Methodology

Data Sources

NCES CCD School Universe

Common Core of Data — school-level directory with enrollment, grade span, locale, Title I status, and demographics

Format: csv · Source

NCES EDGE Geocodes

School-level geographic coordinates and census tract linkage

Format: csv · Source

EDFacts Chronic Absenteeism

School-level chronic absenteeism counts and rates by subgroup (all students, race/ethnicity, disability, economic disadvantage)

Format: csv · Source

California CDE Chronic Absenteeism

California-specific chronic absenteeism data with finer grade-level and demographic breakdowns for richer pilot analysis

Format: csv · Source

Panel Construction

The system builds a longitudinal school-year panel by joining three datasets:

1. NCES CCD — school directory with enrollment, grade span, locale code, demographics
2. EDFacts — chronic absenteeism rates and counts by subgroup
3. NCES EDGE — geographic coordinates for map visualization

Schools are matched across years by their NCES school ID (NCESSCH). The panel spans school years 2016-17 through 2022-23 (subject to data availability). Schools with fewer than 20 students are excluded.

A district-year panel is aggregated from the school panel by grouping on LEA ID (LEAID), summing enrollment and averaging rates.

Peer Group Construction

Schools are grouped into peer cohorts based on shared context so that comparisons are fair. Peer group dimensions:

Attainable Frontier

For each peer group, the 25th percentile of chronic absenteeism rate defines the "attainable frontier" — the rate achieved by the better-performing quarter of peers in similar contexts.

The gap to frontier is the difference between a school's actual CA rate and its attainable rate. A positive gap means room for improvement.

Recoverable Attendance Days

The central metric translates the rate gap into a concrete, actionable quantity:

This represents the total student-days of attendance that could be recovered if the school matched its attainable frontier. A school with a 10pp gap and 1,000 students has 18,000 recoverable days per year.

Subgroup Disparity Score

Measures the spread in chronic absenteeism rates across demographic subgroups (Black, Hispanic, White, Economically Disadvantaged, Students with Disabilities):

Higher disparity indicates larger equity gaps. A disparity of 15pp means the most-affected subgroup has a 15 percentage point higher CA rate than the least-affected.

Composite Intervention Priority

Each component is min-max normalized to [0, 1] then combined using weighted sum:

Rolling Backtest

The model is validated using a rolling temporal backtest:

1. Train peer groups and frontiers on data from years < t
2. Predict top-100 highest-need schools for year t
3. Compare against actual chronic absenteeism in year t
4. Repeat for each test year

Metrics: Precision@100 (fraction of predicted schools that were actually high-absence), Recall@100 (fraction of actual high-absence schools captured), and average CA rate of predicted set. Compared against a naive baseline that ranks by raw CA rate alone.

Limitations

• EDFacts data suppresses small cell sizes for privacy; some school-subgroup combinations are missing
• NCES ID changes (mergers, closures, reopenings) may break longitudinal tracking for a small fraction of schools
• Peer groups using coarse bins may group dissimilar schools; finer demographic controls would improve precision
• COVID-era years (2019-20, 2020-21) distort rates and may not be representative
• Attainable frontier assumes the 25th percentile is achievable; structural barriers may differ
• Subgroup disparity uses a simple max-min spread; more sophisticated inequality measures exist
• District-level aggregation averages over within-district variation