A new paper proposes to help “take into account the planned presence of well-specified, well-justified noise in data releases based on the 2020 Decennial Census” thanks to disclosure avoidance. Erica Groshen and Daniel Goroff reviewed “strategies, trade-offs, and rationales associated with processing and releasing the decennial results” and offer recommendations. They specifically urge the Census Bureau to:
in addition to publishing official tables, the Census Bureau also make either the noisy measurements file (NMF) or unbiased estimates of released table entries available for research purposes. To create official counts, the Census Bureau applies processes to restore face validity to privacy-protected counts (that is, they eliminate disturbing features such as negative and fractional counts). These processes also introduce statistical bias and intractable distortions that researchers may wish to avoid whenever possible. By contrast, the NMF entries do not suffer from the statistical ills added by restoring face validity, and can be easily interpreted by trained analysts. Our other recommendations address critical needs for input to Census Bureau decisions from researchers, for development of suitable statistical tools that work with privacy-protected data, for expanded options with regard to microdata, and for steps to improve the accuracy of decennial census data overall.
- Disclosure Avoidance and the 2020 Census: What Do Researchers Need to Know? by Erica L. Groshen and Daniel Goroff. Jun 24, 2022. https://hdsr.mitpress.mit.edu/pub/tgi3gol2/release/1