I-1 POST 2000 CENSUS IN FRANCE

Michael Isnard, L'Institut National de la Statistique et des Etudes Economiques (INSEE)
michel.isnard@insee.fr

INSEE is investigating a new Census methodology, to be implemented September 2001. This method must give, on an annual basis, the same kind of results as those obtained from a traditional census: official population at the municipality level, and statistical data at the 2000-inhabitant level. It must also give fresher data than the classical census system.

To be more precise, at the end of each year, INSEE will publish two sets of data: the first being related to the beginning of the year will be available for France and at least at a 1,000,000-inhabitant level; the second data series will comprise statistical information covering the last five years and will be available at least at the 2000-inhabitant level. Official population will also be published at the municipality level.¹

Due to budgetary constraints, the system has been designed so that the cost of the new system on 8 years will be the same that the cost of 1999 Census in France (214 M Euro). Due to legal impacts of census results, a new law must be voted by French Congress to allow the change in methodologies.

The proposed method will use a combination of surveys, censuses and administrative records. Whether a given municipality is surveyed or completely enumerated will depend on its size. A 5-year cycle will be at the core of the data collection process.

For small municipalities, a classical census will be undertaken every fifth year and one in five municipality will be censused every year. Estimation of the size of the population during the intercensal years will use administrative records as complementary inputs (mainly occupancy tax and social security files).

Larger municipalities will be divided in 5 roughly equal sets of buildings, each of them representative of the municipality itself. Every year, a census of the dwellings will take place in one set, and a population survey will be conducted in this set. The mean sample rate in the set will probably be 40%. Extrapolation of results will use the same administrative records as in smaller municipalities.

In larger municipalities, a building register will be built and updated continuously with administrative records (building permits, for example). It will be connected to the INSEE Geographic Information System which looks more or less like TIGER.

¹ There are 36,000 municipalities in France. Half of them have less than 350 inhabitants and half of the population is living in municipalities over 10,000 inhabitants.

I-2 POPULATION ESTIMATES FOR SMALL AREAS IN THE UK: PERFORMANCE AND PROMISE

Ludi Stephen Simpson, Cathie Marsh Centre for Census and Survey Research,
University of Manchester and Research Section
ludi@man.ac.uk

In England, Wales, Scotland and Northern Ireland government offices produce accepted population estimates for Districts of population 50-500 thousand population. Available data from birth and death registration, and medical registration proxies for migration, allow a cohort component accounts method giving a detailed age and sex structure for these areas.

The Estimating with Confidence project 1993-1996 evaluated the accuracy of existing methods to produce population estimates for smaller areas, of population 1-30 thousand. A mid-term enumeration- a local census - was proved the only winning strategy, but a number of cheaper desk-top methods employing administrative data gave reasonable results in most areas. The accuracy of these methods has been tabulated and attributed to characteristics of the areas estimated and the estimation strategies employed.

Proven strategies for reducing the largest errors include identifying special populations, and averaging the estimates from more than one method. A series of training sessions and publications have pooled the knowledge of demographers from business, government and academics.

A strategy to make consistent (a) previous census data with (b) current administrative health and electoral data and (c) the component estimates for larger areas has been developed to employ many of the lessons from the Estimating with Confidence project. Its promise is a library of population estimates for very small areas (postcodes, the UK equivalent of local zipcodes, with fifteen households in each unit) which, with measured reliability, can be aggregated to larger administrative and community areas.

The strategy and alternatives are under active development with co-operation between the academic sector, sub-national government, and national statistical offices.

I-3 ESTIMATING A POPULATION TOTAL USING A RANKED SET SAMPLE

Arun K. Sinha, Department of Statistics, Patna University

Ravi R. Sinha, Population Research Centre, Patna University
arunksinha@usa.net

Ranked set sampling (RSS) is a cost-effective sampling technique that provides more efficient estimators of several population parameters as compared with simple random and other classical sampling methods. This refinement is obtained because it induces stratification on a population through samples. To carry out the procedure the randomly selected units are put in small subsets, and then the units of each subset are ranked separately with respect to the characteristic of interest. Finally only one unit is quantified from each subset. Depending upon the variability of the measurements equal or unequal allocation is employed. For ranking purpose this method uses inexpensive and easily available information like visual perception, prior information, ranking correlated covariates etc., or a combination of these information. It is worthwhile to note that the exact quantification are not required for this purpose. This paper illustrates the technique to estimates population total of urban places of Bihar, a state of India. This approach appears to be of particular interest where a registration system of vital events does not work effectively. Unfortunately, this is a common feature in most of the developing and under-developed countries.

I-4 FUTURE POPULATION ESTIMATES IN DESTABLE POPULATIONS

Subrata Lahiri, Department of Public Health and Mortality Studies, International Institute for Population Sciences
subrata@hawaii.edu

In the classical method of forward projection, the population estimates 5 or 10 years later are usually obtained by multiplying the population counts at quinquennial ages in a particular census date with a suitable set of 5 or 10-year census survival ratios. This process of estimation not only requires the knowledge of life table survival ratios but also inherently assumes the equality between the census survival ratios and the corresponding life table survival ratios. It has been shown elsewhere by author using generalized population model developed by Preston and Coale (1992) that such an equality holds good only when the population under study is either stationary or stable( Lahiri, 1999).

