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 76

76 A Guidebook for Using American Community Survey Data for Transportation Planning
The opportunity to improve upon current descriptive analyses by taking advantage of more
timely data and important caveats to using ACS data for descriptive analyses in terms of data
availability are discussed in the next section.
5.2 Benefits and Limitations of ACS
for Descriptive Analyses
In discussions with transportation planners, the main perceived benefit of ACS identified was
the timeliness of the data. For descriptive analyses of larger geographies, ACS will provide ana-
lysts with more current data and pre-made Census Bureau descriptive analysis products at a
quicker rate than the decennial Long Form data collection approach.
With a Long Form data collection approach, an analyst interested in finding the percentage
of households with no vehicles in their state could obtain a year 2000 estimate beginning in
August or September of 2002. If a traditional decennial census Long Form data collection effort
were to be performed, there would be no new estimate for the state available until the summer
of 2012, when new year 2010 data became available to users. With the ACS, the analyst can
obtain this estimate for the previous year as early as late summer of the next year. These data
will be available for 2005 and beyond. If the analyst needed the data in the summer of 2011, with
the ACS they will be able to get a 2010 estimate (as well as similar estimates for 2005 through
2009, depending on the geographic size of the area under study). By comparison, with the Long
Form data collection approach they would still only have the year 2000 estimate in the summer
of 2011.
The migration to ACS, however, does raise a few challenges for data users performing descrip-
tive analyses. First, the ACS sample sizes are significantly smaller than those of the census Long
Form. While it is very common for analysts to present Long Form estimates as point estimates,
without regard to sampling error, users probably will want to show or report the larger standard
errors in the ACS estimates.
In addition, with smaller geographic areas, analysts will need to rely on data accumulated from
multiple years. For areas with populations of more than 65,000, annual ACS estimates will be
available. For areas with populations of between 20,000 and 65,000, analysts will need to rely on
three-year averages. For areas with populations of less than 20,000, including tracts and block
groups, analysts will need to rely on five-year averages.
For many descriptive analyses, the averaging across years of the estimates will not have too
much of an effect. However, for some analyses of population and household characteristics that
can vary over relatively small periods of time, it will be much more difficult for analysts to under-
stand the characteristics of interest.
The analysis of smaller geographic areas also will be complicated by the Census Bureau's
disclosure avoidance procedures. The ACS's smaller sample sizes and the Census Bureau's
stricter rules on avoiding publication of estimates where there is a possibility that an individual
can be identified, will make descriptive analyses of many small areas more difficult.
5.3 Descriptive Analysis Case Studies
The following case studies illustrate how a data user might compile descriptive analyses using
ACS data, and provide a step-by-step description of how to obtain the data, do the computa-
tions, and present the results.

OCR for page 76

Policy Planning and Other Descriptive Analyses Using ACS Data 77
For this purpose, assume that you are a transportation analyst working in a Metropolitan
Planning Organization (MPO). Your manager has asked you to compile descriptive statistics
on commuting-to-work characteristics for a county in the area served by the MPO. In the
first analysis, you will develop a change estimates profile. In the second analysis, you will
compile a ranking profile. In the third analysis, you will develop descriptive statistics for a
"state of the system" report.44 Section 3 of this guidebook provides detailed instructions on
downloading ACS data, and Section 4 describes the basic procedures that are applied in the
case studies.
5.3.1 Analysis 1: Change Estimates Profile
You have been asked to produce a profile of Lake County showing selected commuting-to-
work characteristics and how they have changed from 2002 to 2003. For presenting the results of
the change profile analysis to your MPO director, it is important to show the year-to-year differ-
ences and an assessment as to whether the change is statistically significant. It is not necessary to
include the technical details of the standard error and confidence interval computations.
You could produce some charts and graphs to help visualize the change in commuting char-
acteristics between 2002 and 2003. Figures 5.4 and 5.5 are bar charts showing the distribution of
workers by means of transportation to work and by departure time to work in 2002 and 2003,
respectively. If additional detail is needed, you can also show the actual numbers in tabular for-
mat, such as the estimates in each of the years, the percentages, the difference in percentages, and
whether the difference is statistically significant (see Table 5.4).
Percentage of Workers
100
Year 2002
90 Year 2003
80
70
60
50
40
30
20 Significant Difference
10
0
Drove Carpooled Public Motorcycle Bicycle Walked Other Worked at
Alone Transportation Means Home
Figure 5.4. Distribution of workers by means of transportation to work
in Lake County, 2002 and 2003.
44 The data used for the first two example case studies are from ACS estimates for Lake County, Illinois, for 2002
and 2003, and on fictitious hypothetical estimates for the same county. The data used for the third case study are
from ACS estimates for the San Francisco Bay Area, California, for 2000 through 2003. The Bay Area case study
is based on a poster presentation at the TRB Census Data for Transportation Planning Conference in Irvine, Cal-
ifornia, in May 2005 developed and presented by Shimon Israel of MTC.

OCR for page 76

78 A Guidebook for Using American Community Survey Data for Transportation Planning
Percentage of Workers
30
Year 2002
Year 2003
25
20
Significant
Difference
15
10
5
0
12:00 a.m. 5:30 a.m. 6:30 a.m. 7:30 a.m. 8:30 a.m. 10:00 a.m. 12:00 p.m. Worked
to 4:59 a.m. to 5:59 a.m. to 6:59 a.m. to 7:59 a.m. to 8:59 a.m. to 10:59 a.m. to 3:59 p.m. at Home
5:00 a.m. 6:00 a.m. 7:00 a.m. 8:00 a.m. 9:00 a.m. 11:00 a.m. 4:00 p.m.
to 5:29 a.m. to 6:29 a.m. to 7:29 a.m. to 8:29 a.m. to 9:59 a.m. to 11:59 a.m. to 11:59 p.m.
Figure 5.5. Distribution of workers by departure time to work in Lake
County, 2002 and 2003.
The following conclusions could accompany the graphs:
· For mode to work, except for "walk" and "other means," there were no significant changes
between 2002 and 2003 in the percentage of workers using any given transportation mode for
commuting to work. For workers who walked to work, the difference in percentages is 0.63
percent (increase), and it is a statistically significant increase not caused by sampling error.
For workers who used "other means" to work, the difference in percentages is 0.27 percent
(increase), and it also is a statistically significant increase.
· For departure time to work, except for the time period from 6:00 A.M. to 6:59 A.M., there were
no significant changes between 2002 and 2003 in the percentage of workers departing from
home to work in any given time period. The increase in the percentage of workers departing
during the time period from 6:00 A.M. to 6:29 A.M. was 1.87 percent in 2003 relative to 2002,
and this increase is statistically significant. The decrease in the percentage of workers depart-
ing from 6:30 A.M. to 6:59 A.M. was 1.31 percent in 2003 relative to 2002, and this decrease is
statistically significant.
Available Data Since Lake County has a population greater than 65,000 (according to Cen-
sus 2000, the total population of Lake County is 644,356), ACS data for Lake County will be
released annually. Since Lake County was one of the test sites during the demonstration phase
of the ACS, both the 2002 data and 2003 detailed data tables are available on the American
FactFinder website. Beginning with the full implementation of ACS, these data will be available
via the website for any county of this size. Table 5.1 shows selected commuting-to-work charac-
teristics for Lake County for years 2002 and 2003. The data files and estimates included in the
Census Bureau detailed tabulations are released with a lower bound and an upper bound corre-
sponding to the 90 percent confidence interval. The 90 percent confidence interval means that
90 times out of 100 the true value of the parameter for that area falls between the lower and upper
bounds of an estimate derived from a sample like the one taken.
Analysis Steps The first part of this discussion summarizes the steps you would need to fol-
low for analyzing the change in the percentage of workers who used public transportation to
work between 2002 and 2003. The second part provides a means to determine if this change is
statistically significant.

