Comments at Census Conference on Estimates, July 19, 2006

by Linda Gage
State of California

The Census Bureau’s Basic Principles Paper

“The U.S. Census Bureau’s Intercensal Population Estimates and Projections Program Basic Underlying Principles” paper is thoughtful and thorough and makes a tremendous contribution to estimates users by describing and rationalizing the Bureau’s current methods and strategies for producing intercensal estimates. The clear statement of principles allows discussion and evaluation of the decision points and processes guiding the production of the estimates and projections.

The principles are reasonable. While the majority of the principles are endorsed, or at least accepted; some will be emphasized, some questioned, and others suggested.

The State of California and State Needs

The Census Bureau notes that the estimates products are used for a variety of purposes such as controls for surveys such as the Current Population Survey (CPS) and the American Community Survey (ACS), distribution of federal funds, in the calculation of rates and program planning. States may also distribute state funds based their own independent estimates or on the Census Bureau’s estimates. Population estimates are used for program evaluation, needs assessment, qualification for grants, to describe the population size and characteristics of an area and to compare areas. Within the state, estimates are used not only by state government agencies but also by councils of governments, cities, counties, and special districts, the academic community, private sector researchers, and public, non-profit organizations, utilities, the press and the public.

State needs for population estimates are consistent with those of other data users. The basic need is for population estimates that are accurate. That need is paramount to the desired attributes of credible, stable, and timely. States needs may be satisfied or thwarted by the timing and content of population estimates publications but are not necessarily correlated with how the Bureau produces the estimates. Only a critical core of state users will engage in discussions of the elegance and efficiency of the actual methodology, issues of evaluation and whether the estimates can be replicated. This is true of state-produced estimates as well.

The State of California instituted a Population Research Unit in 1951 in response to a need for official population figures for the state budget. The population estimating program continues to prepare population and housing estimates for the state, counties and cities based on a variety of administrative records and data from local governments. These population estimates are mandated in the California State Constitution and Codes and used for:

  • State agency budgeting

  • Planning, research, and evaluation

  • Appropriation limits (restricting local property tax increases in excess of the percentage increases in the Consumer Price Index and the prior year’s population estimate)

  • Proposition 98 Guarantee (funds guaranteed to public k-8 schools)

  • Distribution of State funds to cities and counties and for air pollution, health, and transportation programs

  • To set compensation for county and city officials

  • To determine the number of liquor licenses available in each county

  • Regulatory programs (The State’s population estimates are referenced in several Code sections including Administrative, Alcoholic Beverage, Business & Professions, Education, Elections, General Government, Health & Safety, Insurance, Penal, Public Health and Safety, Public Contract Public Resources, Public Utilities, Revenue & Taxation, Streets & Highways, Vehicle, Water, and Welfare & Institutions)

The State of California’s Driver License Address Change method for producing population estimates has been used for over 35 years. This method for estimating population using driver licenses to assess migration patterns has consistently yielded more accurate results than other methods used by other organizations, including the U.S. Census Bureau. We use data from about 20 federal and state agencies ranging for the California Departments of Education and Corrections to the federal Internal Revenue Service and the Department of Homeland Security.

The estimates are prepared by professional demographers, devoted to producing the highest quality estimates for the state, employing tested methods. California, like some other states that produce independent population estimates, employs state-specific data to enhance the accuracy of the estimates, especially the domestic migration component.

In the 1980’s the Census Bureau averaged their estimates with those produced by states with approved data sources and methodologies. Since the 1990s the Census Bureau and the State of California estimates begin diverging soon after the decennial census benchmark. There was a difference of 6,000 persons in the two estimates for July 1, 1990, both based on the April 1, 2000 census. The difference was magnified to 890,000 by the July 1, 1999 estimates. The overwhelming reason for the difference is due to the estimation of net domestic migration. Similarly, in the course of producing population estimates for the 2000s, the two sets diverge by 867,000 in July 1, 2005 (well above the 569,000 difference in July 1, 1995). Again, the predominant cause is the domestic migration estimate.

The State of California estimates are very visible. In addition to two population press releases each year, every city and county receives a monthly check from the State based on the California-produced population estimates. The discrepancies between the State’s and the Bureau’s estimates are a source of great consternation and confusion among policy-makers, local government officials, data users, and those who report the news. The differences raise questions about the credibility of the Census Bureau estimates and hamper the use of the American Community Survey which is controlled to the Bureau’s population estimates.

