This has probably been asked and answered on this forum before somewhere, but there's a question I've had for a while: Did this whole differential privacy thing originate within the Census Bureau (okay, maybe not within, but concerned citizens just mentioned the issue with no legal threats involved and Census Bureau officials decided on their own to adopt it; I understand that development began before the 2010 census but wasn't ready then), OR was there a lawsuit or specific threat of a lawsuit on 14th Amendment grounds (right to privacy) that led to the Bureau taking their current course?
These really don't answer why, but provide a little more information.
https://www2.census.gov/about/cic/Differential%20Privacy.pdfhttps://hdsr.mitpress.mit.edu/pub/dgg03vo6/release/1https://www.census.gov/programs-surveys/decennial-census/2020-census/planning-management/2020-census-data-products.htmlI'm making all this up as I write ... don't read it if you are on a low salt diet.
There is a statute that says the Census Bureau can't release individual data. This might be rather old law that meant census forms were confidential.
In 2010, data swapping was employed. IIUC, this involves swapping (same-sized?) households between nearby census blocks. This would mean census block counts were accurate but racial characteristics would be off - but perhaps might be accurate at a block group or census tract level.
There has probably been academic research on the subject and likely a flow of people between academia and the Census Bureau.
The Census Bureau found that they were able to undo their 2010 swapping based on using all their published data.
I'm not sure I understand what information can be gleaned from the Census Data alone. Race, sex, age, household relationship, rent/owned. Maybe they can also add in the ACS data.
They have run a test using the 2010 Census Data but applying their differential privacy algorithm. But they then had to run various fix-up to make sure counts were not negative, etc. Those changes had a systematic bias against less populated areas.
They can also apparently set parameters such that more privacy means less accuracy, more accuracy means less privacy. But I have heard of an anecdote where some towns in Utah were 30% off.
Are they going to have to tweak these parameters for the 2020 Census?
Will their be sufficient accuracy for Hudson wards, or Maine House districts?