Model Selection
Candidate expected-growth models are compared with repeated cross-validation, with the readiness-augmented model selected as the simplest non-benchmark option within one standard error of the best RMSE.
Project Brief
Assessment Growth Analytics in R
Public-safe education analytics project that evaluates BOY-to-EOY assessment growth and identifies section-level improvement signals after accounting for starting performance, readiness, attendance, course track, grade level, and school-year context.
Growth Analytics Surface
Adjusted section signals, decision support
The project uses base R to generate public-safe assessment data, compare expected-growth model families, validate holdout performance, and translate section-level growth signals into instructional review guidance.
Evidence Packet
- 1,737 paired BOY/EOY assessment records
- 174 section-year groups
- Readiness-augmented expected-growth model
- Context-only, linear, quadratic, piecewise, readiness, and spline candidates
- Repeated 5-fold cross-validation with 6 repeats
- Holdout RMSE, MAE, and R-squared diagnostics
- Reliability-weighted section signal flags
- Teacher/course summary views for instructional review
- PDF report, executive brief, model card, methodology, and public-safety notes
Validation
What reviewers can inspect
Decision Metrics
The evidence packet reports average raw BOY/EOY gain of 5.72 points across 1,737 paired records, holdout RMSE 4.399, holdout R-squared 0.181, and 8 section-year groups above expected growth with 5 below expected growth.
Growth Diagnostics
The report compares raw gains with adjusted gains, reviews residual diagnostics, checks section sizes, and separates pattern-finding views from personnel decisions.
Public-Safe R Pipeline
The repo uses synthetic records only, requires no packages or network access, and includes validation scripts that check for private-source leakage.
Inspect In Source Repo
Reproducible R evidence path
The project is intentionally lightweight: R scripts generate data, fit models, render reports, build figures, and validate the public boundary from a `make all` pipeline.
Safety Boundary
Synthetic records, human review
The project is not a real teacher-evaluation, student-placement, grading, or personnel system. It is a public-safe demonstration of statistical modeling judgment, validation, and decision-support communication for instructional review.