Several states produce their own population estimates, some are also discrepant from the Census Bureau’s, some also collect and incorporate state-specific measures of migration, all participate in the Census Bureau’s Federal/State Cooperative Program for Population Estimates (FSCPE) under a jointly signed Memorandum of Agreement.

The Federal/State Cooperative Program for Population Estimates (FSCPE)

Informal cooperation between the Federal Government and the states in the area of local population estimates existed as early as 1953. In 1966, the National Governor’s Conference, in cooperation with the Council of State Governments, initiated and sponsored the ‘First National Conference on Comparative Statistics,’ held in Washington, D.C. This conference gave national recognition to the increasing demand for subnational population estimates. Between 1967 and 1973, a group of Census Bureau and state employees charged with developing annual subnational population estimates, formalized the Federal-State Cooperative Program for Local Population Estimates. During this period, the FSCPE members identified several objectives that were later codified in the FSCPE BY-LAWS:

  • promotion of cooperation between the states and the U.S. Census Bureau;

  • preparation of a set of consistent and jointly prepared county and subcounty estimates with complete state coverage;

  • assurance of highest quality estimates through the use of established methods, comprehensive data review and thorough testing;

  • reduction of duplication in the production of population estimates and improvement of communication among the groups compiling population figures;

  • improvement and advancement of techniques and methodologies and the encouragement of joint research efforts, and

  • enhancement of the recognition of local demographic work.

FSCPE agencies operate under signed joint statistical agreements. The agencies provide input data: births, deaths, group quarters population change; review Census Bureau preliminary files: housing units, estimates and characteristics; and receive press briefings and releases on an embargoed basis before or in tandem with the media to allow them to prepare to provide expert commentary from a local perspective.

In theory, state FSCPE agencies, each designated by their respective governors, work in cooperation with the Census Bureau’s Population Estimates Branch to produce subnational population estimates. In reality, member agencies supply requested input data to the Census Bureau which may or may not be used in their estimates.

The Population Estimates Branch begins the process of preparing population estimates by updating population information from the most recent census with information found in the annual administrative records of Federal and state agencies. The Federal agencies provide tax records, Medicare records and some vital statistics information. The FSCPE agencies supply vital statistics and information about group quarters like college dorms or prisons. The Census Bureau and FSCPE members use statistical models that combine the census and administrative records information to produce current population estimates consistent with the last decennial census counts. After the Population Estimates Branch produces estimates, they are sent to the FSCPE agencies for review.

However, it is difficult to provide review comments that result in changes. The FSCPE agencies do not have access to all input data or to uncontrolled as well as raked results. It is not possible for those outside the Bureau to thoroughly review the estimates. There is currently little flexibility to incorporate responses revealing contradictory data or alternate input data.

At this time, except for group quarters information, the Census Bureau relies exclusively on Federal statistics for input data to the estimates. The review materials provided don’t include the original data, such as the current IRS migration file or Medicare enrollment data, the potential person-based migration data, merged administrative records, or the impact of any rakes of the data. Review comments on migration must necessarily be based on older data that have been released or local sources that the Bureau does not consider. Without input data it is not possible to even check for data entry errors. A side issue is that the data provided to the FSCPE for review are more detailed than what is publicly available. This limits discussion about the estimates with our local governments. That leads to transparency problems both for our review and the level of detail provided to the public about the estimates. A principle should address this issue to either change or affirm this approach.

The Census Bureau does not inform the estimates process with evaluation of state-generated estimates. Though the estimates and data are not directly comparable on a methodological basis they could be used as a way of assessing the accuracy of Bureau estimates and discovering alternate data and method. Many states provide input data to the Census Bureau and some prepare independent estimates. At present, the Census Bureau practices of controlling state-supplied vital events to the numbers reported at the national level for the prior calendar year by the National Center for Health Statistics (NCHS) and the modeling of race data severely distort more current though slightly incomplete data from the states. This, along with different estimates of domestic migration, results in “dueling estimates” in the public domain and perennial confusion.

