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

Thesis

Subgroup data with small N is suppressed for student privacy. Across our metrics, ~7-10% of subgroup cells are suppressed. The suppression is non-random — it falls on the smallest, most-distinctive subgroups. What can we infer about the SCHOOLS where the most data is hidden?

Supporting findings

• Across school_year_metrics, % rows with suppressed=true: chronic absenteeism 7%, test scores 22% (test results have more subgroup cells).
• Suppression hits SWD, ELL, race subgroup cells hardest at schools where those subgroups are small.
• Schools with the most suppressed cells are typically the most demographically-homogeneous (since the small subgroups have small N).
• Information loss: when we can't see a subgroup, we can't see how the school is serving them.
• Identify the schools where most subgroup cells are suppressed — they're the schools where 'the average looks fine' but individual subgroup outcomes are hidden.
• Is the suppression bar (n<5? n<10?) appropriate? Federal guidance varies.
• Compare with NYS's reporting: do they suppress at the same threshold?
• What can be done? Multi-year rollups can reduce suppression (combine 3 years of small cohorts).
• Aggregated borough/district-level reporting can fill in gaps.
• Frame: 'Privacy is good. But privacy without context becomes invisibility for the kids who most need to be seen.'

Reporting directions

• Compute suppression rate per school per subgroup.
• Identify the schools where 50%+ of subgroup cells are suppressed.
• Compare with NYS, US Ed Dept guidelines on small-N reporting.
• Profile the kids hidden by suppression — interview an ELL coordinator at a school with 4 ELL kids and a suppressed cell.