Development Process

The SafeTab-P algorithm was developed by Tumult Labs with input and review from Census experts on Disclosure Avoidance. The specification of the privacy parameters used by SafeTab-P were a result of iterated experimentation and review with subject matter experts in the Bureau

Tumult developed the core algorithm using their own simulated versions of the CEF (Census Edited File) which comes out of the 2020 Decennial survey operations. Point releases of the algorithm were then provided to the Census Bureau which ran them on the actual CEF data under different privacy configurations. The subject matter experts from the Bureau's Demography Division then reviewed the outputs. Based on this review, both the core algorithm and privacy parameters were adjusted to better balance accuracy and disclosure prevention.

Unlike the microdata produced by the TDA for the bureau's earlier 2020 products, the Tumult microdata is not normalized or otherwise subject to numeric constraints. Negative counts, especially at small geographies, are relatively common counts at higher geographic levels may not be consistent with lower levels. In some cases, results at small geographies were suppressed and in others, adaptive design grouped age values into coarse ranges rather than providing exact values.