|
Current-year forecasts and five-year projections provided by Percept
are developed by Claritas using current and historical data, as well
as economic modeling and data pooling statistical techniques. This
process enhances demographic analysis in two important ways. First,
it utilizes all current data and information to accurately estimate
the current location of the population, households and income.
Second, it defines the relationships between each demographic
variable and the appropriate economic, cyclical, and migratory
factors that cause their movements over time. |
|
|
|
|
|
Population Characteristics |
|
|
Population by Age/Sex |
|
|
Population by age/sex composition is
estimated and projected using cohort survival methods. Cohort
survival is a major factor in changing age structures, and is driven
by the reality that, for example, persons age 35 in 2000 who survive
another five years, will be age 41 in 2006. Accordingly, a
population with a large proportion of 35 year olds in 2000 can expect to have large proportions of 41 year olds in 2005. It is this process that has swelled the U.S. age structure at progressively older age categories
as the baby boom generation (or birth cohort) has aged. |
|
|
The Claritas cohort survival method is executed first at county
level, then for tracts, and finally block groups, with each set of
estimates controlled to the results at the next higher geographic
level. To enhance consistency with Census Bureau age/sex estimates,
the county estimates begin with the Census Bureau’s most recent
county age/sex estimates. |
|
|
Note: The
Census Bureau county age estimates contain a known problem in some
counties with large college populations living in households (not in
dormitories). After consulting with the Census Bureau, Claritas
completed a project to identify counties where this problem had the
greatest impact, and effective with the 2006 Update, used the Census
2000 county age data as the starting point for estimates in these
counties. |
|
|
|
|
|
Tract and block group estimates begin with Census 2000 age/sex
estimates. At all levels, the method starts with five-year age/sex
categories—separating persons in households from those in group
quarters. Because Census 2000 data (and the Census Bureau age/sex
estimates) do not provide full age/sex detail for household versus
group quarters populations, Claritas estimates the detail required
to execute the cohort survival method. The cohort survival process
is set into motion with the application of age/sex-specific
five-year survival rates to the census age/sex data described above.
Each round of cohort survival ages the population of each block
group ahead five years. |
|
|
For example, the process projects the number of 30-34 year olds in a
block group who will survive to become 35-39 years old (and so on
for all five-year age categories) by 2005. The initial survivals
yield projections of age/sex composition for April 2005 (short of
the January 1, 2006 estimate date), so a second survival is
performed to 2010, and the results interpolated to January 2006. In
the case of county estimates starting with July 2004 Census Bureau
age/sex estimates, the single survival extends to July 2009, and the
results are interpolated to January 2006. |
|
|
Accounting for Births |
|
|
As part of each round of cohort
survival, the population less than age five is “survived” to age
5-9, so an estimate of births is required to fill the vacated 0-4
category. Births are estimated using the child/woman ratio—defined
as the population age 0-4 divided by
females age 15-44 (childbearing age). |
|
|
The child/woman ratio is an indirect
measure of fertility specific to each small area, but more
important, it is sensitive to projected changes in the number of
women of child bearing age—itself a byproduct of the cohort survival
process. An increase in the
number of child bearing women will result in an increased number of
births even if fertility rates (or child-woman ratios) remain constant. The child/woman ratios applied in the Claritas age/sex estimates and projections are derived from the 2000 census, but
reflect slight increases evident in the Census Bureau’s
post-2000 estimates. |
|
|
Exceptions to Cohort Survival |
|
|
The
cohort survival process is at work in all areas, but in some areas
its effects are overridden by migration. In the absence of
authoritative age-specific migration data for small areas, the
Claritas method defaults to the assumption that the age/sex
composition gained or lost through migration is similar to the
area’s “survived” population. |
|
|
However, because of migration, the cohort survival process is often not applicable to populations
living in group quarters facilities such as dormitories, military
quarters, prisons, and nursing homes. These populations have high
turnover, and therefore age/sex compositions which tend to be
stable, reflecting the nature of the
facility. For this reason, cohort survivals are applied only to the
population living in households. Group quarters populations are
estimated separately and their age/sex compositions held constant
with those measured in the census. |
|
|
Claritas also identifies segments of the household population (such as concentrations of college students
in off-campus housing) for which cohort survival is not applicable.
