New Construction Cost Indices to Improve

Highway Management
Emily Wong 1 and Omar Swei, Ph.D. 2
Downloaded from ascelibrary.org by The University of Queensland Library on 03/31/22. Copyright ASCE. For personal use only; all rights reserved.

Abstract: Construction cost indices support planners, engineers, and policymakers in improving the management of highways. This study
details the creation, analysis, and application of four new quarterly highway construction cost indices disaggregated by material (i.e., asphalt
concrete and portland cement concrete) and activity (i.e., general construction and maintenance and rehabilitation). The contributions of this
research are twofold: distilling for managers differences in cost growth trends across various highway activities and materials and demon-
strating the impact of these findings on their highway budgeting decisions. The results from the time-series analysis demonstrate that asphalt
concrete (AC) and portland cement concrete (PCC) highway construction have exhibited distinctive annual cost growth trends of 3.5% and
2.1%, respectively. Conversely, annual maintenance and rehabilitation cost growth is similar between AC and PCC pavements, with historical
rate increases of 2.6% and 2.7%, respectively. To demonstrate the value of these insights, the resulting forecast models are used to project
future public highway expenditures in the US. The case findings show that projected expenditures are $6–$8 billion lower when relying on
these disaggregate indices compared with the state of the art. These findings highlight that this study’s cost indices and its resulting insights
can be leveraged by practitioners to improve the management and capital planning of highways. DOI: 10.1061/(ASCE)ME.1943-
5479.0000924. © 2021 American Society of Civil Engineers.
Author keywords: Asphalt concrete (AC); Fisher index; Highway construction cost index; Maintenance and rehabilitation; Productivity;
Portland cement concrete; Time-series analysis.

Introduction Construction cost indices are not only valuable for engineers and
contractors in planning their project operations, but are also impor-
Managers, engineers, and contractors rely on accurate forecasts tant for policymakers, in particular those interested in improving the
of construction costs to plan the procurement and maintenance of construction sector’s productivity. Productivity measures the effi-
infrastructure projects. To provide decision makers with reliable ciency in transforming inputs to outputs. Improvements in produc-
forecasts and improve program budgeting (Creedy et al. 2010), tivity are a major determinant of employee wage growth (Dearden
researchers have developed several innovative methodologies et al. 2006) and the success of an individual firm or entire segment
over the last several decades aimed at improving the fidelity of cur- of the economy (Syverson 2011). According to the US Bureau of
rent forecasts. These methodologies include advanced univariate Economic Analysis (BEA), between 2000 and 2018, labor produc-
(Ashuri and Lu 2010; Ilbeigi et al. 2017; Moon et al. 2018; Touran tivity across the US economy increased annually by just 1.4%; for
and Lopez 2006) and multivariate (Shahandashti and Ashuri 2016; the construction sector, labor productivity decreased by −0.9% (US
Swei 2020; Wong and Ng 2010; Xu and Moon 2013) methods to Bureau of Economic Analysis 2020). The noted productivity decline
accurately predict future construction and material prices. in construction has motivated a debate among statistical agencies
To train their models, each of these cited studies relied on con- around whether this slowdown is in fact real or, perhaps, reflects
struction cost indices, which track aggregate changes in construc- the inadequacy of current measures (Sveikauskas et al. 2016,
tion costs over time. Implicitly, both the efficacy and fidelity of 2018). The US Bureau of Labor Statistics (BLS) recently launched
these forecasting models depend on utilizing indices that correctly an initiative aimed at revising the measurement of productivity
represent historical construction costs. For the highway sector, re- growth in construction. Their approach, which serves as the cur-
searchers (e.g., Shahandashti and Ashuri 2016; Nassereddine et al. rent state of the art, involves tracking over time (1) labor input via
2016; Shrestha et al. 2017b) primarily make use of the Federal employee-hours worked; and (2) output in terms of revenues
Highway Administration’s (FHWA) National Highway Construc- (Moulton 2018; Syverson 2011). Because the value of output
tion Cost Index (NHCCI), which tracks temporal changes in aggre- (i.e., revenues) will change due to inflation, revenues must be de-
gate highway construction costs. flated (i.e., the inflation effect is removed) by relying on a producer
price index (PPI) that quantifies average changes in producer costs
1
Research Assistant, Dept. of Civil Engineering, Univ. of British over time (Moulton 2018).
Columbia, 6250 Applied Science Lane, CEME 2001, Vancouver, BC, The key takeaway from the BLS initiative is that the meas-
Canada V6T 1Z4. Email: [email protected] urement of productivity growth is highly sensitive to the selected
2
Assistant Professor, Dept. of Civil Engineering, Univ. of British PPI (Sveikauskas et al. 2016, 2018). Through revising the tradi-
Columbia, 6250 Applied Science Lane, CEME 2004C, Vancouver, BC, tional deflators used in previous productivity studies, BLS has
Canada V6T 1Z4 (corresponding author). Email: [email protected]
estimated that residential, multifamily, and industrial construc-
Note. This manuscript was submitted on March 22, 2020; approved on
January 28, 2021; published online on April 20, 2021. Discussion period tion have all exhibited positive productivity growth over the last
open until September 20, 2021; separate discussions must be submitted for 3 decades. The only subsector that has exhibited zero to negative
individual papers. This paper is part of the Journal of Management in productivity growth (since 2002) according to the authors is high-
Engineering, © ASCE, ISSN 0742-597X. way, roads, and bridge construction. Their analysis of highway,

© ASCE 04021030-1 J. Manage. Eng.

J. Manage. Eng., 2021, 37(4): 04021030

 roads, and bridge construction utilized FHWA’s NHCCI as their Literature Review
deflator.
Simply put, construction cost indices are vital for the capital Review of Techniques Used to Develop Cost Indices
planning of infrastructure and the measurement of productivity
within the construction industry. FHWA’s NHCCI currently serves Because construction cost indices reflect construction cost growth
as the benchmark index for highway infrastructure and has been over time, they are a critical tool for state highway agencies, project
used to support both applications. Having said that, an unanswered planners, and other stakeholders when determining capital budget-
question in the literature is whether the availability of further dis- ing requirements. Consequently, cost index development and fore-
aggregated indices would alter (and ultimately improve) capital casting methods are prominent areas of study in the management
planning decisions and/or the measurement of productivity. There and construction literature. Shahandashti and Ashuri (2016) iden-
is compelling evidence that over the last few decades (1) cost tified useful explanatory variables of the NHCCI to develop a more
accurate and rigorous forecasting model. To better understand peri-
Downloaded from ascelibrary.org by The University of Queensland Library on 03/31/22. Copyright ASCE. For personal use only; all rights reserved.

