TTG, Incorporated: Professional Territory Design Software Since 1988

Estimates and Projections

Content

The Estimates and Projections (E&P) database is the most extensive update available, covering a broad range of demographic characteristics for the current year, and 5-year projections. Variables include:

• Population
• Population by household type (family, non-family, group quarters)
• Households
• Households by type (family, non-family)
• Households by size of household
• Households by age of head of household
• Household type (e.g. lone parent male family with children)
• Average Household Size
• Population by age (19 age breaks)
• Population by age and sex (38 breaks)
• Population by sex
• Population by race
• Population by Hispanic origin
• Population by race and Hispanic origin (e.g. white Hispanic, white non-Hispanic)
• Population by Marital Status
• Population by Educational Achievement
• Labor Force Employment Status
• Labor Force Employment Status by sex
• Aggregate Income (family, non-family households, group quarters)
• Household income distribution (15 breaks)
• Family income distribution (15 breaks)
• Extended Upper-Income distributions
• Median and average income (family, household)
• Households by disposable income
• Age of head of household by income
• Median income by age of head of household
• Vacant Dwellings
• Tenure
• Length of Residence
• Total Vehicles Available
• Households by number of vehicles available
• Households by number of vehicles available and tenure

2008 Update Considerations
Estimates for the gulf coast area affected by Katrina continue to improve, as the post office household counts continue to better reflect current rather than former delivery statistics. Projections continue to show a rebound of this area over the coming decade. For 2008, we undertook a considerable effort to improve estimates in rapid growth areas, especially those areas which were unpopulated at the time of the 2000 census and therefore lack internal geographic coding (that is, there were few or no census blocks defining these areas specifically). This permits a more robust allocation of the growth into these specific areas rather than into neighboring areas. An example of this is the old Denver airport, in which growth was misallocated to adjacent areas in past years but is now correctly allocated to this redevelopment area.

2007 Update Considerations
With respect to the estimates and projections in the Katrina affected areas, the postal and ADVO data has finally begun to reflect post-hurricane changes and subsequently, at the block group level we have been able to undertake a much more thorough and data driven update. This, we believe will result in much better small area estimates along the Gulf Coast, but we do note that overall, the population estimates at the parish level remain relatively close to our estimates of last year.

We made changes to the Age by Income distributions. The income distributions for each age group have been extended to 15 from 12, providing better detail within some of the lower income groups. It is vital that you review any existing variable extractions and reports which might utilize this data, as the sequencing of the variables has been changed. For example, the variable HINCYT2512 is no longer “Households age 25-34 with Income >$200,000” but is now income $100000-$124999. See the AGS Variable List (v2) for a complete list of variables.

2006 Update Considerations
AGShas been working very hard to find data on the current whereabouts of the residents of the gulf coast. We have been busy gathering intelligence on how many people left the area affected by Katrina, where they went, and how many have come back.

The data from standard sources – e.g. postal service NCOA (“National Change of Address”) file, ADVO delivery counts, etc. – were suppressed or missing for the affected areas.

We obtained data on FEMA registrations by ZIP code of original residence and ZIP code of temporary or new residences. These were as of late fall of 2005 and reflect the situation at its worst. We have made some assumptions about the return percentages and adjusted the demographics accordingly.

A Census database which indicates the % of housing declared unlivable by Census Block for the Gulf States was used in conjunction with detailed elevation and flooding maps which were downloaded into a GIS system in order to identify fully the census block groups which were affected. We then excluded those block groups which were not affected.The detailed elevation and flood maps were obtained from federal government sources.

The Census has since released a preliminary Katrina report which details the expected population of the affected gulf counties and the AGS numbers agree very well with those published in the report.

When the Census final report comes out in August, we will again review our numbers in comparison and, if necessary, make appropriate adjustments.

2005 Content Changes
While we have not undertaken any changes to the core variable list, we have begun to incorporate the Census Bureau “American Community Survey” (ACS) data into our procedures. This data source now covers over 60% of the population and is to be updated on a rolling basis over time, eventually serving as the replacement for the decennial SF3 tables.

