Administrative Records Can Improve Local Accuracy

The opportunity to correct the differential undercount is not limited to parole and probation records. According to the Bureau, "children were much more likely than adults to be undercounted in the 1990 census ... they accounted for 52 percent of the undercount."59 The Bureau can find undercounted children in 2000 using administrative records. Several state and local programs maintain timely, accurate records with names and addresses. For instance Medicaid, Temporary Assistance to Needy Families (TANF), and Food Stamps are federal aid programs maintained locally by states and cities, and are specifically designed to aid children. According to officials in the District of Columbia, enrollment rates among the District’s poorest children in minority communities top 90 percent.

People may not have time or interest enough to fill out a census form. But no matter how busy, and no matter how economically and educationally disadvantaged a low income mother may be, if her child gets sick, that mother knows how to get her child to a doctor. In Washington D.C. and elsewhere, that child must be enrolled in Medicaid to see a doctor.

For example, local Medicaid files could be especially useful in reducing the differential undercount of children in the District of Columbia’s eighth voting precinct. Ward 8 is a predominantly Black, low-income voting precinct just minutes southeast of the nation’s Capitol. The Bureau estimates the 1990 census missed more than 1,800 children in Ward 8.60 In May 1998, District officials listed the name and address of nearly 12,000 child Medicaid recipients in Ward 8.

The Bureau could use those records to locate children most likely to be missed in the census in Ward 8. Adding people to the census using Medicaid or similar records is more accurate than adding people through statistical adjustment, which would increase, erroneously, minority population counts in Ward 3 – an overwhelmingly non-Black and non-Latino area.

High undercount areas are not randomly distributed. Instead, undercounted blocks tend to cluster in neighborhoods and larger areas such as Ward 8. The way to reduce both racial and area population coverage differentials, then, is to concentrate on those relatively few geographic areas with higher undercount rates.

The Bureau’s research has identified specific areas that were hard-to-enumerate in 1990, and probably will be again in 2000. A Bureau report indicates that most of the census undercount tends to be clustered in relatively few geographic areas. The Bureau anticipates that fully one third of the hardest-to-count census tracts in 2000 will be located in just three cities: New York, Chicago and Los Angeles.

The Bureau reports that almost half of the hardest-to-count census tracts in the country are located in just 11 cities. Those hard-to-count tracts in those 11 cities represent a mere two percent of all the census tracts in the country. The Bureau concludes that, "These data demonstrate that [hard-to-count] tracts are very concentrated geographically and can be targeted."61 Sensible efforts to find undercounted populations will target these hard-to-count areas.

Administrative records can locate people in very specific geographic areas. This approach would concentrate work in the areas where the administrative record sources indicate the largest potential undercounts in the census.

Consider the example of Ward 8 Medicaid files. The files are updated monthly and are computerized. After all non-response follow-up is completed, the Bureau could review the January 2000 Medicaid files, begin matching, and identify people from Ward 8 enrolled in Medicaid but absent in the census. (Matching records statewide or nationally, high undercount areas like Ward 8 will show up as areas with more unmatched Medicaid cases.)

The Bureau could then double-check using April’s Medicaid files. If Medicaid shows someone living at an address in Ward 8 in both January and April, whom the census has not found, add that person to the census at that address. Unmatched persons who have moved could be added to the census at their April address.

The vast majority of this work can be done rapidly and accurately using the Bureau’s matching software. Records can be sorted by geographic areas to work from areas with the most nonmatches, to areas with fewest nonmatches.

Double-checking two sets of Medicaid files answers one of the most significant concerns of the census, noted by Director Prewitt and others: validating occupancy on April 1, 2000, Census Day.

This approach might be implemented with less error than any proposed statistical adjustment. Administrative records offer fewer opportunities for failure. The work flow plan for the statistical adjustment the Bureau intended before January 25 was characterized by separate steps, hand-offs, and a tight timeframe. As outlined above, an administrative records approach can involve a much simpler set of tasks and fewer points of potential failure. After all, the statistical adjustment plan (rejected by the Supreme Court) proposed five times the workload, performed in roughly the same amount of time, as the 1990 PES. Using administrative records is more consistent with the Census Bureau strategy to "Keep it Simple."