Concentrations of these “hidden group quarters” populations are
identified through their distinctive imprint on small area age
compositions, and are similarly exempted from the cohort survival process. |
|
|
Five Year Projections |
|
|
Five
year projections of age/sex composition are produced with the same
method used for the current year estimates. For example, in the 2006
Update, the 2006 estimates of population by age/sex were the
starting point for five year survivals to 2011. As with the current
year estimates, age/sex projections are produced
first for counties, followed by tracts and block groups, with adjustments ensuring consistency between geographic levels. |
|
|
Population by Race and
Ethnicity |
|
|
There are no universally accepted
definitions of race and Hispanic ethnicity. The census currently
defines “Hispanic or Latino” as an ethnicity, not a race. Race and
Hispanic ethnicity are separate census questions, so in census
tabulations, persons of Hispanic ethnicity can be of any race.
Hispanics are included in each race
category, possible combinations of two or more races. When
cross-tabulated by Hispanic/non-Hispanic, there are 126
race-by-Hispanic categories. |
|
|
Short
of presenting data for all 63 race categories, there are two basic
tabulation options—single classification and all-inclusive.
The
single classification options are: |
|
|
White Alone |
|
Black or African American
alone |
|
American Indian and
Alaska Native alone |
|
Asian alone |
|
Native Hawaiian and
Other Pacific Islander alone |
|
Some other race alone |
|
Two or more races |
|
|
|
|
This option identifies the number of persons marking each race
category by itself, and then provides a seventh category identifying
the number marking two or more races. The tabulation is similar to
those used in the past, and sums to total population. However, it
provides no information about the race of persons in the “two or
more” category, so it is not possible to determine the total number
of persons identifying with a given race. The total number of
persons marking a given race category is revealed by the following
all-inclusive categories: |
|
|
White Alone or in
combination |
|
Black or African American
alone
or in combination |
|
American Indian and Alaska
Native alone or in combination |
|
Asian alone or in
combination |
|
Native Hawaiian and Other
Pacific Islander alone or in combination |
|
Some other race alone
or in combination |
|
|
|
|
This option identifies the total number of persons marking each race
category—either by itself or as part of a combination of two or more
races. However, because persons marking two or more races are
counted two or more times, the table sums to totals larger than
total population. |
|
|
The Claritas Update provides estimates
and projections for both the single-classification and all-inclusive
tabulations. Estimates for the seven single-classification
categories (by Hispanic/not-Hispanic ethnicity) are produced first,
and all-inclusive estimates are then derived from the
single-classification numbers. |
|
|
Estimates and Projections of Race and Hispanic Ethnicity
|
|
|
At the
county level and above, estimates of race and Hispanic ethnicity are
based on the Census Bureau’s estimates of population by race and
ethnicity at the county level. |
|
|
The application is not
straightforward, since the Census Bureau’s race estimates reflect a
modified definition, in which persons marking “Some other race” were
re-assigned (with imputation techniques) to a specified race
category. This reassignment increases the numbers in the specified
categories, making them inconsistent with the census definition race
counts reported in
standard Census 2000 products. |
|
|
For this reason, the Claritas method
applies the Census Bureau’s estimated rates of change from the most
applicable modified race category to the relevant Census 2000 race
counts. For example, the census estimates might suggest a 4.2
percent increase in the percent of a county’s population that is
(modified) “Asian not Hispanic.” The Claritas estimate is
established by applying this rate of change to “percent Asian
not-Hispanic” from the 2000 census. Estimates are produced for the
seven not-Hispanic race categories. Percent Hispanic or Latino
population is estimated
separately based on the rate of change in percent Hispanic population suggested by the Census Bureau estimates. The Hispanic or
Latino estimates are then distributed to race based on county
specific percentages from the 2000 census. The estimates for the 14
race/ethnicity categories are then finalized by applying
estimated percent race/ethnicity to the previously completed estimates of total population for each county. |
|
|
Race/ethnicity estimates below the
county level are based on 1990-2000 census trends in the percent of
population in each race/ethnicity category. Again, the method
focuses on the percent of population in each category. Estimates are
produced first for tract level (with adjustments to county level),
then for block groups (with adjustments to tract level). The
projection of inter-censal trends is not a preferred method, but the
approach was an achievement made possible by the conversion of 1990
data to 2000 geography, and the bridging of 1990 race to 2000 race
definitions. |
|
|
Race Bridging |
|
|
The current race definitions make it
impossible to identify definitive race trends between the 1990 and
2000 censuses. However, to estimate 1990-2000 trends, Claritas
“bridged” 1990 census race data to the 2000 definitions.
