Powerful, and handled with care
The data-driven model complements the theory-driven one: instead of expert rules, it learns pathways from data. That power raises the bar for how it is validated and deployed, which is why it was released only after disclosure review.
What the tool runs
Personalized Plan is built on two optimal dynamic treatment regime (ODTR) models, fitted on restricted-use HSLS:09 data with Targeted Maximum Likelihood Estimation and ensemble machine learning: one optimizing weighted math GPA (Math Achievement) and one optimizing the chance of enrolling in college (College Enrollment). The fitted models are large and stay offline; the tool runs small decision trees distilled from them, for both a plan for Grade 9 only and a plan for all four years (Grade 9–12). How it works explains the distillation and how closely the trees follow the full models.
Data and disclosure review
Both models are trained on the High School Longitudinal Study of 2009 (HSLS:09), a nationally representative study that followed more than 20,000 students through high school. Our analysis uses a subset of about 12,640 of them. Only the distilled decision rules are in the tool; the underlying data never leaves the secure analysis environment, and nothing you enter in the tool is stored or sent anywhere.
Because the models are fitted on restricted-use data, the tool was released only after it passed the Disclosure Risk Review of the Institute of Education Sciences (IES).
How it is served
A recommendation from a statistical model that affects minors' course placement is a deterministic mapping, and it stays one. Rather than serving the fitted models, the tool runs compact, inspectable decision trees distilled from them, entirely in your browser, so the same answers always give the same recommendation.
- It carries the same framing as the theory-driven model: a planning aid that supports advising conversations, never an automatic placement decision.
- Feasibility constraints keep recommendations to pathways a school could actually offer.
A planning aid, not a placement decision. These recommendations come from decision trees distilled from models fitted on restricted-use HSLS:09 data, released after IES Disclosure Risk Review. Use them as a starting point for conversations with the student's teachers and counselor. Their goals, circumstances, and the school's course offerings come first.
Prefer rules you can read line by line? The theory-driven model is fully transparent.