Scientific Data Documentation
Postcensal Population Estimates, County Level 1990-1994
DSN: CC36.CENSUS.POSTCEN.Y9094.COUNTY
Introduction
Estimates of the Population of Counties
by Age, Sex and Race/Hispanic Origin: 1990 to 1994
These data are estimates of the resident population of the 3,143
counties in the United States as defined in 1994, by 5-year age
groups (ages 0 to 4, 5 to 9,...80 to 84, 85 and over), sex (male,
female), and modified race/Hispanic origin (White non-Hispanic;
White Hispanic; Black; American Indian, Eskimo and Aleut; Asian and
Pacific Islander; and Total Hispanic) for July 1, 1990 through
1994. These estimates are consistent with: 1) the estimates of the
population of States by age, sex, race, and Hispanic origin: July
1, 1990 to 1994 and 2) the 1991 through 1994 postcensal estimates
of the total population of counties (available under separate cover
as PE-21).
The county estimates included in this release are developed in a
two-step procedure. First a set of state estimates by single years
of age (Ages O, 1, 2,...85 and over), sex (male, female), modified
race (White; Black; American Indian, Eskimo, and Aleut; Asian and
Pacific Islander), and Hispanic Origin (Hispanic Origin, non-
Hispanic Origin) are developed. These state estimates are
developed using a cohort-component technique. A complete
description of that methodology is included as Appendix A.
The estimates of the population of counties by age, sex, and
race/hispanic origin are developed in a second step using a ratio
method. The ratio method is a technique for adjusting data to sum
to a pre-determined total. It consists of multiplying each element
of the data by the ratio formed by dividing the desired total by
the sum of the data. When there are multiple totals to which we
wish to adjust our data, as with the county estimates, we first
partition the data into groups which correspond to the desired
totals, then construct and apply ratios for each group using the
same method as in the single-total situation. Applying the ratio
method to a data set is referred to as raking.
The detailed state estimates by age, sex, race, and Hispanic Origin
developed in step 1 are aggregated by state to five year age
groups, sex, and race/hispanic categories. These aggregated state
estimates along with the estimates of the total population of
counties serve as the control totals for the estimates of the
population of counties by age, sex and race/hispanic origin. The
methodology used to develop the estimates of the total population
of counties is described in "Subnational Estimates of Total
Population by the Tax Return Methodology" by Michael Batutis
(available from the Population Division, US Bureau of the Census).
The April 1, 1990 modified census counts for counties by age, sex,
and modified race/hispanic origin released as computer tape file
MARS, STF-S-3 are used as the starting point. For July 1, 1990,
the census level county-level MARS data were raked to: 1) the
estimates of State population by five-year age group, sex, race and
Hispanic origin developed in step 1; and 2) the 1990 estimates of
the total population of counties. The process was repeated for the
1991 through 1994 estimates adjusting the data cells to the
appropriate set of state and county estimates.
It is important to note that our regularly published county and
State estimates have been updated since the productions of these
estimates. Consequently, there will be a slight discrepancy
between these estimates and the most recently released county
population estimates and State estimates by age and sex, since
these estimates were controlled to an earlier version of the county
and State estimates.
These data were developed as part of an ongoing project to develop
postcensal population estimates of states and counties by age, sex,
race, and Hispanic origin. Though the method we employed produced
estimates which were fully disaggregated with respect to race and
Hispanic origin, the limitations of this approach are such that we
felt it necessary to combine the fully disaggregated race-Hispanic
origin categories into the categories used here. Work is
continuing on methods and data sets that can be use'd to more
directly estimate the age, sex, race, and Hispanic origin
distributions of the State and county populations. As additional
steps are completed, we plan to prepare new estimates for
subsequent years and revise the existing series back to 1990.
This data set contains population estimates disaggregated by five-
year age groups, sex and race/Hispanic origin for each county.
However the limitations of our methodology are such that we do not
consider these data to be accurate for each individual cell.