growth has varied considerably across different types of highway

construction activities; and (2) there has been a considerable shift in ods of substantial volatility, Joukar and Nahmens (2016) and Moon
how agencies spend their resources. For example, Mehrota et al. et al. (2018) examined the Engineering News-Record (ENR) CCI,
(2020) estimated that, between 1990 and 2006, the cost to construct which is a composite index of 20-city average prices of construc-
a new mile of interstate increased fivefold whereas the cost of resur- tion activities. Touran and Lopez (2006) employed the ENR build-
facing only doubled. Furthermore, between 2002 and 2017, the US ing cost index in a model that considers random variations in cost
Congressional Budget Office noted that operation and maintenance escalation.
expenditures for highways grew from 37% to 47% (CBO 2018). Researchers have also focused on material indices. Faghih and
This research hypothesizes that the availability of disaggregate Kashani (2018) examined the producer price indices of individual
indices will allow decision makers to capture important structural construction materials to more accurately predict price movements,
shifts when estimating their future capital needs. In addition, the and Wang and Liu (2012), Ilbeigi et al. (2017), and Swei (2020)
availability of disaggregate indices will enable policymakers to bet- focused on characterizing asphalt price index trends. Although this
ter understand productivity growth. Contractors tend to specialize discussion does not encompass the entire literature around cost
in specific materials (e.g., bituminous pavements versus cement forecasting, it nevertheless highlights the importance of identifying
pavements) or activities that reflect the evolving needs of transport robust CCI calculation and forecasting methods to develop mea-
agencies. The noted decline in productivity by BLS for highway sures of construction cost growth trends and volatility.
construction may simply reflect a shift in the labor force from more There are three well-known forms of price indices used in high-
productive (i.e., new capital outlays) to less productive (i.e., main- way construction: Laspeyres, Paasche, and Fisher. Due to conven-
tenance and rehabilitation) tasks (Allen 1985; Sveikauskas et al. tion, these indices are referred to as price indices rather than cost
2016). The development of disaggregate indices will allow BLS indices. Shrestha et al. (2016, 2017a) provided an extensive over-
and others to use deflators that better reflect inflation for different view around the utilization of these indices by both FHWA and
producers. Improved measures of productivity, in turn, allows individual state DOTs. The Laspeyres index, L, shown in Eq. (1),
policymakers to enhance their models to predict the future produc- is a common fixed-base index that models changes in item prices
tive capacity of the economy, economic growth, and possible infla- between a historical base period (i.e., t0 ) and the current period
tionary pressures. (i.e., t), with individual item weights based on quantities in the base
Based on this introduction, the goal of this study is the creation period (Nassereddine et al. 2016). In other words, the Laspeyres
of four new cost indices: (1) asphalt concrete (AC) highway con- index at time t is a function of the base price, p0 , for each individual
struction; (2) portland cement concrete (PCC) highway construc- i item, the current price, pt , and the base quantity, q0 . The sum-
tion; (3) AC highway maintenance and rehabilitation (M&R); and mation aggregates the prices and quantities of all n item baskets
(4) PCC highway M&R. This study differentiates between AC and used to consolidate bid items into categories
PCC pavements for multiple reasons. First, based on a recent sur- Pn
pt q0
vey of 18 state departments of transportation (DOTs), AC and PCC Lðp0 ; pt ; q0 Þ ¼ Pni¼1 0i i0 ð1Þ
pavements are among the most common item categories of interest i¼1 pi qi
to planners when assembling a highway construction cost index
(Shrestha et al. 2017a). Second, data from BLS strongly indicated Up until the 1990s, the Laspeyres index was the standard ap-
that the price of material inputs for these two technologies have proach used by statistical agencies to measure inflation. An issue
evolved differently from one another (BLS 2018a, b). Third, there with the Laspeyres index, however, is that it ignores the substitution
is tremendous variance between individual state DOTs around their effect (i.e., when an item’s usage declines due to its relative price
utilization of AC and PCC highways. Additionally, this paper increase and vice versa) (Reinsdorf 1998; Diewert 2004). The
distinguishes M&R from overall highway construction given the Boskin Commission, which was appointed in 1995 by the US
growing focus of agencies around system preservation (Shrestha Senate to study the measurement of inflation, highlighted the sub-
et al. 2017b). stitution effect as a considerable source of bias in price indices
As part of this study, the authors analyze the trends in these data (Boskin et al. 1996). The Paasche index, P, shown in Eq. (2), is
via time-series methods in order to glean differences in their price a current-weighted index that uses quantities transacted in the cur-
growth. The described time-series methods are subsequently used rent period as weights for all periods. The Paasche index at time t is
as part of a case study to forecast future highway expenditures in a function of the base price, p0 , the current price, pt , and the current
the US under two different scenarios. The results of the case study quantity, qt
highlight that the anticipated cost to manage highway infrastructure Pn
pt qt
differs when relying on disaggregate, rather than aggregate, cost ðp0 ; pt ; qt Þ ¼ Pni¼1 0i it ð2Þ
indices. Decision makers can leverage the indices generated as part i¼1 pi qi
of this paper, which will be maintained annually by the authors
and made accessible via the link provided in the Data Availability Because it also neglects the substitution effect, the Paasche index
Statement. will generally underestimate price changes (Diewert 2004). As the

© ASCE 04021030-2 J. Manage. Eng.

J. Manage. Eng., 2021, 37(4): 04021030

 geometric mean of the Laspeyres and Paasche indices, the Fisher One important shortcoming of these indices is their limited data
index, F, shown in Eq. (3), averages the effect of substitution bias coverage. State DOTs tend to only select a few important bid items
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
Pn Pn and ignore items that are either infrequent or low in cost (Shrestha
pt q0 pt qt et al. 2017a). Shrestha et al. (2017a) surveyed eight different state
Fðp ; p ; q ; q Þ ¼ Pni¼1 0i i0 × Pni¼1 0i it
0 t 0 t
ð3Þ
i¼1 pi qi i¼1 pi qi
DOTs, noting that all of their indices rely on less than 50% of con-
struction costs and, in fact, can depend on as little as 14% of avail-
The Fisher index emerges as the only homogeneous, symmetric able cost data. Shrestha et al. (2017b) aimed to overcome this issue
average of the Laspeyres and Paasche indices where the index ratio via the integration of a dynamic item basket generation process as
between any two time periods does not depend on the choice of part of the creation of highway cost indices. Through a case appli-
base period (Diewert 2004). The index itself has long been coined cation with the Montana DOT, Shrestha et al. (2017b) were able to
the Fisher “ideal” index given that it corrects for the noted biases in successfully increase the reliability of their data sample.
Downloaded from ascelibrary.org by The University of Queensland Library on 03/31/22. Copyright ASCE. For personal use only; all rights reserved.

