1
ADDING A FREIGHT NETWORK TO A NATIONAL
INTERSTATE INPUT- OUTPUT MODEL:
IMPLICATIONS FOR CALIFORNIA
Peter Gordon,
JoongKoo Cho,
James E. Moore,
JiYoung Park,
Harry W. Richardson, and
SungSu Yoon.
University of Southern California
July, 2009
2
ADDING A FREIGHT NETWORK TO A NATIONAL INTERSTATE INPUT- OUTPUT
MODEL: IMPLICATIONS FOR CALIFORNIA
Abstract
The state of the nation’s infrastructure is the subject of widespread discussion and comment
because it is thought to include many deteriorating and unsafe bridges. And ever since the
terrorist attacks of 9/ 11, there has been increasing concern over the extent to which an attack on
infrastructure could cause serious economic disruption. This research develops a model by
which the economic consequences of an attack on any major element of the highway network
can be analyzed. We do this by adding a freight network to a national multiregional economic
impact model and making freight traffic flows endogenous. We base our approach on the
National Interstate Economic Model ( NIEMO) and refer to the elaboration as TransNIEMO
( Transportation network and the National Interstate Economic Model). The new model enables
us to study the state- specific and industry- specific economic impacts of any significant changes
in the nature of highway freight movements. We tested the model for selected freight
movements in and out of California. The results are entirely plausible and encourage us to
elaborate and test the model for hypothetical disruptions of freight traffic throughout the U. S.
3
1. Introduction
The construction of canals and railroads in the 19th century connected the central lowlands of the
U. S. with the outer world, facilitating regional specialization, trade and economic growth and
establishing the U. S. as a major supplier of agricultural production to much of the world ( U. S.
Department of Transportation, Federal Highway Administration, 2004). A century later,
President Eisenhower saw the importance of a national inter- connected highway system.. Since
then the Interstate Highway System has served most of the nation’s freight movements,
facilitating continuing regional economic specialization and long- term the development of the
U. S economy.
Innovations in transportation technology and expansions of transport infrastructure bring
substantial changes ( Muller, 2004). Goods are now moved intermodally via diverse routes.
Estimating freight movements and shipping costs between regions is essential for making
investment plans throughout the economy. The same holds for investments by the various public
agencies that manage the highway network system.
Improved analysis is now possible because the Federal Highway Administration ( FHWA) has
integrated various shipments data into the Freight Analysis Framework ( FAF). The FAF contains
commodity flows among sub- state regions ( U. S. DOT, FWHA, 2006a). However, the FAF data
only provide origin- destination shipment data among these sub- state regions and exclude data on
the highway networks along which freight flows move. While FAF was the basis of important
capacity information, our operational multi- regional input- output model, the National Interstate
Economic Model ( NIEMO) was the source of interstate shipment demand data. Our research
plan was to use both tools to find ways to allocate commodity- specific interstate trade to the
national highway network. Any major flow disruption could then be diverted to second- best
routes, the costs of the diversion could then be estimated and NIEMO could be used to determine
a much fuller inventory of economic impacts.
To now there has been something of a divorce between two important branches of spatial
modeling: transportation ( and often land use) and economic impact analysis. Integration between
4
these two approaches is important because changes in economic activity have consequences for
transportation while changes in the transportation network have implications for economic
development.
This study is part of METRANS’ focus: We study commercial goods movement and
international trade in metropolitan areas and its physical economic cost impact in any case of
terrorist attack. Even though this research is initially a California experiment, the success of our
proof- of- concept application now makes it possible to broaden tests and applications of
TransNIEMO to the rest of the U. S.
While NIEMO is spatially disaggregated only to the state level, the transportation nodes for each
mode are the major metropolitan areas, which are the dominant centers of economic activity
( based on U. S Census employment data). Furthermore, in most states there is one or more major
metropolitan area: The non- metropolitan regions in selected cases also account for a high
proportion of state gross domestic product and freight OD movements.
In previous work ( Gordon et al. ( 2006), computing the indirect and induced effects of impacts
associated with capacity losses at the twin ports of Los Angeles- Long Beach showed that two-thirds
of the impacts leak outside the region. Without a model such as NIEMO, we would have
no idea of where those leakage effects might occur. In this sense, the current effort to integrate a
major transportation network system with NIEMO links interindustry transactions of freight
movements between the nation and California. We plan to extend the research to achieve a
parallel integration of the national and regional railway networks.
2. NIEMO
NIEMO is a 47- sector- 52 region input- output model that is fully operational. The idea for such a
model has a long history stretching back to Isard’s suggestion of the “ ideal interregional model”
( Isard, 1951, 1960) and Leontief’s valiant but failed attempt to operationalize a variant of the
model in the 1960s To say that NIEMO has succeeded where Leontief failed is not an immodest
5
statement, but rather a reflection of the improvements in databases and computing capacity over
the past thirty years. However, building bridges among the various data sources remained a
substantial task.
NIEMO is not merely a replica of the original design as conceived by Isard and Leontief. Rather,
NIEMO rests on the successful integration of state- level input- output information with data from
the Commodity Flow Survey ( CFS). Since 1993, the Commodity Flow Survey ( CFS) has been
the largest single nationwide data source for freight movement flows ( USDOT, Research and
Innovative Technology Administration & Bureau of Transportation Statistics, 2006b). The
Bureau of Transportation Statistics and the Census Bureau collect CFS data from a sample
survey of industries through the Economic Census. Although the CFS provides wide range of
commodity shipments and multimodal movement data with five- year cycle updates of 1997 and
2002, some user groups have not been satisfied with its content details when they apply the data
to their specific projects ( USDOT, RITA & BTS, 2005). The most commonly addressed
weaknesses of CFS are its incomplete coverage by commodity sectors and regional detail, and its
inability to fully capture imported goods trade ( Southworth, 2005; Park et al. 2007, Giuliano,
2007). Capelle ( 2000) showed that the CFS estimates only cover less than 75 percent of all the
freight tons moved annually in the U. S. because the survey drops many establishments classified
as farms, forestry, fisheries, construction, transportation, governments, foreign establishments,
services, and most retail activities.