Although we do not have measures of error, we believe that
aggregating the individual cells to larger groups will reduce the
level of error. We include the separate data for your convenience
in aggregating to various groups. Although the data shown in this
data set are unrounded, we do not consider these data to be
accurate to the last digit.
Technical Contact:
Larry Sink
Population Division
301-457-2461
Appendix A:
Estimates of the Population of States
by Age, Sex, Race and Hispanic Origin: 1990 to 1994
These data are estimates of the resident population of the 50
States and the District of Columbia by single years of age (age O,
1, 2,....85 and over), sex (male, female), race (White; Black;
American Indian, Eskimo, and Aleut; Asian and Pacific Islander)
and, Hispanic origin (Hispanic origin, non-Hispanic origin) for
July 1, 1990 through 1994. These estimates are consistent with the
postcensal estimates for the Nation and States by age, sex, race
and Hispanic origin for 1990 through 1994 published in Current
Population Reports, Series P-25, No. 1127.
The State estimates included in this release are developed using a
cohort-component method whereby each component of population change
- births, deaths, domestic migration, and international migration
is estimated separately for each birth cohort by sex and race.
The cohort-component method is based on the traditional demographic
accounting system:
P1 = P O + B - D + NDM + NMA
where:
P1 = population at the end of the period
PO = population at the beginning of the period
B = births during the period
D = deaths during the period
NDM = net domestic migration during the period
NMA = net migration from abroad during the period
To generate population estimates with this model, we first
developed separate data sets for each of these components. The
procedures by which these data are developed by single year of age,
sex, race, and Hispanic origin are described in the following
sections. Once the data for each component were developed, the
estimates could be produced simply by adding the components
together, with the exception of internal migration. The reason for
this exception and our procedure for dealing with internal
migration are explained in the internal migration section below.
This overall approach is similar to that used in the development of
the experimental set of state and metropolitan area estimates by
race and, Hispanic origin. The Current Population Reports, Series
P-25, No. 1040-RD-I, provides a detailed discussion of this general
approach.
Starting Population
The April 1, 1990 Census data files which were used as the starting
points in this methodology represent the modified age, race, sex,
and Hispanic origin (MARS) census data released as computer tape
file (MARS, STF-S-3). The modification methodology is outlined in
Census Report, CPH-L-74.
In order to develop the desired July 1, 1990 starting point from
the April 1, 1990 data, we used the ratio method to make the April
1 data consistent with the July 1, 1990 national population
estimates by age, race, sex, and Hispanic origin and consistent
with the July 1, 1990 state population estimates by age and sex.
The ratio method is a technique for adjusting data to sum to a pre-
determined total, which consists of multiplying each element of the
data by the ratio formed by dividing the desired total by the sum
of the data. When there are multiple totals to which we wish to
adjust our data, as with the state age-sex estimates, we first
partition the data into groups which correspond to the desired
totals, then construct and apply ratios for each group using the
same method as in the single-total situation. Applying the ratio
method to a data set is referred to as raking.
Vital Statistics
The data for births and deaths used in these estimates are based on
1) detailed data available from the National Center for Health
Statistics (NCHS); 2) estimates of births and deaths for counties
developed by the member agencies of the Federal State Cooperative
Program for Population Estimates (FSCPE); and 3) estimates of
births and deaths by detailed demographic characteristics developed
as part of the Population Division program for national population
estimates.
Births-- Extracts of detailed individual-record data on births
from NCHS for calendar years 1990 through 1993 are used to
construct the state-level births. The race and Hispanic
origin codes on the individual NCHS records were converted
into our four race and two-Hispanic origin system as shown on
the attached chart. The individual records for events
occurring July 1, 1991 through June 30, 1993 are aggregated by
year to the county level and adjusted to the county-level data
provided by the FSCPE member agencies. These results are
further adjusted to agree with national-level race-Hispanic
origin estimates developed as part of the national population
estimates program (see ref. 4, p5).