OCR for page 76

Policy Planning and Other Descriptive Analyses Using ACS Data 79
Table 5.1. Selected commuting-to-work characteristics for lake county, 2002 and
2003 ACS data.
2002 Data 2003 Data
Estimate Lower Bound Upper Bound Estimate Lower Bound Upper Bound
Total Population 654,067 ***** ***** 663,721 ***** *****
Total Households 222,841 221,092 224,590 226,074 224,274 227,874
Mode to Work:
Total Workers 16+ 314,647 309,478 319,816 316,525 312,408 320,642
Car, Truck, or Van: 282,426 276,882 287,970 282,407 277,642 287,172
Drove Alone 252,516 247,114 257,918 249,687 244,713 254,661
Carpooled 29,910 27,176 32,644 32,720 29,361 36,079
All Public Transportation: 13,829 12,087 15,571 13,299 11,567 15,031
Bus or Trolley Bus 1,860 1,114 2,606 2,527 1,686 3,368
Streetcar or Trolley Car 117 0 307 138 6 270
Subway or Elevated 541 241 841 776 401 1,151
Railroad 11,110 9,508 12,712 9,557 8,121 10,993
Ferryboat 39 0 103 43 0 115
Taxicab 162 0 330 258 48 468
Motorcycle 249 72 426 72 0 155
Bicycle 728 314 1,142 916 360 1,472
Walked 3,459 2,390 4,528 5,459 4,167 6,751
Other Means 1,315 804 1,827 2,176 1,567 2,785
Worked at Home 12,641 10,994 14,288 12,196 10,652 13,740
Departure Time to Work
Total Workers 16+ 314,647 309,478 319,816 316,525 312,408 320,642
Did Not Work at Home: 302,006 296,960 307,052 304,329 300,014 308,644
12:00 a.m. to 4:59 a.m. 8,499 6,950 10,048 10,439 9,008 11,870
5:00 a.m. to 5:29 a.m. 11,426 9,827 13,025 11,089 9,688 12,490
5:30 a.m. to 5:59 a.m. 17,732 15,472 19,993 16,619 14,774 18,464
6:00 a.m. to 6:29 a.m. 29,941 27,374 32,508 36,024 33,579 38,469
6:30 a.m. to 6:59 a.m. 38,229 35,096 41,362 34,311 32,047 36,575
7:00 a.m. to 7:29 a.m. 50,545 47,247 53,843 47,632 44,422 50,842
7:30 a.m. to 7:59 a.m. 37,398 34,357 40,439 37,845 35,023 40,667
8:00 a.m. to 8:29 a.m. 31,061 28,089 34,033 30,954 28,449 33,459
8:30 a.m. to 8:59 a.m. 16,582 14,818 18,346 15,951 14,473 17,429
9:00 a.m. to 9:59 a.m. 18,903 16,639 21,167 19,434 17,463 21,405
10:00 a.m. to 10:59 a.m. 7,585 6,315 8,856 7,202 5,982 8,422
11:00 a.m. to 11:59 a.m. 2,227 1,595 2,859 3,112 2,390 3,834
12:00 p.m. to 3:59 p.m. 17,191 15,223 19,159 18,379 16,159 20,599
4:00 p.m. to 11:59 p.m. 14,687 12,893 16,481 15,338 13,531 17,145
Worked at Home 12,641 10,994 14,288 12,196 10,652 13,740
Source: American FactFinder web site.
Note: An `*****' entry in the lower and upper bound columns indicates that the estimate is controlled. A statistical test is not appropriate.
In any given year i, the estimate of the proportion of workers P ^ i who used a particular mode
^
to work is equal to the estimate of the number of workers X i who used that mode divided by the
estimate of the total number of workers Y ^ , as given by the following equation:
i
^
^i = X i
P (5.1)
^i
Y
The difference in the percentages of workers who used a particular mode to work between two
years is given by
DIFF = 100% × P (
^final year - P
^initial year ) (5.2)

OCR for page 76

80 A Guidebook for Using American Community Survey Data for Transportation Planning
where P^ initial year Pfinal year and are the proportions of workers who used the given mode to work
in the initial year and final year, respectively.
For this analysis of public transportation, the proportions of workers who used public trans-
portation to work in 2002 and 2003 are
13, 829
P2002 = = 0.0440
314, 647
13, 299
P2003 = = 0.0420
316, 525
The difference in the percentage of workers who used public transportation to work is given by
(
^2003 - P
DIFF = 100% × P )
^2002 = -0.20%
Now that the difference in percentages has been determined. To know whether this difference
in percentages of workers who used a certain mode to work is statistically significant, the steps
described below should be applied. These steps are based on the documents released by the Cen-
sus Bureau on the accuracy of the data and the change profiles, and are summarized in Section
4 of this guidebook.45
Step 1--Compute the standard errors of the numerator X ^ and denominator Y ^ of the propor-
tion P^ , given their lower and upper bounds.
^ ) of an estimate X
The standard error SE(X ^ is computed as follows:
i
X ( )
^ i - LB X
^i
( )
^i =
SE X
1.65
(5.3)
where LB(X ^ ) is the lower bound of the 90 percent confidence interval for the characteristic
i
^
X i, and 1.65 is the critical value of the t-statistic associated with a 90 percent confidence interval.
For example, the standard error of the number of workers who used public transportation to
work in 2002 is
^ 2002 ) = SE(public transportation 2002 ) = 13, 829 - 12, 087 = 1, 056
SE( X
1.65
The standard error of the total number of workers in 2002 is:
^2002 ) = SE(total workers 2002 ) = 314, 647 - 309, 478 = 3,133
SE(Y
1.65
Similarly, for year 2003, these standard errors are given by:
^ 2002 ) = SE(public transportation 2003 ) = 13, 299 - 11, 567 = 1, 050
SE( X
1.65
and
^2002 ) = SE(total workers 2003 ) = 316, 525 - 312, 408 = 2, 495
SE(Y
1.65
45
See "Change Estimates" at www.census.gov/acs/www/Downloads/ACS/accuracy2002change.pdf and "Accu-
racy of the Data (2003)" at www.census.gov/acs/www/Downloads/ACS/accuracy2003.pdf.