Nevertheless, Park et al. ( 2007b) managed to construct interstate trade flows, applying an
Adjusted Flow Model and a Doubly- constrained Fratar Model. The approach depends on 1997
CFS and 2001 IMPLAN data. To reconcile different definitions and classifications of the
commodities among multiple data sources, they created a new commodity sector scheme,
referred to as the “ USC Sectors.” Several applications of the initial version of NIEMO
( excluding interstate service trade) show that the state- to- state trade flows and the flows between
the states and the rest of world for the 29 USC commodity sectors are all readily computable
( Park and Gordon, 2005).
6
3. Freight and Network Flows Information
Freight movements provide fundamental information on economic structure and
relationships among economic and geographic entities. This means that network activity and
other economic activity are integrated. They cannot be studied in isolation. Yet, this is what
most analysts do. The challenge has been to assemble the necessary information to facilitate
integrated modeling.
To overcome the shortcomings of CFS, efforts have been made to combine it with other
data sources. To further understand freight activity, the U. S. Department of Transportation
created the Freight Analysis Framework ( FAF) in 2002. FAF is a comprehensive database for
policy analysis ( U. S. DOT, FHWA, 2002a) that provides more comprehensive information on
freight flows by mode, based on the same sector codes as the CFS. The framework also
forecasts future freight activities. The USDOT’s continuous effort to improve the FAF data set
made it possible to release the latest version of FAF 2.2 in November 2006 ( U. S. DOT, FHWA,
2006a). The FAF 2.2 release includes origin and destination commodity flows among 138
integrated MSAs or equivalent sub- state regions on the basis of 2002 data expressed in tons and
dollar values by transportation mode and by Standard Classification of Transported Goods
( SCTG) sector ( U. S. DOT, FHWA, 2006b). Unfortunately, there are still discrepancies with
other data sources due to the inclusion of service values into the data for commodity sectors.
In addition to government efforts, various research groups have also tried to create better datasets.
Ham et al. ( 2005) integrated interregional, multimodal commodity shipments and transportation
network flows, formulating an alternative to the traditional four- step travel forecasting approach
of trip generation, trip distribution, mode choice, and network assignment. Based on 1993 CFS
data, the research was conducted for 11 commodities plus construction and service sectors.
Additionally, they constructed a network consisting of 167 nodes and 532 links useful for
analyzing the truck commodity flows based on the USDOT’s 1997 National Transportation Atlas
Database ( NTAD).
7
Giuliano et al. ( 2007) estimated Los Angeles intraregional freight flows combining
various sources on trade and transportation datasets. In addition to the 1997 CFS, the research
reconciled IMPLAN, WISERTrade, Caltrans’ Integrated Transportation Management Systems
( ITMS), Transportation Analysis Zone ( TAZ), Waterborne Commerce of the U. S. ( WCUS) and
airport data from RAND. Although this work was limited to the highway network within one
area, the results suggested the potential for expanding the approach to other metropolitan areas.
It is also important to note the availability of highway network data. Freight transportation is
overwhelmingly an interstate activity, accounting for 73 percent of total ton- miles ( U. S. DOT,
FHWA and BTS, 2002). The data for this research is primarily related to the highway .
However, rail, air, and water networks cannot be ignored ( Yu, 2006), and we lanln to address
integrating the other modes in future research. The National Highway Planning Network
( NHPN) has about 452,000 miles of roads, of which the Freight Analysis Framework ( FAF)
contains 245,500 miles. This includes 46,380 miles of Interstate Highways, 162,000 miles of
National Highway System ( NHS) roads, 35,000 miles of other national roads, and 2,125 miles of
urban streets and rural minor arterials.
8
4. Methodology
This application of TransNIEMO involves two major steps:
i. Estimation of increased costs due to a disaster based on the constructed
highway network system, and
ii. State- by- state economic impact analysis applying NIEMO due to decreased
household consumption, resulting from price increase of products shipped via
second- best routes.
The application of TransNIEMO starts with the estimation of increased costs on the highway
network system for a plausible scenario, e. g., destruction of a bridge. Figure 1 shows the
framework for our research model. Step- by- step methodological explanations for estimating
increased shipping costs follow. The test described here is for a one- year disruption. The
linearity of the models makes it possible to study any period.
We established the basic dataset by obtaining the national highway network data for Freight
Analysis Framework, available via download from the Federal Highway Administration
( FHWA). We pruned the small arterials and local by- ways because this reduced computational
burdens without sacrificing realism. Most truck drivers delivering freight minimize costs by
using higher design facilitiess such as the Interstate Highways ( for example, I5, I10, I405, etc.),
U. S. Route ( for example, US101, US395, etc.), or State Route ( for example, SR- 11, SR- 125,
etc.) to the maximum extent possible. To conduct the pruning procedure, we implemented
spatial selection using the GISDK function of TransCAD software. We then built a base national
map for this application that includes only the three major highway networks.
9
• I: Interstate highway, U: U. S. Route, and S: State Route.
Figure 1 TransNIEMO: Data and Modeling Process
10
Because the sectorization scheme used in NIEMO does not break out truck use information, we
established the proportions in the OD data correspond to the truck mode based on the FAF 2.2
data. Trade information from NIEMO was multiplied to the proper proportions and then truck
OD requirements computed. This was to exclude the services sector information in the FAF 2.2
OD pairs. Although FAF 2.2 involves 138 sub- state regions, we excluded 17 FAF international
gateways and 7 FAF foreign trade regions and considered only domestic sub- state regions. Our
examples involving California, Arizona, and Nevada thus do not include any freight gateways
( U. S. DOT, FHWA, 2006b). The sub- state regions are largely equivalent to Metropolitan
Statistical Areas ( MSAs) and remainder non- MSAs.
We generated geographical centroids for the 114 sub- state regions to combine the truck OD data
with a GIS map file. However, because the centroids do not reflect the locations of economic
activities associate with each region, we adjusted the geographic centroids using city- level
employment figures available from the U. S. Census Bureau. The adjusted economic centroids
better represent the economic activities in the MSAs and remainder areas.
One problem with representing the national economy with only 114 economically weighted
centroids is that this limits the number of relevant paths between each origin and all destinations
to only 114. This is extremely unrealistic and defeats some of the objectives of the research, i. e.,
identifying changes in freight flows affecting alternative routes and the cost and activity changes
that result. Further, applications of TransNIEMO present the risk of infeasible solution if
network disruptions are sufficient to isolate supply or demand locations.