The effects of this inclusion are most strongly noticed in some of the detailed tables where intercensal estimates are either inconsistent or non-existent.

In addition, we adjusted our household estimates downward from the 2004 release on the basis of the ACS and other sources. The ongoing housing boom has resulted in over-construction and an increase in the number of vacant dwellings.

2004 Content Changes
Improvements were made to the base population and household models which more accurately incorporate changes to the postal delivery counts, which will be most noticeable in new growth areas.

The following was added:

• estimates of population by detailed age and sex, by single year of age to 20

2003 Content Changes

There are substantial changes to this variable content for this release of the estimates and projections database, including:

• We have, with the exception of base population and household counts, eliminated the ten year projections from the database.
  
• The 2000 Census racial classification has been fully implemented, which includes the “multiple” category and the split of the Asian and Hawaiian/Pacific Islander groups. Variable names are consistent with past releases; however, there are differences in the meaning of each of the variables with the addition of the multiple race response and the split of the Asian and Hawaiian/Pacific Islander fields.
  
• We have provided additional detail on the vehicles available table, by breaking it down by tenure (as on the census) and by providing additional detail in the 2+ vehicle categories. The VPHCY and VPHPY variables have been retained as formulas (e.g. NONE, 1, GT1)
  
• The high income classification (200K+) has been fully integrated into the standard income distributions, which have been reclassified to comply with the 2000 Census definitions in the lower income categories. This has resulted in the renaming of the upper income variables and the elimination of the 200+ income category (which is available as a formula in the library but should not be used within median calculations). In addition, greater detail has been added in the $50-$100K categories by estimating at $5K increments.
  
• Aggregate and average disposable household income is now included in addition to the distributions, the disposable income distribution classes match the standard income distribution classification.
  
• The household structure table has been revised to eleven categories in keeping with the tables provided in the 2000 Census and these have been remodeled to agree more closely with other tables (e.g. size of household).
  
• We have, for the current year, not included employment by occupation or employment by industry. We will revisit these tables on a yearly basis. At present, the Bureau of Labor Statistics does not report current year estimates nor projections using the updated definitions used by the Census 2000 tabulations. These are effectively incomparable for many categories and are unsuitable as the basis for undertaking current estimates and projections. We will review the availability of source data on an annual basis with the intent of reinstituting these tables when it becomes feasible.
  
• Additional detail has been added on labor force participation by splitting the basic labor force status data by sex. This will permit updated unemployment and labor force participation calculations for both males and females, an important aspect for many daytime based studies.
  
• The age distributions have been adjusted to more closely align with most users expectations of five year cohorts. This affects most age groups under age 25, although we have provided this age distribution in both the old and new formats. We encourage conversion to the new age distributions over the next year.
  

There are some significant methodological changes which impact the accuracy of the basic household and population data. Specifically, we have integrated delivery counts from both the USPS and commercial sources by linking the ZIP+4 delivery statistics to the census block level of geography. By using comparisons of these delivery statistics and the census data for an equivalent point in time (April 2000), the discrepancies between the census and “postal” counts can be adequately accounted for.

Methodology and Data Sources

AGS uses a wide range of data sources in constructing its estimates and projections, including:

• Census tabulations from 1970, 1980, 1990 and most recently, the release of the 2000 Census
  
• USPS and commercial source ZIP+4 level delivery statistics
  
• Census Bureau estimates and projections of population characteristics at various levels of geographic detail, including the latest estimates of population at the city level.
  
• The Census Bureau’s American Community Survey results, which cover over 60% of the national population and serve as an increasingly important attribute base at the county, metro, and state levels.
  
• Bureau of Labor Statistics estimates and projections of employment by industry and occupation at the county level
  
• Medicare eligible population counts at the ZIP code level, including population by sex and 5-year age cohorts, provided by the Health Care Financing Administration of Social Security. These counts provide a very accurate local count of the population aged 65 and higher.
  
