Measuring And Adjusting For Frame Undercoverage Of The State And ...
Measuring and Adjusting for Frame Undercoverage of the State and
Local Value Put-in-Place (VIP) Survey
T. Trang Nguyen1 and Shadana R. Myers
U.S. Census Bureau
4600 Silver Hill Rd., Washington, DC 20233 / thuy.trang.ta.nguyen@census.gov and shadana.r.myers@census.gov
1. INTRODUCTION
The U.S. Census Bureau conducts the monthly State and Local (S&L) VIP survey to measure the value of construction put in
place for building and nonbuilding structures owned by S&L governments. The Bureau also collects fiscal year data on
similar construction in the Annual Survey of State and Local Government Finances (ASGF). Conceptually, these estimates
should be comparable on a fiscal basis; nevertheless, they have continued to differ during the past decades. Figure 1 presents
a time series plot of the total construction expenditures for S&L VIP and ASGF. The S&L VIP estimates are consistently
lower than the ASGF estimates. The major difference is attributed to the undercoverage of the S&L VIP frame as reported
by Luery, Asanuma, Vu, McDonald, and Newman-Smith (1992).
Figure 1: Time Series Plot of Total Construction Expenditures for S&L VIP and ASGF
(In Millions)
250000
200000
150000
100000
ASGF Estimates
50000
VIP Estimates
Weighted Expenditures
0
1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004
Year (May - April)
To adjust for the deficiencies in the S&L VIP frame, the VIP estimates were increased by five percent then further
benchmarked to the ASGF estimates. The existing undercoverage adjustment factor of 1.05 was no longer current. It needed
to be updated because its development was unknown and the 18% overall undercoverage rate derived from the study by
Luery et al. (1992) was not reflected in this factor. The benchmarking of the VIP to the ASGF was originally intended to be
a temporary solution (Luery et al. 1992). Benchmarking of the VIP to the ASGF needed to be discontinued because of the
crucial timing and definitional differences between the two sets of estimates:
Forecasting – ASGF estimates are annual and lag the monthly VIP estimates by more than two years. Thus, we
forecasted the ASGF estimates two years ahead and then developed the benchmarking factors based on these
forecasts. Table 1 shows the percent change in the monthly benchmarking factors of current year forecasts to the
revised forecasts from the previous year. Forecasting led to large revisions for most types of construction (TC) and
1 This report is released to inform interested parties of research and to encourage discussion. Any views expressed on
statistical, methodological, technical, or operational issues are those of the authors and not necessarily those of the U.S.
Census Bureau.
the revisions varied from year to year. Refer to U.S. Census Bureau (1990) for the development of the forecast
models used in the benchmarking.
Classification – ASGF estimates are functional type (for example, a highway administration builds an office
building, this structure would be classified as “highway”), where VIP estimates are project based (same example as
above but now the structure would be classified as “office building”). Assumptions and adjustments were made in
order to align the VIP and ASGF estimates when developing the benchmarking factors for the monthly S&L VIP
estimates.
Reporting period – the fiscal year reporting of the ASGF is not always within the VIP data time frame (July through
June) since the fiscal year definition varies across the S&L government agencies. A different time configuration of
the VIP estimates was used when developing the benchmarking factors (May through April).