This is not an unusual problem modeling user behavior in networks. The simplest
meaningful way to predict economic behavior is to assume perfect information and rationality on
the part of economic agents. This is usually a reasonable and useful approach, and is so here.
However, it is sometimes necessary to better account for limitations on travelers’ states of
information or the complexity of travelers objectives when predicting network flows. This calls
for a stochastic network equilibrium approach in which the probability of route choice varies
smoothly with route cost, rather than being represented as a binary outcome. In this case, the
11
problem is somewhat simplified because the distances involved dictate that line haul costs
dominate congestion costs.
To account for the relevance of multiple second shortest paths between sub- states regions, we
developed an approach for selecting and associating nodes representing physical intersections on
the highway network with OD pairs. Estimated shortest paths will generally pass through one or
more of these physical intersection nodes, the majority of which are in the vicinity of economic
centroids associated with MSAs. These locations are used to represent the sites at which
economic demand for transportation is imposed on the network. To select the intersection nodes
as starting points of truck freight shipments, we defined buffer- zone boundaries around
economic centroids, differentiated by the size of the sub- state region. If intersection nodes are
located within the defined boundary, we selected and labeled them “ network centroids.” These
greatest level of spatial detail associated with the transportation network.
Table 1 shows the approach to defining the buffer- mile boundaries for sub- state regions and the
number of network centroids by region. For the MSA regions, we applied 35- mile buffer
boundaries. For the reminder areas, we applied several criteria: 35, 50, 75, 100, and 200 miles of
buffer boundaries, combining geographic size of each MSA and mathematical analysis as
described in Table 1. For the special cases of Massachusetts and Maryland, we adapted 100- mile
boundaries because of peculiar geographic shapes.
< Table 1, here>
These criteria identify and intractably large number of network centroids, leading to a relatively
detailed network representation that is computationally unwieldy. To reduce the computing
burden, we randomly selected 10 percent of the network centroids associated with each
economic centroid. A study about generalized coefficients of buffer miles by region is on our
research agenda. Figure 2 shows that this approach effectively represents the original set of
network centroids. This 10 percent sample of network centroids can reasonably be expected to
define representative sets of alternative paths to accommodate freight flows between economic
centroids.
12
< Figure 2, here>
At the metropolitan level, it is standard practice to use an equilibrium cost flow model to predict
network flows. Applying such a minimum cost flow model requires supply and demand nodes,
link congestion functions, and shipping volumes among OD pairs. The OD volumes on the
network links are determined along with the equilibrium cost link flows. In the case of the FAF
OD data, however, the standard equilibrium cost flow model is inappropriate because interurban
and interstate line- haul costs dominate congestion costs. Finding shortest paths between OD
pairs becomes a combinatoric problem. In this research, we enumerate the number of paths
between two substate regions by defining paths between all possible network centroid pairs
involving the two regions. We distributed truck OD volumes from economic centroids onto
sampled network centroids using even weights of 1/ n, where n is number of enumerated paths
linking the two regions. The cumulative volume of freight so distributed on a highway is
considerably less than the capacity of the highway links.
To demonstrate our approach we simulated a plausible scenario: We eliminated a critical bridge
that plays a key role connecting California to adjacent states. Due to loss of the bridge, major
alternative highways have to be identified to accommodate the affected freight volumes. Critical
bridges on six major routes are shown in Figure 3: These are part of Interstate Highways 5, 8, 10,
15, 40, and 80 ( I5, I8, I10, I40, and I80, respectively).
< Figure 3, here>
The increased miles associated with shipping commodities before and after removing the bridge
from the network are used to estimate increased trucking costs for each industry sector. After
calculating truck costs per mile by industry sector, we multiplied these costs by the increased
miles travelled to estimate the increased cost in each industry. The increased shipping cost, in
turn, drives increased prices, and induces a decrease in consumer expenditures. Modeling this
effect is approached as follows. With the assumption that all increased truck costs are passed
forward as price increases, we utilized a supply- driven I- O model. Dietzenbacher ( 1997) showed
13
that the supply- driven I- O model is to the more meaningful formulation for estimating price
increases than the Leontief price I- O model when absolute costs in value- added sectors are
available. As a result, these increased prices lead to reduced final demands. NIEMO is then
applied to estimate the state- by- state economic impacts resulting from these reductions in final
demands
5. Application
The goals of this research were not merely to develop an appropriate database resource for
adding a transportation network for NIEMO ( TransNIEMO), but to test the empirical
implications of tying freight flows to physical infrastructure. In addition, our aim was to trace
the transportation and output impacts in California following the induced final demand changes
in any location within the national economy – as well as the transportation and output impacts in
each state following from final demand changes in California. In short, these dual objectives are
1) to better reconcile economic impact and transportation modeling, and 2) to integrate the
analysis of a regional economy ( in this case, California) and the national economy.
For future research, the number of potential applications for this approach is almost limitless. A
particularly useful example would involve Los Angeles- Long Beach ports, primarily because of
the spatial reach of their economic importance throughout the national ( and indeed the global)
economy, and their enormous transportation implications for both the regional and national
economy. There are many specific scenarios within this framework, but in the view of the long-term
trend involving trade deficits, an interesting example might be to model an exogenous
increase in imports balanced by an exogenous decrease in exports within individual sectors. We
could then explore the implications of such a scenario for both the national and regional highway
network.
However, largely because of its relevance to homeland security, we selected a scenario based on
eliminating access to highway bridges. We selected freight destinations in two states outside
California: Nevada and Arizona. In response to the disruption scenario for the I15, most of the
14
freight flows on I15 moved to I40. The bridge disruption scenario for the I80 caused most of the
freight flows on the highway move to an adjacent highway, U. S Route 50. Figures 4 and 5 show
these movements graphically. Alternative shortest paths ( coded 0_ 1 and 0_ 2) are highlighted in
the circled areas.
