Below are the first 10 and last 10 pages of uncorrected machine-read text (when available) of this chapter, followed by the top 30 algorithmically extracted key phrases from the chapter as a whole.
Intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text on the opening pages of each chapter.
Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.
Do not use for reproduction, copying, pasting, or reading; exclusively for search engines.
OCR for page 6
CHAPTER 2
American Community Survey
The Census Bureau's stated goals for ACS are to
· Provide federal, state, and local governments with an information base for the administration
and evaluation of government programs;
· Facilitate improvement of the 2010 Census by allowing the decennial census to focus on
counting the population; and
· Provide data users with timely demographic, housing, social, and economic statistics updated
every year that can be compared across states, communities, and population groups.
The Census Bureau began developing the ACS in the mid 1990s. In the first few years, while
the program was just beginning, preliminary ACS data were collected for a few test sites. In 1999,
the number of test sites was increased to 31 locations, comprising 36 counties and representing
a broad range of communities that were selected to provide different combinations of popula-
tion sizes, population characteristics, population growth levels, and difficulty of enumeration.
Table 2.1 lists the ACS test sites and the annual sampling rates that were used for each.
The data collection effort for the 31 test sites has been performed annually since 1999, and
among the most important outputs of the ACS testing phase has been the compilation of three
complete years of data for 1999, 2000, and 2001. In addition to the test site program, the Census
Bureau performed a large-scale (1,203 counties) operational test of ACS methods in the year
2000, entitled the Census 2000 Supplementary Survey (C2SS). These data allow for the compar-
ison with Census 2000 decennial data along several dimensions.
The Census Bureau and other researchers have performed a wide range of analyses related to
the ACS test data. Many of these efforts are summarized below. In addition, this section describes
the key elements of the plan for implementing the full-scale ACS, and how these elements of the
ACS program may affect ACS data analyses--particularly those analyses related to transporta-
tion planning applications.
2.1 ACS Implementation
The Census Bureau first described their plans for fully implementing ACS in the American Com-
munity Survey Operations Plan and in associated website documents.1 These plans have evolved as
a result of ACS testing and federal appropriations processes. The schedule for the transition to full
implementation has slipped due to limitations and uncertainties in the appropriations process, but
the operational components of the program seem to be established and are documented in an ACS
1
U.S. Census Bureau, American Community Survey Operations Plan: Release 1 (March 2003). www.census.gov/
acs/www/index.html.
6
OCR for page 7
American Community Survey 7
Table 2.1. American community survey test sites.
1999-2001 1999-2001
County Annual Sampling Rate County Annual Sampling Rate
Pima County, Arizona 5% Multnomah County, Oregon 5%
Jefferson County, Arkansas 5% Fulton County, Pennsylvania 5%
Tulare County, California 5% Schuylkill County, Pennsylvania 5%
Upson County, Georgia 5% Sevier County, Tennessee 5%
Miami County, Indiana 5% Starr County, Texas 5%
Black Hawk County, Iowa 5% Zapata County, Texas 5%
DeSoto Parish, Louisiana 5% Petersburg City, Virginia 5%
Calvert County, Maryland 5% Yakima County, Washington 5%
Hampden County, Massachusetts 5% Ohio County, West Virginia 5%
Madison County, Mississippi 5% Oneida County, Wisconsin 5%
Iron County, Missouri 5% Vilas County, Wisconsin 5%
Reynolds County, Missouri 5% San Francisco County, California 3%
Washington County, Missouri 5% Broward County, Florida 3%
Flathead County, Montana 5% Lake County, Illinois 3%
Lake County, Montana 5% Bronx Borough, New York 3%
Douglas County, Nebraska 5% Franklin County, Ohio 3%
Otero County, New Mexico 5% Fort Bend County, Texas 1%
Rockland County, New York 5% Harris County, Texas 1%
technical paper on design and methodology.2 This section describes the ACS operations and imple-
mentation schedule based on the descriptions provided by the Census Bureau.
