Short answer. There is no single numeric minimum sample size in 6 RCNY §§5-301–5-302 that automatically determines whether an audit is valid. Small samples can make rates and impact ratios unstable, and §5-302 permits reliance on test data when insufficient historical data is available to conduct a statistically significant bias audit.
What the rule says
Historical data is the default data basis. Under §5-302(b), test data may be used when insufficient historical data is available to conduct a statistically significant bias audit. If test data is used, the published summary must explain why historical data was not used and how the test data was generated and obtained.
What the rule does not specify
- No universal minimum number of records.
- No universal minimum number within each demographic category.
- No rule stating that every result above or below a particular ratio is statistically significant.
Why small samples matter
When a category contains very few observations, one selection or one score can materially change the reported rate and impact ratio. The arithmetic can still be correct while the result remains unstable or difficult to interpret.
Professional review questions
- Is the historical population complete for the defined use case?
- Are small cells caused by the audit period, job structure or demographic missingness?
- Would combining distinct populations create a misleading result?
- Does the available history support a statistically meaningful audit, or does §5-302(b) need to be considered?
- What limitations should be disclosed in the technical report?
Do not confuse this with the 2% rule
The under-2% provision in §5-301(d) is a specific permission concerning certain required impact-ratio calculations. It is not a general substitute for statistical-significance analysis.
Last legally reviewed: September 27, 2026.