< Figure 4, here>
< Figure 5, here>
Table 3 shows the detailed changes in the dollar value of shipping on highways from California
to Nevada due to the disruption of flows that would otherwise be on I80 or I15. In Table 3,
columns 1 ( I15) and 10 ( I80) show large decreases in the number of available paths ( many paths
use these routes) and in the dollar value of shipping along these paths. Similarly, the alternative
routes identified in columns 0_ 1, 0_ 2, and 9 show increases in paths and shipping values.
< Table 3, here>
To complete a more detailed impact analysis, we performed calculations for the disruption
scenario on the I10 for the case of California and Arizona. This case is one of the possible
scenarios as shown in Table 2. Figure 4 describes the changes in alternative pathways that result
from the loss of a key bridge serving I10. The disrupted I10 route is coded as route 3 in the
highlighted circle. These results can be examined in detail in Table 4, which shows the changes
in the number of available paths and in the dollar values of shipping on highway shortest paths.
We calculated the entries in the columns labeled “# of paths” and “$ value” at the network cuts
where shortest paths cross these states’ borders. While the number of paths and dollar value
shipping associated with I10 decreased dramatically, State Highway 60 ( S60, route 4) received
most of this shipping volume and also accounted for increases in the number of paths in use.
< Table 4, here>
15
6. Results
If a major bridge linking California and Arizona on the I10 is eliminated, most shortest- path
travel would move to the adjacent S- 62 freeway. Based on this disruption scenario of I10 and
the diversion to shortest paths, we estimated the increased distance that the freight is shipped.
Table 5 shows increased miles between origins and destinations. The last column “ Change in
Miles,” shows the percentage increase in path miles, that is, % Change. The most impacted OD
pair is the “ Los Angeles to Phoenix,” combination, which shows an 11.88 percent increase in
path- miles.
< Table 5, here>
The increased distances were multiplied by the shipping cost per mile to estimate increased
shipping cost by USC Sector. The shipping cost per mile was calculated from the costs for
purchasing truck mode services as a fraction of the total input value in California, also by USC
sector. We used raw data from the 2001 IMPLAN commodity balance sheet for California to
estimate these values. This permits us to calculate the values in the columns “ Purchase of Truck
Service” ( PTS) and “ Total Input Value” ( TIV) in Table 6. These give ratios corresponding to the
shipping cost per value by industry, shown in the column “ Truck Cost per Value” ( TCV).
< Table 6, here>
The baseline shipping values using truck services are multiplied by the TCV ratios. This
produces the estimated truck service cost for each shipping value. Also, using baseline miles
between the network centroids in California and Arizona, we computed the miles of truck
services cost associated with each shortest path. The top three sectors with respect to average
truck service costs per mile are USC Sectors 16 ( Logs and other wood in the rough & Wood
products), 15 ( Plastics and rubber), and 8 ( Nonmetallic minerals). These sectors have average
truck service costs ranging from $ 53 to $ 119 per mile.
16
Baseline shipping values produce baseline shipping costs. Changing miles traveled due to the
disruption increases shipping costs. This perspective is most valid for a short- run analysis. The
final results are the changes in shipping costs obtained relative to the baseline conditions. These
values appear in the column labeled “ ! cost” in Table 7. Overall the increased shipping cost
relative to the baseline case was 0.16 percent for the I10 scenario, and ranged from 0.02 to 0.33
percent depending on the industry.
< Table 7, here>
In the short run, the increased shipping costs will be passed through as price increases. The
increased costs will boost product prices. Table 8 shows the increased prices by USC Sector due
to the increased shipping costs in the column labeled “ Amounts Resulted from price increases.”
[*** Should be “ Quantities Resulting from Price Increases.”]. These increased product prices
reduce expenditures by final users. We assumed that the increased prices will decrease the total
final demands by Arizona final users by a corresponding amount. Using the demand- driven
version of NIEMO, we calculated the state- by- state economic impacts resulting from the
decreases in final demands specific to Arizona. See the column labeled “ Decreases in
expenditure of Arizona.”. The total decrease in expenditures of $ 30 million accounted for $ 62
million in total output losses nationwide.
< Table 8, here>
Table 9 shows the associated state- by- state economic impacts. As expected, Arizona is the most
seriously affected state. The adjacent states, California and Texas, show over $ 1 million in
economic losses from the bridge services eliminated in the I10 scenario. The indirect economic
impacts are spread throughout the nation. Even though we estimated approximately $ 62 million
direct and indirect economic impacts based on the I10 bridge scenario, this is only a partial loss.
We also have to consider analysis of shipping movements to other destinations using the I10
route. An expanded area analysis will increase the economic impacts considerably, and state- by-state
impacts and may be very different.
17
< Table 9, here>
7. Conclusions and discussion
We have constructed the TransNIEMO prototype and run an initial test involving trade between
California and Arizona. This research examines transportation network and economic impact
analysis based on experimental scenarios. We specified an interstate network system based on
FHWA and FAF nation- wide freeway network data, and distributed freight demand from
economic centroids onto the network system. We calculated the state- by- state economic impacts
resulting from increased shipping costs accruing as a consequence of transportation service
disruptions associated with the elimination of key bridges.
Further extensions require further elaboration of the nation- wide network. This first step
experimental analysis of state- to- state economic impacts includes several restrictive assumptions.
However, once the nation- wide transportation network system model is completed, we can
estimate more precise impact results and results for scenarios involving network links in other
States. This will permit us to complete spatially disaggregate economic impact analyses for case
studies involving any natural disaster or hypothetical terrorist attacks on critical infrastructure.
The most important next step is to improve the model’s computability. This proof- of- concept
exercise shows that NIEMO can be extended into a TransNIEMO formulation and that the model
can be exercised to compute widely distributed economic impacts associated with local
disruptions in infrastructure services. However, the model is computationally intractable in its
present form. The example summarized here could be expanded to include a sub- national region
consisting of several states, but desktop computational resources are not sufficient to permit a
fully national implementation of the model.
The model’s complexity is apparent the formulation is compared to corresponding metropolitan
level models. Urban transportation planning models might include in the neighborhood of 3,000
traffic analysis zones differences, whereas at first inspection TransNIEMO appears to have only
18
114. At the urban level, zone centroids are virtual locations that are connected to network links
to provide a mathematical mechanism for loading transportation demand onto the network. Each
centroid is connected to a small number of physical links. In TransNIEMO, economic centroids
are the centers of gravity for national transportation demands associated with an MSA or similar
region. Relating the spatial identity of an economic centroid to entire metropolitan network is an
abstract objective. Imposing these demands on the network at physical locations corresponding
to a 10 percent sample of physical intersections still produces a highly redundant network in the
vicinity of economic centroid. This makes it unlikely that loss of discrete infrastructure capacity
in urban areas will produce significant changes in path flows, and this is a realistic result.
