The U.S. Census Bureau produces estimates of the total population at the national, state, county, and subcounty levels, and total housing unit estimates are provided at the state and county levels. Population estimates are also produced by demographic characteristics (age, sex, race, and Hispanic origin) for the nation, states, and counties. The estimates are used for the distribution of federal funds by state and local governments, and as controls for Census Bureau and other surveys. As part of an ongoing effort to improve the accuracy of these estimates, the Census Bureau undertook a comprehensive research effort to evaluate both its current methods and alternative methods. This effort included research from experts internal and external to the Census Bureau. The results of this research will inform methodological decisions for the decade after the 2010 Census by providing information that will help answer the five questions below. This paper examines the accuracy of the Administrative Records (ADREC) method, which is currently used by the Census Bureau to produce county-level population estimates, and several alternative series of housing unit-based population estimates. Specifically, this paper provides information to help answer the following research questions.
1) What level of accuracy is obtained through the use of the ADREC method at the county level for different size and percent change categories?
2) Does the ADREC method produce results that are substantially better than the housing unit method at the county level?
3) Can the use of data from the American Community Survey (ACS) improve the housing unit method estimates to a level where they are comparable to the ADREC method estimates?
Two other research questions are being addressed separately by contracted researchers external to the Census Bureau.
4) Does the ADREC method have the potential to serve as an alternative or a replacement for the housing unit method at the subcounty level?
5) Can multiple indicators of population size and/or percent change be used in a regression-based method to produce population estimates that could serve as an alternative to the ADREC method?
First, we provide measures of accuracy that give historical context to the performance of the county estimates produced using the ADREC method. Then, we provide additional measures by size and percent change categories to allow for an in-depth assessment of the accuracy of the estimates produced last decade. We provide and compare measures for ADREC estimates produced with and without the incorporation of challenges from local governments and Special Censuses.
Finally, we compare the accuracy of housing unit-based estimates developed using 1, 3, and 5year estimates of occupancy rates and persons-per-household (PPH) from the ACS to the accuracy of the ADREC estimates. The Census Bureau began to assess the accuracy of estimates produced using this approach with the work of the Housing Unit Based Estimates Research Team (HUBERT) (U.S. Census Bureau, 2008). These comparisons are a continuation of the assessment of the opportunities the ACS offers for more accurate housing unit-based population estimates.
The measures we use to assess differences between the estimates and the census counts in this paper were selected through a collaborative effort with members of the Federal-State Cooperative for Population Estimates (FSCPE) and other external researchers. These measures were selected prior to the comparison work for transparency. When considered together, they are intended to provide an accuracy profile where each captures a different aspect of estimate accuracy. These measures include the mean absolute percent error (MAPE), mean algebraic percent error (MALPE), root mean squared error, percent difference thresholds, and total absolute error of shares. A brief explanation and the formula for each of these measures are provided in Appendix 1 of this document.
Population estimates are developed by starting with the census count and measuring population change since the last census. Estimates of change between two censuses, both nationally and locally, reflect true change and change in net coverage between the two censuses. In comparing our estimate of change to the differences between two censuses, we are not able to distinguish between these two types of change. In this evaluation, we assume the change between Census 2000 and the 2010 Census reflects true change and assess accuracy as the degree of closeness between the population estimates and the 2010 Census counts.