OCR for page 76

Policy Planning and Other Descriptive Analyses Using ACS Data 81
^ of the proportion P, given the standard errors of the
Step 2--Compute the standard error SE(P)
^ is given by the following equation:
numerator and the denominator. The standard error SE(P)
^2
( )
^i = 1
SE P
^i
Y
( )
^ 2 - X SE Y
SE X i Y^i 2
^i 2
( ) (5.4)
If the value under the square root in Equation 5.4 is negative, the minus sign under the square
root is replaced by a plus sign, which results in a conservative estimate of the standard error.
In the above example, the standard error of the proportion of workers who used public trans-
portation to work in 2002 and in 2003 is given by
^2002 ) = SE(proportion public transportati
SE(P ion 2002 ) =
1 13, 8292
[1, 056]2 - [ 3,133] = 0.00333
2
314, 467 314, 647 2
^2003 ) = SE(proportion public transportati
SE(P ion 2003 ) =
1 13, 2992
[1, 050]2 - 2
[ 2, 495 ] = 0.00331
2
316, 525 316, 525
Step 3--Compute the standard error of the difference in percentages. This is given by the
following equation:
( ) ( )
2 2
SE(DIFF ) = 100% × ^ ^
SE Pfinal year + SE Pinitial year (5.5)
In the above example, the standard error of the difference in the percentages of workers who
used public transportation to work in 2002 and in 2003 is given by:
( ) ( )
2 2
SE(DIFF ) = 100% × ^ ^
SE P2003 + SE P2002
= 100% × [0.00331]2 + [0.00333]2 = 0.47%
Step 4--Compute the 90 percent margin of error of the difference in percentages. This is given
by the following equation:
ME(DIFF ) = 1.65 SE(DIFF ) (5.6)
In the above example, the 90 percent margin of error of the difference in the percentages of
workers who used public transportation to work is given by:
ME(DIFF ) = 1.65 0.47% = 0.77%
Step 5--Compute lower and upper bounds of the difference in percentages. These bounds are
given by the following equations:
LB(DIFF ) = DIFF - ME(DIFF ) (5.7)
UB(DIFF ) = DIFF + ME(DIFF ) (5.8)

OCR for page 76

82 A Guidebook for Using American Community Survey Data for Transportation Planning
In the above example, the lower and upper bounds of the 90 percent confidence interval cor-
responding to the difference in the percentages of workers who used public transportation to
work are given by:
LB(DIFF ) = -0.20% - 0.77% = -0.97%
UB(DIFF ) = -0.20% + 0.77% = 0.57%
Step 6--Determine the statistical significance of the difference in percentages according to the
following rules:
· If either the lower bound or upper bound is equal to zero, then the difference is not statisti-
cally significant;
· If the lower bound is negative and the upper bound is positive, then the difference is not
statistically significant;
· If both the lower and upper bounds have the same sign (that is both are positive or both are
negative), then the difference is statistically significant; and
· If ME(DIFF) is undetermined (e.g., due to an estimate of zero for the denominator (universe)
or if the numerator and denominator estimates are controlled), then the significance cannot
be computed.
For the above example, since the lower bound is negative (-0.97 percent) and the upper bound
is positive (0.57 percent), the change in the percentage of workers who used public transporta-
tion to work from 2002 to 2003 is not statistically significant at the 90 percent level of confidence.
The difference in the percentages is attributed to sampling error.
The computation details for the full set of commuting variables are shown below. Tables 5.2
and 5.3 show the calculations of the standard errors of the estimates, proportions, and standard
errors of the proportions for the 2002 and 2003 data, respectively.
Table 5.4 shows the calculations related to the difference between the 2002 and 2003 percent-
ages. Specifically, it shows the 2002 and 2003 percentages, difference in percentages, standard
error of the difference, margin of error of the difference, lower and upper bounds of the 90 per-
cent confidence interval, and whether the difference is statistically significant at the 90 percent
level of confidence.
5.3.2 Analysis 2: Ranking Profile
You have been asked to produce a profile of Hypothetical Lake County showing the percent-
age of zero-vehicle households in year 2010 by county subdivision and how the different subdi-
visions compare to the county average.46 Understanding how vehicle availability varies across the
county is important, for example, for determining whether transit service is adequate in areas
with higher concentrations of zero-vehicle households.
This section presents some different graphical options for presenting the results and conclu-
sions from the analysis of the percentage of zero-household vehicles. If additional detail is
needed, the actual numbers also can be shown in tabular format (as appears later in Table 5.9).
Figure 5.6 shows the ranges of the percentage of zero-vehicle households in each of the county
subdivisions, where each range is colored differently. Figure 5.7 compares the county subdivi-
46
This case study relies on synthetic data similar to what will be available from future ACS data releases. These
synthetic data are used because actual full-implementation ACS data with three- and five-year averaging are not
available at the time that this report is being written.

OCR for page 76

Policy Planning and Other Descriptive Analyses Using ACS Data 83
Table 5.2. 2002 calculations worksheet.
Given Data Calculations
Standard Standard
Lower Upper Error of Error of
Estimate Bound Bound Estimate Proportion Proportion
Total Population 654,067 ***** *****
Total Households 222,841 221,092 224,590 1,063
Mode to Work
Total Workers 16+ 314,647 309,478 319,816 3,142
Car, Truck, or Van: 282,426 276,882 287,970 3,370 0.8976 0.0059
Drove Alone 252,516 247,114 257,918 3,284 0.8025 0.0067
Carpooled 29,910 27,176 32,644 1,662 0.0951 0.0052
Public Transportation: 13,829 12,087 15,571 1,059 0.0440 0.0033
Bus or Trolley Bus 1,860 1,114 2,606 453 0.0059 0.0014
Streetcar or Trolley Car 117 0 307 71 0.0004 0.0002
Subway or Elevated 541 241 841 182 0.0017 0.0006
Railroad 11,110 9,508 12,712 974 0.0353 0.0031
Ferryboat 39 0 103 24 0.0001 0.0001
Taxicab 162 0 330 98 0.0005 0.0003
Motorcycle 249 72 426 108 0.0008 0.0003
Bicycle 728 314 1,142 252 0.0023 0.0008
Walked 3,459 2,390 4,528 650 0.0110 0.0021
Other Means 1,315 804 1,827 311 0.0042 0.0010
Worked at Home 12,641 10,994 14,288 1,001 0.0402 0.0032
Departure Time to Work
Total Workers 16+ 314,647 309,478 319,816 3,142
Did Not Work at Home: 302,006 296,960 307,052 3,067 0.9598 0.0018
12:00 a.m. to 4:59 a.m. 8,499 6,950 10,048 942 0.0270 0.0030
5:00 a.m. to 5:29 a.m. 11,426 9,827 13,025 972 0.0363 0.0031
5:30 a.m. to 5:59 a.m. 17,732 15,472 19,993 1,374 0.0564 0.0043
6:00 a.m. to 6:29 a.m. 29,941 27,374 32,508 1,560 0.0952 0.0049
6:30 a.m. to 6:59 a.m. 38,229 35,096 41,362 1,905 0.1215 0.0059
7:00 a.m. to 7:29 a.m. 50,545 47,247 53,843 2,005 0.1606 0.0062
7:30 a.m. to 7:59 a.m. 37,398 34,357 40,439 1,849 0.1189 0.0058
8:00 a.m. to 8:29 a.m. 31,061 28,089 34,033 1,807 0.0987 0.0057
8:30 a.m. to 8:59 a.m. 16,582 14,818 18,346 1,072 0.0527 0.0034
9:00 a.m. to 9:59 a.m. 18,903 16,639 21,167 1,376 0.0601 0.0043
10:00 a.m. to 10:59 a.m. 7,585 6,315 8,856 772 0.0241 0.0024
11:00 a.m. to 11:59 a.m. 2,227 1,595 2,859 384 0.0071 0.0012
12:00 p.m. to 3:59 p.m. 17,191 15,223 19,159 1,196 0.0546 0.0038
4:00 p.m. to 11:59 p.m. 14,687 12,893 16,481 1,091 0.0467 0.0034
Worked at Home 12,641 10,994 14,288 1,001 0.0402 0.0032
sions' percentage of zero-vehicle households to that of the entire county, using different colors
to show subdivisions that have a smaller rate and those that have a larger rate.
Figure 5.8 shows another method for presenting the conclusions. For every county subdi-
vision, this graph shows the estimate of the percentage of zero-vehicle households and its lower
and upper bounds. The confidence interval for the percentage of zero-vehicle households at
the county level is shown as dotted lines (lower bound, estimate, and upper bound).
It would be common to accompany these graphs with conclusions like the following:
· Waukegan has the largest percentage of zero-vehicle households (8.1 percent);
· West Deerfield has the smallest estimated percentage of zero-vehicle households (0.8 percent); and
· There are 11 county subdivisions where the percentage of zero-vehicle households is smaller
than the county average (3.1 percent), and 7 county subdivisions where the percentage of zero-
vehicle households is larger than the county average.

