Acting Early to Expand College Pathways
What if schools could predict, as early as 9th grade, whether a student is on track to qualify for competitive scholarships or meet college admissions criteria? And what if those predictions could trigger timely interventions—like course recommendations, test prep, enrichment opportunities, or targeted advising—to change that student’s path?
That’s the goal of a new predictive modeling initiative at Education Analytics (EA), supported by the multi-state Ed-Tech Collaboratory (ETC) with funding from philanthropy.
The ETC is a multi-state initiative dedicated to building modern, data standards-aligned infrastructure for public education data systems and software ecosystems. Its vision is rooted in interstate collaboration, open standards, and shared technical assets.
Why this is Important. Across the country, school systems are working to ensure students graduate not just with a diploma, but with the skills and opportunities needed to thrive in college and career. Yet, too many students—especially those from historically underserved communities—miss critical college-readiness milestones due to barriers like limited access to scholarships, advising, or rigorous coursework.
Meanwhile, education agencies face growing urgency to direct resources and policies where they are most needed, and counselors and schools face increasing caseloads and limited bandwidth to identify which students are on track for postsecondary success.
This predictive modeling work lays the foundation for what early college readiness tracking could look like across states and demonstrates how bringing predictive insights to the table can complement and accelerate efforts like Georgia Match to connect students to appropriate college pathways. The modeling builds on a foundational belief: that data should be used not to label students, but to unlock potential and expand opportunities.
What We’re Building: A Model to Predict Scholarship Eligibility
EA is working with state partners to develop predictive models that forecast a student’s likelihood of meeting the eligibility requirements for state-based scholarship programs. These models are trained on real data from students in South Carolina—including variables like GPA, course-taking history, course rigor, end-of-course exams, and attendance.
We’re using an ordered probit model to estimate the likelihood that a student will meet the minimum criteria for each scholarship, based on data available by the end of 9th grade. We use this model because we’re predicting scholarship eligibility among a set of scholarships that are ordered from least to most selective. The ordered probit model takes in all available student data and calculates the likelihood that a student belongs to each of the ordered outcome categories (i.e., meeting the criteria for each scholarship, ranked from least to most selective), based on patterns learned from the data. Our core predictors focus on course grades—grouped by subject and difficulty level—and end-of-course exam performance.
Our technical approach relies upon statistical modeling, informed by domain experts and practitioners. We’ve also built flexibility into the models, allowing different paths to success—for example, excelling in science can offset lower performance in another subject. As part of this initiative, we’re exploring machine learning and other modeling techniques to compare with our current expert-driven approach. Our goal is to understand where these methods align, where they diverge, and how they might complement and inform future research and modeling efforts.
Laying the Foundation: Why Infrastructure Matters
To scale these models, we bring our rigorous research and statistical expertise together with our technical expertise and open data infrastructure. Enable Data Union (EDU) is a secure, standardized data layer that transforms data from multiple sources (including Ed-Fi operational data stores) into analytics-ready formats using open standards like CEDS.
Building a model across districts—and particularly across states—requires significant effort to navigate different state and district systems, address missing data challenges, and align core elements like course codes, GPA weighting systems, and assessment structures. For example, we’re navigating different policies around course rigor and transcript coding while also building analytics in anticipation of data pipelines coming online in different states. Leveraging our data infrastructure and these open standards gives us a leg up in navigating these challenges by providing a common language upon which to build these types of advanced analytics and scale them across states.
From Insight to Action: A Statewide Early College Readiness System
Analytics alone won’t drive student outcomes—people do. That’s why we’re focused on building predictive tools that are not just accurate, but usable and actionable by the people closest to students: counselors, administrators, and other educators.
That means models that refresh at regular, intuitive intervals; tools that are user-friendly; and interfaces that offer multiple ways to interpret results—whether it’s a probability score, an index from 0–100, or visual “on-track" lanes like:
- Very Competitive
- Competitive
- Promising
- Needs Support
- Needs Significant Support
It also means creating governance structures around how these predictions are shared, who has access, and how they are used to support—not limit—students’ aspirations.
We’re building the infrastructure to refresh these predictions regularly and integrate them into platforms like the Rally Analytics Platform (used by educators to support students and families) and Podium or Power BI (used by school and district leaders). This shift allows educators and counselors to be proactive rather than reactive—intervening early, rather than realizing too late that a student missed a key milestone.
Let's Review Some Examples
Imagine a student who is taking mostly high-level courses across core subjects, earns high grades, and scores in the highest range on end-of-course exams. Each of these data points contributes to an overall index score. Course credits attempted are weighted, and course rigor is factored into this score. For instance, earning a “B” in an advanced math class carries a higher weight compared to earning the same grade in a medium-level math course. This student ends up with a 91 out of 100 on the index—translating to a 96.5% chance of qualifying for at least one scholarship and a 52% chance of meeting the most selective criteria. This student is categorized as Very Competitive.
Another student earns mostly “Bs” in mid-level courses and has mid-level end-of-course exam scores. They earn 67.5 points, putting them in the Promising lane. They have a 67.5% chance of qualifying for at least one scholarship, but their chance of hitting the top scholarship criteria is under 10%. This student might benefit from early intervention—if we can flag their potential and provide targeted supports by 10th or 11th grade.
A final student earns lower grades in less advanced courses and has lower end-of-course exam scores. Their index is 35, placing them in the Needs Significant Support lane. Their probability of qualifying for any scholarship is less than 11%. Having this predictive insight—available early—can help schools direct resources where they’re needed most.
Here’s how this model could be used in practice:
- A counselor sees that a student is close to qualifying for a top-tier scholarship but is trending low in math. They connect the student with tutoring or recommend an alternate course plan.
- A school leader identifies a group of students who are just below the threshold and launches a targeted test prep cohort.
- A district data analyst reviews model outputs across multiple schools and identifies that particular subgroups of students are underrepresented in the “Competitive” tier despite strong academic records. They collaborate with counselors and administrators to review support structures, course access, and advising practices to support all students and address systemic barriers.
Looking Ahead: Gathering Feedback and Shaping What’s Next
This summer, we’ll conduct user interviews with state education leaders, school counselors, and other stakeholders to learn:
- What current processes or tools are in place to guide postsecondary readiness?
- What formats are most useful to counselors and educators to act?
- Would it help to predict university entrance likelihood in addition to scholarship eligibility?
- How can these tools align with existing advising practices?
We’ll use this feedback to refine the models, prioritize enhancements, and guide how we integrate predictions into EA’s products—ensuring these tools work not just technically, but in practical, educator-friendly ways.
Closing Thoughts: Predicting Potential, Supporting Pathways
At Education Analytics, we believe readiness isn’t one-size-fits-all. Students have different strengths, paths, and dreams. Our role is to use data to illuminate those paths, not narrow them.
Predictive modeling, when done right, can help ensure more students get the opportunities they’ve earned—and make sure no student misses out because of timing, visibility, or access.
We invite you to follow this work, contribute your insights, and help shape a future where every student has the tools they need to succeed.