PROCESS
AND QUALITY REQUIREMENTS
The Census Bureau is the largest statistical agency of the Federal
Government. While best known for the decennial census of population
and housing, it conducts other surveys and censuses that measure
changing individual and household demographics and the economic
condition of the Nation. The Census Bureau is responsible for quinquennial
censuses of manufactures, retail trade, wholesale trade, service
industries, finance, insurance, real estate, transportation, communication,
utilities, mining, and governments. The Census Bureau also conducts
approximately 200 surveys per year. It is the source of household
demographic surveys sponsored by other Federal agencies, as well
as by the Census Bureau. It is also the source of the country's
official population estimates and projections that are used as the
basis for allocating federal funds each year. Census Bureau economic
surveys provide a majority of the information the Bureau of Economic
Analysis (BEA) uses to update Gross Domestic Product estimates,
the data used by the Bureau of Labor Statistics (BLS) in reporting
Producer Price changes, and the data used by the Federal Reserve
Board (FRB) as input to indices of industrial production and capacity
utilization. However, most of the methods used rely on very spatially
and temporally limited information, which consume great quantities
of time and expense. We have investigated the utility of remote
sensing imagery as a very helpful, easier, and cheaper way to provide
input on change and rate of change to the population estimation
DSS.
USES/IMPACTS
OFCENSUS BUREAU DATA
- Input to the BEA to update GDP estimates
- Input to the BLS to report Producer price changes
- Used by the FRB as input to indices of industrial production
- Determine the appointment of Congressional seats
- Used to distribute hundreds of billions of $$ in federal funding
- Inform about education, income, health insurance coverage
- Used by National, state, and local governments to formulate
policy
- Used by large corporations and local businesses to devise their
business plans
The Office of Management and Budget (OMB), in its February 22,
2002 issuance of government-wide information quality guidelines,
recognizes that Federal statistical organizations provide a substantial
variety of data. Accordingly, while the Census Bureau is part of
a joint Federal statistical agency notice on information quality
guidelines, it presents its specific response to the OMB directive
on quality, including utility, objectivity, and integrity.
The Census Bureau’s
Implementation of the OMB Information Quality Guidelines:
(Federal register, Vol.67, No.36, February 22,2002)
- The Census Bureau shall ensure that information disseminated
to the public shall be useful to its intended users. The requirements
of utility are ongoing for a Federal statistical agency like the
Census Bureau, which must be engaged in the continual development
of more useful data.
- The Census Bureau maintains ongoing contact with a broad spectrum
of users to ensure that its information continues to remain relevant.
Information collected by the Census Bureau is designed to provide
measures that are relevant. These measures are released to the
public as official statistics. Relevance is the degree to which
information products provide useful information for both current
needs and anticipated future needs.
- The Census Bureau disseminates statistical information products
to the public in a timely manner. Timeliness encompasses frequency
of data dissemination, as well as the closeness of the release
to the data's reference period. Efforts are made to collect and
publish data in a time interval that allows high quality data
to be disseminated to the public and also ensures that the information
is usable.
- The Census Bureau disseminates statistical information products
to the public in a manner that allows them to be accessible to
a broad range of data users with different requirements for data
availability and understandability. Accessibility is the ease
of access or effort needed for customers to acquire statistical
data, products, or services. The Census Bureau conducts usability
tests to ensure that its statistical products are accessible and
understandable to its data users.
The Census Bureau strives for ongoing improvements to meet our
customers' expectations for ease of access, quick turnaround times,
simple interface mechanisms, and comparability among different
data sources. We also continually enhance the quality of our products
and services through greater functionality in data collection
instruments as we migrate to e-commerce and computer-assisted
technologies.
- The Census Bureau shall provide information that is accurate,
reliable and unbiased and shall ensure that its information products
are presented in an accurate, clear, complete and unbiased manner.
This objectivity is achieved by using reliable data sources and
sound analytical techniques and by using highly qualified people
to prepare data products that are carefully reviewed. In the area
of statistical information, objectivity also requires acknowledging
that errors in statistical estimates are unavoidable. These areas
generally fall under the categories of "sampling" and
"nonsampling" errors. Sampling errors result when estimates
are based on a sample and not a complete canvass of the population
of interest (as in a census).
- The Census Bureau bases its information products on reliable,
accurate data that have been validated. The Census Bureau assumes
responsibility for determining sources of data (including administrative
records and other data sources), measurement methods, and methods
of data collection and processing for its censuses and surveys
while minimizing respondent burden.