Specifically, Claritas estimated what the 1990 census race data
might have looked like had it been
collected using 2000 categories, and the option of marking two or more races. |
|
|
All race bridging was accomplished
separately for the Hispanic or Latino and not-Hispanic populations
(preserving race by Hispanic cross-tabulation options) for all block
groups nationwide. The first step was the bridging of 2000 race to
1990 definitions. After
combining the Asian and Native Hawaiian and Other Pacific Islander
categories (whether alone or part of combinations) to the 1990 Asian
or Pacific Islander Category, counts from the remaining
multiple-race categories were distributed to single 1990 race
categories. This distribution was accomplished with equal fractions
assignments in most cases (combinations of two races distributed
half to one category and half to the other, combinations of three
races distributed by thirds, and so forth), but National Health
Interview Survey proportions were used for selected combinations.
These include: |
|
|
White or Black or African
American |
White and American Indian
or Alaska Native |
White and Asian |
Black or African American
and American Indian or Alaska Native |
|
|
|
|
|
|
The bridged 2000 race data suggests
how many persons would have been added to each “race alone” category
had multiple-race response not been an option in 2000. For example,
bridging 2000 data to 1990 definitions added some persons from
multi-race categories to “Black or African American alone” to
estimate the 1990 “Black” category. From the reverse perspective,
the data
suggests the proportion of the bridged “Black” population that would
be lost to race combinations when transitioning back to the 2000
“Black or African American alone” definition. The 2000 bridged data
suggests such percentages for all 1990 race categories, and these
percentages were applied to the 1990 census race data (converted to
2000 block groups) to estimate
the number that would have been lost from each category to multiple race responses in 1990. Census 2000 patterns then were used to distribute the estimated 1990 “two or more races” population to the 57 categories reflecting combinations of two or more 2000 census race categories. |
|
|
The bridging
project produced a set of 1990 census population data distributed to
the 126 Census 2000 race by Hispanic categories, and converted to
2000 census block groups. This data—collapsed to single-assignment
race—provided a basis for estimating race/Hispanic population trends
for census tracts and block groups. |
|
|
Five-Year Projections |
|
|
Five
year projections of race/ethnicity are produced with similar
methods—projecting the current year estimates (of percent
race/ethnicity) to the five-year projection date. Again, projections
are made for percent race/ethnicity distributions, and applied to
previously completed projections of population. Counties are
projected first, followed by tracts and block groups, with
adjustments ensuring consistency between geographic levels. |
|
|
All-Inclusive Race |
|
|
Estimates and projections for all-inclusive race/ethnicity are
produced as derivatives of the single-classification estimates and
projections. For each race/ethnicity category, the 2000 census ratio
of all-inclusive race/single-classification race is identified, and
applied to the estimate or projection of single-classification
race—with adjustments made in some areas to ensure consistency with
the number of persons estimated (or projected) to be of two or more
races. Because the all-inclusive estimates and projections are
derived from data already adjusted to county controls, the
all-inclusive estimates and projections are produced only at the
block group level, and summed to higher levels. |
|
|
Population by Age/Sex by Race/Ethnicity |
|
|
Estimates and projections also are
provided for the cross-tabulation of population by
age/sex/race/ethnicity. These estimates start with the completed
estimates of population by age/sex and population by race/ethnicity
at the block group level. Census 2000 does not provide
age/sex/race/ethnicity detail at the block group level. For this
reason, age/sex/race/ethnicity distributions for census tracts are
used as “seed values” for component block groups, and iteratively
adjusted to conform with the previously completed estimates of
population by age/sex and population by race/ethnicity. This
application of IPF produces block group estimates consistent with
estimated age/sex and race/ethnicity, as well as the statistical
relationship between these characteristics observed for the census
tract in the 2000 census. |
|
|
|
|
|
Household Characteristics |
|
|
Households by Income |
|
|
All Claritas income estimates are
expressed in current year dollars using the money income definition
reported in the 2000 census. The estimates and projections reflect
household income, which includes the income earned by all persons
living in a housing unit (i.e., all household members). In contrast
to the 2000 census, which reported income for the previous calendar
year(1999), Claritas income estimates are for the calendar year relevant to each set of estimates and projections. For example, the
2006 estimates would reflect 2006 income for 2006 households. |
|
|
The method starts by estimating
rates of change in median household income for each area. Based on
this rate of change, household income distributions from the 2000
census are advanced to current (or projection) year. As with the
population estimates and projections, data was first produced for
large areas, then for progressively smaller areas, with successive
ratio adjustments ensuring consistency between levels. Aggregate and per capita income numbers were derived from the resulting income distributions. Claritas estimates household income for all 16 income
categories reported by the 2000 census in Summary File 3 (SF3). |
|
|
Income
Estimation Method |
|
|
Income change at the national level is estimated based on national
estimates of income change from the Current Population Survey and
the American Community Survey. The final estimates reflect an
average of estimates based on the two sources. The national income
distributions serve as a target for the state estimates, rather than
a control total. |
|
|
State income estimates are based on
IRS wage and salary data, and BEA estimates of per capita income.
