2020 Census Data Products: Data Needs and Privacy Considerations Proceedings of a Workshop (2020) / Chapter Skim
Currently Skimming:
11 Census Bureau's Responses and Own Analyses of 2010 Demonstration Data Products
11 Census Bureau's Responses and Own Analyses of 2010 Demonstration Data Products
Pages 145-156
The Chapter Skim interface presents what we've algorithmically identified as the most significant single chunk of text within every page in the chapter.
Select key terms on the right to highlight them within pages of the chapter.
From page 145... ...
As with the Census Bureau's first presentation block on the workshop's first day (Chapter 2) , the Census Bureau divided the presentation space into two talks, with Matthew Spence (presenting work cocredited to Population Division colleague Marc Perry)
|
From page 146... ...
The Population Division's analysis calculated absolute percent differences between the original 2010 Census data and the 2010 DDP for all units at several geographic levels; the results are summarized in Table 11.1. Counties proved relatively accurate, with a median absolute percent difference of 0.2 bracketed by 25th and 75th percentiles of 0.1 percent and 0.5 percent, respectively.
|
From page 147... ...
Figure 11.1 illustrates the results for Lyon County. Spence said that the age pyramids for large counties look very good and generally match the patterns seen in the 2010 Census tabulations (indicated in the plots by black outline boxes)
|
From page 148... ...
5-year age bins and topcoded age Figure 11.1 Effect of different categorization and topcoding on age pyramids, Lyon County, Minnesota. NOTES: DAS-generated Microdata Detail File values indicated by shaded bars; true 2010 Census values denoted by unshaded boxes.
|
From page 149... ...
. Spence quickly displayed a series of scatterplots, plotting 2010 DDP counts by 2010 Census original data counts for these multirace combinations, at the state and county levels, all showing generally tight clustering for around the 45-degree equality line but some fanout and locally large shifts for smaller states and counties, for instance, the combination of Asian, American Indian and Alaska Native, and "some other race" in North Dakota.
|
From page 150... ...
2010 Demonstration Data Products Figure 11.2 Percentage of total households whose householder race is Native Hawaiian and Pacific Islander alone, 2010 Census published data and 2010 Demonstration Data Products, U.S. counties.
|
From page 151... ...
pretty much everybody has noticed" and that "we really appreciate you calling it to our attention." The first of these is the now-familiar observation that large areas and population groups tend to lose population in the DAS processing while small areas and groups tend to gain. To illustrate the point, he briefly displayed a table grouping the nation's counties by population size and computing median difference and median absolute differences between the original 2010 Census data and the 2010 DDP estimates.
|
From page 152... ...
However, the fully synthetic MDF arising from the 2020 DAS -- which cleaves person and housing unit data apart at the outset -- will not contain the full set of geographic and demographic characteristics needed for the record linkage and estimation processes to work. As brief illustration of the level of error potentially involved, Spence displayed a table showing mean absolute percent error (MAPEs)
|
From page 153... ...
. He said that the Census Bureau looks forward to taking user feedback into account to create 2020 Census data products with maximum utility for data users.
|
From page 154... ...
Adding more detail, Sexton commented that a central issue is that the TDA post-processing relies on solving what the DAS team calls "the L2 optimization problem," or in other words, using nonnegative least squares (NNLS) to find a best-fitting, nonnegative histogram to a set of noisy measurement "queries," measuring closeness through mean squared error.
|
From page 155... ...
For instance, a custom query could be used to target the collection of cells corresponding to geographic areas where combinations of three or more major race groups are very unlikely. Sexton cautioned that such special queries cannot violate differential privacy, which notably precludes looking at the 2020 Census data themselves to define the special queries (for instance, using the 2020 Census data to find those areas where combinations of three or more major race groups are very unlikely to be reported)
|
From page 156... ...
file series. He said that they are confident that implementing the complete design, which was not possible for the 2010 DDP will address these issues, adding that the Census Bureau would be very interested in gaining access to the code and analyses that data users have run on the vacancy rate problems so that the Census Bureau can verify that the problems have been solved.
|
Key Terms
This material may be derived from roughly machine-read images, and so is provided only to facilitate research.
More
information on Chapter Skim is available.