The method, proposed here, neither requires the knowledge of life table survival ratios (borrowed from a life table selected under certain assumptions), nor requires the assumption of equality between the census survival ratios and the corresponding life table survival ratios. The present paper deals with the estimation of 10-year life table survival ratios associated to the population under study, including the 10-year life table survival ratio for the open-ended terminal age-interval from the two census enumerations, not necessarily 5 or 10 years apart (intercensal interval being not necessarily integral multiple of 5) under the assumption that the study population follows a generalized population model. The technique proposed here assumes that the population under study is closed to migration, and it further requires the knowledge of age-specific growth rates (Aggress). The Aggress can be estimated by examining the trend of Aggress based on earlier census enumerations over two or three intercensal periods. If the population under study affected by migration, it is sufficient if the Aggress are adjusted with respect to the-specific net migration rates during the period under study. Population estimates at younger ages can be treated separately under generalized population model.

I-5 POPULATION AND HOUSING ESTIMATES FOR CENSUS BLOCKS: THE SAN DIEGO EXPERIENCE

Jeff Tayman, San Diego Association of Governments
jta@sandag.cog.ca.us

Information demands for planning and strategic development in both the private and public sector often require estimates for very small geographic areas, such as census tracts, census blocks, and zip+4. To satisfy these demands, San Diego Association of Governments (SANDAG) developed an integrated system for preparing annual population and housing estimates by census block. I describe the basic building blocks of this top-down estimation system which makes extensive use of GIS and address-level symptomatic information to pinpoint the location of new housing units. I also evaluate two sources of address-level information - utility hook-ups and assessors parcel file information - that SANDAG has used to develop these estimates, focusing on their accuracy and their strengths and weaknesses as symptomatic indicators of changes in housing location. Finally, I conclude with some general issues involved in estimating block-level population and housing, including maintaining an up-do-date GIS system, data management, temporal consistency, and controlling.

I-6 USE OF PROPERTY TAX RECORDS AND HOUSEHOLD COMPOSITION MATRICES TO IMPROVE THE HOUSEHOLD UNITS METHOD FOR SMALL AREA POPULATION ESTIMATES

Warren A. Brown, Cornell University
wab4@cornell.edu

The Housing Unit Method for estimating small area population can be improved in three ways: improved count of housing units; identification of geographic location of individual housing units; and refined estimates of persons per household. This paper reports on a pilot effort in a single county to use property tax records in place of housing permits for building and demolition to improve count of housing units. In addition the property tax records include an x-y coordinate for spatially locating the housing unit structure. This enables the housing units to be geocoded to Census Blocks, as well as to non-Census geographic areas such as watersheds. Finally, the use of household composition matrices yields improved estimates of persons per household for small areas. The household composition matrices contain a refined measure of persons per household built from cells of persons per household cells by age of person and age of householder. The improved housing unit method is demonstrated for Census Block groups.

I-7 DESIGN ALTERNATIVES FOR BUILDING BLOCK ESTIMATES

Ronald Prevost, Administrative Records Research Staff, U.S. Census Bureau
ronald.c.prevost@ccmail.census.gov

The Census Bureau has been involved in a major research effort directed towards the expansion of administrative records utilization since 1996. We have begun the prototyping of a Statistical Administrative Records System (StARS) to combine several major federal administrative records systems such as the IRS individual tax returns and information returns, HCFA Medicare enrollment data, SSA Numident information, HUD Tract Rental Assistance Certification System, Indian Health Service, and Selective Service files. We have been researching several other federal micro and aggregate data files for Census Bureau statistical uses. This paper provides background information on StARS and proposes design alternatives for the creation of building block (census block and tract) population and housing estimates after 2000.

II-1 SPATIALLY ARRAYED GROWTH FORCES AND SMALL AREA POPULATION ESTIMATES METHODOLOGY

Roger B. Hammer, Paul R. Voss, Robin M. Blakely, Alice E. Magen and Daniel Veroff, Applied Population Laboratory, University of Wisconsin
rhammer@facstaff.wisc.edu

Population growth is a spatial process in which the growth patterns of small areas both influence and are influenced by neighboring areas. Spatial interactions or spatial growth forces are, for example, evident in the suburban growth patterns observed at least since the end of World War II. Over time, growth rates have risen within the boundaries of municipalities and then spilled over into other surrounding suburban municipalities, often along arterial highways, at increasing distances from urban centers. However, traditional population estimation and projection methodologies have universally treated small areas as a spatial, independent units of analysis, with their patterns of growth being based solely on their recent demographic history and the contemporaneous "symptomatic indicators" arising within their boundaries. This research will attempt to quantify the spatial growth forces acting upon minor civil divisions, as well as other small areas such as census tracts and block groups, and incorporate these forces into population estimation methods. Some work in this general area can be found in the regional science literature, yet very little by way of theory can be applied to this research without detailed investigation into the nature, persistence, strength, and direction of these hypothesized growth forces. As a consequence, we will take a highly empirical approach using a long time series of population estimates for minor civil divisions in Wisconsin. We will create a new series of spatially informed population estimates and measure their level of accuracy by comparing them to a set of earlier unadjusted estimates. The standard of accuracy will be reproducing estimates that were adjusted to both the pre-and post-estimate census enumerations.