To estimate the July 1, 1993 through June 30, 1994 period, we
first aggregated the individual records for events occurring
for the six month period July 1, 1993 through December 31,
1993. Because the 1994 data were not available when we
developed the original set of estimates, we used an
alternative method to estimate the detailed data for the
January 1 to June 30, 1994 period. To do this, we adjusted
the six month aggregations for July through December to: 1)
preliminary estimates of births for July 1993 through June
1994 provided by the FSCPE agencies; and 2) to preliminary
estimates of national level births by race, sex and Hispanic
origin for the July 1993 through June 1994 period, developed
as part of the national estimates program (see ref. 4, p5).
Deaths -- The estimates of deaths are developed in the same
manner as that for births. However, because death data have
the additional dimension of age, the resultant number of
national controls was too great to be handled simultaneously
with the county-level controls. Consequently, state-level
death estimates were prepared independently of the county-
level estimates, by summing individual records to the state
level by age, race, Hispanic origin, sex, and period, and
adjusting these totals to national-level age-race-Hispanic
origin controls.
To develop the deaths by age, we examined the age at death and
the date of birth information on the individual record. If
the date of birth information was missing, age was set to the
most recent valid value. The preliminary age at death value
was computed using date of birth information. This
preliminary value was compared to the age at death value on
the individual record. If the difference between these two
values was no greater than two years, the computed value was
used. If not, the age at death value on the certificate was
used.
Internal Migration
The values for internal migration used in these estimates are
developed using a variant of the basic administrative records
method. The development of the data rely upon two basic files - an
annual extract of tax returns provided by the Internal Revenue
Service (IRS), and a 20% sample of information on the Social
Security Administration Application File (NUMIDENT) which includes
Social Security Number (SSN), month and year of birth, race, sex,
and 6 characters of the last name.
The basic Administrative Records method relies upon annual extracts
of tax returns provided by the IRS. In this approach, using the
SSN on the return, we are able to match the tax returns for two
years and obtain state of residence for the two periods. By
comparing the state of residence at the two points in time, we are
able to develop annual measures of migration for states.
Because the standard tax return provides no demographic
characteristics of the tax filer, the basic administrative record
method provides data for the total population only. To extend this
approach to demographic characteristics, we rely upon an extract of
the NUMIDENT file. Because the Census Bureau is able to receive
only a 20-percent sample of this basic NUMIDENT file, we can only
append the demographic characteristics of the primary filer to the
same 20-percent sample of tax returns.
In addition to demographic characteristics of the primary filers,
the model required demographic characteristics of those persons
claimed as exemptions on the tax return. The rules for assigning
demographic characteristics to dependents are straight forward and
rely on basic familial and demographic relationships.
1. Spouses on the tax return are given the age and
race/Hispanic origin of the primary filer. They are
assigned the opposite sex of the primary filer.
2. Dependent children are given the race/Hispanic origin of
the primary filer and all assigned to the age group under
20. We did not attempt to assign a sex to the dependent
children category.
3. Parent exemptions are assigned the race/Hispanic origin
of the primary filer and all assigned to the age group 65
and over.
4. Other dependents are assigned the race/Hispanic origin of
the primary filer and all assigned the age group under
20.
In order to develop an estimate for July 1 of a given year using
the cohort-component method, we need an estimate of the migration
which took peace between July 1 of the preceding year and June 30
of the year in question. However, the migration data we obtain
using the administrative records method pertain to time periods
determined by when the individual tax-payers file their returns,
which, of course, varies from tax-payer to tax-payer. Since most
tax returns are filed between January and April, it is roughly
correct to say, for example, that the migration data obtained as a
result of matching returns from tax years 1989 and 1990 pertains to
a (one-year) period within the interval from January 1990 (the
earliest most tax-payers would file their 1989 returns) to April
1991 (the latest most tax-payers would file their 1990 returns).