Certain exceptions apply, primarily bridges crossing wide bodies of water.
This level of redundancy has advantages but is likely unnecessary. The current model represents
urban networks in too much detail. This combined with a traffic assignment method that
requires pre- enumeration of shortest paths between numerous network centroids produces a
computation burden that can be substantially diminished. Exactly how is a question for
additional research, but the objective is certainly achievable. The high number of alternative
shortest paths being computed share many long- haul links in common. Consequently, there is
little to be gained from adding details to MSA- level networks once sufficient capacity has been
represented in these locations to ensure transshipment of interstate flows, or that identifies the
relatively small number of high design facilities essential to supporting transshipment flows.
The fact the research questions TransNIEMO is built to address include, principally, anticipating
and avoiding the costs of capacity losses rather than the standard metropolitan level objective of
prediction level of service as a function of fixed supply and increasing demand ensures a
relatively high set of computational requirements. TransNIEMO is to some extent a network
design exercise in which the decision is not to add capacity, but to protect existing capacity,
possibly by adding redundancy. Any application arena in which the physical configuration of a
large transportation network is being updated and flows redistributed will never be
computationally cheap to model.
19
Our initial efforts to specify TransNIEMO focused on data availability, use, and reconciliation
and the higher level question of how to translate increased transportation costs into disaggregate
economic impacts. Once we were convinced we had sufficient computing capacity to allow us to
test the basic framework for approach and provide a proof of concept, computational efficiency
took a back seat to the larger question of how transportation costs filter through an economy
represented in spatial detail. As a result, the current research model can be subjected to
considerable computational improvement, and this will allow it to be used more readily for
analysis of a wider array of alternatives.
20
Table 1. Buffer radius by size of remainder areas
State Area size Miles
Number of
Network centroid
( 10% sample)
State Area size Miles
Number of
Network centroid
( 10% sample)
RI 1,045 35 13 VA 28,094 75 58
NJ 1,153 35 37 NY 29,149 75 106
DE 1,981 35 13 ME 31,173 75 13
CT 2,738 35 43 TN 33,793 75 43
MA 1 3,966 100 43 NC 34,257 75 52
MD 1 4,882 100 37 KY 36,821 75 29
HI 5,736 35 - PA 37,330 75 66
NH 9,238 50 11 LA 39,400 75 27
VT 9,596 50 9 FL 40,560 75 62
WV 24,205 75 13 AL 44,260 100 33
SC 24,739 75 50 IL 46,241 100 79
OH 25,710 75 71 MI 47,252 100 43
IN 27,634 75 35 MS 47,700 100 24
GA 48,325 100 85 KS 79,341 200 57
AR 53,147 100 24 ID 83,393 200 13
WI 54,052 100 70 NV 84,246 200 14
IA 56,181 100 32 AZ 90,225 200 21
OK 56,632 100 31 OR 92,270 200 27
WA 56,905 100 25 CO 94,864 200 35
MO 59,534 100 38 WY 97,634 200 16
UT 66,240 200 26 CA 104,473 200 54
ND 70,484 200 25 NM 121,620 200 22
MN 72,449 200 60 MT 146,646 200 13
SD 76,956 200 22 TX 219,897 200 98
NE 77,249 200 36 AK 572,122 200 -
Notes: 1. Due to the narrow and long shape of these regions, 100 mile boundaries are applied
for the remainders of MA and MD.
of MA and MD is 63.0 and 69.9 respectively.
2. Each buffer radius is selected as follows:
,
where = an index identifying each MSA remainder area and S i is the size each region.
i) 35 miles if " 75.7 ii) 50 miles if 96.1 " " 98.0
iii) 75 miles if 155.6 " " 201.4 iv) 100 miles if 210.4 " " 244.0
v) 200 miles if 257.4 "
3. Hawaii ( HI) and Alaska ( AK) are excluded from the current version of TransNIEMO, but
included in NEIMO.
21
Figure 2. Sampled Network Centroids: CA and AZ
22
Figure 3. Location of Scenario Bridges on Interstate Freeways
23
Table 2. Candidate bridge destruction scenarios
( Interstate) Highway Combinations Affected
I10 I15 I18 I10+ I15 I10+ I18 I15+ I18 I10+ I15+ I80
AZ O
NV Destinations O O O
AZ+ NV O O O O O O O
Note: Shading identifies the scenario cases summarized in Table 4.
24
Figure 4. Shortest Paths Before ( Upper Diagram) and After ( Lower Diagram) Elimination of a
Key Bridge on Highway I15: CA to NV
25
Figure 5. Shortest Paths Before and After Elimination of a Key Bridge on Highway I80: CA to
NV
26
Figure 6. Shortest Paths Before ( Upper Diagram) and After ( Lower Diagram) Elimination of a
Key Bridge on Highway I15: CA to AZ
27
Table 3. Changes in the number of paths and dollar values of shipping on highways resulting from the elimination of major bridges: CA to NV
Route 0_ 2 0_ 1 1 2 3 4 5 6 7 8 9 10 11 12 Total
Highway S78 I40 I15 S190 S168 U6 S167 S182 U395 S88 U50 I80 S36 S299
# of paths - - 212 5 60 158 19 83 26 100 8 60 13 12 756
Baseline
$ value( A) 1
-
-
7431 25 580 1837 210 811 256 1029 175 1105 194 151 13804
# of paths - - 212 5 60 159 19 83 26 124 24 19 13 12 756
$ value( A 1 ) 1 - - 7431 25 580 1846 210 811 256 1359 557 384 194 151 13804
D=( A 1 ) - ( A) 1 - - 0 0 0 8 0 0 0 330 382 - 721 0 0 0
I80
% change - - 0.0% 0.0% 0.0% 0.7% 0.0% 0.0% 0% 38.7% 182.6% - 60.1% 0% 0% 0%
# of paths 5 137 0 23 60 210 19 83 26 100 8 60 13 12 756
$ value( A 2 ) 1 23 7031 0 114 580 2125 210 811 256 1029 175 1105 194 151 13804
D=( A 2 ) - ( A) 1 - - - 7431 89 0 287 0 0 0 0 0 0 0 0 0
Bridge Scenario
I15
% change - - - 100.0% 360.0% 0.0% 15.6% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0% 0%