Table 1: Percent Change of Current to Last Year Benchmarking Factors for Housing, Transportation, and Power
Construction
Housing Transportation
Power
Last
Last
Last
Date
Year
Current
Percent
Year
Current
Percent
Year
Current
Percent
Factors
Factors
Change
Factors
Factors
Change
Factors
Factors
Change
Jan 2004
3.076
3.654
18.8
1.603
1.524
-4.9
4.121
4.549
10.4
Feb 2004
3.065
3.661
19.4
1.609
1.529
-5.0
4.113
4.564
11.0
Mar 2004
3.051
3.662
20.0
1.614
1.534
-5.0
4.136
4.600
11.2
Apr 2004
3.035
3.664
20.7
1.620
1.541
-4.9
4.163
4.640
11.5
May 2004
3.024
3.657
20.9
1.628
1.550
-4.8
4.223
4.697
11.2
Jun 2004
3.018
3.658
21.2
1.634
1.558
-4.7
4.276
4.756
11.2
Jul 2004
3.002
3.656
21.8
1.640
1.566
-4.5
4.333
4.815
11.1
Aug 2004
2.995
3.665
22.4
1.646
1.574
-4.4
4.370
4.855
11.1
Sep 2004
2.981
3.677
23.3
1.650
1.580
-4.2
4.407
4.901
11.2
Oct 2004
2.973
3.690
24.1
1.655
1.586
-4.2
4.467
4.922
10.2
Nov 2004
2.965
3.712
25.2
1.659
1.592
-4.0
4.516
4.926
9.1
Dec 2004
2.963
3.739
26.2
1.662
1.598
-3.9
4.554
4.926
8.2
Jan 2005
2.954
3.766
27.5
1.665
1.603
-3.7
4.572
4.908
7.3
Feb 2005
2.953
3.799
28.6
1.666
1.607
-3.5
4.593
4.878
6.2
Mar 2005
2.950
3.827
29.7
1.667
1.610
-3.4
4.609
4.851
5.3
Apr 2005
2.953
3.861
30.7
1.667
1.612
-3.3
4.605
4.805
4.3
May 2005
2.950
3.900
32.2
1.667
1.614
-3.2
4.500
4.776
6.1
Jun 2005
2.950
3.931
33.3
1.667
1.616
-3.1
4.500
4.724
5.0
Jul 2005
2.950
3.976
34.8
1.667
1.618
-2.9
4.500
4.681
4.0
Aug 2005
2.950
4.005
35.8
1.667
1.619
-2.9
4.500
4.643
3.2
Sep 2005
2.950
4.038
36.9
1.667
1.621
-2.8
4.500
4.605
2.3
Oct 2005
2.950
4.057
37.5
1.667
1.622
-2.7
4.500
4.582
1.8
Nov 2005
2.950
4.077
38.2
1.667
1.623
-2.6
4.500
4.551
1.1
Dec 2005
2.950
4.098
38.9
1.667
1.624
-2.6
4.500
4.535
0.8
This paper reports the results of the State and Local Undercoverage Evaluation (SLUE) that was conducted (from 2002
through 2006) to reevaluate the coverage of the S&L VIP frame by matching known population units against units on the VIP
frame. The SLUE study was modeled after the studies by Luery et al. (1992) and Asanuma and Newman-Smith (1993). We
derived up-to-date undercoverage adjustment factors and began using them to adjust for the VIP frame undercoverage
beginning with the May 2007 press release.
The paper is organized as follows: Section 2 provides an overview of the S&L VIP Survey, Section 3 discusses the
subsampling of government agencies from the ASGF for inclusion in the SLUE, Section 4 discusses the collection of
construction project information from the agencies for the SLUE, Section 5 discusses the matching of projects collected from
the agencies to the projects on the S&L VIP frame to determine the frame’s coverage, Section 6 discusses the results of the
matching, Section 7 discusses the application of the undercoverage adjustment factors, and Section 8 discusses the
outstanding issues.
2. S&L VIP SURVEY
The VIP is a Manufacturing and Construction Division (MCD) survey measuring the total value of construction activity
performed in the United States. The VIP estimates come from the monthly Construction Progress Reporting Survey (CPRS)
augmented with estimates of a non-CPRS component based on regulatory filings, phasing of other Census data,
administrative records, trade association data, and other sources. Table 2 lists the VIP major categories, their sizes based on
weighted values, and the source of the data (CPRS and/or non-CPRS). Based on year 2006 annual data, state and locally
owned construction ranked third behind Private New Residential and Private Nonresidential categories with 20% in weighted
value; and was the largest from the CPRS source. Note that the SLUE study only measured the undercoverage of the S&L
VIP frame. Other components of the VIP have their own undercoverage adjustment methods. For example, there is a similar
ongoing study that measures and adjusts for the coverage of the Private Nonresidential frame of the CPRS (Asanuma and
Newman-Smith 1993).