the Laspeyres and Paasche approaches (Diewert 1992). Within the Despite recent progress by researchers, there remains a lack of
number theory community, the Fisher index has been proven to be a disaggregate cost indices for the highway construction sector. As
“superlative” index (Caves et al. 1982; Diewert 1976, 1992, 2004). noted by Shahandashti and Ashuri (2016), national-level measures
A superlative index is one that can provide a second-order approxi- of highway costs are valuable for both the federal government and
mation to an arbitrary cost, production, or utility function. Super- state DOTs given that such indices tend to be robust and do not face
lative indices such as Fisher have been recommended by the the same data availability challenges as state-level measures. Re-
International Labour Office (ILO), the World Bank, BLS, and other searchers and practitioners have a clear affinity toward modeling
organizations as the preferred technique to compute a price index costs from a granular perspective. There is a considerable literature
(Dalton et al. 1998; International Labour Office 2004). around the analysis and forecasting of state-level material prices
Shrestha et al. (2016, 2017a) noted that the Fisher approach is (Wang and Liu 2012; Ilbeigi et al. 2017; Swei 2020), as well as
the most commonly employed method to generate a construction asphalt and cement producer price indices from the US Bureau
cost index for highways. For example, the Fisher approach is cur- of Labor Statistics (Faghih and Kashani 2018). Of course, these
rently implemented by the FHWA, and the states of California, studies fail to capture relevant ancillary costs such as transportation
Florida, Ohio, and Colorado. In addition, states DOTs such as of materials, labor, and capital equipment given the limited avail-
Wisconsin (Nassereddine et al. 2016), North Dakota (Shrestha et al. ability of robust granular indices for the research community.
2016), and Montana (Jeong et al. 2017) have engaged with re- Capturing these ancillary costs is particularly important in the
searchers to shift from a fixed-base index (in particular Laspeyres) context of creating high-quality deflators for the measurement of
to a Fisher index. productivity given that material price changes reflect only one source
FHWA as well as some other state DOTs implement chaining in of inflation. Whereas the FHWA’s NHCCI captures overall inflation
deriving their price index (Shrestha et al. 2016). A chained index is for highway construction, there is a strong need for more granular
an index in which its value at any time period is related to its im- PPIs that are more reflective of inflation for a sector in which pro-
mediate predecessor. This is distinct from Eqs. (1)–(3), where the ducers (i.e., contractors) tend to specialize in certain technologies.
index values depend on a fixed base that is either the current time As a result, this study develops general construction and M&R
period, t, or the initial time period, t0 . Notable advantages of chain- cost indices for both AC and PCC pavements. By generating such
ing are that because the base is updated each period, the base does indices for the predominant construction activities found in the
not become so far removed from the current period, and the spread highway sector and maintaining them on an annual basis, this re-
between the Laspeyres and Paasche indices is minimized (Diewert search aims to provide decision makers with more granular infor-
2004; International Labour Office 2004). By forming a time series mation around highway construction cost movements. Although
linking the relative indices that measure the price movement be- there do exist disaggregate indices for certain state DOTs, these
tween two consecutive periods, chaining considers the prices and indices exhibit shortcomings due to the evolving nature of highway
quantities from all intermediate periods (Nassereddine et al. 2016). construction projects and sampling. Unlike state-level indices,
The US Bureau of Economic Analysis, which provides macroeco- which use less than 50% of cost data, this study’s indices utilize
nomic statistics on major industries including construction, cur- more than 80% of cost information from more than $170 billion
rently computes chained indices. Following the enactment of the worth of projects.
Tax Cuts and Jobs Act of 2017, BLS’s chained consumer price in- Due to their sample size, these disaggregate indices will provide
dex (CPI) is now used to adjust income tax brackets due on infla- agencies with more robust estimates of cost growth than existing
tion (Moulton 2018). The use of chaining by these important ones, potentially leading to their adoption by planning agencies.
governmental agencies charged with monitoring inflation for The availability of these indices is an important contribution to
policymaking highlights the recognized value of chaining for the the existing literature, particularly if they alter either capital plan-
study of price movements. ning decisions and/or the measurement of productivity. Following
the creation of these indices, this paper implements advanced time-
series methods to comment on differences in their historical trends
Previous Efforts Aimed at Creating Disaggregate and/or seasonality. Those insights are subsequently leveraged in a
Highway Cost Indices case application to anticipate cost inflation for highway construc-
It has been well documented in the academic literature that state tion when using disaggregate indices rather than aggregate ones
highway agencies value granular cost indices (Shrestha et al. such as the NHCCI.
2017b). Shrestha et al. (2017b) noted that disaggregate indices en-
able state DOTs to segment the market into smaller subsectors,
allowing them to understand prevailing conditions with higher Methodology
granularity and ultimately improve their decision-making proc-
esses. There are several state DOTs (e.g., Mississippi and Florida) This study generates its four indices by relying on data sourced
that have created their own subindices for activities such as pave- from Oman BidTabs (Oman Systems Inc. 2019), which the FHWA
ment surfacing (Shrestha et al. 2017a). uses to calculate its own NHCCI. This nationally representative bid

© ASCE 04021030-3 J. Manage. Eng.

J. Manage. Eng., 2021, 37(4): 04021030

 Table 1. Sample data available from Oman BidTabs
Unit cost
State Bid date Contractor Item description Unit Item type Quantity ($/ yard2) Total cost ($)
SD April 4, 2007 AI100 6-in. PCC approach pavement S.Y. PAVE_MAINLINE 90 45.75 4,117.5
SC October 2, 2009 BE002 Roller compacted concrete pavement S.Y. PAVE_MAINLINE 27,050 33.43 904,281.5
SC October 2, 2009 BE002 Diamond grinding of concrete pavement S.Y. PAVE_ANCILLARY 27,050 4.74 128,217
SC September 29, 2009 391 Full depth concrete pavement patch-8-in. S.Y. PAVE_REPAIR 9,000 106.88 961,920
Note: SD = south dakota; S.Y. = square yards; and SC = south carolina.

data set encompasses AC and PCC paving projects within the con- to capture purely M&R bid items. For the PCC data set, M&R bid
Downloaded from ascelibrary.org by The University of Queensland Library on 03/31/22. Copyright ASCE. For personal use only; all rights reserved.

tiguous US except for the state of New Jersey, whose contracts are not items are identified as those that are either repair-type or ancillary-
maintained in their system. The data set spans 2005–2017 and con- type. For the AC data set, this study uses, repair-type, ancillary-
tains details used to identify different bid items, such as the bid date, type, and overlay-type items. This additional segmentation process
job number, item key, item description, units, item type, and position differentiates this study’s indices from the existing NHCCI.
(i.e., a contractor’s rank among all bidders for a project). Table 1 pro-
vides a sample of the Oman BidTabs data used as part of this study. Outlier Detection
An important advantage of this data source is that it contains Outlier detection is critical to ensure that the data used to generate
a breadth of highway projects that vary in terms of both their these four indices are reliable and representative of changes in
geographic location and fundamental characteristics. This variation quantities and prices across time. In a competitive market, one
allows this paper to generate cost indices that capture the diversity would expect bid prices for an individual item to be fairly close
of projects found in the field. For example, the average nominal across competing contractors. Based on this assumption, this paper
cost for projects analyzed as part of this work is $3.1 million, implements a fairly simple criteria rule [Eqs. (4) and (5)] to
and the 10th and 90th percentile values range from $240,000 to detect potential outliers. Specifically, the authors inspect whether
$6.8 million, respectively. Similarly, although the average number the ratio for a given project of the lowest bid price received by a
of bidders for a project is 3.82, the 10th and 90th percentile values state DOT for a given item, p1 , and the second lowest bid price, p2 ,
are two and seven bidders. falls outside of an acceptable tolerance, α. In layman’s terms, a data
Because states vary in their size, certain DOTs are better repre- point is preserved if the lowest bid, p1 , is no more than α times
sented in this data set than others; for example, Vermont constitutes lower than p2
only $100 million in annual expenditures, whereas a state such as p1 1
Texas includes more than $4.4 billion worth of projects annually. 1−α< < ð4Þ
p2 1 − α
For the most part, the majority of projects have been procured via
design-bid-build, with less than 1% of projects delivered via an al- 0≤α<1 ð5Þ
ternative procurement model such as design-build. The availability
of a range of projects facilitates the calculation of an index that can Part of the reasoning for this criteria rule is that although a rel-
represent average cost growth for highways in the US. There is a ative comparison with other available statistics such as the median
tremendous opportunity for future work to leverage these indices bid price may be more robust, many of the projects in this data set
and comment on the effect of changing project characteristics on received only two bids, rendering the definition of median unclear.
the trends underlying these indices. As part of this criteria rule, it is tremendously important to deter-
The development of the AC and PCC highway construction and mine a threshold value, α, that provides reasonable data coverage
M&R cost indices requires that the data are aggregated into item and removes potential outliers. Through (1) private discussions
baskets, allowing the quantity and price data to be compared across with industry stakeholders representing state DOTs, private con-
52 quarters between 2005 and 2017. The upcoming sections high- tractors, and trade associations, and (2) a sensitivity analysis of
light the key differences between the proposed chained-Fisher in- the data set, this research found that setting α equal to 0.4 achieved
dices and FHWA’s NHCCI. The cost indices generated as part of this goal. More specifically, stakeholder input suggested that values
this research are subsequently analyzed to inform practitioners of of α greater than 0.4 would capture bid values outside of what one
their historical trends to support their decision making. Fig. 1 pro- would typically expect, and from the sensitivity analysis (presented
vides a high-level overview of the steps required to compute this in the “Results” section) at least 80% of the data entries across all
paper’s cost indices. Time-series methods are implemented to elicit indices are retained with this threshold value, which is considerable
trends underlying each index, and those insights are applied in a larger than the previously cited approaches.
case study to evaluate the anticipated inflation for highway expend- The proposed outlier detection technique differs considerably
itures using disaggregate versus aggregate cost indices. from that used by many other state agencies and the FHWA. Jeong
et al. (2017) highlighted that state DOTs in Minnesota, California,
and Wisconsin do not consider price data for projects worth less
Data Processing than $100,000. Colorado DOT removes all bid data in which unit
prices are below and above the 5th and 95th percentile, respec-
Segmentation tively, for a given item category (Jeong et al. 2017). Although
As mentioned previously, the goal of this research is to generate FHWA does not detail their outlier detection process, it has been
highway construction cost indices across different materials (PCC documented that the federal government approves of the outlier de-
and AC) and activities (general construction and M&R). Although tection technique used in Colorado. The Ohio DOT compares the
the data sourced from Oman BidTabs differentiate between PCC- deviation of all unit prices in a given time period with that of the
and AC-relevant activities, additional steps are required to do the median (Jeong et al. 2017). Swei et al. (2017) noted that the dom-
same between general construction and M&R. Because the Oman inant source of variation for unit-price bid data is related to the size
BidTabs data set includes an item type, this study exploits this fact of the project and resulting economies of scale. As a result, the