Hypothetically, at least four spatial growth or decline forces can be delineated. Perhaps the strongest of these and the most closely associated with the growth patterns with which we are concerned can be portrayed as a growth pull factor in which a more densely populated, more rapidly growing area influences its neighbors by pulling their growth rates and population densities upward towards its own. We hypothesize that under certain conditions, a decline pull factor might also operate with a lower population density area experiencing population decline exerting a downward pressure on its neighbors, such as in some central cities and aging suburbs. Finally, it is possible that both growth and decline push factors exist. In the former, a high growth area with lower population density might exert an upward force on its more densely populated but less rapidly growing neighbors. We will begin by measuring these forces in a directional pattern, using a simple "queen's convention" adjacency neighborhood definition. In the queen's case on a chess board each adjacent square connected either by a point (i.e. sharing a common corner) or line (i.e. sharing a common boundary) is included as a neighbor and potentially exerts an independent force on the area of interest. In future iterations of this research, definitions of queen's neighborhood more complex than adjacency including radii and buffering with explicit consideration of the friction of distance will be incorporated into the measures. Each of the four force types will be calculated separately for each of the eight directions comprising the queen's neighborhood and then summed separately as well to create a measure of each of the four forces for each area. Interaction among these forces may emerge in the empirical model. For example, the growth patterns for an area with both strong upward pull and downward pull forces acting upon it might be disproportionately affected by those countervailing forces. Initially, the upward pull forces will be defined as follows with the other forces being defined in a similar manner.

II-2 DEVELOPMENT OF A NATIONAL ACCOUNTING OF ADDRESS AND HOUSING INVENTORY - A BASELINE INFORMATION FOR POST-CENSAL POPULATION ESTIMATES

Ching-li Wang, Population Division, U.S. Census Bureau
Ching.Li.Wang@ccmail.census.gov

The population estimates program to update population counts for all levels of geography in the nation will be very critical in the next decade. The Supreme Court's decision on sampling method for the census 2000 posts one challenge for the estimates program - how to determine appropriate population base for population estimates. In developing population estimates, one needs to take into account the accuracy and efficiency - time, effort, and cost. The advanced computing technology in recent years has changed the ways estimates were produced in the past. This paper is to examine the feasibility of developing a national accounting of address and housing inventory as the baseline information for population estimates.

The Census Bureau has spent so much effort to compile the address file and is working with local governments to carry out the Local Update of Census Addresses to ensure it completeness. The Census Bureau has developed the powerful TIGER system to link each living quarter to a space location and each location to a specific geographic area with mapping capacity. During the census, the completeness and accuracy of the address file will be further tested. After the census, the characteristics of housing units or living quarters associated with the addresses will become available. In order words, the MAF and TIGER has already provided the basis information about where household population live for all levels of geography. If we develop a system to constantly update the MAF and TIGER based on this existing tremendous effort and resources, we should be able to update the baseline information for population estimates in the post-censal years.

MAF Update

The address change over time due to construction of new housing units, demolition of old housing units, new designation of addresses, etc. The net increase of housing units between 1997 and 1998 in the U.S. is about 1.5 million. For efficiency purpose, the current housing unit method uses the number of permits reported by the reporting government units, in which there is no verification or quality control on where the new or demolished units are exactly located. The proposed approach is to collect the addresses of building permits, demolitions, and conversions. The GIS software has become so powerful that the census geocodes can be assigned to new addresses in a friction of time. This new approach not only can update the MAF, but also allocate the new housing units to specific geographic areas based on the geocodes of the addresses in the TIGER system.

Housing Unit Method Update

Once the MAF and housing unit count are updated, we can apply the housing unit method to estimate population based on vacancy rate, persons per household, and type of housing in terms of single unit or multiple unit. However, during the intercensal years, these variables for the housing unit method must be updated. The upcoming American Community Survey will be an important source for updating these parameters for all levels of geography. The cumulation of the information from ACS over time along with the census results allows us to examine characteristics of vacant housing units and further identify uncounted populations in the households.

Operation - Federal-state-local cooperative program

The National Accounting of Address and Housing Inventory requires the coordination of the Population Division, Housing and Household Economic Statistics Division, and Geographic Division to update the MAF and TIGER on a regular basis. More importantly, it is necessary to enhance the cooperative program between Census Bureau and states and local units of government. Building on the existing LUCA program and Housing Inventory Program, the Census Bureau collects the addresses associated with building permits submitted by local units of government. The Federal-State Cooperative Program's state agencies coordinate the compilation of the addresses, and provide assistance to their local units of government.

Benefits

The National Accounting of Addresses and Housing Inventory can benefit not only the population estimates but also the on-going surveys conducted by the Census Bureau. The cumulative effort and results of this project can provide the update address file for the next census, instead of starting over again every 10 years.

In addition, the involvement of local units of government in updating addresses and housing units would provide educational opportunity for localities to build up and maintain their capacity in collecting local information.

As the sampling method to supplement the census is banned, the ability to identify hard-enumerated populations in local government becomes more critical. Through collecting and reviewing addresses in their own localities, local governments would enhance their role in the census process. Therefore, the long term benefit of the National Accounting of Addresses and Housing Unit is not only for the population estimates program but also for the decennial census and other surveys using addresses for sampling.

II-3 A DISCUSSION AND EXAMINATION OF TECHNIQUES TO ESTIMATE TOTAL POPULATION AND POPULATION BY AGE FOR 400 JURISDICTIONS IN FLORIDA

William O'Dell, Marc Smith, and June Nogle, Shimberg Center, University of Florida
billo@ufl.edu

Introduction

The Florida legislature passed growth management legislation in 1985 that, among other components, required all cities and counties in the state to prepare and adopt comprehensive plans. The plans were submitted to the state according to a rolling schedule with the first submissions coming in 1988 and the last in 1991. One of the required elements of the comprehensive plan was a housing element. In response to problems identified in the housing elements submitted, the legislation was passed in 1993 to establish a uniform methodology and data source for housing element. Florida's land planning agency contracted with the Shimberg Center to create the affordable housing needs assessment methodology. The charge in developing the needs assessment methodology was that it be based on readily available statewide data sources, be simple and flexible, and provide the opportunity for the inclusion of locally-generated data.