Because national IRS and BEA income data tends to reflect more rapid
income growth than the national estimate, these sources are used to
monitor each state’s income growth relative to the national
level—change in the ratio of state income to national income. The
final rates of change reflect the average of such ratios based on
IRS and BEA data, as well as a projection of the ratio based on
1990-2000 census trends. |
|
|
County
income rates of change are estimated with similar
procedures—this time applying county/state ratios of IRS and BEA
income data to 2000 census county/state income ratios. Again, the
final estimated rates of change reflect the average of ratios based
on IRS and BEA data, and the projection of 1990-2000 census trends. |
|
|
Income change at the tract level also is estimated with alternative sources, with the final estimated rate reflecting the average of these rates. The first estimate is based on historical performance.
Specifically, tracts were estimated to outpace or lag behind county
income growth in proportion to their performance relative to county
during the 1990 to 2000 census period. The second is based on
post-2000 trends in income estimates aggregated from the Equifax TotalSource consumer household database. The TotalSource income
estimates are modeled for all individual
household records on the database. The third is based on trends in
the Equifax ACE-Geosummary database, which provides tract level
summaries of consumer financial information from the Equifax Consumer Marketing Database (ECMD). Although not a direct measure of
income, the ECMD data item “Average sum of credit limits for bank,
national credit card, savings & loan, and credit union revolving
accounts” is strongly associated with income at the tract level, so
change in this variable is used to track change in income at the
tract level. |
|
|
The approach with all three sources is to track change in the
tract/county ratios—or the performance of tract income relative to
county level. Income change at the block group level is estimated
with sources and methods similar to those described for census
tracts above. |
|
|
For all geographic levels, the
estimated rate of income change is used to advance, or shift, the
2000 census distribution of households by income forward to current
year. This procedure involves the estimation of the number of
households advancing from one income category to another—based on
the area’s estimated rate of income growth. |
|
|
The resulting current year
distribution is adjusted to conform with that estimated for the next
higher geographic level. For example, the estimated household income
distribution for states is adjusted to the national distribution,
the county estimates are adjusted
to the final state distributions, and so forth. |
|
|
Five Year Projections |
|
|
Five year projections of income begin
with the projection of current year median household income to the
projection year, and the advancing of the household income
distribution to reflect the projected change. Median incomes for
sub-national areas are produced by projecting estimated income
change to the projection
year, and advancing the current year estimated income distribution
to reflect that rate of change. As with the current year estimates,
the initial income distributions are adjusted to the final
distributions for the next higher geographic level. State
projections are adjusted to national, county to state, and so forth. |
|
|
Family
Household Income |
|
|
A family household is one in which the
householder is related to one or more persons living in the
household. Family households also include any other non-related
persons living in the same housing unit. Family household income
includes the income of all persons living in a family household. |
|
|
Family household income is estimated
by applying the final estimated household income growth rates (1999
to current year) to the 2000 census distribution of family
households by income—advancing the family household income
distribution to reflect the relevant rate of income growth. Five
year projections were produced by trending the estimated rate of
family income change out five years, and advancing the current year
distribution to reflect the projected change. Because the estimates
and projections of family household income are derivatives of the
completed household income estimates—which already reflect control
totals for large areas—they are estimated and projected at block
group
level only, and summed directly to higher levels. |
|
|
Income
by Age of Householder |
|
|
The cross-tabulation of household
income by age of householder is valuable because income and life
cycle stage, when combined, are so strongly associated with consumer
needs and behavior. The Claritas income by age updates are produced
after the estimates of population by age and households by income
have been completed. The data constitutes a 198 cell table defined
by 18 categories of household income and 11 categories of
