the “CORE Data Collaborative”) serving approximately two million students. This system integrates a range of academic and non-cognitive metrics into a single user-oriented platform to drive innovation and improvement for districts and schools. Using this system, CORE has begun developing a comprehensive, longitudinal CCR predictive analytics system to provide CCR indicators at all levels of the K-12 continuum. Specifically, CORE stakeholders have prioritized indicators of high school graduation, college enrollment, and college persistence.
The CCR predictive models include both classical regression models and modern machine learning approaches; such a dual approach enables multiple strands of decision making at the student, program, school, and district level. The advantage of regression models is that it is easier to understand the effect that each indicator has on the outcome of interest, which makes the data more actionable. However, in recent years, the development of machine learning techniques has often proven to provide more accurate predictions, the tradeoff being that the mechanism of the models tends to be more difficult to understand.
The predictive models draw upon a broad swath of data, including data from the National Student Clearinghouse (e.g., college enrollment, college persistence); standardized assessments (e.g., SBAC, AP exams, PSAT, SAT); students’ self-reported social-emotional learning (e.g., growth mindset, self-efficacy); measures of school culture and climate (e.g., school safety, sense of belonging); and administrative data (e.g., attendance, disciplinary records, course grades). Leveraging these data, the system provides grade-by-grade, student-level “on track” indicators for CCR—not only annually, but also at key moments during the school year (e.g., when attendance data update monthly, or course grades update quarterly).
This approach allows us to identify the indicators that are most predictive students’ CCR in late high school, and then “daisy chain” the analytics backwards to determine what indicators in earlier years predict if a student is likely to meet those criteria in high school. This means we can determine a proxy CCR metric all the way back to elementary school, without collecting extensive longitudinal data that tracks a cohort from kindergarten through college. In this way, CORE’s ongoing work stands to make a crucial contribution to the field’s growing understanding of how to develop indicators of CCR throughout students’ schooling, in order to maximize the informative and actionable nature of data.

Author

• Libby Pier, Education Analytics