Disclosure Avoidance for the #2020Census...
TL;DR?
Yes. It's rly dense material. Let me summarize it for you:
Good news! but also bad news.
Here's my thread...
1/n
TL;DR?
Yes. It's rly dense material. Let me summarize it for you:
Good news! but also bad news.
Here's my thread...
1/n
https://twitter.com/censusSDC/status/1410704538235936768
First, some recap: #DifferentialPrivacy is process that Census chose to distort #2020Census data tables. DP adds "random noise" to (actual) pop counts 👉🏽They do this to frustrate any1 who'd want to "sudoku solve" the age, race, & who's-related-to-who... in your neighborhood. 2/7
The GOOD news: Setting the data accuracy "budget" higher (announced in June) allows Census more accuracy (better than prev'ly discussed) for most counties & cities (all but the smallest). And almost-OK accuracy for other geogs of concern (AIAN areas, lege districts). 3/7
Now the bad news: Many external stakeholders (like me) asked for more postprocess rules to head-off ridiculous published results at the most local level of geog: Blocks.
But Census execs didn't act on our recommendations; they Do Not Want more postprocess steps. 4/7
But Census execs didn't act on our recommendations; they Do Not Want more postprocess steps. 4/7
@uscensusbureau is acknowledging, yes, random noise treatment can cause "impossible and improbable" results.
And they will measure the incidence of "impossible and improbable" in the published data! But they're not going to prevent it. 5/7
And they will measure the incidence of "impossible and improbable" in the published data! But they're not going to prevent it. 5/7
For example: The table on avg household size will show many places w/ less than 1 person per HH.
@uscensusbureau says they won't solve that. But they *will* report that this happens in 308,000 blocks. (yes, realy, that's the number for that specific problem.)
6/7
@uscensusbureau says they won't solve that. But they *will* report that this happens in 308,000 blocks. (yes, realy, that's the number for that specific problem.)
6/7
TL;DR? The takeaway is: the #2020Census data to be published this Aug-Sept may not be useable at the Block level for all purposes you or I would want.
/end/
/end/
• • •
Missing some Tweet in this thread? You can try to
force a refresh
