Thread by @HistDem

There has been so much happening on the Census Bureau’s disclosure avoidance plans that it can be hard to follow. Here’s a quick guide to key developments over the past three months. /1

Back on March 10, the State of Alabama filed a lawsuit objecting to the use of differential privacy in the census, arguing that the infusion of deliberate errors into the data is unconstitutional./2

The lawsuit also argues that the decision to implement differential privacy and to delay the redistricting numbers is “arbitrary and capricious” in violation of the Administrative Procedure Act./3

The lawsuit is being reviewed by a three-judge panel. A decision is expected soon. Regardless of the outcome, any appeals will be referred directly to the Supreme court. /4

My testimony is here: users.hist.umn.edu/~ruggles/censi…
Other documents related to the case are here: brennancenter.org/our-work/court… /5

The Census Bureau argues that differential privacy is needed because of the threat of “database reconstruction.” /6

I argue in this thread that the Census Bureau grossly exaggerated the threat, and that there is no realistic threat to privacy from published tables. /7

https://twitter.com/HistDem/status/1393640235809001476

I produced a working paper with @dcvanriper that showed that the great majority of purported Census Bureau re-identifications could be explained by chance. /8
assets.ipums.org/_files/mpc/wp2…

On April 28, the Census Bureau released their final “demonstration product,” a noise-infused version of the 2010 census redistricting data so that outsiders could evaluate how the new disclosure avoidance procedures would affect usability of the data./9

Interested parties had exactly one month to provide feedback to the Census Bureau. Several feedback documents have now appeared publicly, and the results are devastating. /10

For example, @Chris_T_Kenny et al. find “Our analysis finds that the DAS-protected data are biased against certain areas, depending on voter turnout and partisan and racial composition.” /11 alarm-redist.github.io/posts/2021-05-…

Similarly, @AABeveridge argues that the noise infused data violate the traditional redistricting principles required by the courts. /12
dropbox.com/s/i9sfutxwj8f1…

@JTomMueller and @AppDemography “find the method introduces significant error into growth ratesat the county level for all groups except the total and non-Hispanic white population,” especially in rural areas. /13
osf.io/preprints/soca…

@ipums found “pervasive biases and inconsistencies, high levels of inaccuracy in the counts of minority populations, and isolated large errors in the population counts for particular communities.” /14
users.hist.umn.edu/~ruggles/Artic…

In the midst of all this, on May 20 the Census Bureau gave a presentation at the ACS data users conference in which they announced that the ACS public use microdata files would be replaced with fully synthetic data by 2024. /15

I tweeted about why this would be terrible. /16

https://twitter.com/HistDem/status/1395774558096039938

On May 27--a few days after Twitter erupted in protest--Acting Census Director @jarmin_ron backtracked from the ACS presentation, tweeting “no decisions have been made or are imminent on introducing a synthetic ACS PUMS.”/17

The newest development will occur on June 4, when the Census Bureau will present a webinar to demonstrate why they cannot revert to traditional methods of statistical disclosure control for the 2020 census. /18

They argue that even if you swap 50% of the households while altering 50% of household sizes and perturbing 70% of tracts, it somehow does not have a significant impact on the percentage of “re-identifications.” /19

I argue that this is because the overwhelming majority of their reidentifications are fake. /20

https://twitter.com/HistDem/status/1400169810512822281

https://twitter.com/HistDem/status/1400525776542384129

That brings us up to the present. The saga continues tomorrow. /end