2.1.1 Operational Components of ACS3
Once fully implemented, the ACS will sample about 3 million addresses from the Master
Address File (MAF) each year and about 2.5 percent of group quarters populations. This annual
sample will be systematically divided into 12 monthly samples for interviewing, and the sampled
units will then be contacted to provide data. The collected data will then be processed and
refined, and made available to data users on an annual basis. Depending on the size of the
geographic area under study and the analyses being performed, data users may need to combine
multiple years of ACS data to analyze specific geographic areas--the more detailed geography,
the greater the number of years that will need to be combined.
To accomplish the ongoing implementation of ACS, the Census Bureau will need to contin-
uously perform the following functions:
· Address list development and updating to provide the sample universe;
· Implementation of sample selection protocols to obtain a sample each month;
· Implementation of the following data collection:
Mail out/mail back data collection phase;
Computer-assisted telephone interviewing (CATI) data collection phase; and
Computer-assisted personal interviewing (CAPI) data collection phase.
· Implementation of data entry and telephone follow-up procedures for mail returns;
2
U.S. Census Bureau, Design and Methodology: American Community Survey, Technical Paper 67 (May 2006) U.S.
Government Printing Office, Washington, D.C.
3
This entire section relies heavily upon the ACS Design and Methodology and Operations Plan documents.
OCR for page 8
8 A Guidebook for Using American Community Survey Data for Transportation Planning
· Data processing as follows:
Coding, editing, and imputation procedures; and
Weighting, disclosure editing, and tabulation,
· Data product dissemination.
These elements of the ACS process are discussed below.
Address List Development and Update The Census Bureau maintains the MAF, a national
address sampling frame for the decennial census and other census data collection activities.
Maintaining the quality of this database will be an essential element of successful implementa-
tion of the ACS program. Therefore, the Census Bureau is actively engaged in efforts to improve
the database and to maintain it into the future.
The MAF was developed for Census 2000 using the previous decennial census list, the U.S.
Postal Service's Delivery Sequence File, and address data supplied by local governments. The
MAF is linked with the Census's Topologically Integrated Geographic Encoding and Referenc-
ing (TIGER) database.
The TIGER system and the MAF currently are being updated by the Census Bureau in prepa-
ration for the 2010 decennial census.4 One update process is called the MAF/TIGER Accuracy
Improvement Project, or MTAIP. The project, expected to be complete by 2008, will improve
the positional accuracy of street centerlines in the TIGER database.
The update process is using existing data sources whenever possible, including
· State/local/county/tribal GIS files;
· Commercial GIS files; and
· Existing imagery.
If existing data are not available, new sources, such as imagery and field collection, are used.
Although the project will result in spatially more accurate TIGER/Line files, the TIGER/Line
identifiers will not change. Attribute data will be conflated to the new geometry. For new seg-
ments, if city-style addresses are present in the file, they will be transferred to TIGER.
The MAF/TIGER Accuracy Improvement Project (MTAIP) process is focused on Census 2010
and is expected to be useful to ACS after 2008. A pilot study of acquiring coordinates for resi-
dential structures also is being conducted, where attributes, including feature names, address
ranges, and address lists as appropriate from state/local/tribal/county GIS files are collected.
The MAF is kept up to date by use of the U.S. Postal Service's Delivery Sequence File for both
residential and non-residential addresses. The update takes place twice a year for those blocks
that are completely city-style residential addresses. In addition, ACS field representatives note
any address corrections found in visiting housing units during the personal visit non-response
follow-up data collection phase. The Census Bureau also performs systematic listing and map-
ping of selected areas to support several of their data collection efforts. Finally, to address qual-
ity concerns relating to areas with high concentrations of non-city-style addresses, the Census
Bureau has initiated a program called the Community Address Update System (CAUS).