Statistical products are accompanied by descriptions of, or references
to descriptions of, the methods and procedures used in their development,
and other information about the data that may affect its use. The
information on methodology provided or referenced permits the user
to determine whether the data adequately approximate what they wish
to measure, and whether the estimates they wish to use were produced
with tolerances acceptable for their intended purpose.
Objectivity in analytic results is achieved by ensuring disclosure
of the specific quantitative methods and assumptions that have been
employed, and the disclosure of error sources affecting data quality.
Statistical information products disseminated to the public by the
Census Bureau must be transparent and reproducible following prescribed
methodology. Reproducibility means that there is the capability
to use the documented methods on the same data set to achieve a
consistent result.
In summary, the quality criteria for complying with the OMB Guidelines
appear to be as follows:
- Utility
- Relevance
- Timeliness
- Accessibility (Ease of access, Quick turnaround, Friendly interface)
- Objectivity (Accurate, Complete, Unbiased, Acknowledging errors,
Use of reliable data sources, Informing users of data quality
and methodology, and Transparency and reproducibility)
- Minimizing respondent burden
Description of the
DSS –
There are several DSS being used within the FSCPE to generate the
estimate of population for a given geography. The two to which our
effort is targeted are the State/County system as well as the Sub-County
system. It should be pointed out that one of the inputs into the
Sub-County system is the estimate from the State/County system.
The inputs to the current State/County population estimates decision
support system are listed below along with the current source for
that information.
- Births and deaths (NCHS, FSCPE)
- Population (Census 2000)
- International Migration (INS, State Department)
- Internal Migration (IRS)
- Internal Migration (SSA NUMIDENT)
- Medicare (CMS)
- Group Quarters Updates (FSCPE)
Outputs from the State/County DSS go to support the development
of: analytical tables, media and government briefing tables, denominators,
monthly survey controls, and dissemination products of the U.S.
Census Bureau and FSCPE.
The Sub-County DSS is used to produce estimates for 40,630 cities,
towns, villages, and township governments. The estimates have been
produced for the following historic period: 1992, 1994, 1996, 1998,
1999, and 2002. It should be noted that the distributive housing
unit method replaced administrative records method in 1996. The
inputs to the current Sub-County population estimates DSS are listed
below along with the current source for that information.
- Housing Units and Population (Census 2000)
- Building Permits (Census)
- BAS Data (Census)
- Vacancy / PPH (Census 2000)
- Group Quarters Updates (FSCPE)
- Special Census Outputs
- County Population Estimates (State/County System)
Outputs from the Sub-County DSS go to support the development of:
analytical tables, media and government briefing tables, MSA designation
updates, HUD funds allocation, and various other dissemination products.
DSS product support
- This effort is evaluating the addition of a remote sensing based
parameter of growth that is currently not in these DSS’s.
Remote sensing data and information products (e.g., change detection
products) can benefit the production of population estimates by
addressing the following needs:
- Increase robustness of estimates
- Have more reliable empirical data
- Have additional unbiased/objective data source
- Enhance usefulness/utility
- Reduce error (MAPE) in estimates
- Minimize respondent burden
- Improve reproducibility/consistency
- Data attributes: timeliness; accuracy; accessibility; ease
of use; usability.
Objectives
The population estimation products are in response to the requirements
of the U.S. Census Bureau’s Federal-State Cooperative Program
for Population Estimates (FSCPE). The current chair of the research
component of the FSCPE is the Missouri State Demographer. The FSCPE
has wanted to evaluate new approaches to the determination of parameters
for use in their population estimation tasks.
The research arena of the population estimation was to develop a
methodology for local governments that would allow them to more
easily update population data using remote sensing. The results
of this research will benefit the decision-making for natural resources
management, natural environment monitoring and social economy developing
etc.
The research-based objectives of this project were:
- Create the multitemporal land-cover database and change-detection
products.
- To build a model for population estimation based on residential
information available in remote sensed imagery.
- To compare result from different remote sensing image and select
the most suitable remote sensing data for the estimation.
- Integrate remote sensing to Census Bureau’s traditional
population estimation model to increase the accuracy of estimation.
User Community of
Focus
Missouri State Office of Administration is the user community for
this product. This product helps the planners to understand the
population growth and land cover changes in Missouri State.
The Missouri Census
Data Center (MCDC) Program is a cooperative program operating
under a memorandum of understanding between the Office of the Secretary
of State and the U.S. Bureau of the Census. The Missouri State Library
in the Office of the Secretary of State is the agency responsible
for the program. MCDC holdings consists of more than 2,000 public
machine readable data files and numerous printed reports.
