docs/design/reference-scaffold/config/covariates.yaml
Served verbatim from the project repository. Internal working document conventions apply: documents may reference file paths, branch names, and findings-ledger anchors from the repo.
# Covariate sets, keyed by grade band.
# numeric -> continuous controls (demographic shares, 0..1; enrollment as count)
# categorical -> dummy-encoded controls (first level dropped as reference)
#
# Note on collinearity: race/ethnicity shares are compositional and sum to ~1,
# so we deliberately OMIT pct_white (it becomes the reference) to avoid a
# perfectly collinear design.
sets:
es_base:
numeric: [pct_econ, pct_ell, pct_swd, pct_black, pct_hispanic, pct_asian, enrollment]
categorical: [admission_bucket]
ms_base:
numeric: [pct_econ, pct_ell, pct_swd, pct_black, pct_hispanic, pct_asian, enrollment]
categorical: [admission_bucket]
hs_base:
numeric: [pct_econ, pct_ell, pct_swd, pct_black, pct_hispanic, pct_asian, enrollment]
categorical: [admission_bucket] # method1 bucket; dominant predictor at HS