Sample Selection Protocols According to the Design and Methodology document, when
ACS is fully implemented, each year the Census Bureau will select a systematic sample of addresses
(3 million addresses per year or 250,000 addresses per month) from the most current MAF. Ini-
tially, this sampling rate will be equivalent to 2.5 percent of households each year, but this rate will
decrease over time as the nation's population increases. In addition, about 2.5 percent of the
4 Robert Lamacchia, U.S. Census Bureau, "TIGER/MAF Update Process." Presentation to U.S. DOT on July 9,
2004, as part of the ACS FAIP Program.
OCR for page 9
American Community Survey 9
people in group quarters facilities will be included in the ACS. The sample will be selected from
each county in the United States. No address will receive the ACS questionnaire more than once
in any five-year period.
To improve the reliability of estimates for small governmental units (such as small counties or
American Indian reservations) with less than 1,200 addresses, some areas will be over sampled
similar to what was done for the census 2000 Long Form design. For 2005, the actual sampling
rates are expected to range from 1.6 percent to about 10 percent each year.
In the future, the Census Bureau also will consider additional over sampling of certain counties
to try to improve the reliability of estimates of geographically dispersed small minority population
groups (such as Native Hawaiians and other Pacific Islanders, Asians, or American Indians and
Alaska Natives) living in urban areas, but these changes would not be made until the current over
sampling scheme for mail survey response is fully analyzed.
2.1.2 Questionnaires
The current ACS questionnaire (ACS 2003) is the result of several iterations of questionnaire
implementation and revision. After the ACS demonstration period testing (1996-1998) and prior
to the comparison period testing (1999-2001), the ACS questionnaire was modified. In addition,
the ACS questionnaire was modified after the comparison period, and again for the 2003 ACS.
The same questionnaire is being used for the 2003-2007 ACS efforts.
In preparation for the 2008-2012 period, the Census Bureau has been conducting the 2006
ACS Content Test to evaluate potential reworded and reformatted questions and to try new
questions related to marital history, health insurance coverage, and veteran's service-related
disabilities. Some potential rewording of questions related to work status would have the great-
est impact on transportation planners. It is expected that the results of the 2006 test will lead to
the 2008 questionnaire, and that this questionnaire will remain the same through 2012.
Appendix A summarizes differences between the current ACS questionnaire and the decen-
nial census Long Form (1990 and 2000) for housing and population questions. The ACS and
Census 2000 questionnaires roughly have the same questions (in different question order). The
differences in the data collection protocol, however, lead to a few key differences in the popula-
tion questions, as described below.
Residence Rules The ACS uses different residence rules than have been used in past decen-
nial censuses. Decennial censuses and most surveys use the usual residence concept. The usual
residence concept requires that respondents have only one place as their usual residence--most
often the place where they spend the most time. The usual residence rule does not count people
who are staying somewhere other than their usual residence as occupants of that place. For exam-
ple, people who spend their winters in Florida and the rest of the year in Vermont, so called
"snowbirds," have in the past been enumerated in the census as residents of Vermont, not Florida.
The ACS, in contrast, uses the current residence concept and the Two-Month Rule. Under the
Two-Month Rule, anyone who is living for more than two months in a survey unit when the unit
is contacted (either by mail, telephone, or personal visit) is considered to be a current resident
of that unit.
Persons who are away from a residence for two months or less, regardless of their temporary
location or the purpose of their travel, are considered to be "in residence" at the residence. If a
residence does not have any occupants for more than two months from when it is sampled for
the ACS, it is classified as a vacant housing unit. If a residence is occupied only by individuals
that stay there for two months or less, and who have another permanent address, the residence
OCR for page 10
10 A Guidebook for Using American Community Survey Data for Transportation Planning
is classified as a temporarily occupied housing unit. Only limited housing unit data are collected
for vacant and temporarily occupied housing units (no household or person data).
The ACS Two-Month Rule has the following exceptions:
· Children (kindergarten through Grade 12) away at boarding schools are considered residents
of their parental home. (College students' current residency is based on the Two-Month Rule.)
· Children living in joint custody and who frequently move between separate residences are
considered to be residents of the sampled residence if they are present at that residence when
contact is initially made.