We have assumed that this is a reasonable approximation to the
interval needed for our 1991 estimates and, similarly, that the
data from the tax years 1990-1991, 1991-1992, and 1992-1993 match
are appropriate for our 1992, 1993, and 1994 estimates,
respectively. These assumptions are based on our research which
indicates that state-to-state migration rates change on average
only about 15% a year, and these changes tend to offset one another
when out-rates and in-proportions are calculated.
The migration data yielded by the method described above consists
of counts of those tax-filers whose SSNs were in the 20% sample
plus the dependents claimed by those filers, disaggregated by state
of origin, state of destination, age, race, sex, and, Hispanic
origin. The first step in converting these counts into the
statistics actually used in our estimates is to construct state-to-
state migration rates by demographic characteristic (i.e. age,
race, sex, and, Hispanic origin). This is done by summing all the
counts for a given origin by characteristic and then dividing each
count by the sum for that characteristic.
Because of the potentially large number of origin-destination-
characteristic combinations, it is necessary to collapse age into
categories to avoid stretching the data too thin. The categories
we selected are 0-19, five-year categories from 20-24 to 60-64, and
65 and over. The 0-19 and 65+ categories were selected because we
have no age information for dependents, and we assign all dependent
children to the 0-19 category and all dependent parents of filers
to the 65+ category. Once the state-to-state rates are used to
derive out-rates and in-proportions using the method described
below, the out-rates and in-proportions are converted from age-
groups to single years by giving each year the value of the group
to which it belongs except for those years which form age-group
boundaries (e.g. 19 and 20), which are averaged. This is done to
prevent drastic changes between years.
An additional problem is created by the fact that the SSA data have
only three race categories: White, Black, and Other. In order to
convert from this system into the four race system used in these
estimates, it is necessary to split the "Other" category into
American Indian, Eskimo and Aleut (AIEA) and Asian & Pacific
Islander (API). This split is based on the relative sizes of the
total, AIEA and API populations in the origin state for the out-
rates and on the relative sizes in the destination state for the
in-proportions. The racial composition of migration flows depends
upon the racial composition of both the origin and the
destination, so that in reality this "Other" group probably has a
different composition for each of the 2550 different state-to-state
flows, but the numbers involved are too small to permit separate
analysis for each flow. By combining the state-to-state rates into
out-rates and in-proportions, with the method described below, we
greatly increase the number of observations underlying each of our
statistics and have the ability to base our rate calculations
solely on origin characteristics and our proportions solely on
destination characteristics. These separate rates and proportions
for the four race groups were applied only to the non-Hispanic
population. One set (by age and sex) of migration rates and
proportions are computed for Hispanics without regard to race and
applied to all Hispanic race groups.
The creation of out-rates and in-proportions from the state-to-
state migration rates involves converting the origin-destination-
characteristic-specific rates into origin-characteristic-specific
rates and destination-characteristic-specific proportions. In this
process, all calculations are performed separately for each
combination of demographic characteristics. The state-to-state
migration rates are multiplied by our starting population estimate
for the appropriate group to obtain an estimate of the total
migration flow between the states in question for this group.
These flows are summed to get total out- and total in-migration by
characteristic for each state. Each state's out-migration totals
are divided by their respective populations to obtain out-migration
rates and the out-migration totals are summed across states to
obtain the national total of migration by demographic
characteristic. These national totals are divided into each
state's in-migration totals to obtain the in-migration proportions.
It should be noted that the population figures used in these
calculations are our starting population estimates, since at this
stage of the process we do not have population estimates for the
periods to which the migration rates pertain (except for the 1991
estimates, for which our starting population is the beginning-of-
period population). The out-rates and in-proportions are converted
into actual estimates of migration within the process which
produces the finished population estimates, since it is only at
this point that we have the population figures needed. This
conversion is accomplished by multiplying each state's out-rates
by the respective beginning-of-period population to obtain our
estimate of that state's out-migration, which is then summed across
states to obtain national-level migration. Finally, each state's
in-proportions are multiplied by the national-level migration to
get that state's in-migration estimates.
