Cross-publisher (iii-b) — FLDOE race shares vs NCES CCD
Verdict: PASS (2018-19, 2022-23, 2023-24).
| share | pairs | within process bound (consistency) | within 2 pp | median Δ (FLDOE − CCD) | mean |Δ| | mean |Δ| equal split | status |
|---|---:|---:|---:|---:|---:|---:|---|
| pct_black | 1533 | 1533 (100.0%) | 1385 (90.3%) | +0.00 | 0.88 | 1.38 | PASS |
| pct_hispanic | 1533 | 1533 (100.0%) | 1375 (89.7%) | +0.00 | 0.77 | 1.27 | PASS |
| pct_white | 1533 | 1533 (100.0%) | 1350 (88.1%) | +0.00 | 0.93 | 1.23 | PASS |
Served FLDOE-derived share vs NCES CCD race subtotal. (a) CONSISTENCY, not accuracy: |Δ| within the process bound min(9 × hidden cells, unallocated) / total + 0.5 pp for ≥ 99% of school-years, per share, with hidden cells counted as the loader counts them — any allocation that respects the cap and sums to the total satisfies it (the plain equal split included), so a failure means a loader defect, not a poor prior; (b) no school-year served with all seven shares equal; (c) ACCURACY against the no-information reference: the served share's mean |Δ| from CCD is no larger than the equal split's (the two coincide except where the capped adjacent-year prior applies, 2022-23). The prior was chosen among five candidates on CCD 2022-23, the same reference reported here (in-sample selection; the held-out split-half check is in docs/cities/miami/probe/race_imputation_prior.py). Mean |Δ| and the within-2-pp share are informational.
Uniform-split school-years served: 0
| year | hidden cells (loader rule) | tokens resolved as zero | share | pairs | within bound | within 2 pp | mean abs Δ | mean abs Δ equal split | lift |
|---|---|---|---|---|---|---|---|---|---|
| 2018-19 | 5700 | 0 | pct_black | 500 | 100.0% | 90.4% | 0.91 | 0.91 | +0.0% |
| 2018-19 | 5700 | 0 | pct_hispanic | 500 | 100.0% | 88.4% | 0.82 | 0.82 | +0.0% |
| 2018-19 | 5700 | 0 | pct_white | 500 | 100.0% | 87.2% | 0.94 | 0.94 | +0.0% |
| 2022-23 | 15982 | 1482 | pct_black | 515 | 100.0% | 91.1% | 0.89 | 2.39 | +62.8% |
| 2022-23 | 15982 | 1482 | pct_hispanic | 515 | 100.0% | 92.2% | 0.66 | 2.15 | +69.5% |
| 2022-23 | 15982 | 1482 | pct_white | 515 | 100.0% | 90.3% | 0.84 | 1.73 | +51.8% |
| 2023-24 | 5961 | 0 | pct_black | 518 | 100.0% | 89.6% | 0.83 | 0.83 | +0.0% |
| 2023-24 | 5961 | 0 | pct_hispanic | 518 | 100.0% | 88.4% | 0.83 | 0.83 | -0.0% |
| 2023-24 | 5961 | 0 | pct_white | 518 | 100.0% | 86.7% | 1.03 | 1.03 | +0.0% |
pct_black — largest differences (school, year, FLDOE, CCD, bound)
- 7861 2022-23: 19.64 vs 62.5 (bound ±100.5)
- 4651 2018-19: 33.33 vs 72.73 (bound ±82.32)
- 7835 2018-19: 43.75 vs 67.31 (bound ±100.5)
- 7892 2023-24: 50.0 vs 72.22 (bound ±100.5)
- 7840 2018-19: 66.67 vs 87.18 (bound ±100.5)
- 7811 2022-23: 39.39 vs 57.89 (bound ±100.5)
- 7805 2023-24: 75.0 vs 92.86 (bound ±100.5)
- 8012 2023-24: 45.45 vs 27.78 (bound ±100.5)
pct_hispanic — largest differences (school, year, FLDOE, CCD, bound)
- 7861 2022-23: 73.81 vs 37.5 (bound ±100.5)
- 7855 2022-23: 50.0 vs 81.82 (bound ±100.5)
- 6997 2018-19: 55.56 vs 82.61 (bound ±100.5)
- 5417 2023-24: 33.33 vs 56.25 (bound ±100.5)
- 8012 2023-24: 45.45 vs 66.67 (bound ±100.5)
- 6997 2023-24: 56.47 vs 77.27 (bound ±75.95)
- 7840 2018-19: 33.33 vs 12.82 (bound ±69.73)
- 7811 2022-23: 30.68 vs 10.53 (bound ±100.5)
pct_white — largest differences (school, year, FLDOE, CCD, bound)
- 7855 2022-23: 25.0 vs 9.09 (bound ±100.5)
- 7892 2023-24: 25.0 vs 11.11 (bound ±100.5)
- 8181 2023-24: 23.43 vs 9.9 (bound ±62.88)
- 6997 2023-24: 29.02 vs 17.27 (bound ±75.95)
- 7863 2018-19: 14.29 vs 4.55 (bound ±41.41)
- 7855 2023-24: 25.0 vs 15.38 (bound ±100.5)
- 1019 2022-23: 9.52 vs 18.57 (bound ±23.36)
- 8121 2018-19: 12.65 vs 3.7 (bound ±33.83)