OCR for page 76

84 A Guidebook for Using American Community Survey Data for Transportation Planning
Table 5.3. 2003 calculations worksheet.
Given Data Calculations
Standard Standard
Lower Upper Error of Error of
Estimate Bound Bound Estimate Proportion Proportion
Total population 663,721 ***** *****
Total households 226,074 224,274 227,874 1,094
Mode to work
Total workers 16+ 316,525 312,408 320,642 2,503
Car, truck, or van: 282,407 277,642 287,172 2,897 0.8922 0.0058
Drove alone 249,687 244,713 254,661 3,024 0.7888 0.0072
Carpooled 32,720 29,361 36,079 2,042 0.1034 0.0064
Public transportation: 13,299 11,567 15,031 1,053 0.0420 0.0033
Bus or trolley bus 2,527 1,686 3,368 511 0.0080 0.0016
Streetcar or trolley car 138 6 270 80 0.0004 0.0003
Subway or elevated 776 401 1,151 228 0.0025 0.0007
Railroad 9,557 8,121 10,993 873 0.0302 0.0027
Ferryboat 43 0 115 26 0.0001 0.0001
Taxicab 258 48 468 128 0.0008 0.0004
Motorcycle 72 0 155 44 0.0002 0.0001
Bicycle 916 360 1,472 338 0.0029 0.0011
Walked 5,459 4,167 6,751 785 0.0172 0.0025
Other means 2,176 1,567 2,785 370 0.0069 0.0012
Worked at home 12,196 10,652 13,740 939 0.0385 0.0029
Departure time to work
Total workers 16+ 316,525 312,408 320,642 2,503
Did not work at home: 304,329 300,014 308,644 2,623 0.9615 0.0033
12:00 a.m. to 4:59 a.m. 10,439 9,008 11,870 870 0.0330 0.0027
5:00 a.m. to 5:29 a.m. 11,089 9,688 12,490 852 0.0350 0.0027
5:30 a.m. to 5:59 a.m. 16,619 14,774 18,464 1,122 0.0525 0.0035
6:00 a.m. to 6:29 a.m. 36,024 33,579 38,469 1,486 0.1138 0.0046
6:30 a.m. to 6:59 a.m. 34,311 32,047 36,575 1,376 0.1084 0.0043
7:00 a.m. to 7:29 a.m. 47,632 44,422 50,842 1,951 0.1505 0.0060
7:30 a.m. to 7:59 a.m. 37,845 35,023 40,667 1,716 0.1196 0.0053
8:00 a.m. to 8:29 a.m. 30,954 28,449 33,459 1,523 0.0978 0.0047
8:30 a.m. to 8:59 a.m. 15,951 14,473 17,429 898 0.0504 0.0028
9:00 a.m. to 9:59 a.m. 19,434 17,463 21,405 1,198 0.0614 0.0038
10:00 a.m. to 10:59 a.m. 7,202 5,982 8,422 742 0.0228 0.0023
11:00 a.m. to 11:59 a.m. 3,112 2,390 3,834 439 0.0098 0.0014
12:00 p.m. to 3:59 p.m. 18,379 16,159 20,599 1,350 0.0581 0.0042
4:00 p.m. to 11:59 p.m. 15,338 13,531 17,145 1,098 0.0485 0.0034
Worked at home 12,196 10,652 13,740 939 0.0385 0.0029
However, given the sampled nature of ACS data, it is more appropriate to present the results
using the confidence intervals. For instance, one could say with 90 percent confidence that
· Waukegan has the largest percentage of zero-vehicle households (7.5 percent to 8.8 percent);
· West Deerfield (0.7 percent to 0.9 percent), Ela (0.7 percent to 1.0 percent), and Newport (0.8
percent to 1.0 percent) have the smallest percentage of zero-vehicle households;
· Ten county subdivisions have a statistically significant smaller percentage of zero-vehicle
households than the overall county;
· Three county subdivisions (Grant, Waukegan, and Zion) have a percentage of zero-vehicle
households that is statistically higher than the overall county; and
· Five county subdivisions (Antioch, Benton, Moraine, Shields, and Wauconda) have percent-
ages of zero-vehicle households that are statistically the same as the county percentage.

OCR for page 76

Policy Planning and Other Descriptive Analyses Using ACS Data 85
Table 5.4. 2002-2003 change calculations worksheet.
Given Data Calculations
2002 2003 Diff: 2003 % SE ME UB Statistically
Estimate Estimate 2002 % 2003 % -2002 % (Diff) (Diff) LB (Diff) (Diff) Significant?
Total population 654,067 663,721 100% 100%
Total households 222,841 226,074 100% 100%
Mode to work
Total workers 16+ 314,647 316,525
Car, truck, or van: 282,426 282,407 89.76 89.22 -0.54 0.83 1.36 -1.90 0.82 No
Drove alone 252,516 249,687 80.25 78.88 -1.37 0.99 1.62 -2.99 0.25 No
Carpooled 29,910 32,720 9.51 10.34 0.83 0.82 1.36 -0.52 2.19 No
Public transportation: 13,829 13,299 4.40 4.20 -0.20 0.47 0.77 -0.97 0.57 No
Bus or trolley bus 1,860 2,527 0.59 0.80 0.21 0.22 0.36 -0.15 0.56 No
Streetcar or trolley car 117 138 0.04 0.04 0.01 0.03 0.06 -0.05 0.06 No
Subway or elevated 541 776 0.17 0.25 0.07 0.09 0.15 -0.08 0.23 No
Railroad 11,110 9,557 3.53 3.02 -0.51 0.41 0.68 -1.19 0.17 No
Ferryboat 39 43 0.01 0.01 0.00 0.01 0.02 -0.02 0.02 No
Taxicab 162 258 0.05 0.08 0.03 0.05 0.08 -0.05 0.11 No
Motorcycle 249 72 0.08 0.02 -0.06 0.04 0.06 -0.12 0.00 No
Bicycle 728 916 0.23 0.29 0.06 0.13 0.22 -0.16 0.28 No
Walked 3,459 5,459 1.10 1.72 0.63 0.32 0.53 0.10 1.16 Yes
Other means 1,315 2,176 0.42 0.69 0.27 0.15 0.25 0.02 0.52 Yes
Worked at home 12,641 12,196 4.02 3.85 -0.16 0.43 0.71 -0.88 0.55 No
Departure time to work
Total workers 16+ 314,647 316,525
Did not work at home: 302,006 304,329 95.98 96.15 0.16 0.37 0.62 -0.45 0.78 No
12:00 a.m. to 4:59 a.m. 8,499 10,439 2.70 3.30 0.60 0.40 0.67 -0.07 1.26 No
5:00 a.m. to 5:29 a.m. 11,426 11,089 3.63 3.50 -0.13 0.41 0.67 -0.80 0.54 No
5:30 a.m. to 5:59 a.m. 17,732 16,619 5.64 5.25 -0.39 0.56 0.92 -1.30 0.53 No
6:00 a.m. to 6:29 a.m. 29,941 36,024 9.52 11.38 1.87 0.67 1.10 0.76 2.97 Yes
6:30 a.m. to 6:59 a.m. 38,229 34,311 12.15 10.84 -1.31 0.73 1.20 -2.51 -0.11 Yes
7:00 a.m. to 7:29 a.m. 50,545 47,632 16.06 15.05 -1.02 0.86 1.42 -2.44 0.41 No
7:30 a.m. to 7:59 a.m. 37,398 37,845 11.89 11.96 0.07 0.78 1.29 -1.22 1.36 No
8:00 a.m. to 8:29 a.m. 31,061 30,954 9.87 9.78 -0.09 0.74 1.21 -1.31 1.12 No
8:30 a.m. to 8:59 a.m. 16,582 15,951 5.27 5.04 -0.23 0.44 0.72 -0.95 0.49 No
9:00 a.m. to 9:59 a.m. 18,903 19,434 6.01 6.14 0.13 0.57 0.94 -0.81 1.08 No
10:00 a.m. to 10:59 a.m. 7,585 7,202 2.41 2.28 -0.14 0.34 0.56 -0.69 0.42 No
11:00 a.m. to 11:59 a.m. 2,227 3,112 0.71 0.98 0.28 0.18 0.30 -0.03 0.58 No
12:00 p.m. to 3:59 p.m. 17,191 18,379 5.46 5.81 0.34 0.57 0.93 -0.59 1.28 No
4:00 p.m. to 11:59 p.m. 14,687 15,338 4.67 4.85 0.18 0.49 0.80 -0.62 0.98 No
Worked at home 12,641 12,196 4.02 3.85 -0.16 0.43 0.71 -0.88 0.55 No
Available Data Table 5.5 shows synthetic ACS estimates of population for each of the county
subdivisions, as well as for the entire Hypothetical Lake County in each of the years 2005 to 2009
(although these annual population estimates are not released for each county subdivision, they are
shown in this table to determine which types of ACS estimates are released in a given year). The
Census Bureau releases the following types of estimates based on the population of a given area:
· For areas with population greater than 65,000, annual estimates, as well as three- and five-year
average estimates are available;
· For areas with population between 20,000 and 65,000, three- and five-year average estimates
are available; and
· For areas with population less than 20,000, only five-year average estimates are available.