Table 2: VIP Major Categories
VIP Category
Percent of the Total VIP
Data source
Private New Residential
40
CPRS and Non-CPRS
Private Nonresidential
25 CPRS
and
Non-CPRS
State & Local
20
CPRS
Private Residential Improvements
14
Non-CPRS
Federal
1
CPRS
The sampling frame for the S&L VIP is a list of construction projects in the United States valued at $75,000 or more that
have started or will start construction within 60 days. MCD purchases the list from McGraw-Hill Construction (MHC). The
projects are stratified by six construction value groups and 12 general TC categories based on the information provided to us
by MHC. Each of the 72 strata is assigned a specific sampling rate with currently 15 of the strata being certainty (that is
having a sampling rate of 1-in-1). An independent systematic sample of projects is selected in the remaining 57 strata. There
are about 11,000 S&L projects in the survey at a given time. These include newly selected projects along with projects
selected from previous months that are still active. For each sampled project, monthly construction progress reports are
requested from the owner or builder until the project is completed.
The monthly estimate of the total S&L VIP is a weighted sum of the value of work done on all projects underway during the
month of interest, regardless of when construction on the individual project started or when payment was made to the
contractors or builders. Each project is first multiplied by the final weight then adjusted for (1) undercoverage of projects not
reported by MHC and (2) projects that have already started but have not yet been selected with the late selection factor. The
final weight is a product of the selection weight or the inverse of the probability of selection, outlier adjustment factor to
reduce the influence of extreme values on the VIP, and adjustment factor for architectural, engineering, and miscellaneous
costs. There is also an imputation procedure used for item nonresponse. For more information on the methodology of the
VIP survey, see U.S. Census Bureau (1995 and 2007).
3. SUBSAMPLING FROM THE ASGF
In conducting the SLUE study to evaluate the S&L VIP frame’s coverage, a list of projects from an independent source was
needed to match against projects on the S&L VIP frame. Because such a list was not readily available, we had to survey
S&L government agencies to acquire the information. The frame for the SLUE survey was the ASGF sample of S&L
government agencies.
Governments Division of the U.S. Census Bureau provided MCD the ASGF sample file for fiscal year 2000. The file
contained 201,060 records for 25,336 government agencies. There were nine different categories of expenditures and 50
function codes. We only wanted those records with construction activity, thus the number of records was reduced to 29,644
(16,986 agencies). In preparation for the stratification of the agencies, we reclassified the function codes to coincide with the
12 S&L VIP general TC categories. The expenditures represented the total amount of a particular construction type spent by
an agency. Of the 16,986 remaining agencies, 81% reported one construction type and 19% reported multiple construction
types. Of the agencies with more than one type of construction expenditures, only the construction type with the largest
weighted value was kept in order to create a value-classifier for stratification purposes. We used both reported and imputed
records in the agency subsampling.
The government agencies were stratified by three value groups and 12 general TC categories2. The value groups were based
on weighted construction expenditures from the ASGF. The three groups were $10 million or more, $1 million to less than
$10 million, and less than $1 million. Refer to Table 5 for a list of the general TC. Each of the 32 strata was assigned a
specific sampling rate with 12 of the strata being certainty or take every. Two TC: conservation & development and not
elsewhere classified (NEC) had one stratum each, with a sampling rate of 1-in-1. An independent systematic sample of
agencies was selected in the remaining 20 strata. There were 4,026 agencies selected from a sample of 16,986 agencies that
met our criteria for selection.
4. DATA COLLECTION FOR SLUE
We utilized a mailout/mailback strategy with telephone follow-up (TFU) to collect data from the S&L government agencies.
A total of 5,352 forms were mailed to 4,026 agencies that were selected, in two separate waves three months apart. Roughly
half of the forms were mailed out July 1, 2004 requesting all contracts awarded and force account3 work started during
second quarter 2004; and the other half were mailed out October 1, 2004 requesting similar information for the third quarter.
An eligible respondent was asked to provide contract award date, project description, physical location of the project, name
of the general contractor, and the contract value for each contract awarded for a construction project valued at $75,000 or
more. For contracts of less than $75,000 and force account projects, total amounts were requested from the respondent. The
majority of agencies (3,888 out of 4,026) were single-mailing agencies, which received one form each. The remaining 138
agencies were multiple-mailing agencies, which received multiple questionnaires addressed to different entities or
departments within the agency. To help minimize respondents’ burden, different reporting modes were offered to the
agencies: mail back, fax back, email back, and phone in. We conducted TFU for nonresponse or delinquent cases
immediately after all returned forms were keyed.