© ASCE 04021030-4 J. Manage. Eng.

J. Manage. Eng., 2021, 37(4): 04021030

 1. Data Collection
• Collect Oman BidTabs data from 47 states for both general highway and
M&R activities

2. Data Processing
a. Segmentation
• Segment PCC and AC construction activities
• Segment general construction and M&R activities according to item type

b. Outlier Detection
• For each bid item, track winning unit-price, p1, and second lowest bid, p2
• Detect outliers by finding activities that do not meet the below condition: 1
– α < p1 / p2 < 1/(1 – α)
Downloaded from ascelibrary.org by The University of Queensland Library on 03/31/22. Copyright ASCE. For personal use only; all rights reserved.

• Discard outliers

c. Item Basket Selection

• Identify key phrases across 5,000+ item descriptions, which form basis for
item baskets
• Track the most common measurement unit if multiple are used
• Categorize data into item baskets
3. Generate Indices
• Conduct chained Fisher computation for each index per:

4. Time-Series Analysis

• Transform indices into log space

• Test for stationarity and serial correlation in log indices
• Estimate statistical model
• Conduct inference to comment on long-run trends and possible seasonality

Fig. 1. Methodology employed to generate and analyze disaggregate indices.

proposed outlier detection method is able to capture a range of cost indices. As highlighted in the “Introduction” section, the
project types and does not necessarily restrict itself to a certain high- chained Fisher index is regarded as the strongest evenly weighted
way construction project size. average of the Laspeyres and Paasche indices; the Fisher index
minimizes substitution bias, and the effect of chaining captures
Item Basket Selection the shift in item quantities over time. The computation of a chained
Once the data have been filtered appropriately, it is critical to select Fisher index requires three values as shown in Eqs. (6)–(8). First,
item baskets that generate sufficient data across the predetermined the baseline index value, C0 , at the start of the analysis must be set
number of periods. Although state DOTs generally use broad terms to a constant. As shown in Eq. (6), C0 is set equal to 100 in order to
to classify their item baskets, such as earthwork, aggregate, or be consistent with current construction cost indices. Second, Rt , the
asphalt, more specific item categories must be defined for the de- relative shift between consecutive time periods in the chained
velopment of material-specific highway construction cost indices. Fisher index, Ct , follows the same process as Eq. (3), but using
This study uses a phrase frequency analyzer by Adamovic (2009) values from time t and t–1. Here, p and q represent the unit-price
to determine the most commonly occurring key phrases across and quantities for the ith item basket out of the n total baskets.
more than 5,000 item descriptions. These item descriptions can be The chaining process is shown in Eqs. (7) and (8)
further consolidated by comparing, storing, and merging items with
similar descriptions. The reason why there are many variants of the C0 ¼ 100 ð6Þ
same item description is that state DOTs will oftentimes vary in
their terminology to describe similar tasks and items. The phrase sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
Pn Pn ffi
t t−1 t t
frequency analyzer by Adamovic (2009) enables the quick determi- i¼1 pi qi i¼1 pi qi
Rt ¼ Pn t−1 t−1
× Pn t−1 t
ð7Þ
nation of similar item descriptions and their grouping into a single i¼1 pi qi i¼1 pi qi
item basket. To ensure consistency of units, the predominantly oc-
curring unit for each item basket is identified, and an item is catego-
Ct ¼ Ct−1 × Rt ð8Þ
rized only if its units match. The proposed item basket classification
approach differs from FHWA’s NHCCI, which classifies the compo-
nents of highway construction into 29 categories. Time-Series Analysis
Generation of Indices As mentioned previously in this paper, construction cost indices
After processing the available data, the chained Fisher approach have been widely used within the management and construction
was used to develop the four disaggregate highway construction literature to forecast capital requirements and future expenditures.

© ASCE 04021030-5 J. Manage. Eng.

J. Manage. Eng., 2021, 37(4): 04021030

 Many of these studies have relied on univariate time-series methods

200
Government Expenditures (Billions of $'s)
(e.g., Ashuri and Lu 2010; Ilbeigi et al. 2017; Moon et al. 2018) to
characterize cost trends and ultimately forecast future values. This
paper similarly relies on these types of time-series methods to sup-

150
port its case application: the projection of future highway construc-
tion expenditures across the US.
Principal to the development of time-series models is the
assumption of (1) stationarity (i.e., constant statistical properties),

100
and (2) uncorrelated residuals across time (Box et al. 2008).
Because the generated time series used in this research trend up-
ward, it is assumed that each index either follows a trend-stationary
Downloaded from ascelibrary.org by The University of Queensland Library on 03/31/22. Copyright ASCE. For personal use only; all rights reserved.

50
(TS) or difference-stationary (DS) process. As shown in Eq. (9), a
TS model assumes that the previously defined cost indices at time t, 2005 2008 2011
Year
2014 2017

Ct , in their log form are composed of both a deterministic drift, μ,

and time trend component, β. The residuals, εt , can follow an au- Capital
Total
Operations and Maintenance