International Migration
The international migration component in these estimates is an
aggregation of four separate parts: 1) alien immigration, refugees,
and net undocumented migration; 2) legal emigrants; 3) net movement
between Puerto Rico and the mainland; and 4) net movement of
federal civilian citizens.
Immigration (including refugees and undocumented). We
utilized legal immigration data developed from the
Immigration and Naturalization Service public use
microdata, refugee data drawn from unpublished reports of
the Office of Refugee Resettlement, and net undocumented
immigration files developed as part of the national
estimates program (see ref. 5, pp24-29). The legal
immigration and refugee files both possess full
demographic detail and state-level geography for both
years in question. The file on net undocumented
immigration contains full demographic detail at the
national level, but no state-level information. We gave
undocumented immigration the state-level distribution of
the emigrants by raking the emigrant file to the
undocumented's national distribution.
Legal emigration. We began with state-level data on
legal emigration with race and ethnic detail (see ref.
6), to which we applied the age-sex distribution from the
emigration data used in our state projections (see ref.
2, p.xxvii). The resulting data were raked to national-
level controls developed for the national estimates
program (see ref. 5, p38). Emigration was assumed to
remain constant over the estimation period.
Net Puerto Rican migration. We utilized a national-
level file on net Puerto Rican migration with full
demographic detail for both years developed as part of
the national estimates program (see ref. 3, p6). This
net migration was distributed to the states based on
their respective portions of the Puerto Rican migration
developed from past research (see ref. 6).
Net federal citizen migration. We utilized a national-
level file on net federal citizen migration with full
demographic detail for both years developed as part of
the national estimates program (see ref. 3, pp6-7).
State-level distributions were obtained using the IRS-SSA
data employed in the internal migration estimates, which
also contains data on movements to and from foreign
countries. The Other races distribution was used for
both, AIEA and API, and the below 20 and above 65 age
distributions were taken from the national distribution.
These state-level distributions were raked to the
national-level distribution to yield the final data.
Consistency with Previous Estimates
Once we had used the cohort-component method described above to
produce preliminary population estimates, we used the ratio method
to make these estimates consistent with previously published State
and National estimates (for a description of past State estimates,
see ref. 1). This was done by raking each year of the new
estimates to the corresponding year's National estimates by age,
sex, race, and, Hispanic origin and to the corresponding year's
State estimates by State, age, and sex. Because this procedure
produces fractional numbers, a special rounding routine was applied
which transforms all individual estimates into integers while
preserving their consistency with the State and National controls.
It is important to note that our regularly published State and
national estimates have been updated since the production of these
estimates. Consequently, there will be a slight discrepancy
between these estimates and the most recently released national
estimates and State estimates by age and sex, since these estimates
were controlled to an earlier version of the State and national
estimates.
Limitations
These data were developed as part of an ongoing project to develop
postcensal population estimates of states and counties by age, sex,
race, and, Hispanic origin. These estimates represent an
intermediate step in this overall project. Work is continuing on
methods and data sets that can be used to more directly estimate
the age, sex, race, and Hispanic origin distributions of the state
and county populations. As additional steps are completed, we plan
for subsequent years and revise the 1990.
This data set contains population estimates disaggregated by single
year of age, sex, race, and, Hispanic origin for each state.
However, the limitations of our methodology are such that we do not
consider these data to be accurate for each individual cell.
Although we do not have measures of error, we believe that
aggregating the individual cells to larger groups will reduce the
level of error. We include the separate data for your convenience
in aggregating to various groups. Although the data shown on this
diskette are unrounded, we do not consider these data to be
accurate to the last digit.
Technical Contact:
Larry Sink
Population Division
(301) 457-2461
References
1. Batutis, Michael J., "Subnational Estimates of Total Population
by the Tax Return Methodology", Population Division, U.S.