Of the total number of forms sent minus the 309 out-of-scope cases, the survey had an overall unweighted response rate of
73%. Table 3 provides the response rates for the different response modes and the TFU’s response rate. It is also of interest
that 75% of the eligible agencies responded to the SLUE survey.
Table 3: Response Rates Prior to TFU by Response Modes and After TFU
(Total Number of Cases = 5,043)
Response Mode
Frequency
Percent
Mail Back
2,165
43
Fax Back
203
4
Email Back
9
< 1
Phone In
13
< 1
Total (no TFU)
2,390
47
Total TFU
1,310
26
Total Response
3,700
73
Agency nonresponse was adjusted by reweighting the design weights within the stratification cells. Thus 32 nonresponse
adjustment factors were computed. For the purpose of computing agency nonresponse adjustment factors, we defined a
2 The same 12 general TC categories as the monthly S&L VIP.
3 Force account projects are projects that were done entirely by the agency’s own employees.
response for both single and multiple mailing agencies as an eligible agency’s completion of at least one questionnaire in its
entirety, with the exception of missing force account expenditures. The total number of eligible agencies was the 4,026
agencies in the SLUE sample minus the 17 out-of-scope agencies. The nonresponse adjustment factor NFs is given by
p +
∑qa y
k
k
k =
NF =
1
s
∑pa y
k
k
k =1
Where s represents the stratum number (s = 1, 2, 3 …, 32), p is the number of responses for the s stratum, q is the number of
nonresponses for s stratum, yk = construction expenditures from the ASGF (unweighted) for the kth agency, and ak = (ASGF
sampling weight) x (agency subsampling weight) for the kth agency.
5. MATCHING OF PROJECTS – CHECKING FOR FRAME COVERAGE
To reduce the workload for the matching and verification of the matches, we subsampled 2,098 contracts from 6,845
contracts (valued at $75,000 or more) that were awarded for construction projects and reported to us by SLUE respondents.
The contracts were stratified by three value groups, 12 general TC4, and four Census regions. The value groups were based
on unweighted contract award value reported by the SLUE respondents. The three groups were $10 million or more, $750
thousand to less than $10 million, and $75 thousand to less than $750 thousand. The four Census regions were Northeast,
Midwest, South, and West. Each of the 144 strata was assigned a specific sampling rate with 104 of the strata being
certainty. An independent systematic sample of agencies was selected in the remaining 40 strata.
The selected projects/contracts5 were sent to MHC to match against projects that should be on the S&L VIP frame. MCD
reviewed all projects matched by MHC manually. We reviewed the information for each project and decided that a project
was a match if the SLUE information was similar to the MHC information on equivalent variables. Each project had six
matching variables. We first compared each of the variables individually before concluding on an overall project status. The
guidelines that we used to decide whether each variable was referring to the same project or not are as follows:
1) Owner of project – name of the government agency, city, and state
Same – The information referred to the same government agency.
Different – The information referred to two different government agencies.
2) Project title/description
Same – The project title/description reported by SLUE respondent was similar to the MHC project title.
Different – The two titles/descriptions described different projects.
3) Physical location of project – street address or site boundaries, city, and state
Same – The two addresses described the same location or locations within close proximity.
Different – The two addresses described different locations.
4) Name of general contractor6
Same – The information referred to the same contractor.
Different – The information referred to two different contractors.
5) Date
Same – The issue date from MHC was between April 1, 2003 and September 30, 2005.
Different – The issue date from MHC was prior to April 2003 or after September 20057.
4 The same 12 general TC categories as the monthly S&L VIP and ASGF subsampling. Each contract was manually assigned
a general TC code based on the reported project description.
5 Contract is surrogate for construction project and vice versa. From here on, contract and project will be used
interchangeably.
6 This variable was omitted during verification if MHC did not provide a contractor name.
6) Value8
Same – SLUE contract value was approximately equal to the MHC value (and other variables listed above were
referring to the same project).
Different – SLUE contract value was very different from the MHC value (and other variables listed above
seemed to be referring to a different project).
For the majority of the projects, it was clear whether a project was a match or not. For those projects that fell in the gray area
or where we disagreed with MHC, we sent them back to MHC for additional information/explanations, performed searches
on our end with the information that we had available, and/or as a last resort called the respondent to verify the information.