toregressive moving-average (ARMA) model to account for pos-

sible serial correlation
Fig. 2. Historical government expenditures on highways in nominal
ln Ct ¼ μ þ βt þ εt ð9Þ dollars. (Data from CBO 2018.)
If the statistical properties (in particular, variance) of Eq. (9) are
nonstationary across time, a model can be fitted to the differenced
data set, Δ ln Ct . Differencing the time series and fitting a model to lag 4 to both comment on possible seasonality and include it in the
the resulting DS process is a technique commonly employed in final model specification.
practice to deal with the issue of nonstationarity. Per Eq. (10),
the DS model includes only a stochastic drift term, μ 0 , and resid-
uals, εt0 , which can similarly be structured as an ARMA process. Case Application: Forecasting Future Government
The estimate of β in Eq. (9) or μ 0 in the case that the time series Highway Expenditures
should be modeled per Eq. (10) reflects the likely upward trend in Federal, state, and local governments in the US spent $177 billion
nominal construction costs over time in 2017 to manage their extensive highway system (CBO 2018).
Approximately 53% of that spending was allocated to major capital
Δ ln ln Ct ¼ μ 0 þ εt0 ð10Þ
investments (i.e., $94 billion), and the remaining 47% was appro-
priated to maintenance and operations (i.e., $83 billion). Fig. 2 plots
To test for whether the indices follow a TS or DS model, this
capital investments, maintenance and operations expenses, and total
study relies on the augmented Dickey-Fuller (ADF) test, a common
highway expenditures across all three levels of government between
approach used in research (e.g., Wong and Ng 2010; Ashuri and Lu
2005 and 2017 in nominal dollars (CBO 2018). Although there has
2010; Shahandashti and Ashuri 2016; Faghih and Kashani 2018)
been an increase in maintenance and operations spending since
to determine the structure of a time series. As can be observed in
2005, capital investments have remained fairly consistent in nominal
Eqs. (9) and (10), each data set is modeled in log space. There are
dollars. A challenge facing policymakers as part of their appropria-
multiple advantages when analyzing the logarithm of a time series;
tion process is the determination of future investment levels required
in addition to linearizing possible trends, taking the natural loga-
to effectively manage the existing highway system.
rithm of a series allows one to exploit the mathematical property
The generated cost indices from this paper can be used to ad-
described in Eq. (11) when an arbitrary value, x, is sufficiently
dress this challenge by assuming that the total cost, TC, to generate
small (i.e., much less than 1). In other words, estimates of β for
a certain amount of capital investments at two time periods, a base
Eq. (9) or μ 0 in the case of Eq. (10) approximately characterize
period (i.e., t ¼ 0) and some future time period, t, as follows:
quarterly cost growth. Estimates of quarterly cost growth, q, can
be viewed from a year-over-year perspective, y, through simple TCc0 ¼ β c0 Qc0 ð13Þ
compounding per Eq. (12)

ex ≈ 1 þ x ð11Þ TCct ¼ β ct Qct ð14Þ

where c simply denotes that all values are related to the production
y ¼ ð1 þ qÞ4 − 1 ð12Þ of major capital investments; Q = total amount of production;
β = unit-cost of production; and 0 and t index the time period
To determine the correct ARMA structure for the residuals, this of these calculations. A similar cost function can be generated for
research analyzes their autocorrelation function (ACF) and partial operations and maintenance in which the only difference between
autocorrelation function (PACF) to characterize the correlation be- Eqs. (13) and (14) and Eqs. (15) and (16) is the use of m rather than
tween different time lags. The major advantage of developing quar- c to denote operations and maintenance
terly, rather than annual, indices is that they can be leveraged to
characterize possible seasonality in the construction market that TCm m m
0 ¼ β 0 Q0 ð15Þ
would be beneficial information for decision makers. Should sea-
sonality exist, one would expect that there would be significant TCm m m
t ¼ β t Qt ð16Þ
correlation between data points four time periods apart from one
another. Because the PACF summarizes the relationship between Based on the findings of Sveikauskas et al. (2016, 2018), a rea-
current and previous outputs with the effect of intermediate lags sonable assumption is that there has been no significant change in
removed, the authors inspect the PACF for large correlations at the productivity of highway construction firms and this trend will

© ASCE 04021030-6 J. Manage. Eng.

J. Manage. Eng., 2021, 37(4): 04021030

 continue into the future. If firms are not becoming more efficient differences in outlier detection and item categorization used by
over time, one would expect that the unit cost of production would FHWA.
grow at some inflation rate that captures changes in labor, equip- One key challenge as part of this analysis is that the authors are
ment, energy, and material input prices and quantities over time. In unable to access the specific highway construction activities that
other words, the historical rate of inflation for highways would re- the Congressional Budget Office (CBO) has classified as “capital”
flect in this paper’s chained Fisher indices. Forecasts of year-over- investments and “operations and maintenance.” For the purpose of
year unit-price increases, y, would be based on stationary estimates this analysis, the authors assume that the general construction in-
via the time-series analysis [i.e., Eq. (12)]. Based on the preceding dices capture cost growth for capital investments and the M&R in-
description, the unit-cost of production, β, at time period t follows dices characterize inflation growth for operations and maintenance.
A second challenge is that the disaggregated indices segment cost
β ct ¼ β c0 ð1 þ yc Þt ð17Þ growth by not only activity but also material. In an ideal setting,
Downloaded from ascelibrary.org by The University of Queensland Library on 03/31/22. Copyright ASCE. For personal use only; all rights reserved.

one would conduct this analysis with knowledge around total

βm m m t
t ¼ β 0 ð1 þ y Þ ð18Þ spending and future output across different materials. To deal with
this issue, a set of capital investments and operations and mainte-
where c and m = major capital investments and operations and nance cost indices are created. These indices are computed by con-
maintenance, respectively. ducting the same process as was done for the aggregate cost index
Policymakers may seek to increase or decrease future produc- but now only doing so for relevant activities (e.g., relying on the
tion Q as they balance appropriating funds across competing needs. PCC highway M&R and AC highway M&R indices for operations
If one assumes that δ represents the annual rate that policymakers and maintenance).
aim to increase or decrease total output within the highway sector,
then
Results
TCc0
Qct ¼ Qc0 ð1 þ δ c Þt ¼ ð1 þ δ c Þt ð19Þ
β c0
Data Processing
TCm 0 Using bid data from across the US, this paper developed four in-
Qm m m t
t ¼ Q0 ð1 þ δ Þ ¼ ð1 þ δ m Þt ð20Þ
βm
0
dices for submarkets within the highway paving sector: (1) AC
highway construction, (2) PCC highway construction, (3) AC high-
As a result, the total cost to produce a certain amount of major way M&R, and (4) PCC highway M&R. Table 2 presents the sen-
capital investments as well as operations and maintenance in the sitivity of the outlier detection processing for values of α ranging
future are shown in Eqs. (22) and (23). Eq. (21) demonstrates from 0.3 to 0.5 for both the AC and PCC data sets. With α set equal
the intermediate steps required to generate each cost function to 0.4 (i.e., the second-lowest bid must be no more than 67% higher
than the winning bid), more than 725,000 entries are preserved
TC0
TCt ¼ β t Qt ¼ β 0 ð1 þ yÞt ð1 þ δÞt ¼ TC0 ð1 þ yÞt ð1 þ δÞt across both materials, ensuring that there is sufficient data to gen-
β 00 erate robust results. Although not presented in this section, the re-
ð21Þ sulting indices were consistent across these three different values
of α.
Table 3 subsequently provides summary information around
TCct ¼ TCc0 ð1 þ yc Þt ð1 þ δ c Þt ð22Þ
the final item basket coverage. The total monetary value of trans-
actions used to generate these four indices exceeds $170 billion.
TCct ¼ TCm m t m t
0 ð1 þ y Þ ð1 þ δ Þ ð23Þ Although the FHWA’s NHCCI is composed of approximately 30
item baskets, this paper refines those item baskets into upward
In other words, under the assumption of no productivity gains, of 160 instances in the case of the AC construction index. This
the total cost to generate a certain amount of output in the future is a further refinement allows the proposed indices to capture more
function of three variables: (1) the current amount of expenditures, granular price and quantity shifts across time than existing aggre-
TC0 , which is known per Fig. 2; (2) the rate of inflation to produce gate indices.
a type of highway, y, which is based on Eq. (12); and (3) the annual
increase or decrease in total output, δ. This case study evaluates two
scenarios: (1) policymakers aim to maintain both capital investment
Table 2. Percentage of total costs covered for different values of α
as well as maintenance and rehabilitation output at their current
levels (i.e., δ ¼ 0); and (2) capital investment output remains Number of data
unchanged whereas maintenance and rehabilitation output in- Data set α ¼ 0.3 (%) α ¼ 0.4 (%) α ¼ 0.5 (%) entries (α ¼ 0.4)
creases by 1.7% per year so that in 2030, it is 25% higher than AC 76 83 88 624,000
it was in 2017. PCC 71 80 86 110,000
The goal of this analysis is to compare anticipated highway
expenditures by 2030 when using disaggregate measures of high-
way cost growth rather than a single aggregate measure that
assumes that the cost of both capital investments and operations Table 3. Summary of item basket selection
and maintenance inflate at the same rate. This paper creates an ag-
Number of
gregate cost index to benchmark the analysis results by (1) tracking
Cost index item baskets Total value ($)
relative changes in each quarter for all four indices; (2) weighting
those relative changes by the amount of spending in that quarter for (1) AC highway construction 160 $124 billion
each index; and (3) computing a weighted change of aggregate (2) PCC highway construction 38 $24.5 billion
costs. Although one could rely on FHWA’s NHCCI to benchmark (3) AC highway M&R 84 $18.3 billion
(4) PCC highway M&R 22 $5.57 billion
this analysis, the authors have selected this approach given the

© ASCE 04021030-7 J. Manage. Eng.