A single methodology was needed that could be used in cities below 2,500 population, unincorporated areas of counties, and cities such as Miami and Tampa.

There are two elements of the needs assessment, a supply-side and a demand-side. The demand-side is founded on estimates and projections of households by age, tenure, size, and income. Population and population by age estimates and projections drive the household estimates and projections.

Affordable Housing Needs Assessment Methodology in Florida

Critical decisions in developing the methodology were necessitated by several factors. First, the methodology must be designed to be used by all jurisdictions regardless of size (and it provides housing needs numbers at the jurisdiction level, not for smaller geographic areas). Second, data sources need to be accessible and low cost so that all jurisdictions can implement the methodology themselves if they so choose (and the methodology should also recognize that some jurisdictions may have better data and should be able to incorporate that data). Third, many communities have already developed population projections, but to have a uniform methodology across the state the methodology develops population projections.

Issues in Analyzing Housing Demand

Small Area Population Projections - Population projections for jurisdictions and the unincorporated portions of counties are based on trends since 1980, using a series of methodologies as described below, and are adjusted to University of Florida's Bureau of Economic and Business Research figures (the state demographer). The methodology uses 1995 as the base year and provides projections for the years 2000, 2005, and 2010. The BEBR's 1995 population estimate for each jurisdiction is used as the base year population. To estimate and project housing demand, the next step is to divide the population into households. Finally, these households are allocated across tenure classes, size, and income groups. The way in which the population divides itself into households is related to a number of economic and social factors including income, housing prices, governmental assistance, marriage and divorce rates, and the mobility of the population. While household sizes declined significantly in the 1970s and continued to decline in the 1980s, the rate of decline slowed and household size has been relatively constant across the state between 1990 and1995. Further, factors that lead to changes in household size are not exhibiting a clear pattern that would lead to a convincing conclusion as to the direction of change in the future. The methodology therefore assumes that household formation rates and the distribution of household characteristics remain constant in their 1990 proportions. However, changes in the age distribution of the population would be expected to lead to shifts in average household size as different age groups have different propensities to form households. Therefore, the number of households is estimated using age-specific headship rates to reflect the projected changing age structure. The changing age structure is also a component in projecting the changing real income distribution. Income, prices, and rents are assumed to be measured in constant 1990 dollars (for income, 1989 figures as reported in the 1990 Census). This is the last date for which measures of income and rents are available.

Data requirements are minimal and include jurisdiction and total county population for base and launch years (1980, 1990, 1995), using census data or BEBR estimates (for 1995). For target years (2000, 2005, 2010) BEBR medium range county projections are used. (If a community believes that a different BEBR projection better reflects their community then they may propose that alternative.)

The four basic projection techniques used in the methodology include the linear, exponential, share and shift methods. The linear and exponential techniques use the mathematical extrapolation approach; they take the jurisdiction's population from the base period and extrapolate it into the future. The shift and share methods use the ratio approach; they express the data as ratios or shares of the larger, parent population, for which a projection already exists. Therefore, these techniques require a county or parent population projection. The linear and share techniques use both 5 and 15 year base periods, resulting in a total of six projections. We attempt to minimize the influence of those techniques by removing the high and low projections for each jurisdiction, averaging the remaining four and then controlling that average to the BEBR county projection.

Householder by Age and Tenure Projections: Households are the fundamental unit of demand for housing, and are the way in which the population divides itself to occupy housing units. One member of a household is considered the representative of that household and is referred to as the householder. The percentage of the population in a given age group that are householders is the headship rate in that age group, or the propensity of persons in that age group to be household heads. Therefore, headship rates allow the conversion of the population of an age group into households. Different age groups have different propensities for forming households, so that as the age structure of the population shifts, the number of households that a given population would yield would also change. Estimates and projections of households are therefore based on age-specific householder (headship) rates. These headship rates are applied to the age-specific population projections calculated in the previous section.

The remaining sections of the needs assessment build on estimates and projections of households. Householder rates are held constant at 1990 rates, as the rate of household formation within age groups has been relatively constant in recent years and the underlying social and economic factors affecting household formation rates do not indicate a direction of change. These underlying factors include marriage and divorce rates, incomes, housing costs, and government transfer payments (welfare and Social Security, for example). The projection of householder by age and tenure (headship) builds on the age group projections.

II-4 USING THE MASTER ADDRESS FILE TO ESTIMATE POPULATION FOR SMALL AREAS

Patricia C. Becker, APB Associates
pbecker@umich.edu

In recent years, the Census Bureau has moved from population-based methods to housing unit methods for purposes of estimating sub-county population. The reason is clear: population-based methods have suffered from seriously inadequate data sources and high levels of geocoding error. Housing units, already located on the ground, provide a better data set with little geocoding error.

Efforts in the 1990's have depended on local reporting of new construction and demolition counts, which are added to and subtracted from 1990 housing unit counts to provide an estimate of the housing stock. After applying estimated vacancy and population per household rates to the counts, we derive the household population. Added to an estimate of the group quarters population, this provides the requested estimate of population at the place and minor civil division level.

These methods are flawed as well. The biggest problem, aside from any question of the accuracy of the 1990 count, comes from the quality of the building permit (construction and demolition) data set. The paper will discuss some of these issues.