6. RESULTS OF THE MATCHING
Of the 2,0239 SLUE contracts, 67% were found on the VIP frame. Figure 2 shows the breakdown of the coverage for
contracts greater than or equal to $75,000. MHC matched 11% of the SLUE contracts, but equivalent MHC projects were
never submitted to MCD for inclusion in the monthly S&L VIP frame. Therefore, these contracts were also treated as
nonmatches in the computation of the match rates. The projects found on the VIP frame represented 83% of the weighted
value of all contracts awarded in two quarters in the United States (valued at $75,000 or more), with 2% standard errors. No
matching was attempted for force account projects and projects less than $75,000 since these projects are part of the monthly
S&L VIP nonsampled stratum.
Figure 2: Coverage of Contracts at $75,000 or More
(Of
Count)
No match
found
On VIP frame
Not on VIP
22%
67%
frame
33%
Matched but not submitted
11%
The SLUE contract award and force account values were weighted by multiplying the amount by the final weight. The final
weight for a contract at $75,000 or more was a product of the ASGF weight, agency subsampling weight, agency
nonresponse adjustment factor, and contract selection weight. The final weight for a contract less than $75,000 and a force
account project was similar to the final weight of a contract at $75,000 or more but without the contract selection weight.
The estimated match rate rˆ for projects with contracts valued at $75,000 or more is given by
∑mw xjj
j=
r =
1
ˆ
∑
m+n
w x
j
j
j=1
Where m is the total number of matches, n is the total number of nonmatches, wj = final weight of the jth project, and xj =
SLUE contract value of the jth project.
7 Since there were only a handful of projects that fell outside the matching time frame, with a date of October 2005 or after,
we did not reject these as matches.
8 The SLUE contract value and the MHC value could be very different for a particular project and still be a match.
9 There were 2,098 contracts selected for matching. 75 projects were out-of-scope because they were privately or federally
owned, not covered by our definition, abandoned or delayed, or duplicates (kept one and deleted others).
The variance for the match rate was computed using the Jackknife method with 20 replicates. The match rate variance
2
estimator ˆ
σ is given by
r
20
∑(t − t)2
i
ˆ 2
1
=
σ = i
r
20
20
20
(∑
m
mi
w x −
w x )
∑
j
j
∑ j j
t
19
i
j=1
j=1
=
with
t =
and
1
t = i
i
∑
m+n
20
w x
j
j
j=1
Where i represents the replicate number (i = 1, 2, …, 20), m is the total number of matches, n is the total number of
nonmatches, mi is the number of matches for the ith replicate, wj = final weight of the jth project, and xj = SLUE contract value
of the jth project.
Table 4: Match Rates of the Weighted Contract Values and Standard Errors by Value Groups
(In
Percent)
Value Group
Match Rates
Standard Errors
$10 million or more
91
3
$750 thousand to less than $10 million
84
3
$75 thousand to less than $750 thousand
54
1
Table 5: Match Rates of the Weighted Contract Values and Standard Errors by General TC
(In Percent)
General TC
Match Rates
Standard Errors
Housing 44
3
Office 86
6
Health care
81
11
Education 88
4
Public safety
91
7
Amusement & recreation
89
5
Transportation 69
4
Highway & street
84
4
Sewer system
76
4
Water supply system
91
9
Conservation & development
84
6
NEC (Not elsewhere classified) – includes
39 3
Hotel/motel, Commercial, Religious,
Communication, Power, and Industrial
Tables 4 and 5 present the estimated match rates of the weighted contract values and the associated standard errors by value
groups and general TC, respectively. In general, MHC coverage was better for large projects than for small projects. This is
similar to what Luery et al. (1992) and Asanuma and Newman-Smith (1993) found in their coverage studies of the VIP
frames. Tests of significance10 at the 10% level indicated that the match rates for the three value groups were significantly
10 Test of significance used in the analysis of the match rates of weighted values were based on student’s t, with Bonferroni
adjustment of type I error for the multiple comparisons.