OCR for page 76

86 A Guidebook for Using American Community Survey Data for Transportation Planning
Figure 5.6. Percentage of zero-vehicle households by county subdivision
based on the point estimates for Hypothetical Lake County.
Figure 5.7. Percentage of zero-vehicle households by county subdivision
as compared to the county average based on the point estimates for
Hypothetical Lake County.

OCR for page 76

90 A Guidebook for Using American Community Survey Data for Transportation Planning
Table 5.7. Three-Year average estimate (2007-2009), ACS data for
Hypothetical Lake County.
Total Zero-
Occupied Lower Upper Vehicle Lower Upper
County Subdivision Households Bound Bound Households Bound Bound
Antioch Township 9,773 9,046 10,500 291 244 338
Avon Township 19,812 18,687 20,937 485 412 558
Benton Township
Cuba Township
Ela Township 14,654 13,712 15,596 116 95 137
Fremont Township 9,222 8,522 9,922 94 76 112
Grant Township
Lake Villa Township 13,199 12,316 14,082 179 148 210
Libertyville Township 19,794 18,670 20,918 493 419 567
Moraine Township 14,209 13,284 15,134 470 399 541
Newport Township
Shields Township 11,730 10,910 12,550 343 289 397
Vernon Township 25,694 24,402 26,986 259 216 302
Warren Township 24,984 23,710 26,258 596 510 682
Wauconda Township
Waukegan Township 31,727 30,299 33,155 2,444 2,177 2,711
West Deerfield Township 12,237 11,395 13,079 92 74 110
Zion Township 8,399 7,742 9,056 490 417 563
County Total 245,745 244,910 246,580 7,129 6,542 7,716
Table 5.8. Five-year average estimate (2005-2009), ACS data for
Hypothetical Lake County.
Total Zero-
Occupied Lower Upper Vehicle Lower Upper
County Subdivision Households Bound Bound Households Bound Bound
Antioch Township 9,613 9,056 10,170 304 266 342
Avon Township 19,487 18,624 20,350 507 449 565
Benton Township 7,036 6,586 7,486 192 166 218
Cuba Township 6,709 6,273 7,145 87 74 100
Ela Township 14,388 13,666 15,110 121 104 138
Fremont Township 9,064 8,528 9,600 98 84 112
Grant Township 7,614 7,138 8,090 284 248 320
Lake Villa Township 12,969 12,292 13,646 187 162 212
Libertyville Township 19,475 18,612 20,338 515 456 574
Moraine Township 13,996 13,286 14,706 490 433 547
Newport Township 1,706 1,550 1,862 16 13 19
Shields Township 11,549 10,920 12,178 358 314 402
Vernon Township 25,265 24,273 26,257 270 236 304
Warren Township 24,586 23,607 25,565 622 553 691
Wauconda Township 6,729 6,292 7,166 230 200 260
Waukegan Township 31,314 30,214 32,414 2,551 2,337 2,765
West Deerfield Township 12,025 11,380 12,670 96 82 110
Zion Township 8,285 7,781 8,789 512 453 571
County Total 241,810 241,147 242,473 7,440 6,972 7,908

OCR for page 76

Policy Planning and Other Descriptive Analyses Using ACS Data 91
Table 5.9. Proportion and standard error computations for the example.
Total Zero- SE(Total SE(Zero- Percentage of Zero-Vehicle Households
Occupied Vehicle Occupied Vehicle Estimate
County Subdivision HHs HHs HHs) HHs) (%) SE (%) LB (%) UB (%)
Antioch Township 9,613 304 339 23 3.2 0.2 2.8 3.5
Avon Township 19,487 507 523 35 2.6 0.2 2.3 2.9
Benton Township 7,036 192 273 16 2.7 0.2 2.4 3.1
Cuba Township 6,709 87 264 8 1.3 0.1 1.1 1.5
Ela Township 14,388 121 438 10 0.8 0.1 0.7 0.9
Fremont Township 9,064 98 325 8 1.1 0.1 1.0 1.2
Grant Township 7,614 284 288 22 3.7 0.2 3.3 4.1
Lake Villa Township 12,969 187 410 15 1.4 0.1 1.3 1.6
Libertyville Township 19,475 515 523 36 2.6 0.2 2.4 2.9
Moraine Township 13,996 490 430 35 3.5 0.2 3.1 3.9
Newport Township 1,706 16 95 2 0.9 0.1 0.8 1.1
Shields Township 11,549 358 381 27 3.1 0.2 2.8 3.4
Vernon Township 25,265 270 601 21 1.1 0.1 0.9 1.2
Warren Township 24,586 622 593 42 2.5 0.2 2.3 2.8
Wauconda Township 6,729 230 265 18 3.4 0.2 3.0 3.8
Waukegan Township 31,314 2,551 667 130 8.1 0.4 7.5 8.8
West Deerfield Township 12,025 96 391 8 0.8 0.1 0.7 0.9
Zion Township 8,285 512 305 36 6.2 0.4 5.6 6.8
County Total 241,810 7,440 402 284 3.1 0.1 2.9 3.3
The percentage of zero-vehicle households for the entire Hypothetical Lake County is 3.1 per-
cent. Table 5.9 shows the percentage of zero-vehicle households for all county subdivisions.
The percentages of zero-vehicle households computed as shown above are point estimates.
They can be compared across county subdivisions as well as with respect to the overall county
percentage of zero-vehicle households. In addition to examining the point estimates, it also is
important to examine the standard errors of these estimates to see whether the conclusions are
significantly altered.
First, the standard errors of total occupied households and zero-vehicle households are com-
puted given the estimates and their lower and upper bounds, using Equation 5.3. For example,
for Antioch, the standard error of total occupied households is
(9,613 - 9,056)/1.65 = 338
The standard error of zero-vehicle households is
(304 - 266)/1.65 = 23
Second, the standard error of the percentage of zero-vehicle households is computed using
Equation 5.4. For example, for Antioch, the standard error of the percentage of zero-vehicle
households is
1 304 2
[ 23]2 - [339]2 = 0.00212 = 0.2%
9, 613 9, 6132