An evaluation of several states test estimates for April 1, 2000, along with those of the Census Bureau, showed that some states are producing more accurate estimates and distributions of the population within their states than the Bureau. We recommend a principle of conducting joint research to determine whether the Census Bureau methods can be designed to accommodate additional data and techniques to improve the accuracy of the intercensal estimates. The goal of the Census Bureau and the states is to produce the most accurate estimates possible.

The Basic Principles

The FSCPE Research and Methods Subcommittee offered suggestions on an early draft of the "Guiding” or Basic Principles of the Census Bureau's estimates program with the knowledge that the principles would be the focus of a future meeting of stakeholders. After much discussion and consideration, the subcommittee made these comments:

  • It may be helpful to partition the principles according to estimate type and geographic level. Different assumptions and production constraints govern national, state and local estimates.

  • The principles need to be consistent with the FSCPE Bylaws and Operational Guidelines or conversely the governing documents of the FSCPE are in accord with the affirmed basic principles.

  • The principles should reflect new procedures to incorporate FSCPE involvement in the production and review of estimates. They should reflect a provision for formal FSCPE participation on the Estimates Change Control Board.

  • The principles need a section that addresses a routine process for evaluating the estimates, whether in relation to decennial census data, survey data, or other measures. The section should include provisions for sharing the evaluations with the FSCPE and stakeholders.

  • The principles should account for the necessity of periodically revising estimates series, whether through methodological change or input data improvement. This is an issue for many, and should be treated prominently in the principles.

  • While this may be a specific methodological point which may be too detailed to find its way into the principles, it bears attention. If the estimates are to be used as controls in the American Community Survey, it is important to resolve or publicize differences in residency rules between the ACS and the decennial population estimates base.

  • The principles need to address the population estimates challenge process. The number of challenges has grown from 3 for the 2001 vintage when the population was adjusted by 12,979 to 38 for the 2004 vintage when population was changed by 345,392. If jurisdictions can provide compelling data that result in a population change then the Bureau should explore collecting and incorporating such data in the basic estimates process and methodology. The principles should address methods to adjudicate successful challenges that do not negatively affect all non-challenging jurisdictions.

At that time, we did not embrace or reject any of the Census Bureau’s current estimates methods. In the absence of a proven superior method or methodological framework, the “Guiding Principles” for now must adopt the current methods and general framework. Since that time, substantial progress that has been made by the Census Bureau and the FSCPE Research and Methods Subcommittee regarding the capability of states to report more timely vital statistics data and the processing of the data to preserve the integrity of the input data. Future research could explore the addition of data sets to assess migration patterns:

  • The California Driver License Address Change monthly report provides information on licensed drivers who apply for a new license or address change within the United States. Driver licenses surrendered by former Californians are returned to the state’s Department of Motor Vehicles. Information on the driver license includes date of birth, gender and prior address. The state’s demographers use these data to estimate the domestic migration of the 18 to 64 year old population both to and from the state and between California counties. The State also supplements the data Census Bureau collects on building permit activities with a survey asking for permits, certificates of completion or occupancy and demolitions.

  • The Alaska Permanent Fund Database covers 95% of Population and includes date of birth, gender, physical address, alien status. It does miss in-migrants for the past year, transients, about half of the military and dependents, and prisoners. The State of Alaska uses this database when producing their official county, place, Census Designated Place, and special area estimates as well as age/sex characteristic estimates and migration estimates.

  • The Utah FSCPE uses a roster of Latter Day Saint’s members to augment domestic migration estimates. The data indicate size and movement of a substantial portion, about 70 percent, of the state’s population.

Emphasis, Challenges, Additional Views, and Missing Links


  • Improve the accuracy of the annual migration estimates by age, sex, race, and Hispanic origin for counties by maximizing the efficient use of available administrative data files, Census 2000 data, and the American Community Survey (ACS) data. (Include the use of state administrative data files.)

  • Use current, direct measures that are of known good quality to the extent possible

  • Use the most effective and understandable model possible in favor of a more complex but not more accurate procedure.

  • Develop procedures to systematically incorporate participation by State FSCPE Agencies in the production of state and county population estimates


  • Question the Principle of a top down methodology with the national estimate as a control, but if the methodology is affirmed, concentrate efforts on improving the accuracy of national and state estimates rather than refining the population characteristics. The initial April 1, 2000 demographic analysis estimate of the U.S. resident population implied a net census overcount of 1.8 million persons. The validity of the components of change and population characteristics is evaluated in reference to the accuracy of the total estimate and independent data or observation.

