According to the bureau, people fail to respond to the census questionnaire for a variety of reasons: illegal or questionable citizenship status, failure to make child- support payments, welfare cheaters -- the list is long. Kelly Grieve, programmer analyst with the Sacramento Area Council of Governments (SACOG), pointed out that recent immigrants may fear census questionnaires. "In 2000, about 25 percent of Californians are going to be immigrants, many of them from repressive societies where government agents asking questions is cause for alarm."
Statistical Sampling
To improve the accuracy of the 2000 census, in terms of numbers and demographics, Congress created a panel of experts from the National Academy of Sciences to find a solution to the problem of nonresponding households. The panel concluded that, by supplementing head counting with statistical sampling of the number and characteristics of the nonresponding households, it was possible to achieve a more demographically inclusive census with greater accuracy and for less money. The accuracy of the sampling, the panel pointed out, can be checked by comparing the results obtained with those achieved through repeated follow ups by highly trained enumerators (census takers), until all households are accounted for.
Dress Rehearsal for Census 2000
To test the accuracy of the new method, the Census Bureau in 1998 is conducting a Census 2000 "dress rehearsal" in selected parts of the country. Three areas were invited to take part: Sacramento, Calif.; the Menominee Indian Reservation in Wisconsin; and portions of north and central South Carolina. Census questionnaires have already gone out in these areas. Where only questionnaires were used, field workers will repeatedly return to nonresponding households until all have responded. As a last resort, they will bang on neighbors' doors to get the information.
In Sacramento, where the bureau is testing only the sampling method, 90 percent of the assigned addresses will be accounted for by returned census questionnaires, while the number and character of the remaining 10 percent will be determined from statistical sampling. The accuracy of the sampling will be checked against results obtained from the follow-up enumeration.
In South Carolina, 11 contiguous counties make up the largest of the three areas selected to take part in the dress rehearsal. According to the bureau, the counties are an ideal mix of rural and urban populations for testing both traditional and statistical sampling methods. In 1997, the selected areas of the state in the dress rehearsal were first requested to conduct a local review of their address lists and compare it with the Census Bureau's Master Address File (MAF) and the Topologically Integrated Geographic Encoding and Referencing (TIGER) file, developed by the bureau as a mapping source for the 1990 census and related survey programs.
However, several rural jurisdictions with limited staff and technical resources were forced to turn to the state's Office of Research and Statistics (ORS) for assistance in carrying out the task. Fortunately, ORS was in the middle of its own three-year Addressing and Geocoding Project (AGP), updating the TIGER file for the entire state in preparation for Census 2000 and simultaneously developing a statewide geographic database from E911 and other government services. As a result, ORS was able to help five dress-rehearsal counties evaluate the completeness of the census address list and TIGER maps.
According to ORS GIS managers David Alexander, Jack Maguire and Jonathan Fisk, technical limitations in several rural jurisdictions precluded local governments from handling files of the size sent by the Census Bureau. Using Environmental Systems Research Institute's ArcView, however, GIS personnel at ORS worked around the problem by offering telephone technical support and/or translating the fixed-width files to formats local governments could work with. They were also able to help several counties assign Census 2000 collection blocks to address records and add them to the census list. New local addresses were geocoded to the TIGER, already updated by the state's AGP. ArcView assigned geographic codes to local address records in conjunction with special Census 2000 collection block information.
With the assistance of ORS, the five counties completed the task and, in the process, discovered that up to 30 percent of their addresses had been missing or were in error. Should other counties turn up similar results, locally updated census address lists and TIGER files for South Carolina will undoubtedly have far-reaching benefits for the state.
Updating TIGER
Another factor in the Census 1990 undercount was TIGER itself. TIGER files were based largely on data contained in the old Dual Independent Map Encoding system, developed for the 1980 census. As a result, street addresses in suburban and rural areas developed in the following decade were not included. "This first version of TIGER," said Grieve, "was highly inaccurate in terms of addressing and topology; but the bureau has since been updating the TIGER with street addresses, particularly in rural areas, and ensuring that each address is geocoded and linked with the correct census block." The Census Bureau, local government and the private sector have also been developing automated methods for transferring information from remote spatial databases to TIGER, a process that has led to shared responsibility for accuracy of the file.
To ensure an accurate TIGER, the bureau -- in partnership with state and local government, and the private sector -- is currently conducting the Local Update of Census Addresses (LUCA) Program. In the LUCA Program, each local agency receives the MAF and TIGER file that covers its jurisdiction to check the accuracy and completeness of all addresses. The bureau sends the files in ASCII fixed width, comma delimited or dB-4 formats. (TIGER is not a ready-to-use GIS format; it is one that major GIS systems can interpret.) The local agency then digitally checks its own address list against the MAF. Local jurisdictions draw their addresses from several databases -- county assessor, utility, E911, ZIP code and GPS coordinate. If addresses are missing, incorrect or not coded to the correct census block, the agency makes the corrections and adds any additional addresses or streets needed to bring the file up to date. Each address is then geocoded and linked to a block number (larger geographic unit) representing an address range in the TIGER. "This is a pre-census use of GIS," Grieve explained, "updating your TIGER, making sure it is correct and comparing it with the bureau's data."
In Sacramento, SACOG, the agency responsible for updating the addresses, also uses digital orthoquads and Trimble Pro-XR Kinematic GPS in specially equipped vans to capture addresses, street coordinates and to draw street patterns, particularly in new areas and in mobile-home parks, where streets and addresses are not included in assessor maps. TIGER file errors in the older areas are corrected by layering the TIGER map over geographically aligned, digital orthoquads and checking for differences between the two.
Bureau regional offices analyze the revised addresses and either accept them or send them back for confirmation. Accepted corrections are incorporated into the bureau's integrated MAF/TIGER file. Besides providing addresses to which Census 2000 questionnaires will be sent, the MAF is used to measure the completeness of the census response. It also serves as a national database of housing and businesses.
Statistical sampling (should it prove accurate in the dress rehearsal), together with an accurate, nationwide MAF/ TIGER database, will certainly contribute to a more accurate and demographically inclusive decennial census, and, in turn, more equitable congressional and state legislative apportionment and distribution of federal block grant funds. Although the MAF remains confidential and protected by Title 13 of the U.S. Code, its accuracy and scope will have vastly improved the functionality and application of the TIGER file for both public and private sectors, particularly the GIS community.
For more information, contact Kelly Grieve, programmer analyst, Sacramento Area Council of Governments, 916/457-2264 or David Alexander, GIS manager, South Carolina Office of Research and Statistics, 803/734-3781.
Bill McGarigle is a writer specializing in communication and information technology.
August Table of Contents