• Internal Revenue Service statistics on tax filers and year-to-year migration
  
• The Census Bureau’s Current Population Survey, which provides detailed demographic breakdowns and enables a thorough longitudinal analysis of demographic trends
  
• Experian’s INSOURCE database, a household level credit and demographic database which covers the vast majority of households

INSOURCE is a vast database at the household and individual level that Experian provides to AGS for use in its demographic estimates. The INSOURCE database was aggregated to the ZIP+4 and Block Group levels of geography for analysis and standardized to Census Bureau county level current estimates. A large number of demographic attributes from INSOURCE were utilized in building the current year estimates, including:

• Population
• Population by Age
• Households
• Household Size
• Household Type (presence of children)
• Marital Status
• Income
• Hispanic origin
• Population of Asian origin
• Dwelling Tenure (own/rent)
• Length of Residence

In turn, the AGS demographic estimates are used as the foundation of Experian’s U.S. MOSAIC segmentation system.

Now in its seventh year of use within the AGS estimates methodology, INSOURCE provides an excellent source of small area year-to-year change which greatly improves the quality of local estimates, especially in areas of growth.

The estimates and projections methodology combines the best current and projected information from the data sources noted above. It is supplemented by the extensive experience of Applied Geographic Solutions in creating accurate and reliable estimates and projections. A summary of the methodology for each of the major variable groups is included in the sections that follow.

Population

The current population of the United States is obtained from the monthly Census population estimate. This is a very accurate and current estimate of the population and serves as the basis for projection and estimation at lower levels of geographic detail. The five and ten year projections have been derived from the middle-series projections of the Census Bureau.

The current year estimates rely heavily on the 2000 Census block level population counts, as these provide the most accurate recent data available. These 2000 Census counts replace the 1990 Census counts as the basis for undertaking estimates. In effect, the latest Census tabulation provides a baseline for the estimates and projections.

State and county level estimates are based on the compilation of data from a range of Federal and State authorities, including the latest county population estimates from the Census Bureau, the American Community Survey (ACS), reviews of building permit statistics, the current population survey (CPS), and additional local sources. Where required, the resulting estimates are then ratio-adjusted so that the sum of the county estimates is equal to the state total, and the state estimates equal to the national total. For the five- and ten-year projections, a similar method is employed. However, rather than using simple straight-line techniques, AGS uses straight-line methods only for growing areas. For declining areas, a log-normal extrapolation is used. This has the effect of slowing decline over time, which is characteristic of long-term population decline at the state level.

At the block group level, the population model consists of the application of a non-linear trend model which estimates population given historical patterns, INSOURCE population counts, and the latest Census age distributions (using cohort-survival techniques). Special consideration is given to the population age 65+ by applying ZIP code level counts by age and sex of all Medicare eligible persons. This provides considerable improvement in the estimates of this important segment of the population. The final results are then carefully balanced to the county and city level population estimates to ensure consistency with current Census Bureau estimates.

The result is a comprehensive set of population estimates and projections which includes the knowledge of State, County, and private agencies about their detailed areas but also ensures that the total population is consistent with the Census Bureau estimates, which have proved extremely reliable over time.

Population by Age, Sex, and Race

National and State level Census bureau projections of age by sex and race/Hispanic origin were used as overall controls to ensure consistency with the Census projections. Detailed forecasts by age, sex, and race, as well as Hispanic origin, were obtained from the Census Bureau ‘middle series’ projections.

At the state level, the projections of individual state agencies and ACS estimates were combined with the results of a cohort survival approach to obtain reliable state estimates by age and sex. The block group estimates were compiled using cohort survival methods, then balanced to both the estimated block group population totals and to the state level control totals. Consistency checks with the annual CPS (Current Population Survey) are used to ensure the validity of the resulting age/sex distributions. Further, INSOURCE population by age summaries were used to adjust local estimates for the adult population, with further adjustments applied using the ZIP code level Medicare eligibility statistics.