J. Manage. Eng., 2021, 37(4): 04021030

 Visually, there is a distinct trend difference between the AC
200
indices (solid) and the PCC indices (in gray) in Fig. 3, suggesting
that AC highway construction has exhibited higher cost growth
Highway Cost Index (2005Q1 = 100)

than PCC highway construction over the last 13 years. Although

not shown in this figure, the AC highway construction index tracks
150

similarly to FHWA’s NHCCI, which is anticipated given that it con-

stitutes a larger relative portion of state expenditures. The PCC
highway construction index also experiences higher volatility
than AC highway construction index, which may be attributed
100

to its lower sample size. The clear differences between the AC

and PCC highway construction indices demonstrate that aggregate
Downloaded from ascelibrary.org by The University of Queensland Library on 03/31/22. Copyright ASCE. For personal use only; all rights reserved.

construction indices may conceal valuable information for decision

makers that only become apparent when leveraging disaggregate
50

2005 2008 2011 2014 2017 indices.

AC Highway Construction PCC Highway Construction
For the most part, both the AC highway M&R and PCC high-
way M&R indices in Fig. 4 have tracked their material-specific
highway construction cost index. This outcome is somewhat sur-
Fig. 3. AC (solid) and PCC (gray) highway construction indices. prising given that M&R activities have likely experienced much
lower productivity growth than new highways and reconstruction.
Empirical research, including that related to construction (Swei
2018), has suggested that activities with low productivity growth
tend to experience high cost growth. Although this paper provides
possible hypotheses around this outcome in the “Discussion” sec-
250

tion, the structure of the current data set limits one’s ability to char-
Highway Cost Index (2005Q1 = 100)

acterize the exact mechanism leading to this result. This, however,

200

should not detract from the value of this work but rather point to-
ward the importance of disaggregate indices in discovering new
150

insights to motivate future research within the management and

construction communities.
100

Time-Series Analysis
50

Table 4 summarizes the results of the time-series analysis. Based on

the ADF test, each time series is classified as a TS process given
that each time series rejects the null hypothesis that the indices
0

2005 2008 2011 2014 2017 follow a DS model. Table 3 details the parameter estimates for
AC Highway M&R PCC Highway M&R the deterministic drift (μ) and trend (β) components of the model,
as well as the ARMA structure of the residuals. Values in paren-
theses are the p-values for parameter estimates. All parameters as
Fig. 4. AC (solid) and PCC (gray) highway M&R indices. part of the residual structure reject the null hypothesis that they are
equal to zero at the 5% level.
All four indices exhibit a statistically significant (at the 5%
level) positive secular cost growth trend according to the estimates
Generation of Indices
of β. Based on the PACF of the residual structure, the PCC highway
Figs. 3 and 4 plot the material-specific highway construction and construction index is the only time series that exhibits any signifi-
M&R indices. One should be specifically interested in comparing cant seasonality. This result is accounted for by incorporating a lag
trends between the following relevant pairs of indices throughout 4 parameter estimate (0.37) within the residual structure. Poten-
the “Results” section: (1) AC highway construction versus PCC tially the seasonality for PCC highway construction may be tied
highway construction, (2) AC highway construction versus AC to the limited sample size. Future research should explore if other
highway M&R, and (3) PCC highway construction versus PCC mechanisms (e.g., types of projects conducted in certain seasons)
highway M&R. These three key comparisons provide a more holis- are causing this seasonality for the index.
tic understanding of construction market conditions and variations Table 5 converts the time-trend parameter estimates, β, into
in cost growth across construction types. approximate annual cost growth rates for each index. With a

Table 4. Estimated time-series models for each index

Data set TS or DS μ β Residual structure Seasonality
(1) AC highway construction TS 4.82 (<0.001) 8.69 × 10−3 (<0.001) εt ¼ 0.76εt−1 þ ut No
(2) PCC highway construction TS 4.71 (<0.001) 5.25 × 10−3 (0.014) εt ¼ 0.37εt−4 þ ut Yes
(3) AC highway M&R TS 4.84 (<0.001) 6.31 × 10−3 (<0.001) εt ¼ 0.47εt−1 þ ut No
(4) PCC highway M&R TS 4.85 (<0.001) 6.66 × 10−3 (0.004) εt ¼ 0.44εt−1 þ ut No
TS model: ln ln Y t ¼ μ þ βt þ ϵt
Note: ut = white-noise term following a Gaussian distribution with a mean of zero.

© ASCE 04021030-8 J. Manage. Eng.

J. Manage. Eng., 2021, 37(4): 04021030

 Table 5. Equivalent median annual cost growth rate according to the time aggregate construction are slightly below cost growth for the AC
trend β from Table 4 highway construction index (3.5%). Similarly, cost growth for the
Time trend β Approximate annual operations and maintenance index is identical with that of the AC
Data set (per Table 4) cost growth rate (%) highway M&R index (2.6%).
(1) AC highway construction 8.69 × 10−3 3.5
Fig. 5 presents anticipated expenditures in 2030 for two scenar-
(2) PCC highway construction 5.25 × 10−3 2.1 ios: output for capital investments as well as operations and main-
(3) AC highway M&R 6.31 × 10−3 2.6 tenance remains static between 2017 and 2030 (i.e., Scenario 1);
(4) PCC highway M&R 6.66 × 10−3 2.7 and (2) output for operations and maintenance in 2030 is 25%
higher than it was in 2017, reflecting the growing emphasis of de-
cision makers to preserve their highway system (i.e., Scenario 2). In
both scenarios, anticipated expenditures in 2030 are lower when
year-over-year increase of 3.5%, the AC highway construction
Downloaded from ascelibrary.org by The University of Queensland Library on 03/31/22. Copyright ASCE. For personal use only; all rights reserved.