Coordinating Partners. The State Library
works in conjunction with a coordinating group of partners comprised
of the Missouri Office of Administration’s Division of Budget
and Planning and offices at the University of Missouri that provide
census support services under a contractual arrangement with the
Secretary of State’s Office. From the university side, the
Office of Social and Economic Data Analysis leads a group of other
coordinating partners that include the Geographic Resources Center,
and the Small Business Research Center to provide census support
services such as Internet application development, mapping and geographic
information systems (GIS) services, special studies.
Local Affiliates. In addition to the lead
and coordinating agencies, the Census Bureau allows a maximum of
30 local affiliates in the MCDC Program to receive free Census Bureau
products in exchange for outreach, training, and consulting services
to data users in their respective geographic or service areas.
Business and Industry Development Affiliates.
The Census Bureau also allows a maximum of 30 local affiliates in
the state to be a part of its Business and Industry Development
Center (BIDC) Program.
Local Associate Members. Rounding out
the MCDC Program are many associate agencies, comprised of federal
depository libraries, state agencies, not-for profit agencies, and
other organizations that use and disseminate data.
State FSCPE agencies, designated by their respective
governors, work in cooperation with the Population Estimates Branch
(PEB) to produce subnational population estimates.
PEB begins the process of preparing population estimates by updating
population information from the most recent census with information
found in the annual administrative records of Federal and state
agencies. The Federal agencies provide tax records, Medicare records
and some vital statistics information. The FSCPE agencies supply
state school enrollments, vital statistics, and information about
group quarters like college dorms or prisons. The Census Bureau
and FSCPE members use statistical models that combine the census
and administrative records information to produce current population
estimates consistent with the last decennial census counts. After
PEB produces estimates, they are sent to the FSCPE agencies for
review.
Objectives identified by the FSCPE members:
- promotion of cooperation between the states and the US Census
Bureau
- preparation of a set of consistent and jointly prepared county
and subcounty estimates with complete state coverage
- assurance of highest quality estimates through the use of established
methods, comprehensive data review and thorough testing
- reduction of duplication in the production of population estimates
- improvement of communication among the groups compiling population
figure
- improvement and advancement of techniques and methodologies
and the encouragement of joint research efforts
- enhancement of the recognition of local demographic work.
Product Development
All of the Missouri pilot test sites (St. Louis, Kansas City, Springfield,
and Columbia) were classified using dual-season TM/ETM imagery into
6 classes (water, forest, grass, urban, barren, agriculture) using
an unsupervised classification method. The land-cover products were
created for 1990, 2000, and 2002, and the corresponding change matrices
were created for 1990-2000 & 2000-2002. The change detection
analyses were completed except for some aspects of the statistical
analysis of population change and its relationship to the observed
landscape / land cover changes. ASTER, MODIS, Landsat ETM+, and
in some cases IKONOS image datasets were acquired and assessed to
determine the scalability of the change detection results for the
various resolutions represented by the sensor suite. Related activity
in support of this research included: 1) Conducting MODIS unsupervised
& supervised classification, 2) Developing a fuzzy classification
methodology to improve accuracy, 3) Comparison of several change
detection algorithms (change vector analysis, post-classification
comparison, simple vegetative indexes, etc.), 4) Evaluation of traditional
point-based and polygon-based accuracy assessments, and 5) Conducting
correlation analysis between land cover and population, land cover
change and population change, and population estimation change.
TM 1992 Classification for City of Springfield
 
ETM 2000 Classification for City of Springfield
The classification results were transformed into vector
format. Then, overlaid with city, & county boundary data. Therefore,
the land-cover data is linked to population data at MSA, city &
county level. A series of scatter-maps of urban area vs. population
were plotted to visualize the correlation between land-cover and
population at different administrative level. The plots showed a
strong correlation between total urban areas vs. total population.
Linear regression models were used to estimate population from TM
derived urban area.
The MOD43B4 Nadir BRDF-Adjusted Reflectance (NBAR) Product (resolution
is 1KM) is computed for each of the MODIS spectral bands (1-7) at
the mean solar zenith angle of each 16 day period. Since the view
angle effects will have been removed from the directional reflectances,
this will result in a more stable and consistent product.
We selected the best available MOD43B4 data for our study areas.
The urban areas in 2000 & 2002 for the 4 MSAs were detected
using unsupervised classification. The preliminary results are quite
good for the 1km resolution data. The population estimation was
produced at MSA and county level.