different from each other. Nevertheless, we did not produce undercoverage adjustment factors by values. Our concerns were
(1) the contract value for SLUE was not always close to being equal to the MHC value on an equivalent project. The two
values could be very different when MHC picked up an entire project and SLUE only picked up a portion of the project and
vice versa. (2) The match rates by value groups were biased since more effort was taken to match and verify larger projects
by MHC and MCD. Also, we had more detailed information for projects at $10 million or more (since these were selected
with certainty for the monthly S&L VIP), so we were able to use this additional information to conduct a more thorough
verification than for those smaller projects. The smaller the project values, the probability of having the additional
information for the verification of the matches greatly decreased. The match rates of weighted value across the general TC
categories were also found to be significantly different at the 10% level for some categories. We collapsed the smaller TC
categories if not significantly different from each other to have more reliable estimates since the combined sample sizes are
much larger than the sample sizes of the individual category. The match rates of the three largest TC categories: highway &
street, education, and sewer-water combined were not significantly different from each other. Nevertheless, we did not
combine these TC categories into one because a small change in the adjustment factor could have a great impact on the total
VIP estimates due to their sizable contribution.
7. APPLICATION OF THE UNDERCOVERAGE ADJUSTMENT FACTORS
Undercoverage adjustment factors were derived for seven TC categories: highway & street, education, sewer-water, power,
housing-hotel/motel, transportation, and others. The undercoverage adjustment factor was a product of the inverse of the
match rate of the weighted value and the inverse of the contribution from projects less than $75,000 and force account
projects. For example, the overall adjustment factor F that included all projects is given by
1
1
F =
×
= .
1 27
83
.
.95
Where the factor 1 is to account for all projects greater than or equal to $75,000 and the factor 1 is to account for projects
83
.
95
.
less than $75,000 and force account projects (which represented approximately 5 percent of all projects). Table 6 presents
the factors from SLUE along with the benchmarking factors that were used prior to the implementation of the SLUE factors.
There are six unique SLUE factors (education and others TC categories have the same undercoverage adjustment factor of
1.20). Five of the seven SLUE categories coincided with the benchmarking categories, transportation no longer grouped with
the others category under the SLUE. Note that the benchmarking factors for the education and sewer-water categories
sometimes dropped below one. Thus, we had to exclude some of the VIP expenditures that were collected for these
categories, which was not justifiable.
Table 6: SLUE and Benchmarking Undercoverage Adjustment Factors for the Monthly S&L VIP
Sewer-
Housing-
All
Highway
Education
Water
Power
Hotel/motel
Transportation Others
TC
SLUE Factor
1.25
1.20
1.24
5.85
2.51
1.53
1.20
1.27
Benchmarking Factor
(1994-2003)
Lowest
1.15
0.85
0.74
2.43
2.07
1.31
1.13
Highest
1.24
1.18
1.02
6.66
3.49
1.63
1.27
Last 3 Years Average
1.24
1.09
0.95
5.06
3.41
1.54
1.26
10 Years Average
1.19
1.08
0.90
4.36
2.96
1.44
1.21
Beginning with the 2007 annual revision of the VIP series (released on June 29, 2007), all S&L VIP series were adjusted
using SLUE factors listed in Table 6. We also adjusted the S&L VIP series back to the start of the series (January 1993)
using the same set of SLUE factors listed in Table 6. The latter adjustment was an attempt to improve the historic S&L VIP
series. We found that MHC coverage of projects owned by S&L government agencies appeared to have only changed
slightly since the coverage study by Luery et al. (1992). Table 7 shows the comparison of the SLUE factors and the
undercoverage adjustment factors derived from the study by Luery et al. (1992). The overall factors were 1.27 and 1.24 for
the SLUE and the study by Luery et al. (1992), respectively.
Figure 3 presents a time series plot of the total construction expenditures for S&L VIP with SLUE adjusted and
benchmarking. S&L VIP with SLUE adjusted was higher than benchmarking from 1993 through 1998, but lower from 1999
to present. Luery et al. (1992) pointed out that the ASGF estimates included nonconstruction costs and the S&L VIP
estimates underreported the architectural, engineering, and miscellaneous costs. One would expect the SLUE adjusted S&L
VIP estimates to be lower than the benchmarking; however, we did not see that for the earlier time periods.
Our plan is to update these adjustment factors every five years by conducting similar undercoverage evaluation studies.