With the advent of Address List Development (ALD) at the Census Bureau in the 1990s, a new opportunity presents itself. The Bureau's Master Address File is designed to include one record for each housing unit, geocoded down to the census block in which it is located. The MAF is being developed and improved for use in the 1990 census. It is also an integral part of the process labeled continuous measurement, a program which includes the American Community Survey and the Estimates program as well. If the MAF is updated continually, it can provide de facto housing unit counts for any piece of geography in the United States at any point in time.

This paper takes a look at this question. What are the chances that we can use the MAF to provide the counts required for the housing unit method of population estimates? We will discuss the MAF as it arrived, in summer 1998, in Detroit, several suburbs, and some other areas of the nation at the beginning of the LUCA program. We will look at what's happened to it as MAF improvement operations, (to the extent that these answers are known in June 1999). Finally, we'll look at the questions that need to be asked and answered before the MAF can become the definitive and authoritative source of housing unit counts in the new millennium.

As a bonus, we'll also provide some information on the special places part of ALD/LUCA, and look at what that might do for us in estimating the non-household population.

II-5 A LOCAL ALTERNATIVE TO NATIONAL DEMOGRAPHIC DATA VENDOR POPULATION AND HOUSEHOLD ESTIMATES - APPLYING A MODIFIED HOUSING UNIT APPROACH TO INDIVIDUAL HOUSEHOLDS

John McHenry, Demographic Data for Decision-Making, Inc.
demogdat@bellsouth.net

Purchasers of small-area demographic data from national data vendors confront three interrelated problems. First, for standard reporting geographies (blocks, block groups, tracts, zips) national data vendor estimates often vary dramatically from one another. Second, clients interested in small area estimates for non-standard geographies (user-defined circles, polygons, etc) face small-area data aggregation and retrieval problems when vendor geographies do not match user-requested geographies. Third, small-area data purchasers are often in possession of reliable local area information that strikingly contradicts national vendor estimates.

We offer a generalizable solution to these programs in which readily available local data is combined with a modified housing unit approach to produce household-level population estimates which can then be aggregated up to any requested sub-county geography. We test our ideas out on 1998 Miami-Dade County tax assessor records and complementary data. We show that the resulting population and household estimates compare very favorably to estimates sold by national demographic data vendors.

II-6 APPLYING DATA FROM THE AMERICAN COMMUNITY SURVEY (ACS) AND THE MASTER ADDRESS FILE (MAF) TO THE INTERCENSAL POPULATION ESTIMATES PROGRAM

Gregg J. Diffendal, Demographic Statistical Methods Division, U.S. Census Bureau,

Stella U. Ogunwole, Amy Symens Smith, Population Division, U.S. Census Bureau
gregg.j.diffendal@ccmail.census.gov

One of the Census Bureau's primary goals is to provide current, continuous, and timely demographic measures for small areas. The Intercensal Population Estimates Program produces population estimates for counties with age, sex, race, and Hispanic origin detail, as well as total population estimates for places and county subdivisions. One of the Census Bureau's newest programs, the American Community Survey (ACS), is a monthly household survey designed to provide continuous demographic characteristics for states, as well as for cities and counties.

Traditionally, the Census Bureau integrates survey results with estimates by introducing the population estimates as independent controls to the sample survey results. We envision that data collected from the ACS will also enhance the intercensal population estimates program, which serves as the focus of this paper.

Population estimates for both counties and subcounties are vital because they are used for federal and state government funds allocation, denominators for vital rates and per capita time series, survey controls, administrative planning, marketing strategies, and academic research studies (Bryan and Devine, 1999). The reliance on population estimates for planning, decision-making, and research has generated a need for more detail and greater accuracy.

The ACS is a new approach for collecting accurate and timely information on American communities every year instead of every ten years (About the ACS, 1999). The ACS relies on the Master Address File (MAF), a nationwide address listing system that documents the address of every (occupied or vacant) housing unit in the United States and its territorties. The ACS will supply estimates of housing, social, and economic characteristics every year for all states, as well as for all cities, counties, metropolitan areas, and population groups of 65,000 persons or more (Symens, 1999; U.S. Census Bureau, 1999).

This paper will address some of the significant issues involved in integrating the Intercensal Population Estimates program and the American Community Survey. Specifically, two cases studies will detail research underway since the introduction of the ACS with four communities in 1996. In the first case study we shall compare the ACS results to county population estimates by age, sex, race and Hispanic origin. For the second case study we shall compare the ACS results with sub-county population estimates. Both case studies will conclude with a discussion of how the Intercensal Population Estimates program and the American Community Survey can "inform" one another in the future.

III-1 DOMESTIC MIGRATION ESTIMATION IN SUB-COUNTY AREAS: A CASE STUDY IN MASSACHUSETTS

Zongli Tang, Massachusetts Institute for Social& Economic Research (MISER), University of Massachusetts
zongli@miser.umass.edu

A principal challenge in small area population estimation is to find adequate data and appropriate methods for estimating domestic migration. Tax return data collected by the Internal Revenue Service (IRS) are employed by the US Bureau of the Census to conduct population estimates at the county level. Since the IRS data are not available for cities/towns, the Bureau has developed the housing-unit method to estimate population at the sub-county level. However, this method requires a vast amount of information which is not always easy to acquire. Data provided by various agencies are often found to be incomplete and inaccurate. It is especially problematic when this method is applied to areas with a large amount of population of minorities, immigrants, and college students. Moreover, the method is too complex to manage and too costly in time and financial resources when collecting and analyzing data on an annual basis. In preparing the 1991-97 population estimates for Massachusetts at the state and the sub-county levels, we have developed the Census-Administration method, attempting to address the issues mentioned above and to propose a new strategy for producing local population estimates.