householder age. The row and column totals from these tables (the
income and age totals) are commonly referred to as the marginal
totals. |
|
|
The estimates of households by income
serve as the income marginals, but population by age estimates must
be converted to householder by age for use as the age marginals. For
each area estimated, 2000 census data is used to determine
age-specific headship rates, or the percent of persons in specific
age categories who are householders. These headship rates are then
applied to estimated population by age to produce estimated
householders by age. A final adjustment to total households ensures
consistency with that critical base count. |
|
|
With the income and age (row and column) marginal totals estimated,
the final step is to estimate the full cross-tabulation of income by
age of householder. In other words, values must be determined for
each of the 198 income by age categories, or cells. Block group
level income by age cell values from the 2000 census
provide the initial input (or seed values). Within each age
category, the 2000 census income distributions are advanced to
reflect the block group’s (previously) estimated rate of income
growth. This adjustment expresses the 2000 census income by age
distribution in current dollar values. The resulting table is then
adjusted to conform with both the income and age of householder
totals estimated for current year. These adjustments are
accomplished through iterative proportional fitting, which adjusts
the 2000 table to conform simultaneously with the household income
and
householder by age estimates, while preserving the block group
specific statistical relationship between income and age reflected
in the 2000 census income by age data. |
|
|
The income by age estimates are produced at the county, tract, and
block group levels, with adjustments ensuring consistency between
levels. |
|
|
Five year projections are produced using similar methods. Projected
households by income serve as the income marginal totals, and Census
2000 headship rates again are used to convert projected population
by age to projected householders by age.
The income by age table is then advanced to projection year dollar
values, and iteratively adjusted to the projected income and age
marginal totals. |
|
|
Income by
Race and Ethnicity of Householder |
|
|
Estimates and projections of
households by the race and
ethnicity of the householder are produced by applying the estimated/projected rates of change in income for each area to
the income distribution for each race/ethnicity group in the area.
The rates of change are used to project each distribution forward to
the current (or projected) year, and the resulting distributions are adjusted to conform with the householder by race/ethnicity estimates
and projections described above. |
|
|
Householders by Race and Ethnicity |
|
|
Estimates and projections of
householders by (single assignment) race and Hispanic ethnicity are
based on the estimates and projections of population by
race/ethnicity. |
|
|
For each block group, the 2000 census ratio of householders by
race/Hispanic to population by race/ethnicity is identified, and
applied to the current year estimate of population by race/ethnicity. This ratio indicates the percent of persons in each
race/ethnicity category who were householders in the 2000 census.
The final ratio is modified somewhat through the adjustment of
householders by race to total households for each area, and it is
the final current year ratio that is applied to the five-year
projections. |
|
|
Households by Size |
|
|
Working at the block group level, estimates of households by size
(number of persons) are produced for the following categories: |
|
|
1 person |
2 persons |
3 persons |
4 persons |
5 persons |
6 persons |
7 or more persons |
|
|
|
The distribution of households by size from the 2000 census serves
as the base from which the current year estimates are derived. The
2000 distribution is advanced to current year based on estimated
change in persons per household (average household size). Iterative
proportional fitting is then used to ensure consistency with
estimated household totals and average household size. |
|
|
Projections of households by size are based on the 2000 census and
current year estimated distribution of households by size. The
current year distribution is shifted to reflect the growth or
decline in average household size during the projection interval.
Iterative
proportional fitting is then used to ensure consistency with
projected household totals and average household size. |
|
|
Households by Year Moved Into Unit |
|
|
Year moved in survival probabilities
are computed from 1990 and 2000 census data (in this case reflecting
the loss of residents of specific lengths of residence). These
national level probabilities are applied to the 2000 census
distribution of households by “Year Moved In” to establish estimated
and projected distributions.