Table 7: Undercoverage Adjustment Factors From SLUE and Luery et al. (1992)
TC Category
SLUE Factor
Luery et al. (1992) Factor
All TC
1.27
1.24
Highway
1.25
1.23
Sewer and water
1.24
1.24
Housing, lodging, health care, and education
1.22
1.23
Building11 1.28
1.20
Nonbuilding11 2.02
1.48
Figure 3: Time Series Plot of the Total Construction Expenditures for S&L VIP With SLUE Adjusted and
Benchmarking
(In Millions)
s 250000
200000
nditure 150000
pe
Ex 100000
d
VIP w/ Benchmarking Factors
50000
ighte
VIP w/ SLUE Factors
We
0
1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005
Year (January - December)
8. OUTSTANDING ISSUES
Starting November 2006, the monthly S&L VIP frames also have included the ITEM reports for the first time. ITEM is an
MHC acronym for Individual Trade Equipment and Material. Recall that Figure 2 showed 11% of the SLUE contracts were
matched by MHC but were not submitted to MCD for inclusion in the monthly S&L VIP frame. Of this percentage, 7% were
identified as ITEM reports and 4% with various statuses. We need to evaluate MHC coverage of the ITEM reports to decide
when and how to introduce the new set of SLUE boost factors that includes ITEM reports as matches.
11 The building category for SLUE included: office, public safety, religious, industrial, amusement & recreation, and
transportation; and the nonbuilding category for SLUE included: conservation & development, commercial, communication,
and power. We did not know which TC were included in the building and nonbuilding categories for the study by Luery et
al. (1992). The factors from these two categories should not be compared.
In terms of the number of contracts awarded for state and locally owned construction projects by Census regions, SLUE
found that the Northeast had the lowest level of activities followed by the Midwest then the West. The South region awarded
the most number of contracts during the second and third quarters 2004. Table 8 presents the match rates and standard errors
for projects valued at $75,000 or more by regions based on weighted contract values. MHC coverage was not significantly
different across the four Census regions at the 10% level. Even though the match rate for the Midwest was not significantly
different from the other three regions, MHC did not seem to cover this region as well as the others. For the next S&L VIP
sample redesign, we should examine stratifying by regions to see if we can further improve the monthly S&L VIP sample.
Table 8: Match Rates of the Weighted Contract Values and Standard Errors by Census Regions
(In Percent)
Regions
Match Rates
Standard Errors
Northeast 83
5
Midwest 74
4
South 85 4
West 86 3
ACKNOWLEDGEMENTS
We thank all Census Bureau Headquarter staff and staff members from the National Processing Center in Jeffersonville for
their assistance in making this study a success. A special thank you to Josephine Holland for the work that she did in
conducting the screening telephone survey, searching for mailing addresses, and conducting telephone follow-ups for
delinquent cases.
REFERENCES
Asanuma, M. and Newman-Smith, A. (1993). “A Built-In Evaluation System For a Sample Design With an Imperfect
Frame.” Proceedings of the International Conference on Establishment Surveys, American Statistical Association, pp. 448-
453.
Luery, D., Asanuma, M., Vu, L., McDonald, K., and Newman-Smith, A. (1992). “Nonsampling Errors and Data
Improvements for State and Local Governments Value-in-Place (VIP) Survey.” Proceedings of the Section on Government
Statistics, American Statistical Association, pp. 48-53.
U.S. Census Bureau (1990). “Prediction of Governments Annual Construction Outlay Series for Use with the Business
Benchmark Program.” U.S. Census Bureau Memorandum for the Record from Donald M. Luery, Chief, Research and
Methods Staff, Construction Statistics Division.
----------- (1995). “Revised Weighting and Estimation Specifications for the Construction Progress Reporting Surveys
(CPRS) System Redesign.” U.S. Census Bureau Memorandum for Sarah Baumgardner from Brian V. Greenberg, Assistant
Division Chief for Research and Methodology from the Manufacturing and Construction Division.
----------- (2007). “Construction Spending Methodology.” http://www.census.gov/const/www/methodpage.html.
®
®
The analysis of this paper was generated using SAS Software. Version 8.2 of the SAS System for Windows . Copyright ©
1999-2001 by SAS Institute Inc. SAS and all other SAS Institute Inc. product or service names are registered trademarks or
trademarks of SAS Institute Inc., Cary, NC, USA.