· Commuter workers who stay in a residence close to where they work and return regularly to
their primary residence are considered to be residents of their primary residence, not the
work-related one.
The current residence concept suits the ACS, because the ACS continuously collects informa-
tion from monthly samples throughout the year. The current residence concept recognizes that
people can live in more than one place over the course of a year, and that population traits for
some areas may be noticeably affected by these shifts. Although ACS will not capture the seasonal
changes in the population (because ACS estimates are tied to Census Bureau annual estimates for
July 1), ACS can capture the characteristics of the population for the full year.
Reference Date An important difference between ACS data and previous decennial census
data that is brought about by the continuous nature of ACS data collection is the reference period
of the survey. In the decennial census, the questions are referenced to the beginning of April of
the census year, and questions that require retrospective information are tied to the calendar
year. For example, in Census 2000 respondents were asked the location of their places of work
for the week before the April 1, 2000, census date and their household incomes for the 1999
calendar year. For ACS, the questions are referenced to the time the survey is conducted. Respon-
dents are asked at what location they worked last week. The ACS household income reference
period is the 12 months ending in the month prior to the survey.
The ACS's variable reference dates will capture seasonal differences the decennial census could
not capture, but it is important that analysts consider the changed reference date definitions
before using the ACS data, particularly in comparison with previous census estimates.
Other Questionnaire Differences Although data elements between the ACS and decennial
census Long Form are consistent, wording differences (for both the query and the answer cate-
gories) do exist. Such wording changes, however, have been common in the evolution of the
Long Form because the Census Bureau has a program of continuous improvement for ques-
tionnaire items. As in the past, analysts will need to be cautious when trending data elements for
which there have been wording changes. Research of survey methods indicates that the wording
of questions affects the corresponding answers.
For example, in an instruction for the housing questions, the ACS directs "Please answer the
following questions about the house, apartment, or mobile home at the address on the mailing
label." The Long Form instructed, "Now, please answer [the housing] questions about your
household." While the questions seek essentially the same information, some respondents could
have interpreted them differently, thus differences in the population could be identified where
none really should exist. There also are small differences in the ways that the ACS and decennial
census efforts collect data from respondents in larger households.
2.1.3 Data Collection Procedures
The ACS data collection occurs in continuous, three-month cycles using a combination of
mail out/mail back, CATI, and CAPI data collection modes. The data collection protocols were
OCR for page 11
American Community Survey 11
established based on the Census Bureau's experience with the decennial census and their demo-
graphic surveys. Figure 2.1 shows the workflow for the ACS data collection effort.
The data collection process begins with the mail phase. Sampled addresses are evaluated to
determine whether they are accurate and complete. Thus far, over 95 percent of the sample uni-
verse have been eligible for the ACS mail out. Those sampled units with non-mailable addresses
are assigned to the CAPI follow-up. If a sampled unit has a valid address, the Census Bureau mails
a prenotification letter, then the initial mailing package (which includes the ACS questionnaire,
an instruction booklet, and related materials), and then a reminder card. If no response is received
from an address after three weeks, a replacement mailing package is sent. Currently, only English
Sample Drawn from MAF
Valid Mailed Address?
No Yes
Mailed Prenotification Letter
Mailed ACS Questionnaire, Instructions, Related Materials
Mailed Reminder Card
(After Three Weeks as Necessary) Replacement Mailing Package
Response Received?
No Yes
Telephone Number
Available?
No Yes
Telephone (CATI) Phase
Response Received?
No Contact Yes
Quality Control,
Assigned to Subsample Drawn from Coding Editing
Personal Visit CAPI Addresses
(CAPI) Phase
In Subsample? Yes
No Yes
Personal Visit Survey
(CAPI) Phase Response?
No
Figure 2.1. Data collection process for ACS monthly sample panels.
OCR for page 12
12 A Guidebook for Using American Community Survey Data for Transportation Planning
language materials are available for U.S. states, but a Spanish language version is used in Puerto
Rico, and future plans call for the general availability of a Spanish language package. Alternative
language forms will be available upon request.
