When Utah lawmakers had the chance to kick the tires on a new privacy program the Census Bureau is planning for the 2020 data, what they found was troubling.
For this story, TPM spoke to more than a dozen experts who rely on the census data, including state demographers, redistricting consultants, voting rights attorneys and private researchers.
Agreed. This is the first article I’ve read that explains what is going on in the data collection and why it is being held up. It seems like a true quagmire. Perhaps the way of collecting it in the first place needs to be tweaked for next time…
So maybe the answer is to revert to the raw data and use the systems that have worked reliably in the past. Don’t throw out the data; throw out the analysis and start over.
I would hate to think the nation would have to go through the counting again, but we do know that there was some chicanery in how the count took place. Let’s go through the analysis again, using reliable systems with a good history, and see if that’s even necessary.
Well done! Seems to me that this is a job that needs doing (because otherwise yeah, the big data people will be able to de-anonymize census returns with ease), but in some way that the CS people haven’t considered yet. Maybe by restricting certain kind of queries or users.
And this on top of the distortions that the trump administration is injecting into the data for political purposes.
Incompetence has failed to temper malevolence in this case. The only questions being asked by the President and Secretary have been how to stop counting cities where they don’t think the people count. Organizational culture is such that people working on the question the boss is asking, get attention. When the boss is an incompetent putz who doesn’t care about the question in any event, well, the right questions, which it’s management’s job to ask, don’t get asked.
Accuracy can often be improved if the system allows repeated queries. The noise will average itself out because of the statistical law of large numbers. But this averaging strategy can be defeated if the system caches the queries and added noise so that it can output the same noisy result every time. I doubt the Census system would go to those lengths, so the averaging approach is worth a try.
Anne Klinefelter and I made these observations when looking at Facebook’s advertising interface in 2012.
I’d hate to see a recount, as well, but you mentioned “chicanery.” I’ve heard 1st-hand reports from census takers who’ve told me that the data is unreliable to begin with. TPM has reported much the same. It’s no secret. T rumpp’s rush to rig the census count to favor Republicans has led to unreliable data. Add to that the problems @tierney just catalogued here, i’m afraid a complete do-over may be required to adequately fulfill the constitutional obligation to count this nation’s residents. Yes, doing that would be expensive, and there would be delays in using the new census data, but i think that inconvenience outweighs the harms that would be caused over the next decade of using inaccurate data to formulate local, state, regional and national policies. 2 cents.
Tiger Beat On The Potomac continues to recite the “Democrats in disarray” BS that is their hallmark.
Next–Politico speaks to members of the Q Cucks Klan, er, “Proud Boys” to let them spread their plans to come with their guns to DC on January 20th to form a human wall around the White House to keep Trump in office.
Perhaps Biden could order the Census Bureau to use the extremely accurate meta data available to draw statistical inferences to give America an accurate profile of those living here?
Well, to the best of my knowledge, she isn’t wrong. I mean, what kind of training is there for a succession, other than being the longest serving/senority. Most organizations groom (right or wrong) individuals to step in as others leave. Saying it may be apolitical, but that doesn’t mean it isn’t true.
The problem is either the tool or the data. If the tool can’t reflect reality then it’s no good. If Trump found a way to fiddle with the data, then it’s no good.