  • Question that the cohort-component method is the preferred method for development of the national, state, and county-level total population estimates and population estimates by characteristics. The current method should be compared and assessed with other methods as well as independent estimates and other methods should be considered. For example, the Population Association of America members of the Census Advisory Committee of Professional Associations noted that the Census Bureau’s research shows that housing unit based methods of population estimation hold promise as an alternative to the current components of change approach. They recommended that the Census Bureau continue to explore the utility of adopting housing unit based methods of population estimation at county and sub-county levels, as part of an integrated approach to the creation of intercensal population estimates. The Bureau should actively seek to understand and minimize differences between the Census Bureau and FSCPE estimates. The State of California has formally and actively requested such a dialogue for over two years. The most accurate estimate often results from averaging different methods. (This was previously done and should again be explored as an alternate method.)

  • Question the practice that state total population estimates are not developed independently. National population estimates are first developed; then county total population estimates are developed and controlled to the national total population estimates. The state total population estimates are the sum of the “nationally controlled” county total population estimates for the state. (Test developing independent state estimates.)

  • Seriously question the use of a consistent method for entities at the same level of geographic aggregation. (Use the data and method that produce the most accurate estimate for each state.) We can do better than a “one size fits most” method.

  • Two activities to improve estimates of net international migration appear marginal and the benefits should be described. 1. Construct algorithms to estimate the migrant status of the foreign-born populations and 2. Produce estimates of international migrants by migrant status (legal migrants, temporary migrants, quasi-legal migrants, unauthorized migrants, and emigrants). The second activity is especially sensitive to the Census Bureau’s Race and Ethnic Advisory Committee members and members of the Bureau’s Census Information Centers.

Additional Views

  • Develop a stronger collegial role for FSCPE members. Use experience of the state partners who have additional expertise, resources and insights for their states, counties and communities and may have state-specific data and methods that could improve the Bureau’s intercensal estimates. Actively explore the inclusion of state data, methods and estimates. The successful (first-ever after nearly four decades) joint research on vital statistics demonstrates the ability of the FSCPE network to work with the Census Bureau to improve the data and methods for the population estimates. Specifically, on the advice of an FSCPE member and substantiation by the Census Bureau, accept the more current state-supplied vital statistics as direct input data rather than modeling older National Center for Health Statistics data.

  • Develop an alternative method for age estimates. For example, age estimates may be better achieved by using different administrative data for different age groups and by replacing enrollment data as the only migration indicator.

  • Develop age/sex estimates and race/ Hispanic Origin estimates separately, rather than by age, sex, race and Hispanic origin. Those comprehensive breakdowns, especially amid the complex race categories, may be useful but may not be reliable. To separate the race estimates from the age estimates may get better results for both.

Missing Links

  • Provide a compelling rationale for selection of preferred methods, citing prior research and results and describing alternatives.

  • Specify criteria for determining the “method that produces the most reasonable estimates,” “the reasonableness of the total change in the population since the last decennial census,” and “the reasonableness of the annual change in the estimates since the last census.”

  • Operationally define, follow, and monitor the adopted basic principles.

  • Determine or discuss the weight given to each listed criteria for judging a method and describe the process for changing methodology.

  • Include FSCPE partners and other external reviewers in the discussion and decision to change methodology

  • Specify the strict criteria and provide examples of the objective evidence necessary for state members of the FSCPE to follow as they undertake the review of the state and county population totals prior to final production.

  • It appears the immigration data from the Department of Homeland Security’s Office of Immigration Services has been abandoned. The report of newly arrived legal permanent residents and the monthly applications to adjust from a non-immigrant to immigrant status should be assessed along with ACS data.

A Commitment

There are profound issues of equity inherent I the use of population estimates to determine receipt of funds, to measure social, health, and economic conditions; to allocate resources; and to support responsible governance. The Census Bureau is ethically-bound to continuously seek, test, and incorporate the best data sources and methods to produce the most accurate and equitable estimates possible for each state, county, and functioning governmental unit. This is an appropriate time to reaffirm my personal commitment and the commitment and cooperation of the Census Bureau’s FSCPE partners to engage in discussions, joint research, and collaboration that will result in the production of the most accurate and credible intercensal estimates possible.