OCR for page 76

92 A Guidebook for Using American Community Survey Data for Transportation Planning
Third, given the estimate of the percentage of zero-vehicle households and its standard error,
the lower and upper bounds of the 90 percent confidence interval are computed as follows:
LB = Estimate - 1.65 SE(Estimate) (5.9)
LB = Estimate + 1.65 SE(Estimate) (5.10)
For example, for Antioch, the lower bound of the 90 percent confidence interval is
(3.2 0.2 1.65) = 2.8 %
The upper bound is
(3.2 + 0.2 1.65) = 3.5 %
Computations of standard errors and confidence intervals are shown in Table 5.9. For each
county subdivision, the 90 percent confidence interval means that 90 times out of 100 the true
value of the percentage of zero-vehicle households for that area falls between the lower and upper
bounds of an estimate derived from a sample like the one taken.
Once the percentages and standard errors of the percentages are calculated, the differences
between individual subdivisions and the county as a whole can be calculated and compared using
the procedures previously described. The differences in the estimates are calculated directly.
Therefore, for Antioch the difference between the subdivision and county estimate is 3.2 - 3.1 =
0.1 percent.
The standard error of the difference can be calculated using a variant of Equation 5.5 from the
first analysis.
2
SE(DIFF ) = 100% ^ + ^
SE(Pcounty ) SE(Ptownship ) (5.11)
The 90 percent confidence-level margin of error of the difference is
ME(DIFF ) = 1.65 SE(DIFF ) (5.12)
The upper and lower bounds of the difference in percentages are as follows:
LB(DIFF ) = DIFF - ME(DIFF ) (5.13)
UB(DIFF ) = DIFF + ME(DIFF ) (5.14)
For Antioch, the results of these calculations are
SE(DIFF) = 0.2 percent,
ME(DIFF) = 0.4 percent,
LB(DIFF) = -0.3 percent, and
UB(DIFF) = 0.5 percent.
Table 5.10 shows the results of these calculations for each of the subdivisions.
An interesting finding of this analysis is that one county subdivision, Libertyville, is found to
be statistically different from the county average despite the fact that when one compares the esti-
mates, lower bounds, and upper bounds (for instance, see Figure 5.8), the margins of error of that
county subdivision and the county overlap. In order to correctly assess the statistical significance
of the differences, it is necessary to calculate the standard errors of the differences, rather than to
simply inspect the estimates, standard errors, and margins of error of the variable of interest.

OCR for page 76

Policy Planning and Other Descriptive Analyses Using ACS Data 93
Table 5.10. Statistical difference computation results.
Difference Between Township Statistically
and County Zero-Vehicle SE ME LB UB Significant
County Subdivision Households Percentages (Diff) (Diff) (Diff) (Diff) Difference?
Antioch Township 0.1% 0.24% 0.40% -0.3% 0.5% No
Avon Township -0.5% 0.20% 0.33% -0.8% -0.1% Yes
Benton Township -0.3% 0.23% 0.38% -0.7% 0.0% No
Cuba Township -1.8% 0.16% 0.26% -2.0% -1.5% Yes
Ela Township -2.2% 0.13% 0.22% -2.5% -2.0% Yes
Fremont Township -2.0% 0.14% 0.23% -2.2% -1.8% Yes
Grant Township 0.7% 0.27% 0.46% 0.2% 1.1% Yes
Lake Villa Township -1.6% 0.16% 0.26% -1.9% -1.4% Yes
Libertyville Township -0.4% 0.20% 0.34% -0.8% -0.1% Yes
Moraine Township 0.4% 0.25% 0.42% 0.0% 0.8% No
Newport Township -2.1% 0.16% 0.26% -2.4% -1.9% Yes
Shields Township 0.0% 0.24% 0.40% -0.4% 0.4% No
Vernon Township -2.0% 0.14% 0.23% -2.2% -1.8% Yes
Warren Township -0.5% 0.20% 0.33% -0.9% -0.2% Yes
Wauconda Township 0.3% 0.26% 0.43% -0.1% 0.8% No
Waukegan Township 5.1% 0.40% 0.65% 4.4% 5.7% Yes
West Deerfield Township -2.3% 0.13% 0.22% -2.5% -2.1% Yes
Zion Township 3.1% 0.39% 0.64% 2.5% 3.7% Yes
5.3.3 Analysis 3: Monitoring the State of the System
You have been asked to compile various descriptive statistics related to commuting-to-work
characteristics and vehicle ownership to develop a "state of the system" report for the Bay Area
for the years 2000 to 2003. You were asked to use any available data sources and to track changes
over time where data are available.
This section describes different options that might be used for presenting the descriptive sta-
tistics to policymakers. First, one can show some important transportation variables such as
commuting mode shares by various means of transportation, percentage of zero-vehicle house-
holds, and average commute time. Table 5.11 shows a summary of these statistics using Census
2000 data and 2000 to 2003 ACS data.
Second, one can show some of these statistics graphically along with the confidence intervals
for the ACS estimates. This is shown in Figure 5.10 for the number of public transportation com-
muters, along with information on the statistical significance of the difference estimates. Some
conclusions that can be drawn from this analysis include
· Statistically, the 2000 ACS estimate is significantly larger than the Census 2000 estimate;
· Statistically, the 2001 ACS estimate is significantly smaller than the 2000 ACS estimate; and
· Statistically, the 2002 ACS estimate is not significantly different from the 2001 ACS estimate,
and statistically the 2003 ACS estimate is not significantly different from the 2002 ACS estimate.
Third, one could compare trends from various data sources. For example, Figure 5.11 shows
the change in the number of employed civilians over time using Census 2000, 2000-2003 ACS,
and 2000-2003 Bureau of Labor Statistics-Local Area Unemployment Statistics (BLS-LAUS)

OCR for page 76

94 A Guidebook for Using American Community Survey Data for Transportation Planning
Table 5.11. Important transportation variables for the annual MTC San
Francisco Bay area "state of the system" report.
Census American Community Survey
2000 2000 2001 2002 2003
Commute Share by:
Public Transportation 9.7% 10.7% 10.0% 9.6% 9.4%
Bicycle 1.1% 1.1% 1.0% 1.0% 1.0%
Walk 3.2% 2.8% 2.9% 3.1% 3.1%
Drive Alone 68.0% 67.5% 69.2% 69.9% 69.2%
Carpool 12.9% 12.9% 11.3% 11.1% 11.5%
Worked at Home 4.0% 4.0% 4.4% 4.4% 4.7%
Percent Zero-Vehicle Households 10.0% 9.0% 9.2% 8.9% 8.6%
Average Commute Time (Minutes) 29.4 28.5 27.7 27.5 26.7
data. The ACS and BLS-LAUS show a trend of decrease in employed civilians over this time-
frame, but the BLS estimates are larger than the ACS estimates. In the analysis of ACS data, it will
be important for analysts to perform validity checks using other available data sources whenever
possible.
Available Data This section describes the data that were available for this analysis.
Figure 5.12 shows the distribution of places in the Bay Area by population size. This has impli-
cations for the types of ACS data that will be available for each of these areas. Only five-year
average ACS data will be available for 14 percent of places, three- and five-year average ACS data
will be available for 24 percent of places, and annual estimates, three- and five-year average ACS
data will be available for 62 percent of places. All estimates used in this case study use annual
estimates.
Table 5.12 shows the distribution of number of workers by means of transportation to work
using Census 2000 data and 2000-2003 annual ACS data.
Commuters (in Thousands)
450
400
350
300
250
200
150
100
50
0
Census 2000 ACS 2000 ACS 2001 ACS 2002 ACS 2003
Survey Year
Statistically Significant Increase [ACS Relative to Census 2000]
Statistically Significant Decrease
Figure 5.10. Total commuters on public transportation, Census 2000 and
ACS 2000-2003.