Mail survey respondents are provided with a toll-free telephone number that respondents may
use if they have questions, or if they prefer to provide responses by phone. Assistance is provided
in English and Spanish.
About six weeks after the first questionnaire is mailed, the telephone data collection is begun.
The Census Bureau contracts with commercial vendors to obtain available telephone numbers
for the identified addresses. Households that have not responded but for which telephone num-
bers have been obtained are contacted by telephone interviewers. Using the Census CATI
system, interviewers from three call centers perform the ACS (using the same data collection
instrument as for the mail survey) over the phone. The CATI operation makes use of quality
assurance and training procedures being used in the best commercial calling facilities. If a
respondent refuses to participate in the CATI survey, a refusal conversion specialist calls again
and makes a second attempt to complete the interview. The CATI surveys are performed in
English and Spanish.
At the conclusion of the CATI operation (which lasts about four weeks for each sample panel),
the Census Bureau selects a subsample of remaining uninterviewed addresses for CAPI. The CAPI
subsample contains addresses categorized by their geography and whether or not they have mail-
able addresses. The different address categories are sampled at different rates, as discussed below.
Over a four-week period, Census Bureau field representatives visit CAPI subsample addresses and
at each one, verify the existence of the address, determine its occupancy status, and conduct inter-
views if possible. The field representatives collect the data using laptop computers with English and
Spanish translations. The ACS interviewers are more experienced than decennial census interview-
ers as they are continuously employed by one of the Bureau's 12 regional offices. All interviewers are
supervised by senior interviewers with three or more years of experience, and emphasis is given to
recruit bilingual staff to improve the data collection from non-English-speaking households. Unlike
for the decennial census, proxy interviews from a non-sample housing unit resident are not permit-
ted in the ACS. Proxy interviews within sample housing units are permitted.
The ACS schedule means that each monthly sample panel is collected over a three-month
period. As shown in Figure 2.2, the collection of data from the monthly sample panels overlaps
Calendar Month
Sample Panel Feb 2009 Mar 2009 Apr 2009 May 2009 Jun 2009 Jul 2009
Jan 2009 Mail Phone Visit
Feb 2009 Mail Phone Visit
Mar 2009 Mail Phone Visit
Apr 2009 Mail Phone Visit
May 2009 Mail Phone
Jun 2009 Mail
Source: David Hubble, Census Bureau Presentation at [Irvine].
Figure 2.2. Example data collection schedule for ACS monthly sample
panels.
OCR for page 13
American Community Survey 13
Table 2.2. CAPI subsampling rates for the 2005 ACS.
Address and Tract Characteristics CAPI Subsampling Rate5
Unmailable addresses and addresses in remote Alaska 66.7%
Mailable addresses in tracts with predicted levels of completed interviews 50%
prior to CAPI subsampling between 0% and 35%
Mailable addresses in tracts with predicted levels of completed interviews 40%
prior to CAPI subsampling between 35% and 50%
Mailable addresses in other tracts 33.3%
Source: United States Census Bureau, Design and Methodology: American Community Survey, Technical Paper 67
(May 2006) U.S. Government Printing Office, Washington, D.C.
Note: Percentage of addresses with uncompleted interviews prior to CAPI phase that are included in CAPI data
collection.
so that each step of the survey methodology will proceed in each month. This means that data
collection staff can work continuously on their specialty tasks.
Therefore, in February 2009, the Census Bureau's mail phase team will concentrate on the
portion of the ACS annual sample that has been assigned to January. Then, in March 2009, they
will focus on the sample assigned to February, while the phone (CATI) team works on the
January portion of the sample. In April, the mail team will turn their attention to the March sam-
ple; the phone team will work with the February sample; and the field representatives will work
with a subsample of the January sample. This process continues indefinitely.