Trends in the racial distribution and Hispanic populations were used to derive preliminary estimates at the block group level, which were then adjusted to balance with appropriate control totals. This method allows the utilization of the historical changes in race and Hispanic origin distributions and projects those changes into the future while maintaining consistency with national level projections. Again, the CPS is used extensively to assist in the verification of the models.

Households and Household Type

Total households were modeled by:

• projecting trends in the population per household over time at the national level to provide a control total;
• reviewing currently available household size statistics at the State level; and utilizing the current estimates of population by age and sex to determine household formation rates for small areas

The ACS data has been extensively used in order to bridge the gap between population estimates and dwelling/postal delivery counts.

All household based numbers are initially estimated / projected separately for family and non-family households. Non-family households have been growing in number at a higher rate than family households have over the past several decades. Average household sizes for family households have been decreasing for several decades. However, during the 1990’s, the decline has stopped in most areas and has actually reversed in several states.

The group quarters population, that is population that is not in households (such as persons in institutions, military barracks, nursing homes, college dormitories, and homeless persons), is expected to increase slightly during the decade, but remain relatively constant as a percentage of the total population. This is a reflection of two trends: the decreasing armed forces employment since the 1980’s and the longer term increasing elderly population which results in high populations in nursing homes and other institutions which cater to the elderly population. As a result, the total group quarters population has been relatively constant.

Income

Income estimates include aggregate income by household type and income distributions as well as derived measures include per capita income, and various median income measures.

All income estimates produced by Applied Geographic Solutions are in current, rather than constant, dollars. In other words, a projection of income for the year 2010 includes both an inflationary component and a ‘real’ component, the latter being the difference between the change in income and the change in inflation during the period. The ‘real’ component is normally attributed to productivity gains in the economy and to differences in the international competitiveness of the economy.

Aggregate income estimates for the current year are based on an analysis of income information from the SF3 database of Census 2000, supplemented heavily by the 2003 ACS estimates. The projections of aggregate income are based on a review of Bureau of Economic Analysis (BEA) projections, which assume an effective increase of 3.5% per annum in per capita incomes during the next ten years at the national level.

Income distributions are estimated and projected for both family households and non-family households separately. Total household income distributions are simply the aggregate of the two detailed distributions.

Income distributions were derived by using a complex distribution shifting technique which utilizes the changes in per family household and non-family household incomes as a means of adjusting the income distributions over time. The relative ratio between changes in per household average incomes and median incomes were used to adjust for above-average growth in high-income households within some geographic areas. The resulting distributions were then normalized to higher order totals and adjusted to national level expectations and were verified for internal consistency with respect to the mean and median measures.

As of the 2003 release, for the current year estimates, a new set of income breaks are provided for the $150,000+ category, namely $150000-$199999, $200000-$249999, $250000-$499999, and $500000+. Created by using logistic regression techniques that account for the local income distribution, these should be considered as maximum likelihood estimates. Although little data exists to substantiate incomes in these ranges, comparisons have been made to IRS taxation statistics to ensure that the results are consistent. Users are cautioned that these estimates are statistical in nature only.

Other Variables

A number of other variables are also projected within the series. In large part, these are derived by using available current estimates and projections at the lowest possible level of geography as the base for the estimation procedures, relying heavily upon the annual release of the ACS. The CPS is used extensively to track changes using available cross-reference information related to age, race, sex, and income. Where possible, these CPS statistics are supplemented by INSOURCE estimates.

For example, current marital status estimates are available at the state level ACS from the Census Bureau as “control targets”. The ACS is used in conjunction with the annual CPS surveys (both historical and current) are used to track the changes in marital status dependent upon other symptomatic variables such as age, sex, race, and income levels. These “micro-models” are then applied to the block group level changes between the census and the current period. This results in block group level data which is consistent with higher order levels but also reflects changes in marital status owing to shifting local demography.

On the other hand, vacant housing is tracked using state and regional indicators, then adjusted for seasonally vacant dwellings which are a significant component of the marketing landscape in many areas of the country.