Notes 1. Units: $ Millions
2. Routes 0_ 1 and 0_ 2 are generated as alternatives to baseline shortest paths.
3. Entries for “# of paths” and “$ value” are calculated for network crossing the State boarders.
28
Table 4. Changes in the number of paths and dollar values of shipping on highways resulting from the elimination of major bridges: CA to AZ
Route 1 2 3 4 5 6 7 Total
Highway I8 S78 I10 S62 I40 U95 S375
Baseline ( B) 140 72 510 13 321 72 6 1134
After ( A) 377 3 0 316 360 72 6 1134
D=( A)-( B) 237.0 - 69 - 510 303 39 0 0 0
# of paths
% Change 169.29% - 95.83% - 100% 2,330.77% 12.15% 0% 0% 0%
Baseline ( B) 186.1 176.3 11030.1 8.1 282.3 229.7 4.3 11917
After ( A) 2492.7 1.5 0 8819 369.8 229.7 4.3 11917
D=( A)-( B) 2306.6 - 174.9 - 11030.1 8811 87.5 0 0 0
Dollar Value 1
% Change 1239.51% - 99.17% - 100% 108,847.06% 30.98% 0% 0% 0%
Note 1. Unit: $ Millions
29
Table 5. Increases in miles traveled between economic centroids: California to Arizona
Origin Destine Total miles of path Change of miles
ID Name of region ID Name of region
number
of path Before ( 1) After ( 2) D=( 2)-( 1)
%
change 1
8 CA Los Angeles 5 AZ Phoenix 65 24,528 27,441 2,913 11.88%
8 CA Los Angeles 6 AZ Tucson 52 25,742 26,590 848 3.29%
8 CA Los Angeles 7 AZ remainder 156 77,266 79,767 2,501 3.24%
9 CA Sacramento 5 AZ Phoenix 20 15,467 16,053 587 3.79%
9 CA Sacramento 6 AZ Tucson 16 14,256 14,562 305 2.14%
9 CA Sacramento 7 AZ remainder 48 39,754 39,794 40 0.10%
10 CA San Diego 5 AZ Phoenix 45 16,436 16,947 510 3.10%
10 CA San Diego 6 AZ Tucson 36 15,306 15,306 0 0.00%
10 CA San Diego 7 AZ remainder 108 52,006 52,876 869 1.67%
11 CA San Jose 5 AZ Phoenix 45 33,168 34,713 1,545 4.66%
11 CA San Jose 6 AZ Tucson 36 30,771 31,458 687 2.23%
11 CA San Jose 7 AZ remainder 108 86,494 86,628 134 0.16%
12 CA remainder 5 AZ Phoenix 95 57,824 60,205 2,381 4.12%
12 CA remainder 6 AZ Tucson 76 55,203 56,414 1,211 2.19%
12 CA remainder 7 AZ remainder 228 151,493 151,815 322 0.21%
Total 1,134 695,714 710,567 14,853 2.13%
Note: 1. % change = ( D / Before value) * 100
30
Table 6. California costs of purchasing truck mode services as a fraction of the total input value, 2001
USCsec. Description PTS 1 ($ M.) TIV 2 ($ M.) TCV 3
USC01 Live animals and live fish & Meat, fish, seafood, and their preparations 137.3 7892.6 0.0174
USC02 Cereal grains & Other agricultural products except for Animal Feed 328.8 21337.3 0.0154
USC03 Animal feed and products of animal origin, n. e. c. 120.9 4060.2 0.0298
USC04 Milled grain products and preparations, and bakery products 147.9 10401.8 0.0142
USC05 Other prepared foodstuffs and fats and oils 631.1 32475.7 0.0194
USC06 Alcoholic beverages 173.9 9869.6 0.0176
USC07 Tobacco products 0.1 15.0 0.0045
USC08
Nonmetallic minerals ( Monumental or building stone, Natural sands,
Gravel and crushed stone, n. e. c.)
26.0 1194.7 0.0218
USC09 Metallic ores and concentrates 4.1 221.1 0.0185
USC10 Coal and petroleum products ( Coal and Fuel oils, n. e. c.) 117.6 41272.0 0.0028
USC11 Basic chemicals 48.3 3024.6 0.0160
USC12 Pharmaceutical products 85.9 20064.7 0.0043
USC13 Fertilizers 25.3 1070.3 0.0236
USC14 Chemical products and preparations, n. e. c. 201.8 10496.8 0.0192
USC15 Plastics and rubber 459.2 15633.8 0.0294
USC16 Logs and other wood in the rough & Wood products 169.4 6675.9 0.0254
USC17 Pulp, newsprint, paper, and paperboard & Paper or paperboard articles 187.5 7193.4 0.0261
USC18 Printed products 326.8 20040.6 0.0163
USC19 Textiles, leather, and articles of textiles or leather 273.5 21400.1 0.0128
USC20 Nonmetallic mineral products 390.2 8911.2 0.0438
USC21 Base metal in primary or semIfinished forms and in finished basic shapes 123.2 5403.2 0.0228
USC22 Articles of base metal 246.1 17885.7 0.0138
USC23 Machinery 220.0 25935.1 0.0085
USC24
Electronic and other electrical equipment and components, and office
equipment
293.6 154741.2 0.0019
USC25 Motorized and other vehicles ( including parts) 203.2 14856.5 0.0137
USC26 Transportation equipment, n. e. c. 183.0 18052.5 0.0101
USC27 Precision instruments and apparatus 91.5 27226.5 0.0034
USC28
Furniture, mattresses and mattress supports, lamps, lighting fittings, and
illuminated signs
156.7 8729.6 0.0180
USC29
Miscellaneous manufactured products, Scrap, Mixed freight, and
Commodity unknown
142.4 16274.3 0.0087
Total 5515.3 532356.0 0.0104
Source: Raw data was obtained from 2001 IMPLAN software package “ commodity balance sheet.” The authors
aggregated from IMPLAN sectors to USC sectors using conversions developed by Park et al. ( 2007).