It is important to note that only a portion of the sample households that have not participated
in the mail or telephone phases are included in the CAPI subsample. The sampling plan is
designed so that desired sample sizes are achieved without having to complete the field inter-
views with all the households that remain after the mail and phone phases of the effort. The CAPI
subsampling rates were initially established to be 1-in-3 of the mailable addresses that have not
completed the mail or telephone phases and 2-in-3 of the unmailable addresses. Based on the
initial ACS experience, the Census Bureau now applies the subsampling rates shown in Table 2.2.
The actual disposition of households in the 2001 ACS was something like that shown in
Figure 2.3. More than a quarter (28.4 percent in the figure) of the households in the original
ACS sample did not respond to the mail phase of the data collection or the telephone (CATI)
phase of the data collection, but then were never contacted as part of the personal visit phase of
the data collection. Since the data collection effort is not completed for this group, the Census
Bureau uses a weighted response rate that effectively discounts this group in the response rate
calculation.
Figure 2.4 shows the completion results by data collection mode for the 2001 ACS based on the
weighted response rate calculation method. By this definition, the weighted response rate for the
2001 ACS was almost 97 percent. The rates for subsequent years have been similarly high. It is
important to note that the Census Bureau response rates do not reflect the substantial proportion
of households for which the data collection effort is not completed. Thus, the potential for non-
response bias is higher than one would infer from the reported weighted response rates.
Data Entry and Follow-Up The returned mail surveys are sent to the Census Bureau's
processing center, checked in, and reviewed by staff to determine whether they are minimally
complete. If so, the returned survey is keyed and automatically reviewed for completeness and
internal consistency. If problems are detected, the return is subjected to the Census Bureau's
telephone edit follow-up procedures, in which respondents are contacted by phone to clarify
5The CAPI Subsampling Rate column represents the percentage of addresses with incomplete interviews prior
to CAPI phase that are included in CAPI data collection.
OCR for page 14
14 A Guidebook for Using American Community Survey Data for Transportation Planning
Completed Phone
Completed
6.6% Personal Visit
26.0%
Completed Mail
36.6%
Non-Response
2.4%
Nonfollow-up
Disposition Unknown
28.4%
Source: David Hubble, TRB Conference: Census Data for Transportation Planning, May 2005, Irvine, CA.
Figure 2.3. 2001 ACS disposition of sample.
their mailed responses. Because the decennial census process schedule cannot accommodate
this data quality review and verification, the final ACS returns are more complete and inter-
nally consistent than the census Long Form data.
Coding In the coding phase of the ACS data collection, questionnaire fields with write-in
values are coded to a prescribed list of valid values. Manual coding methods are used to assign
codes for industry and occupation, and automated coding programs are used to assign codes for
the following:
· Place of birth,
· Migration,
· Ancestry,
· Language,
· Race,
Phone
9.2%
Mail
51.1%
Personal Visit
36.4%
Non-Response
3.3%
Source: David Hubble, TRB Conference: Census Data for Transportation Planning, May 2005, Irvine, CA.
Figure 2.4. 2001 ACS completion results by data collection
mode.
OCR for page 15
American Community Survey 15
· Hispanic origin, and
· Place of work.
The most significant coding effort is the geocoding of reported work locations.
In the processing of Census 2000 data, the work location was geocoded in a two-phase oper-
ation using both the workplace address and employer name given by respondents on the Long
Form questionnaires.6 The first phase is an automated or computer-match operation. Records
not resolved during this phase moved on to a computer-assisted clerical phase.
In January 2000, the U.S. DOT and Census Bureau cosponsored a program called Work-UP
to improve the quality of the employer file used by the Census Bureau in the automated and cler-
ical coding process. In this program, local agencies (MPOs and state DOTs) used customized
GIS software to examine and update employer locations. This effort resulted in about 75 percent
of the responses being geocoded properly during the first phase.
For records not coded in the first phase, data attributes underwent an allocation process. The
allocation procedures used both trip data and job data to assign workplace locations using
"standard allocation" and "extended allocation."