relying on disaggregate price growth measures rather than a single

index has historically grown much more rapidly than the PCC high- aggregate inflation rate. According to the CBO (Fig. 2), relative
way construction index (2.1%). When considering the effects of government spending on highway operations and maintenance
annual compounding, the cost trend difference between AC and gradually increased from 37% in 2005 to 47% in 2017. During that
PCC highway construction becomes much more apparent. As a re- same time period, highway M&R costs increased at a slower rate
sult, both agencies and policymakers should account for this dis- than major capital investments. As decision makers shift their fiscal
crepancy in their decision making. As mentioned previously, in resources toward system preservation activities, the aggregate cost
comparing the highway construction and highway M&R indices, growth measure is unable to fully account for this reality. In turn,
one may anticipate higher cost growth for the latter due to its lower this fact partially causes the aggregate inflation measure to antici-
productivity growth. This expectation seems to hold true for PCC pate higher expenditures in the future than the model that relies on
paving activities: the approximate annual cost growth rate for the disaggregate indices.
PCC highway M&R index is 2.7%, above the 2.1% estimate for the
PCC highway construction index. Conversely, however, the year-
over-year cost growth rate for AC highway M&R is 2.6%, much Conclusions and Discussion
lower than the 3.5% rate for AC highway construction and running
counter to this initial assumption. Construction cost indices are a fundamental tool in many applica-
tions, such as cost estimation, capital budgeting, and measuring
productivity. By developing AC and PCC indices for both highway
Case Application: Forecasting Future Government construction and M&R, this research has created more granular cost
Highway Expenditures indices to improve the management of the highway system. The
As described in the “Methodology” section, the construction stark difference between AC and PCC highway construction high-
cost indices are weighted to generate measures of cost growth lights the importance of investigating specific submarkets within
for capital investments, operations and maintenance, and aggregate the paving sector. The four indices provide the insights and the
highway construction with terminology consistent with the CBO. A tools to track the prices of two predominant paving materials in the
time-series analysis of these indices for this case analysis indicate context of highway construction as well as highway M&R.
year-over-year inflation rates of 3.3% (capital investments), 2.6% Overall, because projects in the paving sector are typically large
(operations and maintenance), and 3.2% (aggregate construction), and expected to last for decades, even small percentage differences
respectively. These values make intuitive sense given that the ma- in year-over-year cost growth will have considerable long-term
jority of expenditures in the data set per Table 3 are for AC con- effects on the planning and maintenance of facilities. These small
struction. Year-over-year inflation rates for capital investments and percentage differences are also important considerations in the
study of productivity, where a 1% change in productivity is gen-
erally viewed as quite significant. Therefore, a more detailed under-
standing of cost growth across different types of activities will
promote more confidence in decision making. The national high-
way construction cost indices can provide helpful comparisons
with state-level data, and they can also enhance the accuracy of
project-level cost forecasts given the contrasting trends between
AC and PCC highway construction. Through the case study appli-
cation, this research highlights the value of disaggregate indices in
planning future budgeting requirements.
A study of the approximate long-run trends via time-series
analysis concluded that AC highway construction costs, with a
year-over-year increase of 3.5%, have experienced greater cost
growth than PCC highway construction costs, which increase at
2.1%. Comparing between maintenance and overall construction,
Scenario 1 Scenario 2
PCC highway M&R costs have grown notably faster at 2.7% year-
Disaggregate Indices Aggregate Indices over-year. The PCC cost indices thus confirm the initial expectation
of the effect of low labor productivity growth tasks experiencing
rising costs. However, in the case of AC highway construction,
M&R costs have actually evolved notably slower than general
Fig. 5. Anticipated expenditures in 2030 for Scenario 1 (no change in
construction at only 2.6%.
production) and Scenario 2 (increase in operations and maintenance
One potential explanation for this outcome could be the
production by 25%) using disaggregate and aggregate indices.
substantial difference in material price growth between these

© ASCE 04021030-9 J. Manage. Eng.

J. Manage. Eng., 2021, 37(4): 04021030

 two construction types. Between 2005 and 2017, average annual Ashuri, B., and J. Lu. 2010. “Time series analysis of ENR construction cost
price growth for asphalt and cement, the two cost dominant mate- index.” J. Constr. Eng. Manage. 136 (11): 1227–1237. https://doi.org
rials for AC and PCC highway construction, has been 3.7% for /10.1061/(ASCE)CO.1943-7862.0000231.
asphalt (BLS 2018a) and 2.3% for cement (BLS 2018b). Over that BLS (US Bureau of Labor Statistics). 2018a. “BLS Data Viewer: PPI in-
dustry data for Asphalt paving mixture & block manufacturing, not sea-
same time period, annual wage growth across the entire construc-
sonally adjusted.” Accessed June 24, 2019. https://beta.bls.gov
tion industry was 2.4% (BLS 2018c). M&R activities tend to (1) be /dataViewer/view/timeseries/PCU324121324121.
more labor-intensive, and (2) require less materials than general BLS (US Bureau of Labor Statistics). 2018b. “BLS Data Viewer: PPI in-
highway construction. Therefore, although AC M&R may have ex- dustry group data for Cement and concrete product manufacturing, not
perienced high labor cost growth due to low productivity growth, seasonally adjusted.” Accessed June 24, 2019. https://beta.bls.gov
its overall cost growth could have been lower relative to general /dataViewer/view/timeseries/PCU3273--3273--.
construction due to its limited reliance on a material that has wit- BLS (US Bureau of Labor Statistics). 2018c. “BLS Data Viewer: Average
Downloaded from ascelibrary.org by The University of Queensland Library on 03/31/22. Copyright ASCE. For personal use only; all rights reserved.

nessed substantial price growth this last decade. Unfortunately, this hourly earnings of all employees, construction, seasonally adjusted.”
paper is unable to access with its data set the relative contribution of Accessed June 24, 2019. https://beta.bls.gov/dataViewer/view
materials and labor for each individual index to evaluate the validity /timeseries/CES2000000003.
of this plausible explanation, highlighting one of the key gaps and Boskin, M. J., E. R. Dulberger, Z. Griliches, R. J. Gordon, and D.
Jorgensen. 1996. Toward a more accurate measure of the cost of living.
opportunities to enhance this work in the years ahead. Future re-
Washington, DC: Advisory Commission to Study the Consumer Price
search may aim to collect further refined price data from individual Index for US Congress.
firms to understand changes in the relative contributions of materi- Box, G. E. P., G. M. Jenkins, and G. C. Reinsel. 2008. Time series analysis.
als, labor, and capital equipment for these developed indices. Hoboken, NJ: Wiley.
The creation of these new highway construction cost indices Caves, D. W., L. R. Christensen, and W. E. Diewert. 1982. “The economic
provide tremendous opportunities for future research. In the imme- theory of index numbers and the measurement of input, output, and pro-
diate future, research should focus on understanding the impact of ductivity.” Econometrica 50 (6): 1393–1414. https://doi.org/10.2307
evolving design standards, procurement models, and policies used /1913388.
by state DOTs on highway cost growth. Such an analysis may ben- CBO (Congressional Budget Office). 2018. Public spending on transpor-
efit by further refining costs at a state level, which could lead to tation and water infrastructure, 1956 to 2017. Washington, DC: CBO.
a more detailed understanding of the effect of contextual factors Creedy, G. D., M. Skitmore, and J. K. W. Wong. 2010. “Evaluation of risk
(e.g., rural versus urban construction) on costs. In addition, there factors leading to cost overrun in delivery of highway construction proj-
ects.” J. Constr. Eng. Manage. 136 (5): 528–537. https://doi.org/10
are important cost contributors, such as work-zone closure controls,
.1061/(ASCE)CO.1943-7862.0000160.
that have not been adequately incorporated in this study. Practi-
Dalton, K. V., J. S. Greenlees, and K. J. Stewart. 1998. “Incorporating
tioners would benefit greatly from research aimed at capturing a geometric mean formula into the CPI.” Monthly Labor Rev.
these costs. Future work should quantify the implications of certain 121 (10): 3–7.
data processing differences between this study and the FHWA Dearden, L., H. Reed, and J. V. Reenen. 2006. “The impact of training on
(e.g., the utilization of a larger number of item baskets) on the productivity and wages: Evidence from British panel data.” Oxford Bull.
resulting indices. Finally, it is common in creating a price index J. Econ. Stat. 68 (4): 397–421. https://doi.org/10.1111/j.1468-0084
to capture possible quality improvements via a hedonic approach. .2006.00170.x.
Should design standard changes improve the average quality and Diewert, E. 1992. “Fisher ideal output, input, and productivity indexes re-
performance of highways, those enhancements should be properly visited.” J. Productivity Anal. 3 (3): 211–248. https://doi.org/10.1007
captured in the underlying cost index. /BF00158354.
Diewert, W. E. 1976. “Exact and superlative index numbers.” J. Econom.
4 (2): 115–145. https://doi.org/10.1016/0304-4076(76)90009-9.
Diewert, W. E. 2004. “Basic index number theory.” In The consumer price
Data Availability Statement index manual: Theory and practice, 370–402. Geneva: International
Monetary Fund.
Some or all data, models, or code generated or used during Faghih, S. A. M., and H. Kashani. 2018. “Forecasting construction material
the study are available in a repository or online in accordance with prices using vector error correction model.” J. Constr. Eng. Manage.
funder data retention policies: https://doi.org/10.5683/SP2 144 (8): 04018075. https://doi.org/10.1061/(ASCE)CO.1943-7862
/SHMTAG. .0001528.
Ilbeigi, M., D. Castro-Lacouture, and A. Joukar. 2017. “Generalized autor-
egressive conditional heteroscedasticity model to quantify and forecast
Acknowledgments uncertainty in the price of asphalt cement.” J. Manage. Eng. 33 (5):
04017026. https://doi.org/10.1061/(ASCE)ME.1943-5479.0000537.
Financial support for this research was provided by the Natural International Labour Office, OECD, International Monetary Fund, the
Sciences and Engineering Research Council (NSERC) of Canada World Bank, the United Nations Economic Commission for Europe,
and Statistical Office of the European Communities. 2004. Consumer
Discovery Grant Program. The authors are grateful to Leo
price index manual: Theory and practice. Geneva: International Labour
Sveikauskas from BLS and Anuar Onayev from the University of Office.
British Columbia for their comments on previous drafts. The Jeong, H. D., D. D. Gransberg, and K. J. Shrestha. 2017. Advanced
authors are also thankful for the constructive feedback received methodology to determine highway construction cost index. Ames,
from four anonymous reviewers of this manuscript. IA: Institute for Transportation at Iowa State Univ.
Joukar, A., and I. Nahmens. 2016. “Volatility forecast of construction cost
index using general autoregressive conditional heteroskedastic
References method.” J. Constr. Eng. Manage. 142 (1): 04015051. https://doi.org/10
.1061/(ASCE)CO.1943-7862.0001020.
Adamovic, M. 2009. “Text analyzer, online utility.” Accessed June 27, Mehrotra, N., M. Turner, and J. Uribe. 2020. “Does the US have an infra-
2019. https://www.online-utility.org/text/analyzer.jsp. structure cost problem? Evidence from the interstate highway system.”
Allen, S. G. 1985. “Why construction industry productivity is declining.” In Economics of transportation in the 21st century fall 2020.
Rev. Econ. Stat. 67 (4): 661–669. https://doi.org/10.2307/1924811. Cambridge, MA: National Bureau of Economic Research.

