← All stories

Perception gapsmedium

How much data does NYC suppress, and what does it hide?

The question

How much of New York City's school-outcome data is hidden from public reporting for student-privacy reasons, and what's behind the hidden part?

Roughly 27 to 29% of New York City subgroup test-result cells were withheld from public reporting in 2024-25 for student-privacy reasons (typically when fewer than about 5 students fall in the subgroup at the school). Test-result cells are the most heavily suppressed kind of measure because small subgroups in small grades trigger the threshold. The schools with the most suppression are mostly small specialty programs — Special Music School, Ballet Tech, the city's sign-language lower school. At schools where the English-Language-Learner cell is suppressed, the public can't see how that group is doing.

Read the full briefing

Data analysis

Investigation run 2026-05-25.

TL;DR

About 27 to 29% of subgroup test-result cells were withheld from public reporting in 2024-25 for student-privacy reasons — higher than expected. Test results are the most-suppressed measure because small subgroups in small grades trigger the suppression threshold. The schools with the most suppression are mostly small specialty schools (Special Music School, Ballet Tech) and small District 75, charter, and lower-grade schools — places where every subgroup is small by design.

% of metric cells suppressed by metric type (2024-25)

Subgroup cells (by race, ELL, SWD, gender) get suppressed when fewer than ~5 students fall in the cell. At a school with 60 fifth-graders, the ELL subgroup might be 3 kids — and the cell gets suppressed. Across NYC, this hides ~30% of subgroup outcomes.

MetricTotal cellsSuppressed%
ela_grade3_proficiency9253272229.4%
math_grade3_proficiency9293263628.4%
math_grade8_proficiency3997113028.3%
ela_grade4_proficiency9193258928.2%
math_grade5_proficiency9277253727.3%
math_all_proficiency14003373026.6%
ela_all_proficiency14008350125.0%
ela_grade8_proficiency5511125422.8%

Schools with most suppression

SchoolBorough# suppressed cells
PS 308 Clara CardwellBrooklyn189
47 American Sign Language & English Lower SchoolManhattan188
Ella Baker SchoolManhattan172
MS K394Brooklyn171
PS 175 City IslandBronx170
Special Music SchoolManhattan154
Ballet Tech NYC Public School for DanceManhattan154
PS 76 A. Philip RandolphManhattan146
PS 138 BrooklynBrooklyn146
Teachers College Community SchoolManhattan144

Why these schools

Most are small (under 200 enrollment). Small schools have small subgroups. When ELL count is 3 or SWD count is 4, the cell is suppressed for privacy — meaning we know less about how those kids are doing precisely where we'd most want to know. The 47 ASL & English Lower School is a specialty school where every subgroup is tiny by design.

What this conceals

Aggregate school numbers can look fine while specific subgroups (ELL students, students with disabilities) score very differently. At a small school where the ELL cell is suppressed, the data doesn't tell us how ELL students are doing relative to peers. The privacy protection has real costs in information; the trade-off between privacy and information is real and rarely discussed openly.

Mitigation options

1. Multi-year rollups (combine 3 years of small cohorts to get sample sizes above suppression threshold). 2. District-level reporting for tiny-subgroup cases. 3. Loosen the threshold from n<5 to n<3 (some states use n<3). 4. Synthetic data with noise injection — controversial but used in some federal datasets.

Caveats

We're using the suppressed flag as reported by NYCDOE — the threshold isn't documented in our DB but is conventionally n<5 for race/ethnicity, n<10 for SWD/ELL in some NYS contexts.

What's next

Methodology piece: NYC reports ~70% of school subgroup cells. The 30% it suppresses is also the part most likely to show within-school variation. Pair with interview of ELL coordinator at a small school where the ELL outcome is suppressed.

Methodology & replication recipe — the data sources, queries, and steps behind this analysis.

More on perception gaps