This method consists of two major steps. First, converting IRS data by counties into migrants by cities/towns. Migration can occur at three levels: 1) state, i.e., migration among states; 2) county, including migration among states and migration among counties within a state; and 3) subcounty (city/town), including levels 1), 2), and migration among cities/towns within a county. All of the data can be found in the "long form questionnaire" of the 1990 census. Using this information, we can calculate the proportion of county migration for each city/town in Massachusetts. We can also calculate the ratio of migration at the city/town level to the migration at county level for each city/town. These proportions and ratios are assumed to be constant during the entire period of estimation (1991-1997) and are calculated separately for in-migration and out-migration. Once calculated, the proportions are first applied to the annual IRS data between 1991 and 1997 to generate migration estimates at county level for all Massachusetts cities/towns. Then, we convert migration at county level into subcounty level for each city/town using the ratios.

Second, adjusting the migration counts calculated on the IRS data. The IRS data are only available for population under 65 and may undercount some sub-populations. We employ school enrollment data (including elementary/high school and college enrollment data) and Medicare data to adjust the domestic migration numbers derived above.

This method enables us to produce timely population estimates by race, sex, and age at the sub-county level in an efficient and economical manner. We can also generate estimates at state and county levels using bottom-up method. A comparison between our output and that produced by the US Bureau of the Census indicates that the historically problematic population underestimation for Massachusetts has been significantly improved through the application of this new methodology. This method could be employed in population estimates in all areas, especially in New England region.

III-2 COLORADO METHODS OF POPULATION ESTIMATES

Richard Lin, Colorado Division of Local Government
richard.lin@state.co.us

Using the U.S. Census Bureau's (USCB) decennial census counts of population and housing units for Colorado state, counties, municipalities and census blocks as bases, the Demography Section of the Colorado Division of Local Government

(CDLG) has developed unique methods and procedures to produce annual, timely and periodically evaluated estimates of population and households for Colorado state, counties, municipalities, and Conservation Trust Fund (CTF) special districts since 1980.

We use component method (births, deaths, and net migration) to produce Colorado state population estimates and projections to be published by the end of November for the annual Colorado Business/Economic Outlook Forum in early December (i.e. the 34th in 1999). While the vital statistics (births and deaths) are from the Colorado Department of Public Health, the net migration estimates are derived from analysis of demographic and economic indicators and from consultation with field experts. The state estimates are compared to and evaluated against the USCB's states population estimates for Colorado by the end of December.

Then in March of each year we use the Colorado state estimates as control total and apply the following methods and procedures to allocate the estimates to Colorado counties. They are: (1) Regression Residual Method (births, deaths, school enrollments and residuals), (2) Component Method (births, deaths, group quarter population, net migration estimates), (3) Ratio Method (ES202, births, deaths), (4) the annual change of USCB's county estimates, and (5) Local Review Challenges. By May a variant of Housing Unit Method is applied to allocate county population to its municipalities and to the CTF special districts for local reviews and challenges.

First, we process the USCB's annual building permits survey data files and allocate permit units to municipalities for estimating new housing units. Second, we apply population/units ratio to allocating county population to its municipalities. Third, we estimate group quarter population to determine household population and consult annual CPS (Current Population Survey) national household size change to calibrate household sizes of Colorado municipalities for estimating households and vacancy rates. This same method and procedure has been used for estimating population and households of Colorado CTF special districts.

Our methods and estimates have been constantly monitored and evaluated for fine tuning. Our annual state estimates of population have deviated less than 5,000 persons from the USCB's estimates for Colorado. The three-variable (births, deaths, and school enrollments) regression residual method for county population estimates yields the 1980-1990 test reliability with a mean absolute error (MAE) of 2,139 persons and a mean absolute percent error (MAPE) of 6.6%. Although it does poorly for estimates of Colorado CTF special districts with hard-to-define boundaries and allocations, the variant of housing unit method shows very promising results of the 1980-1990 test reliability with an overall MAE of 470 persons and MAPE of 15.1% for municipal estimates and with an overall MAE of 2,037 persons and MAPE of 7.7% for Colorado counties. In addition, this method has produced annual state estimates of households which deviated around 1,000 to 3,000 households from that of the USCB's estimates in the 1980s and our estimates of 1,286,351 households is only about 4,000 households over the USCB's 1990 census counts of 1,282,489 households.

With our improvements in annual municipal boundary updates and cooperative local review procedures in the 1990s, we are very confident that we can produce more accurate estimates of population and households for Colorado counties, municipalities and CTF special districts by using a variant of housing unit method after the Census 2000.

III-3 REVIEW OF ADMINISTRATIVE RECORDS AND RATIO- CORRELATION METHODS FOR PRODUCING POST 2000 COUNTY POPULATION ESTIMATES IN ILLINOIS

Mohammed Shahidullah, Mark Flotow, Illinois Center for Health Statistics
mshahidu@idph.state.il.us

In this paper we review two methods for producing county population estimates in Illinois. We plan to evaluate the estimates for 2000 against 2000 census counts to find out how accurate the estimates will be. In the administrative records (ADREC) method, we will use data on births, deaths, immigration from abroad, nongroup quarters net migration, change in institutional population, change in military population in the barracks and change in medicare enrollees. We will produce population estimates for age groups 0-64 and 65 and over separately. The l990 household census population will be used as the base. We will control the county sum to Census Bureau's produced state population estimate. We will obtain on migration, military population in the barracks and medicare enrollees from the Census Bureau, and the rest of the data will be available from state sources.