Households in excess of those surviving (staying in place) to longer
lengths of residence are those estimated to have moved in following
the 2000 census. Thus, areas with rapid household growth will show
the greatest concentrations of new movers. |
|
|
|
|
|
Housing Unit Characteristics |
|
|
Housing
Value |
|
|
Housing value (often referred to as
home value) is estimated and projected for all owner-occupied
housing units, and is based on the 2000 census measure, which
reflects census respondents’ estimates of how much their dwellings
would sell for, or the asking price of units currently for sale.
Median value is estimated and
projected as well as the distribution of units among the 24
categories of value reported by the 2000 census. |
|
|
The total number of owner-occupied
housing units is estimated by applying the relevant 2000 census
percentage to the completed estimate of total occupied housing
units. The basic rate of change in value is estimated first, and is
used to advance the 2000 census distribution of units by value to
reflect this rate of change. |
|
|
At
the national and state levels, the rate of change in home value is
monitored through the Census Bureau’s American Community Survey
(ACS), and House Price Index data from the OFHEO. Even in its test
phase, the ACS was collecting data on home value from a nationwide
sample of 700,000 households. And the OFHEO House Price Index is a
measure of post-2000 changes in housing value derived from Fannie
Mae and Freddie Mac mortgage transaction data. |
|
|
County rates of change in home value are derived from two sources at
the metropolitan area level. The first is data indicating the change
in median sales price from the NAR. Changes in sales price reflect
only units sold during the time in question, but are strongly
associated with overall change in home value. The second source is
change in the OFHEO House Price Index described above. |
|
|
At the
census tract level, change in home value is tracked with
ACE-Geosummary
data from the Equifax Consumer Marketing database. The Equifax files
do not measure home value directly, but the variable “Average
original balance on mortgage accounts” is strongly associated with
home value. Claritas has compiled tract summaries of this variable
for all census tracts for years back to 2000, and uses trends to
track small area changes in home value. The greater the increase in
mortgage amounts, the greater the increase in home values. |
|
|
Five Year Projections |
|
|
Five year projections of value are based on rates of change derived
form change in median value from 2000 census to the current year
estimate. |
|
|
For each geographic level, the estimated rates of change are used to
advance the 2000 census home value distribution to current year.
Estimates and projections are produced first at state and national
levels, but these estimates serve as targets for the county
estimates and projections, rather than control totals. Starting at
the county level, the estimates and projections serve as control
totals for small areas. |
|
|
Housing
Units by Year Built |
|
|
Estimates and projections of housing units by year built start with
the distributions from the 2000 census. These distributions are
advanced to current year (and five year) targets based on housing
loss patterns exhibited in the 1990 and 2000 censuses. |
|
|
|
|
|
Additional Terminology |
|
|
Block Group Parts |
|
|
Many Claritas methods are executed at
what is technically the block group and block group part level of
geography. Block group parts are defined where block groups are
split by place and/or MCD boundaries, and census data reported for
block groups is reported for these block group parts. Thus, block
group parts
function as a geographic level between block group and block. Because it is more familiar, the term block group level is used throughout this document. However, it is worth keeping in mind that
Claritas block group level applications usually refer to data and
methodologies executed for block groups and block group parts. |
|
|
Consistency
of Complete Count and Sample Census Total |
|
|
Because much census information was
collected on a sample basis using the census long form, the Census
Bureau used weighting techniques to present such data in complete
count form. The weighted sample totals presented in SF 3 often
differ from the SF 1 complete count totals by small amounts. For
example, a census tract with 1,200 (SF 1) households might have an
income
table (from SF 3) summing to 1,206 or 1,197 households. The
differences are statistically inconsequential. |
|
|
Claritas products provide 2000 census
data as published by the Census Bureau. The 1990 census data also is
provided as published, but has been converted to 2000 census
geography. Thus, for both 1990 and 2000 census, the usually minor
discrepancies between sample and complete count totals are
preserved. |
|
|
Adjustment
Techniques |
|
|
The adjustment process is essential to
the production of estimates that use input data at various
geographic levels, and are consistent across all levels of
geography. The Claritas updates are geographically consistent,
meaning that for each data item, block group data always sums to
tract totals, which always sums in
turn to county, state, and national totals. Adjustment techniques
also ensure that characteristic distributions sum to base count
totals (e.g., households by income always sums to total households).