Bureau of the Census, Washington, DC, 1994.
2. Campbell, Paul R., Population Projections for States, by Age,
Race, and Sex: 1993 to 2020, U.S. Bureau of the Census,
Current Population Reports, P25-1111, U.S. Government Printing
Office, Washington, DC, 1994.
3. Deardorff, Kevin E., Frederick W. Hollmann, and Patricia
Montgomery, "U.S. Population Estimates by Age, Sex, Race, and
Hispanic Origin: 1990 to 1994", U.S. Bureau of the Census,
PPL-21, 1995.
4. Hollmann, Frederick W. United States Population Estimates, by
Age, Sex, Race, and Hispanic Origin: 1980 to 1988, U.S. Bureau
of the Census, Current Population Reports, Series P-25,
No.1045, U.S. Government Printing Office, Washington, DC,
1990.
5. ______________. "U.S. Population Estimates, by Age,
Sex, Race, and Origin: 1990 to 1993", U.S. Bureau of
the Census, PPL-8, 1994.
6. Word, David L. "The Census Bureau Approach for Allocating
Internal Migration to States, Counties and Places:1981-
1991", U.S. Bureau of the Census, Technical Working Paper No.
1, 1992.
Conversion of National Center for Health Statistics
(NCHS)
Race, Ethnicity, and Age for State Estimates
RACE
NCHS State Estimates
(1) White ------------------------------------------- White
(2) Black ------------------------------------------ Black
(3) American Indian, Eskimo or Aleut ---------------- American Indian,
Eskimo, or Aleut
(4) Chinese ---------------------------------------|
(5) Japanese --------------------------------------|
(6) Hawaiian -------------------------------------|___Asian and Pacific
(7) Filipino -------------------------------------| Islander
(8) Other API -------------------------------------|
(9) Other Race ------------------------------------|
ETHNICITY
NCHS State Estimates
(00) Non-Hispanic ------------------------------------ Non-Hispanic
(01) Mexican --------------------------------------|
(02) Puerto Rican ---------------------------------|
(03) Cuban ----------------------------------------|___Hispanic
(04) Central of South American --------------------|
(05) Other Hispanic -------------------------------|
(99) Unknown, not asked -------------------------------Allocated
according to
proportion
Hispanic for
appropriate
sub-group in MARS
file.
Estimates of the Population of Counties
by Age, Sex, and Race: 1990-1994
This file contains estimates of the population of the 3,143
counties (1994 geography) and the 51 State totals in the United
States by five-year age groups (ages 0-4, 5-9, ..., 80-84, 85+),
sex (male, female), and modified race/Hispanic Origin for July 1,
1990 through July 1, 1994.
Record length: 153 Characters
Block length: 18360 Characters
Record count: 191640 (12 per county)
Block count: 1597
Character Item Description
1-2 Year ('90', '91', '92', '93', or '94')
3-4 FIPS State Code
5-7 FIPS County Code ('000' = State Total)
8-9 Race/Sex Indicator:
1 = White Not Hispanic male
2 = White Not Hispanic female
3 = White Hispanic male
4 = White Hispanic female
5 = Black male
6 = Black female
7 = American Indian, Eskimo, or Aleut male
8 = American Indian, Eskimo, or Aleut female
9 = Asian or Pacific Islander male
10 = Asian or Pacific Islander female
11 = Total Hispanic male
12 = Total Hispanic female
Ages:
10-17 0 to 4 years
18-25 5 to 9 years
26-33 10 to 14 years
34-41 15 to 19 years
42-49 20 to 24 years
50-57 25 to 29 years
58-65 30 to 34 years
66-73 35 to 39 years
74-81 40 to 44 years
82-89 45 to 49 years
90-97 50 to 54 years
98-105 55 to 59 years
106-113 60 to 64 years
114-121 65 to 69 years
122-129 70 to 74 years
130-137 75 to 79 years
138-145 80 to 84 years
146-153 85 years and over