In the ratio-correlation method to estimate the parameters, the dependent variable will represent the ratio of a county's share of the state total population in 1990 to its share in 1980. The independent variables resident births, grade school enrollment, automobile registration, and federal income tax exemptions will be expressed in a corresponding manner. Also we will stratify the counties on the basis of certain characteristics. Data on all the independent variables except federal income tax exemptions will be available from state sources. We will use the ratio correlation method to produce population estimates for age group 0-64 only.

We will have the following four sets of population estimates for 2000:

(1) using only ADREC method,

(2) using only Ratio Correlation method,

(3) averaging (1) and (2), and

(4) using each stratum of the Ratio Correlation Method.

We will evaluate each set of the population estimates using mean absolute percent errors and mean algebraic percent errors by population size in 1990 and population growth rate during 1990 to 2000. We will also evaluate median absolute errors and proportion of positive algebraic errors. We hope this research will help us in deciding which method(s) we should be using for producing post 2000 county population estimates in Illinois.

III-4 TESTING POPULATION ESTIMATION MODELS IN VIRGINIA

Donna Tolson, Center for Public Service, University of Virginia
dtolson@virginia.edu

Virginia's governmental system of counties and independent cities means that all cities in the state are estimated with county-level estimate methodology. This situation raises some unique data issues for Virginia localities. On the down side, most federally maintained administrative data such as Medicare and IRS tax records contain some level of geographic inaccuracy in Virginia because zip codes often cross jurisdictional boundaries between county and city. On the positive side, most state and local data are closely reviewed for geographic accuracy, and local government data are fairly consistent across the state and generally do not overlap jurisdictions.

Given these conditions, the Virginia FSCPE plans to test many county-level population estimation methods that capitalize on the data strengths of the state and which avoid the use of data sets known to have problems. Examples of "good" data for Virginia include state data on tax returns and exemptions, vital statistics, fall membership in public and private schools, driver's licenses, and local housing data. The methods we plan to test include Component Method II, the Housing Unit Method, the Census Ratio Method, and the Ratio-Correlation Method. In each case, we will test various "wrinkles" to each model. For instance, we plan to test the Housing Unit Method using local data on mobile and manufactured homes, demolitions, and conversions rather than distributing a state or national figure. Within the Ratio-Correlation Method (which we are using this decade), we will test many independent variables including state tax returns and exemptions, births and deaths, estimated housing stock (by several different definitions), school enrollment data, and driver's licenses. We will also test the accuracy of Ratio-Correlation method estimates of total population and net migration, in addition to non-GQ population.

My paper will discuss all of these options in more detail. I will also include a rough schedule for the workplan. We have already begun the data collection phase of the project, and should be able to begin actual method testing late this summer. By its nature, our research will be driven by the unique conditions of our state, however, we believe that much of it may be applicable to county estimates, and possibly to city estimates, in other parts of the nation.

III-5 USING WATER DEMAND TO DETERMINE POPULATION ESTIMATES AND PROJECTIONS FOR KANSAS

Darrel L. Eklund, Tina K. Rajala, Kansas Water Office

Ann H. Durkes, Kansas Office of the Budget
deklund@kwo.state.ks.us

Kansas has developed a unique method for utilizing water use data to determine not only future water use, but also to project population in the state. Additionally, this method will be used to verify the accuracy of the U.S. Census Bureau's sub-county population estimates for Kansas. This method was developed by the Kansas Water Office and approved by the Kansas Water Authority. Additionally, the projections are being considered by the Kansas Division of the Budget for presentation as the official population projections for Kansas. The importance of the Kansas method is its accuracy and similarity to the U.S. Census Bureau's newest methodology for estimating Kansas population. Also of importance is the fact that the Kansas method was conceived, developed, and implemented exclusive of the U.S. Census Bureau's method.

This paper contains the purpose and objective of the Kansas Water Office's efforts to develop and prepare population and water demand projections for Kansas and each of its counties, cities, and rural water districts. The projections were developed using linear regression for calendar years 2000, 2010, 2020, 2030, and 2040. Subject to the constraint that no city or county population decline could be more than 10.0 percent per decade.

Data sources included 1980 and 1990 Decennial Census counts, U.S. Census Bureau population estimates for 1992 and 1994, time series data of active residential service connections from public water suppliers, and extensive on-site interviews with local government officials and other groups. Also included in the methods development was contact with every public water supplier in Kansas for input on perceived in changes population, water use, and water demand occurring in local communities or rural areas.

III-6 OVERVIEW OF CALIFORNIA'S COUNTY POPULATION ESTIMATING METHODS

Melanie Martindale, California Department of Finance
fimmarti@dof.ca.gov

This presentation describes the features, strengths and weaknesses of California's four current methods used to estimate county population proportions: (1) Tax Return method; (2) Ratio-Correlation method; (3) Household method; and (4) Driver License Address change (DLAC) method. Twice yearly three of these methods are used to estimate county proportions of population; resulting distributions are averaged and applied to the independently estimated state total. Two of these four methods are used twice yearly, with the other two used once a year. The presentation will focus on the DLAC method.

