Miami-DadeMethodology and data qualityhard
Adjusting test scores for who enrols makes them agree less, not more, with Florida's own growth measure
Thesis
The pre-registered P2 check asks whether adjusted achievement correlates with Florida learning gains at least as strongly as raw achievement. The linked residual analysis reports the result for each fit. This is an empirical check: a demographic gradient in gains does not mathematically force P2 to fail. Spearman correlations and sampling-error-standardized residuals cannot be ordered using a Pearson covariance identity. Florida's learning-gains rules also differ from a continuous growth model; a student maintaining level 5 can qualify, and the school's lowest quartile can include proficient students under those same rules.
Supporting findings
- Fits use weighted least squares per outcome and year, controlling for race shares, the explicit English-learner lower bound, economic disadvantage, enrollment, level, sector and magnet status; schools with observed outcomes remain eligible even if closed today.
- Read the pre-registered checks and ablation counts from the current packet. Residual-z describes performance relative to measured demographics; it does not identify a school's causal contribution.
- Suppressed English-learner counts create uncertainty. The analysis refits at the upper endpoint on the same schools and reports coefficient, score and flag changes. That comparison is not a full range over every possible allocation of hidden counts.
- Peer groups use k = 40. Stored k = 20 and 30 memberships do not establish that ratings are robust to k; that sensitivity has not been evaluated.
Reporting directions
- Ask the Department of Education how learning gains are computed for students at level 5 and whether the gains component is intended as a growth measure.
- Find a Florida researcher who has estimated school value-added on student microdata and compare their ranking with both public measures.
- Explain to readers why 'beating expectations' and 'growth' are not the same claim.
Where the numbers are
Methodology & replication recipe — the queries and sources behind the findings above.