The simultaneous adjustment of characteristics to higher level
control totals and to total persons or households within each
smaller area is achieved with IPF. The basic techniques
are described below. |
|
|
Ratio
Adjustment |
|
|
Ratio adjustment is used to bring
small area data into conformity with large area totals. For example,
if preliminary block group population estimates sum to a tract total
of 552, but the independent tract estimate is 561, the preliminary
block group estimates are adjusted upward by 1.63 percent (561/552)
to achieve the target tract total. Similar adjustments are made to
bring preliminary distributions (such as age and race) into
conformity with population totals for each geographic unit. |
|
|
Iterative
Proportional Fitting |
|
|
Iterative Proportional Fitting (IPF)
methods are an elaborate form of ratio-adjustment, and are used when
estimates must be adjusted to conform simultaneously with two sets
of marginal control totals—often referred to as the dimensions of a
two-dimensional table. Income by age of householder is a good
example. The estimates must sum to both households by income and
householders by age. |
|
|
IPF methods begin with a table with
target row and column totals, referred to as the row and column
marginal totals. For example, one might have 12 categories of
households by income as the row totals and 11 categories of
householders by age as the column
totals established for a 132 cell (12 by 11) table. The objective is
to produce estimates for the table’s 132 cells that sum to both the
row and column marginals. |
|
|
The execution of IPF methods requires
initial or seed cell values. In the case of income by age of
householder, seed values are obtained from the 1990 census. This
matrix of cell values reflects an intricate set of probabilities
defining the relationship between income and age—as measured for the specific geography in the census.
However, these 1990 census figures sum to neither estimated
households by income nor estimated householders by age. |
|
|
Iterative proportional fitting
achieves this conformity through a series of ratio adjustments to
the row and column marginal totals. Each round (or iteration) of row
and column adjustments brings the seed values closer to conformity
with the marginal totals. The number of iterations required varies
by area, but the values eventually converge on a result that sums,
within rounding error, to the marginal totals. The resulting
estimates not only sum to the desired marginal totals, but preserve
the statistical relationship between the two variables (income and
age) measured for the area by the census. |
|
|
Income
Distributions |
|
|
A source of occasional confusion is
the fact that the 2000 census reported income earned during calendar
year 1999. This is the case whether the data are described as 1999
income or 2000 census income. The one year census lag is logical,
since no one had yet received their 2000 income in April 2000 when
the census was taken. The Claritas update is not constrained by this
reporting
limitation, and therefore presents income for the calendar year
corresponding to the household estimate or projections. For example,
the 2005 update provided estimates of 2005 households by income
earned in 2005. When comparing such estimates against the census,
note that total households represent a five year change since 2000,
while income represents a six year change since 1999. |
|
|
Inflation and
Income |
|
|
A common question is how the effect of
inflation is
accounted for in the Claritas income estimates.
Inflation, as commonly measured by the Consumer Price
Index, reflects changing prices, and a corresponding
change in the value of a dollar. For example, items that
would have cost $100 in 1983, would have cost about $147
by 1993—a 47 percent inflation in prices. Thus, $100 was
not the same in 1993 as it was in 1983. |
|
|
Inflation is not a measure of income change, but the two are
related. Some income sources (such as Social Security and some union
contracts) are indexed by inflation, and workers typically require
and demand more pay to cover the increased costs of living. Although
income tends to follow inflation, it does not move at the same rate.
There are periods when income growth outpaces inflation, and periods
when it lags behind. These income changes relative to inflation are
referred to as real income growth. |
|
|
The
Claritas income estimates and projections are expressed in current
dollar values, which reflect how many dollars are being received at
the relevant year. As such, they reflect both real income growth (or
decline) and the change due to the effect of inflation. Rather than
estimating the effects separately, Claritas
measures the combined or net effect through input sources (such as
the Bureau of Economic Analysis income estimates), which themselves
estimate income change in current dollars. The inflation effect
built into these estimates is implicitly incorporated into the
Claritas estimates. Note that accounting for inflation in this
manner is different from controlling for inflation, which requires
removing the effect of inflation, to produce estimates in constant
dollar values. |
|
|
|
|
|
|
|