OCR for page 76

Policy Planning and Other Descriptive Analyses Using ACS Data 95
Employed Civilians (in Millions)
3.65
BLS-LAUS BLS, % Change, 2000-2003 = -4.7%
ACS, % Change, 2000-2003 = -7.4%
3.60
3.55 175,300
3.50
266,000
ACS
3.45
3.40
145,700
3.35
Census 2000
70,800
3.30
3.25
3.20
2000 2001 2002 2003
Year
Figure 5.11 Employed civilians, Census 2000, ACS, and BLS-LAUS.
Analysis Steps This section describes the computations that were performed to reach the
conclusions presented earlier. First described are the methods for working with confidence inter-
vals when the analysis involves a geography for which ACS data are not directly available. Fol-
lowing this discussion is a description of how to compute the statistical significance of the dif-
ference between two estimates.
When working with ACS confidence intervals, one should note the following two important
rules of thumb:
· The standard error is larger, and confidence intervals are wider (as a percentage of the esti-
mate), for geographic areas with smaller populations and for characteristics that occur less
frequently. For example, the estimate for Bay Area bicycle commuters (a relatively small
percentage of total commuters) has a confidence interval that is proportionately wider than
that for carpool (2+) commuters.
Places <20,000 +
Remainder of County
14% Places 65,000+
62%
Places
20,000-65,000
24%
Figure 5.12. Bay area population for ACS reporting.

OCR for page 76

96 A Guidebook for Using American Community Survey Data for Transportation Planning
Table 5.12. Means of transportation to work, Census 2000 and ACS.
Census American Community Survey
2000 2000 2001 2002 2003
Total: 3,306,100 3,337,500 3,236,400 3,203,700 3,186,900
Car, Truck, or Van: 2,674,600 2,683,800 2,604,100 2,596,300 2,571,300
Drove Alone 2,248,100 2,252,300 2,239,900 2,240,700 2,205,800
Carpooled 426,500 431,500 364,200 355,600 365,500
Public Transportation: 321,100 357,700 324,300 306,100 300,800
Bus or Trolley Bus 178,900 199,400 180,000 178,400 176,900
Streetcar or Trolley Car 14,300 14,300 15,100 12,800 10,400
Subway or Elevated 98,700 107,500 94,800 88,100 85,300
Railroad 20,100 24,600 21,900 19,000 20,400
Ferryboat 5,800 6,700 7,900 4,100 6,100
Taxicab 3,300 5,100 4,500 3,600 1,900
Motorcycle 11,900 13,700 13,800 9,400 9,400
Bicycle 36,000 36,800 31,300 33,000 31,000
Walked 106,100 92,200 92,900 100,800 100,100
Other Means 23,700 19,000 27,600 15,800 23,900
Worked at Home 132,700 134,400 142,500 142,300 150,300
· Performing fewer data calculations generally produces the tightest confidence intervals.
For the Metropolitan Transportation Commission (MTC), error reporting is typically
better if its 9-county ACS estimates are derived by subtracting Santa Cruz PMSA from the
10-county CMSA (fewer calculations) than by summing its 9 counties or 5 PMSAs (more
calculations).
The following example demonstrates how the estimates and standard errors for different
geographic areas may be combined to analyze custom geographies. In this example, we derive
the ACS 2003 estimate and confidence interval for MTC's planning jurisdiction.
MTC's metropolitan planning jurisdiction, the nine-county San Francisco Bay Area, does not
have a single census-equivalent geography for which published ACS datasets are available.
Instead, MTC must derive study data by
· Summing ACS estimates for its nine constituent counties;
· Summing ACS estimates for its five constituent PMSAs; or
· Subtracting ACS estimates for one PMSA (Santa Cruz PMSA) from the ACS estimates for the
San Francisco CMSA (which is composed of 10 counties, or equivalently, 6 PMSAs).
To accomplish any of these tasks, we must derive new standard errors and confidence inter-
vals based on those provided in the available ACS dataset. Table 5.13 shows some of the relevant
PMSA and CMSA data that are available from the American FactFinder website.
With these data, we can develop estimates for the MTC region by either subtracting the last
row estimates (Santa Cruz PMSA) from the CMSA total or by summing the other five PMSA
estimates.
To combine geographies, the estimates may be added or subtracted directly, but we also need
to account for the confidence intervals by calculating combined standard errors from the com-
ponent standard errors. The following four analysis steps are required:
1. Calculate the combined estimate by adding or subtracting the component geography estimates,
2. Calculate standard errors for the component geography estimates,
3. Calculate the standard errors of the estimates for the combined geography, and

OCR for page 76

Policy Planning and Other Descriptive Analyses Using ACS Data 97
Table 5.13. Estimated number of carpool and bicycle commuters for
selected Bay Area geographic areas from 2003 ACS.
Carpool Commuters Bay Area Bicycle Commuters
Lower Upper Lower Upper
Geographic Area Estimate Bound Bound Estimate Bound Bound
San Francisco CMSA 373,018 353,631 392,405 34,561 28,350 40,772
Oakland PMSA 148,912 134,587 163,237 11,635 7,980 15,290
San Francisco PMSA 75,418 67,238 83,598 10,149 8,007 12,291
San Jose PMSA 79,052 68,397 89,707 6,743 2,826 10,660
Santa Rosa PMSA 24,377 18,945 29,809 1,795 372 3,218
Vallejo PMSA 37,723 31,846 43,600 723 29 1,417
Santa Cruz PMSA (outside MTC area) 7,536 5,062 10,010 3,516 1,183 5,849
Source: 2003 ACS Base Table P047: Means of Transportation To Work for Workers 16 Years and Over.
4. Convert the combined geography standard errors to margins of error.
This process shown below is for the carpool estimate.
Step 1--Calculate combined estimates.
For the first approach, in which the non-MTC PMSA estimate is subtracted from the CMSA
estimate, the combined estimate is:
Carpool CommutersMTC Area = 373,018 - 7,536 = 365,482
For the second approach, in which the five MTC PMSAs are summed, the combined estimate is:
Carpool CommutersMTC Area = 148,912 + 75,418 + 79,052 + 24,377 + 37,723 = 365,482
Step 2--Calculate component geography standard errors.
The standard error calculations for each component geography are similar to the previous
translations from census margins of error to standard errors:
Standard error = 90 percent confidence margin of error/1.65
Margin-of-error = max(upper bound estimate, estimate lower bound)
Hence, the standard errors for the carpool estimates are those shown in Table 5.14.
Table 5.14. Standard error calculations for geographic areas that
comprise the MTC study area.
Carpool Commuters Standard Error Calculation
Lower Upper Critical Value 90%
Geographic Area Estimate Bound Bound ME Confidence SE
San Francisco CMSA 373,018 353,631 392,405 19,387 1.65 11,750
Oakland PMSA 148,912 134,587 163,237 14,325 1.65 8,682
San Francisco PMSA 75,418 67,238 83,598 8,180 1.65 4,958
San Jose PMSA 79,052 68,397 89,707 10,655 1.65 6,458
Santa Rosa PMSA 24,377 18,945 29,809 5,432 1.65 3,292
Vallejo PMSA 37,723 31,846 43,600 5,877 1.65 3,562
Santa Cruz PMSA (outside MTC area) 7,536 5,062 10,010 2,474 1.65 1,499

