Thread by @HistDem

The Census Bureau has conducted a new analysis that purports to show that swapping is ineffective for the prevention of reidentification attacks. /1
www2.census.gov/about/partners…

The new analysis completely misses the point, and actually provides a useful demonstration of the gross misrepresentation of the Census Bureau’s “Database Reconstruction Experiment.” /2

The Census Bureau claimed that without swapping, they could “putatively re-identify” 44.60 of the population. /3

As @dcvanriper and I explained, you would expect to find that result 40.9% of the time even if you assign age and sex randomly and then assign the modal race and ethnicity of the block./4 assets.ipums.org/_files/mpc/wp2…

What that means is that this isn’t re-identification at all—one would expect to almost as many matches between the reconstructed data and the real data purely by chance. This would be laughable if it were not so sad./5

The so called “confirmed reidentifications” are even more disingenuous. All they mean is that only about some 38% of the population in commercial data sources can actually be located in the Census file, which means that either the commercial data or the Census are really bad./6

So now in the new experiment they add swapping, and find that it doesn’t change the results. Even if you swap 50% of the population, it hardly affects the percentage of “putative” or “confirmed” reidentifications. How could that possibly be?/7

Could it be that swapping for some mysterious reason is completely ineffective, even when you swap out half of the entire population?/8

The real answer is obvious—they were never reidentifying at all, they were just getting random matches. You could swap 100% and the results would still come out the same!!/9

This is an excellent illustration of how the Census Bureau database reconstruction experiment was egregiously misrepresented, and never represented the slightest threat./10

This is a truly embarrassing moment for the Census Bureau. /end.

P.S. The Census Bureau is presenting this on Friday, in case anyone cares to sho up and ask a few wuestions: census.gov/data/academy/w…