© ASCE 04021030-10 J. Manage. Eng.

J. Manage. Eng., 2021, 37(4): 04021030

 Moon, S., S. Chi, and D. Y. Kim. 2018. “Predicting construction cost index Sveikauskas, L., S. Rowe, J. Mildenberger, J. Price, and A. Young. 2016.
using the autoregressive fractionally integrated moving average model.” “Productivity growth in construction.” J. Constr. Eng. Manage.
J. Manage. Eng. 34 (2): 04017063. https://doi.org/10.1061/(ASCE)ME 142 (10): 04016045. https://doi.org/10.1061/(ASCE)CO.1943-7862
.1943-5479.0000571. .0001138.
Moulton, B. J. 2018. The measurement of output, prices, and productivity: Swei, O. 2018. “Long-run construction cost trends: Baumol’s cost disease
What’s changed since the Boskin Commission? Washington, DC: and a disaggregate look at building material price dynamics.” J. Constr.
Hutchins Center of Fiscal and Monetary Policy, Brookings Institution. Eng. Manage. 144 (7): 04018058. https://doi.org/10.1061/(ASCE)CO
Nassereddine, H., G. Whited, and A. Hanna. 2016. “Developing a chained .1943-7862.0001511.
fisher construction cost index for a state highway agency.” Transp. Res. Swei, O. 2020. “Forecasting infidelity: Why current methods for predicting
Rec. 2573 (1): 149–156. https://doi.org/10.3141/2573-18. costs miss the mark.” J. Constr. Eng. Manage. 146 (2): 04019100.
Oman Systems Inc. 2019. “Bidding software for road construction contrac- https://doi.org/10.1061/(ASCE)CO.1943-7862.0001756.
tors.” Accessed May 20, 2019. http://omanco.com/index.php/product Swei, O., J. Gregory, and R. Kirchain. 2017. “Construction cost estimation:
A parametric approach for better estimates of expected cost and varia-
Downloaded from ascelibrary.org by The University of Queensland Library on 03/31/22. Copyright ASCE. For personal use only; all rights reserved.

/bidtabs/.
tion.” Transp. Res. Part B: Methodol. 101: 295–305. https://doi.org/10
Reinsdorf, M. 1998. “Formula bias and within-stratum substitution bias in
.1016/j.trb.2017.04.013.
the US CPI.” Rev. Econ. Stat. 80 (2): 175–187. https://doi.org/10.1162
Syverson, C. 2011. “What determines productivity?” J. Econ. Lit. 49 (2):
/003465398557375.
326–365. https://doi.org/10.1257/jel.49.2.326.
Shahandashti, S. M., and B. Ashuri. 2016. “Highway construction
Touran, A., and R. Lopez. 2006. “Modeling cost escalation in large infra-
cost forecasting using vector error correction models.” J. Manage.
structure projects.” J. Constr. Eng. Manage. 132 (8): 853–860. https://
Eng. 32 (2): 04015040. https://doi.org/10.1061/(ASCE)ME.1943-5479
doi.org/10.1061/(ASCE)0733-9364(2006)132:8(853).
.0000404. US Bureau of Economic Analysis. 2020. “Interactive data”. Accessed May
Shrestha, K. J., H. D. Jeong, and D. D. Gransberg. 2016. “Current practices 27, 2020. https://apps.bea.gov/itable/index.cfm.
of highway construction cost index calculation and utilization.” In Wang, Y., and M. Liu. 2012. “Prices of highway resurfacing projects in
Proc., Construction Research Congress, 351–360. Reston, VA: ASCE. economic downturn: lessons learned and strategies forward.” J. Man-
Shrestha, K. J., H. D. Jeong, and D. D. Gransberg. 2017a. Advanced meth- age. Eng. 28 (4): 391–397. https://doi.org/10.1061/(ASCE)ME.1943
odology to determine highway construction cost index (HCCI). Helena, -5479.0000094.
MT: Montana DOT. Wong, J. M. W., and S. T. Ng. 2010. “Forecasting construction tender price
Shrestha, K. J., H. D. Jeong, and D. D. Gransberg. 2017b. “Multidimen- index in Hong Kong using vector error correction model.” Constr. Man-
sional highway construction cost indexes using dynamic item basket.” age. Econ. 28 (12): 1255–1268. https://doi.org/10.1080/01446193
J. Constr. Eng. Manage. 143 (8): 04017036. https://doi.org/10.1061 .2010.487536.
/(ASCE)CO.1943-7862.0001307. Xu, J., and S. Moon. 2013. “Stochastic forecast of construction cost index
Sveikauskas, L., S. Rowe, and J. Mildenberger. 2018. “Measuring produc- using a cointegrated vector autoregression model.” J. Manage. Eng.
tivity growth in construction.” Monthly Labor Rev. 141 (1): 1–33. 29 (1): 10–18. https://doi.org/10.1061/(ASCE)ME.1943-5479.0000112.

© ASCE 04021030-11 J. Manage. Eng.

J. Manage. Eng., 2021, 37(4): 04021030