OCR for page 76

98 A Guidebook for Using American Community Survey Data for Transportation Planning
Step 3--Calculate the combined standard errors.
The standard error of a sum or difference can be calculated as
^ +Y
SE( X ^ ) 2 + SE(Y
^ ) = SE( X ^ ) 2
Applying this equation to the two alternative approaches for developing MTC area-specific
estimates, provides the following estimates. For the first approach, in which the non-MTC PMSA
estimate is subtracted from the CMSA estimate, the combined standard error is
SE MTC = [11, 750]2 + [1, 499]2 = 11, 845
For the second approach, in which the five MTC PMSAs are summed, the combined standard
error is:
SE MTC = [8, 682]2 + [ 4, 958]2 [6, 458]2 + [3, 292]2 + [3,5
562 ] = 12, 852
2
Step 4--Calculate the margins of error.
The margin of error for the 90 percent confidence level is
ME MTC = 1.65 SE MTC
Therefore, for the first approach, in which the non-MTC PMSA estimate is subtracted from
the CMSA estimate, the combined estimate is:
Carpool CommutersMTC Area = 365,482 ± 19,544
For the second approach, in which the five MTC PMSAs are summed, the combined estimate
is
Carpool CommutersMTC Area = 365,482 ± 21,206
Although the two approaches have the same central point estimate, the first approach that
combines only two ACS estimates provides a more precise estimate than the second approach
where five separate estimates are combined.
If ACS data were published at the MPO level, the need to combine geographic areas like this
would be obviated, but it is likely that many transportation planners will need to create custom
geographic combinations, so the example will probably remain useful.
Figures 5.13 and 5.14 show the resulting confidence intervals for carpool commuters and bicy-
cle commuters, respectively, using the two methods mentioned above for deriving the MPO-
level 2003 ACS data estimates.
This section shows an example of how year-to-year statistical significance computations, such
as those shown in Figure 5.10, can be accomplished. The standard errors of the individual ACS
estimates are computed using Equation 5.3. For example, the standard errors of the 2000 and
2001 ACS estimates are
357, 661 - 338, 774
SE ( ACS2000 ) = = 11, 481
1.645
324, 287 - 308, 877
SE ( ACS2001 ) = = 9, 368
1.645

OCR for page 76

Policy Planning and Other Descriptive Analyses Using ACS Data 99
2+ Commuters (in Thousands)
450
400 90% Confidence Interval
350
300
250
200
150 365,482 365,482
+ 19,544 + 21,206
100
50
0
San Francisco CMSA less Santa Cruz PMSA Sum of 5 PMSAs
Method for Deriving MPO Data from Census Geography
Figure 5.13. Bay Area carpool (2+) commuters, with confidence intervals,
ACS 2003.
The standard error of the difference in estimates is computed using the following equation:
SE(DIFF ) = SE X1 ( )
^ 2 + SE X
^ 2 2
( ) (5.15)
^1 and X
where X ^2 are the estimates used to compute the difference.
For example, the standard error of the difference between the ACS 2000 and 2001 estimates is
SE ( DIFF2001-2000 ) = (11, 481)2 + (9, 368)2 = 14, 818
The margin of error of the difference is computed using Equation 5.6 as follows:
ME ( DIFF2001-2002 ) = 1.645 × 14, 818 = 24, 376
Bicycle Commuters (in Thousands)
40
35 90% Confidence Interval
30
25
20
15 31,045 31,045
10 + 6,635 + 5,983
5
0
San Francisco CMSA less Santa Cruz PMSA Sum of 5 PMSAs
Method for Deriving MPO Data from Census Geography
Figure 5.14. Bay Area bicycle commuters, with confidence intervals, ACS
2003.

OCR for page 76

100 A Guidebook for Using American Community Survey Data for Transportation Planning
The lower and upper bounds of the difference are computed using Equations 5.7 and 5.8 as
follows:
LB ( DIFF2001-2000 ) = (324, 287 - 357, 661) - 24, 376 = -57, 750
UB ( DIFF2001-2000 ) = (324, 287 - 357, 661) + 24, 376 = -8, 998
Since both the lower bound and upper bound of the difference have the same sign, the differ-
ence between the 2000 and 2001 estimates is statistically significant.
These calculations can be performed for each successive year to track year-to-year changes or
they can be performed on non-consecutive years to measure whether accumulated differences
are statistically significant.
5.3.4 Conclusions from these Analysis Case Studies
Just as with the decennial census Long Form dataset, there are many different types of
descriptive analyses that analysts can produce using ACS data. These case studies demonstrate
how to produce change profiles and ranking profiles for a county and its subdivisions, create
descriptive statistics for compiling a state-of-the-system report for a multicounty area, and
interpret the results in light of the lower and upper bounds of the 90 percent confidence inter-
val that are released with the data.
One of the clear advantages of ACS data with respect to census data is the timeliness of the data.
For example, the first case study example that was described demonstrates that the availability of
ACS data in years 2002 and 2003 for the county in question enabled the analyst to determine the
change in a given characteristic between the two years. With census Long Form data, such an
analysis would be based on data points that correspond to a difference of at least 10 years. The sec-
ond case study also emphasizes the value of having ACS data in years just prior to the decennial
census year (i.e., using 2005-2009 ACS data as opposed to using Census 2000 data for an analysis
conducted in year 2010). The third case study also shows that ACS data will be important in the
identification of key annual trends in transportation-related variables, and in supporting agen-
cies' efforts to advocate for the transportation needs of the elderly, disabled, low-income, and
youth populations. In the absence of ACS data, agencies would have to rely on the decennial cen-
sus, and on non-census releases for intercensal years, which would sometimes produce data dis-
crepancies.
As the second case study shows, one of the ACS analysis challenges is the need to deal with the
varying availability of estimates. When comparing estimates across different geographic areas
where multiple types of estimates are available (annual, as well as three-and five-year moving aver-
age), it is likely that users will need to use the same type of estimate for all geographic areas to main-
tain consistency and to avoid the bias resulting from the different periods of data accumulation.
Some further issues that users of the data should be aware of when doing similar types of
analyses are the following:
· Annual estimates of change (Case Study 1) cannot be computed in cases where only multiyear
average data are available if the multiyear average estimates include data from overlapping years.
This is because when standard statistical procedures are used to test for significant differences
between estimates over time, it is assumed that the two estimates are drawn from independent
samples, an assumption that is violated in the case of two overlapping multiyear averages.
· The averaging of estimates over three or five years increases the survey sample sizes from
which the estimates are derived, and thus reduces the sampling error and the size of the
statistical confidence intervals. However, these statistics do not account for any bias that may