Notes 1. Purchase of truck services ($ M)
2. Total input unit value ($ M.)
3. Truck cost per unit value
31
Table 7. Shipping costs: California to Arizona
Baseline Scenario
USC
Sector
Description
Value of
Shipping ($ M) !
"
Avg. truck
cost per mile 1
($/ Mile)
Total
shipping
cost ($ M) !
#
Total shipping
cost ($ M)
$
! COST ($ M)
% & $ ' #
%
Change 2
% / "
USC01
Live animals and live fish & Meat, fish,
seafood, and their preparations 272.02 9.63 4.73 5.15 0.4139 0.15%
USC02
Cereal grains & Other agricultural
products except for Animal Feed 540.91 16.79 8.33 9.05 0.7167 0.13%
USC03
Animal feed and products of animal
origin, n. e. c. 296.81 17.87 8.85 9.61 0.7618 0.26%
USC04
Milled grain products and preparations,
and bakery products 0.15 0.00 0.00 0.00 0.0001 0.07%
USC05
Other prepared foodstuffs and fats and
oils 8.99 0.40 0.17 0.19 0.0183 0.20%
USC06 Alcoholic beverages 275.24 9.54 4.84 5.22 0.3749 0.14%
USC07 Tobacco products 0.00 0.00 0.00 0.00 0.0000 -
USC08
Nonmetallic minerals ( Monumental or
building stone, Natural sands, Gravel and
crushed stone, n. e. c.) 1278.66 53.23 27.87 30.00 2.1204 0.17%
USC09 Metallic ores and concentrates 1 5 8 . 4 4 6.86 2.93 3.28 0.3484 0.22%
USC10
Coal and petroleum products ( Coal and
Fuel oils, n. e. c.) 278.46 1.62 0.78 0.85 0.0665 0.02%
USC11 Basic chemicals 2 1 4 . 9 5 6.69 3.44 3.67 0.2267 0.11%
USC12 Pharmaceutical products 1 5 7 . 1 7 1.33 0.68 0.73 0.0584 0.04%
USC13 Fertilizers 0 . 2 7 0.01 0.01 0.01 0.0002 0.07%
USC14
Chemical products and preparations,
n. e. c. 147.05 5.54 2.82 3.07 0.2504 0.17%
USC15 Plastics and rubber 1 1 7 5 . 2 6 68.62 34.55 37.34 2.7825 0.24%
USC16
Logs and other wood in the rough &
Wood products 2298.11 113.88 58.37 63.03 4.6576 0.20%
USC17
Pulp, newsprint, paper, and paperboard &
Paper or paperboard articles 144.54 7.76 3.77 4.09 0.3164 0.22%
USC18 Printed products 8 1 9 . 4 6 26.90 13.36 14.49 1.1325 0.14%
USC19
Textiles, leather, and articles of textiles or
leather 2.87 0.05 0.04 0.04 0.0015 0.05%
USC20 Nonmetallic mineral products 3 5 9 . 9 9 30.77 15.77 16.94 1.1708 0.33%
USC21
Base metal in primary or semi- finished
forms and in finished basic shapes 1106.65 47.72 25.23 27.16 1.9252 0.17%
USC22 Articles of base metal 2 1 4 . 7 8 5.89 2.96 3.20 0.2355 0.11%
USC23 Machinery 2 5 4 . 9 3 4.53 2.17 2.36 0.1978 0.08%
USC24
Electronic and other electrical equipment
and components, and office equipment 295.49 1.11 0.56 0.61 0.0459 0.02%
USC25
Motorized and other vehicles ( including
parts) 232.59 6.40 3.19 3.45 0.2621 0.11%
USC26 Transportation equipment, n. e. c. 339.39 6.60 3.43 3.67 0.2375 0.07%
USC27 Precision instruments and apparatus 7 7 1 . 2 0 5.16 2.62 2.83 0.2077 0.03%
USC28
Furniture, mattresses and mattress
supports, lamps, lighting fittings, and
illuminated signs 1.08 0.03 0.02 0.02 0.0005 0.05%
USC29
Miscellaneous manufactured products,
Scrap, Mixed freight, and Commodity
unknown 268.65 4.76 2.34 2.53 0.1904 0.07%
Total 11914.12 233.83 252.56 18.7208 0.16%
Notes 1. Ave. truck cost per mile denotes average cost per mile for shipping from California to Arizona using trucks. In
application, we used the truck cost per mile for each route between CA to AZ to estimate the increased cost.
2. % change =( ! cost / Shipping value) * 100.
32
Table 8. Economic impacts: Price and total impacts by industry sector ($ Millions)
33
Table 9. Economic impacts by State ($ Millions)
34
References
Batten, D. and D. Martellato, 1985. Classical versus Modern Approaches to Interregional Input-
Output Analysis, Annals of Regional Science, 19, 1- 15.
Capelle, R. B. et al, 2000. Freight USA. Highlights from the 1997 Commodity Flow Survey and
other sources. Report prepared by Oak Ridge National Laboratory for the Bureau of
Transportation Statistics, U. S. Department of Transportation, Washington D. C.
Cho, S., Gordon, P., J. E. Moore II, H. W. Richardson, Shinozuka, M., and Chang, S., 2001,
Integrating Transportation Network and Regional Economic Models to Estimate the Costs
of a Large Urban Earthquake, Journal of Regional Science, Vol, 41, No. 1, p. 39- 65
Erik Dietzenbacher, 1997, In Vindication of the Ghosh Model: A Reinterpretation as a Price
Model. Journal of Regional Science 37 ( 4), 629– 651 doi: 10.1111/ 0022- 4146.00073.
Giuliano, G., 2004. Land use impacts of transportation investments: Highway and transit,
Chapter 9 in S. Hanson & G. Giuliano, eds., The Geography of Urban Transportation ( 3 rd
ed.), pp. 237- 273. New York: The Guilford Press.
Giuliano, G., P. Gordon, Q. Pan, J. Park & L. Wang, 2006. Estimating freight flows for
metropolitan area highway networks using secondary data sources.