IV-1 THE MEASUREMENT OF MIGRATION COVERAGE BIAS

Greg Williams, Alaska Department of Labor
greg_williams@labor.state.ak.us

The key variable in the production of county and state estimates under the estimates methodology of the 1990's is IRS migration. Migration is measured as movement between counties either in state or across state boundaries. Where there is differential coverage of persons from tax returns, resulting either from no-filers or first time or last time filers, there is likely to be a cumulative bias in the measured amount of migration. The result is a distortion in county and/or state population estimates. A comparison of migration flows as measured by IRS migration and Alaska Permanent Fund can document this tendency and possibly suggest ways to improve the measurement of migration in the next decade.

IV-2 ASSESSMENT OF IRS TAX RETURNS MIGRATION COVERAGE IN NEW MEXICO

Adelamar N. Alcantara, Bureau of Business & Economic Research, University of New Mexico
dalcant@unm.edu

Using the housing unit method, this research will attempt to evaluate how well the IRS tax return method estimates migration in the three counties in Metropolitan Albuquerque. Data on building permits, manufactured and mobile homes as well as occupancy rates by type of unit will be collected for the period between 1996 and 1997. Results of the estimates based on the housing unit method will be compared with the migration flows implied by the IRS data for the same time period.

IV-3 EVALUATION OF TEXAS POPULATION AND ESTIMATES AND PROJECTIONS PROGRAM'S POPULATION ESTIMATES AND PROJECTIONS FOR 1990

Nazrul Hoque and Steve H. Murdock, Department of Rural Sociology, Texas A&M University
n_hoque@tamu.edu

Population estimates and projections are difficult to complete with accuracy for small areas (Murdock et al., 1991; National Academy of Science, 1980; Ascher, 1978). As a result, it is essential that any ongoing program of population estimation and projection periodically evaluate the results of past estimation and projection efforts against actual counts of the population (Murdock and Ellis, 1991). Only by assessing the accuracy of past efforts it is possible to know the nature of errors made in past efforts and to take steps to improve future estimates and projections. This paper presents the results of the evaluation of the Texas Population Estimates and Projections Program's population estimates and projections for 1990 compared to the 1990 Census Counts for counties and places in Texas. We evaluated ratio-correlation method, component method II, and the housing unit methods. Ratio-correlation techniques utilize multiple regression techniques with the ratio of variable values for adjacent time periods rather than simply the variable values themselves being used as independent and dependent variables. This model used variables of births, deaths, elementary school enrollment, vehicle registration, and voter registration. The component-method II procedure employed data on births, deaths and elementary school enrollment to estimate population. The housing-unit method used is of used standard form, i.e. add new units, subtract demolition units with the base units. The evaluation presented here is being used as a major source of information for the modification of population estimates and projection procedures for the 1990s.

In the remainder of this paper we describe, several basic principles of population estimation and projection, the historical context of population growth from which the Texas program's estimates and projections were made, and the methods used in the evaluation. We also present an evaluation of county and place-level estimates and projections produced by the Texas program. inally, we outline changes to be implemented in the post-1990 estimates and projections as a result of the evaluation.

IV-4 AN EVALUATION OF THE ACCURACY OF U.S. BUREAU OF THE CENSUS COUNTY POPULATION ESTIMATES

Dean H. Judson, Planning, Research, and Evaluation Divison, U.S. Census Bureau,

Michael J. Batutis Jr., Population Division, U.S. Census Bureau and

Carole L. Popoff, Nevada Statistics and Research Methods Laboratory and Social Psychology Ph.D. Program, University of Nevada
Dean.H.Judson@ccmail.census.gov?

There has been much debate on discrepancies found between Decennial Census counts and intercensal county estimates. Numerous anecdotal speculations on the cause of the discrepancies at the state level have been made. A more systematic study by Davis (1994) compared the reliability and accuracy of the Census Bureau's county population estimates by examining the results of four of the Census Bureau's estimation methods against the 1990 Census count. In this study, we focus on biases in the administrative records themselves, which are used to estimate population changes, as clues to finding the possible reasons for over- and under-estimated counties. We examine, in detail, the sources of data used in the Census Bureau's methodology for reasons why there might be a systematic bias based on the circumstances under which data are gathered - either by the agency responsible for the administrative record used or by the person who will report. We test for these biases by using indicative economic and demographic data contained in the Bureau of the Census USA Counties 1994 release. The theory identifies causes of discrepancies in estimates that are systematic to the methodology and suggest the direction and likely magnitude of the discrepancy. In virtually all cases, our results are completely consistent with a prior hypotheses.

IV-5 EVALUATING THE HOUSING UNIT METHOD: A CASE STUDY OF 1990 POPULATION ESTIMATES IN FLORIDA

Stanley K. Smith, Scott Cody, Bureau of Economic and Business Research, University of Florida
sksmith@ufl.edu

The housing unit (HU) method is the most commonly used approach to making small-area population estimates in the United States. This study evaluates the accuracy and bias of HU population estimates produced for counties and subcounty areas in Florida for April 1, 1990. The major findings are that population size has a negative effect on estimation errors (disregarding sign) but no effect on bias; growth rates have a U-shaped effect on estimation errors (disregarding sign) and a negative effect on bias; electricity customer data provide more accurate household estimates than do building permit data; errors in household estimates contribute more to population estimation error than do errors in estimates of average household size or group quarters population; and the application of professional judgment improves the accuracy of purely mechanical techniques. We believe the HU method offers a number of advantages over other population estimation methods and provides planners and demographers with a powerful tool for small-area analysis.