Standard allocation used travel time, residence tract, means of transportation, and industry to
code work locations to a state, county, and place geocode. In addition, many records were allocated
down to the block group and traffic analysis zone (TAZ) level during the standard allocation.
The extended allocation procedure developed for use in CTPP 2000 was targeted at assigning
workplace tract and block codes to workers who could not be coded during the standard alloca-
tion process. Extended allocation was done in two stages. In the first stage, a set of potential des-
tination areas was identified for each recipient, based on trip characteristics (such as mode and
travel time) and residence location. In the second stage, the recipient was matched to a fully
geocoded donor who matches the recipient's industry and occupation characteristics and who
works in any one of the potential destination areas.
Preliminary negotiations are underway between Census Bureau and U.S. DOT on developing a
Work-UP for ACS. The extended allocation system currently is not being used for ACS (because of
cost and the insufficient number of donor records). It is expected that when five years of ACS data
are collected, the extended allocation process may be implemented. Currently, the rate of origin-
destination pairs in ACS is about 75 percent of the successfully geocoded Census 2000. Once the
above improvements are made, a better match between ACS and Census 2000 would be expected.
The coded data for each residence are recompiled and a data file is produced for editing and
weighting.
Editing and Imputation The Census Bureau's edit and allocation rules are used to account
for missing, incomplete, and contradictory responses. As for the Long Form data, the Census
Bureau has established specific rules regarding procedures for supplying values for variables that
are missing. In the ACS, the values are based on other responses provided by the respondent and
on responses for similar households. The editing and imputation procedures allocate the hous-
ing and population variables according to a predetermined hierarchy, similar to that used for the
census 2000 Long Form.
The ACS editing process begins with the determination of whether collected information con-
stitutes a usable interview. Those responses that are deemed to be non-interviews are included in
the later non-response weighting effort. For those responses that are deemed to be interviews, the
6Ed Limoges, Sabre Systems Inc., "Allocation of Missing Place of Work Data in Decennial Censuses and CTPP
2000," CTPP 2000 Status Report, January 2004.
OCR for page 16
16 A Guidebook for Using American Community Survey Data for Transportation Planning
Census Bureau staff use automated procedures to identify inconsistent and missing answers that
require imputation, or the substitution of reasonable values for missing and incorrect data items.
ACS imputation is accomplished through the use of "assignments," which are rule-based
procedures that use established relationships between different data items to fill in or correct the
missing or incorrect items, and through the use of "allocations," which are statistical procedures
(nearest neighbor methods and hot-deck methods) that use other respondents' data to infer rea-
sonable values for missing or incorrect items.
Table 2.3 shows the item imputation rates of several data items for ACS and for the decennial
census Long Form data collection. As the table shows, data item imputation rates are significantly
lower for ACS than for the decennial census, and in many cases, appear to be improving over
time. These improvements are likely the result of the superior survey design and procedures of
the ACS, compared to the decennial census.
It should be noted that although individual transportation-related items show reasonable allo-
cation rates, many of the household items, when combined with person items, show unusual
results. This is probably a result of Census Bureau processing the allocations of household items
and person items separately without any cross-referencing. Table 2.4 shows that a number of
workers from zero-vehicle households were allocated to driving alone for their commute to
work. While 16 percent of the weighted respondents from households without vehicles said they
"drove alone" to work, almost 60 percent of the workers who reported that they did not have a
vehicle and that did not report a mode to work were assigned by Census Bureau procedures to
Table 2.3. Selected ACS imputation rates of interest.