2000 MOD43B4 Nadir BRDF-Adjusted Reflectance for St. Louis Area
The results of the index for population estimation
at these different levels using regression analysis were good. The
strength of relationship is very high. The relative strength index
for regression (r2) is listed in following Table 1 (1990, 2000 population
are from census, 2002 population is from population estimates produced
by the U.S. Census Bureau).
Coefficient of determination for Population Estimation from Urban
Remotely Sensed land Cover
| R2 |
MSA |
County |
City |
| TM1990 |
0.9388 |
0.9281 |
0.9544 |
| TM2000 |
0.9964 |
0.9569 |
0.9434 |
| |
|
|
|
| MODIS2000 |
0.8936 |
0.8134 |
N/A |
| MODIS2002 |
0.8921 |
0.8216 |
N/A |
The products developed to date were created as a proof-of-concept
for the Missouri test sties in response to the FSCPE research chair’s
request for a pilot study. The products relate to the two levels
of estimation work conducted by the FSCPE (‘state / county’
and ‘sub-county’). Specifically we are addressing a
stated need for higher accuracy in estimating populations in fast
growing areas. The accuracy of these estimates is evaluated by the
FSCPE using two measures: Mean Absolute Percent Error (MAPE) and
Mean Algebraic Percent Error (MALPE). We have adopted these as one
component of our performance metrics. We have demonstrated above
that the change extracted from the various sources of imagery evaluated
(TM, ASTER, and MODIS) have significant (predictable) relationships
(high r2) with the observed aggregated population changes for specific
MSAs in Missouri at various resolution groups. Simply stated, these
geographic extent relationships to imagery source are: a) Metropolitan
Statistical Areas – MODIS, TM, b) County – MODIS, TM,
ASTER and c) Municipal (sub-county) – TM, ASTER
MSA level Population Estimation from
TM:
County level Population
Estimation from TM:
City level Population
Estimation from TM:
MSA level Population
Estimation from MODIS:
County level Population
Estimation from MODIS:
This research has successfully demonstrated the ability
to improve population estimates in four Missouri high-growth urban
areas by using remote sensing to set parameter(s) within the FSCPE
decision system related to rate and extent of growth. This research
utilized commercial high-resolution systems for validation purposes
only.
User response to these products
ICREST personnel conducted several meetings and discussions with
the Missouri State Demographer to review the requirements of the
U.S. Census Bureau’s Federal-State Cooperative Program for
Population Estimates regarding the image based population change
and monitoring aspects as they relate to population estimates. From
this and other conversations, the potential role for this application
within their decision making process is very high in terms of identifying
fast growing areas within and/or adjacent to urbanized areas. These
areas are hard to model with their current model inputs and provide
the opportunity for remote sensing to have an impact on the policy
of funding distribution equity that has direct benefit and value
to society and the general citizenry.
User Requirements for FY2004 Activities:
- 1. More research is needed into county characteristics and error.
This was the basic premise upon which we launched this research
activity. The needed ability to monitor ‘growth’ or
‘sprawl’ for the purposes of generating a more accurate
estimate is paramount to FSCPE’s goal of improving the estimates
generated for the nation.
-
Decrease the error of the estimator for those
areas where growth is occurring faster than the trend data collected
through other mechanisms would predict.
-
Expand the population change modeling to
national scope. The overall MAPE for the nation was 3.3%.
This however belies the fact that there exist regional differences
in the accuracy as measured by the estimator. Both the South
and West regions were higher (3.8 and 4.4% respectively) while
the Midwest and Northeast were lower (2.5 and 2.3% respectively).
The largest errors occurred in the following states: Hawaii
(15.4%); Nevada (8.2%); Arizona (7.3%); Colorado (5.8%), and
Florida (5.7%). The robustness of this remote sensing application
needs to be tested at a variety of locations across the country.
The designation of these areas will come from the FSCPE group
as they represent several states and have the knowledge and
data to provide input into the process.
-
Development and implement benchmark , validation,
and verification procedures for the addition of remote sensing
input to the current DSS application. It is our plan to follow
the NASA DSS Guidelines for the National Application assimilation
process.
-
Assess the impact of adding this remote sensing
input to the current DSS application. Through a formal partnership
with Census Bureau’s FSCPE, coordinate the assessment
of the efficacy of the upgraded State 2 population model to
support the distribution of federal funds, policy decision-making,
and human and fiscal resource management.
|