Gordon, P and Q. Pan, 2002. Assembling and Processing Freight Shipment Data: Developing a
GIS- Based Origin- Destination Matrix for Southern California Freight Flows. Los Angeles:
METRANS.)
Gordon, P., J. Moore II, H. Richardson, and Q. Pan. ( 2005), “ The Economic Impact of a
Terrorist Attack on the Twin Ports of Los Angeles- Long Beach”, Chapter 14 in Harry W.
Richardson, Peter Gordon, and James E. Moore II ( eds.), The Economic Impacts of
Terrorist Attacks, Northampton, MA, USA: Edward Elgar Publishing.
Gordon, P., J. E. Moore II, J. Y. Park, and H. W. Richardson, 2007, Constructing a Flexible
National Interstate Economic Model, presented at 46thAnnual Meeting of the Western
Regional Science Association, Newport Beach, CA, February 21- 24.
H. L Hwang, Jane Rollow, 2000, Data Processing Procedures and methodology for estimating
trip distances for the 1995 American Travel Survey ( ATS).
35
Ham, H., T. J. Kim, and D. Boyce, 2005. Implementation and estimation of a combined model of
interregional, multimodal commodity shipments and transportation network flows.
Transportation Research Part B 39 ( 1) 65- 79.
Isard, W, 1951. Interregional and Regional Input- Output Analysis: A Model of a Space Economy.
Review of Economics and Statistics, 33, 318- 328.
Isard, W, 1960. Methods of Regional Science. New York: Wiley.
Leontief, W. W. and A. A. Strout, 1963. Multiregional Input- Output Analysis, in T. Barna, ed.,
Structural Interdependence of the American Economy.
Muller, P. O., 2004. Transportation and urban form: Stages in the spatial evolution of the
American Metropolis, Chapter 3 in S. Hanson & G. Giuliano, eds., The Geography of Urban
Transportation ( 3 rd ed.), pp. 59- 85. New York: The Guilford Press.
Park, J. Y. and P. Gordon, 2005, “ An Evaluation of Input- Output Aggregation Error Using a New
MRIO Model” Paper presented at North American Meetings of the Regional Science
Association International 52nd Annual Conference, Riviera Hotel & Casino, Las Vegas, NV,
November 10- 12.
Park, J. Y., 2006a. Estimation of State- by- State trade flows for service industries. Presented at
2006 North America Regional Science Council ( NARSC).
Park, J. Y., 2006b, The Economic Impacts of a Dirty- Bomb Attack on the Los Angeles and Long
Beach Port: Applying Supply- driven NIEMO, presented at 17 th Annual Meeting of the
Association of Collegiate Schools of Planning, Fort Worth, TX, November 9- 12.
Park, J. Y., 2008, Application of a Price- Sensitive Supply- Side Input- Output Model to an
Examination of the Economic Impacts of Hurricane Katrina and Rita Disruptions of the U. S.
Oil- Industry, Ecological Economics, Forthcoming.
Park, J. Y., P. Gordon, J. E. Moore II, and H. W. Richardson, L. Wang, 2007a, Simulating The
State- by- State Effects of Terrorist Attacks on Three Major U. S. Ports: Applying NIEMO
( National Interstate Economic Model), p. 208- 234, in H. W. Richardson, P. Gordon and J. E.
Moore II, eds., The Economic Costs and Consequences of Terrorism. Northampton, MA,
USA: Edward Elgar Publishing.
Park, J. Y., P. Gordon, J. E. Moore II, and H. W. Richardson, 2007b. A two- step approach to
detailed estimating State- to- State commodity trade flows. The Annals of Regional Science,
under review.
36
Southworth, Frank, 2005, Filling gaps in the U. S. commodity flow pictures: Using the CFS with
other data sources. Paper presented at CFS Conference 2005 Workshop by Oak Ridge
National Laboratory, Boston, MA, July 8- 9, 2005.
U. S. Department of Transportation ( USDOT), Federal Highway Administration, 2002a. Freight
Analysis Framework ( FAF) overview. Washington D. C, October 2002. Available at
http:// www. ops. fhwa. dot. gov/ freight/ freight_ analysis/ faf/ index. htm.
U. S. Department of Transportation, Federal Highway Administration, 2004. Freight
Transportation Improvements and the Economy. Washington D. C, June 2004.
U. S. Department of Transportation, Federal Highway Administration, 2006a. FAF2 User guide.
Washington D. C, November 2006. Available at http:// www. ops. fhwa. dot. gov/ freight/
freight_ analysis/ faf/ index. htm.
U. S. Department of Transportation, Federal Highway Administration, 2006b. FAF2 User guide.
Available at http:// ops. fhwa. dot. gov/ freight/ freight_ analysis/ faf/ faf2userguide
U. S. Department of Transportation, Federal Highway Administration, 2007. Common Issues in
Emergency Transportation Operations Preparedness and Response. Washington D. C,
February 2007. Available at http:// www. trb. org/ news/ blurb_ detail. asp? id= 7470
U. S. Department of Transportation, Federal Highway Administration, 2003. This is a projection
from the 1998 base year, as reported by the Federal Highway Administration's Freight
Analysis Framework.
Available at http:// www. bts. gov/ programs/ freight_ transportation/ html/ transportation. html
U. S. Department of Transportation, Office of Freight Management and Operations, 2002b.
Freight Analysis Framework Highway Capacity Analysis. Washington, D. C: USDOT.
U. S. Department of Transportation and Bureau of Transportation Statistics, 2002. Freight
Shipments in America, Preliminary Highlights From The 2002 Commodity Flow Survey
Plus Additional Data.
U. S. Department of Transportation, Research and Innovative Technology Administration &
Bureau of Transportation Statistics, 2005. Transportation Statistics Annual Report.
Washington D. C, November 2005.
U. S. Department of Transportation, Research and Innovative Technology Administration &
Bureau of Transportation Statistics, 2006b. Freight in America: A new national Picture.
Washington D. C, January 2006.
37
Wei Luo, Fahui Wang, 2003, Measures of spatial accessibility to health care in a GIS
environment: synthesis and a case study in the Chicago region.
Yu, S, 2006. The National Transportation Network. Los Angeles: USC, Viterbi School of
Engineering.