Imputation Ratesa
Description 2003 ACS 2002 ACS 2001 ACS 2000 (C2SS) Census 2000
Number of Vehicles Available 1.0 1.1 1.3 1.6 6.2
Place of Birth 6.2 4.4 4.6 634 10.1
Citizenship 0.4 0.4 0.4 0.5 0.8
b
Previous Residence
Mobility Status 2.2 2.5 2.6 4.0 6.9
Previous Residence Geography 5.9 6.0 7.3 14.9 11.0
(One or More Parts)
Employment Status Recode 3.4 3.5 3.8 6.0 10.9
Place of Work Geography 5.2 4.9 5.3 9.9 10.7
(One or More Parts)
Means of Transportation to Work 3.1 3.0 3.1 4.6 7.6
Private Vehicle Occupancy 4.1 3.9 4.1 5.8 10.0
(Carpooling)
Time Leaving Home to Go to Work 9.6 9.2 9.9 11.3 15.0
Travel Time to Work (Minutes) 7.0 6.9 7.2 8.7 11.8
Imputation Rates for Items of Branch Interest: 2003, 2002, 2001, and 2000 ACS, and Census 2000 compiled by David Hubble,
Census Bureau for the Irvine, CA presentation.
Source: Data are based on 2003, 2002, 2001, and C2SS data from the American Community Survey detailed tabulations and
the Summary File 3 from the Census 2000 detailed tabulations.
Note: Data are limited to the household population and exclude the population living in institutions, college dormitories, and
other group quarters in the ACS tabulations. However, the Census 2000 data include these persons.
a
Base to the imputation rate is the population at risk for the characteristic. For example, the imputation rate for "travel time
to work" is based on "workers 16 years and over who did not work at home."
b
Previous residence is for a one-year interval in ACS and for a five-year interval in Census.
OCR for page 17
American Community Survey 17
Table 2.4. 2002 ACS allocation rates for workers in zero-vehicle
households.
Zero Vehicles in Household
Not Allocated Allocated
Total Workers Not Allocated 5,065,639 34,425
Allocated 380,191 108,451
Drove Alone to work Not Allocated 824,431 (16.3%) 18,624 (54.1%)
Allocated 226,424 (59.6%) 55,458 (51.1%)
Source: U.S. Census Bureau American Community Survey 2002 PUMS data.
the drove-alone-to-work category. Where both data items were missing, more than half of the
respondents that were allocated to households with zero vehicles also were assigned to the drove-
alone-to-work category.
Weighting, Disclosure Editing, and Variance Estimation The coded ACS data for a calendar
year are weighted so that the combined sample units reflect the actual population as well as possible.
Weighting includes the following three adjustments:
1. Initial weights are developed to account for differences in sampling units' probabilities of
selection,
2. Initial weights for interviewed households are adjusted to account for non-interviews by
month and census tract, and
3. Weights are then adjusted to match independent housing unit and population control totals.
Among its many other activities, the Census Bureau develops annual estimates of popula-
tion by race/ethnicity, age, and sex. These are based on the previous decennial census counts
and a range of administrative records databases. These post-census estimates serve as the
weighting targets for ACS and other census surveys. Because of the way they are developed,
the post-census estimates will almost certainly be much more accurate immediately following
the decennial census. Therefore, by the end of each decade, the ACS estimates are somewhat
less likely to reflect the actual population, and it is likely that larger year-to-year differences
will be detected in the ACS data as the post-census estimates are updated with new decennial
census count data.
ACS users will need to understand that the reported changes in the ACS data for decennial
census years and the previous years are likely to be affected by larger than normal changes in the
underlying population estimates by race/ethnicity, age, and sex.
The Census Bureau population estimates for previous years are revised when new decennial
census count data become available, but the ACS estimates will not be revised. The ACS estimates
for each year will be weighted based on the initial Census Bureau population estimates.
Before releasing any ACS data, however, the Census Bureau first edits the database to ensure
it is within compliance with disclosure rules. The Census Bureau's Disclosure Review Board
(DRB) governs the release of census data as described below:
Title 13 of the United States Code authorizes the Census Bureau to conduct censuses and surveys.
Section 9 of the same Title requires that any information collected from the public under the author-
ity of Title 13 be maintained as confidential. . . . The Census Bureau's internal Disclosure Review Board
(DRB) sets the confidentiality rules for all data releases.7
7 See www.census.gov/eos/www/sestats.html.
