Content uploaded by James F. Campbell
Author content
All content in this area was uploaded by James F. Campbell on Jul 07, 2014
Content may be subject to copyright.
Impact of Shippers’ Choice on Transportation System
Congestion and Performance: Integrating Random
Utility with Simulation
Kevin Sweeney, James Campbell, and Donald Sweeney
Abstract
In this research, we show how modeling shippers’ responses to congested
freight transportation on an important segment of the Upper Mississippi
River (UMR) inland navigation system strongly infl uences the measure-
ment of expected economic benefi ts attributed to a range of congestion
mitigation measures. We present a model of the UMR that integrates a ship-
pers’ random utility model with a discrete event simulation model of the
most congested 100-mile segment of the UMR system. The random utility
model recognizes that waterway shippers may opt out of using the UMR in
response to increased congestion and instead utilize alternative transport
modes or destinations. Incorporating the dynamic response of shippers to
changing operating conditions improves existing simulation models by
explicitly accounting for the preferences and values of shippers, thereby
providing a consistent estimate of the direct economic benefi ts associated
with measures designed to reduce congestion and improve system perfor-
mance. The major contributions of our research include demonstrating the
importance of using models that capture shippers’ responses to congestion
in freight transportation systems and illustrating a novel methodology for
quantifying the direct economic benefi ts to users of measures to improve
transportation on the UMR.
Keywords
Congestion, simulation model, random-utility model, Upper Mississippi
River, freight transportation
Kevin Sweeney
Corresponding Author
University of Maryland
Email: ksweeney@rhsmith.umd.edu
James Campbell
University of Missouri–St. Louis
Donald Sweeney
University of Missouri–St. Louis
Transportation Journal, Vol. 53, No. 2, 2014
Copyright © 2014 The Pennsylvania State
University, University Park, PA
TJ 53.2_03_Sweeney-Campbell-Sweeney.indd 143TJ 53.2_03_Sweeney-Campbell-Sweeney.indd 143 27/02/14 2:04 AM27/02/14 2:04 AM
144 / TRANSPORTATION JOURNAL™
Introduction
In many freight transportation systems, traffi c congestion is a signifi cant
problem that causes shippers and carriers to alter the timing, routing,
mode, and/or destination for shipments. More extensive responses to con-
gestion include changes in system operating policies (e.g., scheduling),
regulations, and investments in new or improved infrastructure to increase
the carrying capacity. Mathematical models to facilitate the evaluation of
congestion mitigation measures need to properly integrate the behavior of
system users to accurately estimate the magnitude of the reduced conges-
tion and its corresponding economic value. Models that ignore the behav-
ior of the system’s users, especially when users have viable alternatives to
using the system in response to increased levels of congestion, will likely
overstate congestion and can lead to inappropriate recommendations
and decisions (Bhat 1998 ). In this article, we illustrate the importance of
incorporating users’ behaviors in modeling congested freight transporta-
tion systems by integrating a random utility model of shippers’ modal and
destination alternatives with a detailed simulation model of a congested
segment of the Upper Mississippi River (UMR) navigation system. Our
research leverages recent research on simulation and optimization models
for the UMR (Nauss 2008 ; Smith, Sweeney, and Campbell 2009 ; Sweeney
2004 ) and on random utility models for shippers on the UMR system (Train
and Wilson 2007 ).
The random utility model recognizes that potential system users (ship-
pers who prefer to send their products on the UMR) may, in their self-inter-
est, dynamically “opt out” of using the system in response to anticipated
high levels of congestion, and instead utilize alternative transport modes
and/or shipment destinations. Incorporating the response of shippers to
dynamic operating conditions improves existing simulation models by
explicitly accounting for the behavior of self-interested agents and thereby
provides a consistent foundation for estimating the direct economic ben-
efi ts associated with implementing measures designed to improve system
performance. In this article, we model UMR system performance under a
wide range of traffi c levels with three congestion mitigation measures con-
sidered by the National Research Council ( 2004 ).
The major contributions of our research include (a) demonstrating the
importance of including shippers’ responses to congestion when model-
ing complex freight transportation systems, (b) providing a novel meth-
odology for incorporating shippers’ alternatives and dynamic responses
to congestion in a discrete event simulation model, and (c) illustrating a
TJ 53.2_03_Sweeney-Campbell-Sweeney.indd 144TJ 53.2_03_Sweeney-Campbell-Sweeney.indd 144 27/02/14 2:04 AM27/02/14 2:04 AM
Sweeney, Campbell, and Sweeney: Integrating Random Utility with Simulation \ 145
methodology for quantifying the direct economic benefi ts to users of
different congestion mitigation measures in a transportation system. The
remainder of this article is organized as follows. The next two sections
discuss relevant related research and provide background information on
UMR traffi c and operations, respectively. Then we describe our simulation
model and provide details on integrating a random utility model of ship-
pers’ choice. Next, we present and discuss results with three proposed UMR
congestion mitigation measures: improved scheduling by resequencing
queues at congested UMR locks, employing helper boats to reduce lockage
times, and constructing new larger locks. Finally, we present our conclu-
sions and identify future research.
Related Research
Quantitative analyses of the benefi ts and costs for transport projects and
infrastructure investments have long been used to help inform transporta-
tion policy in areas such as road pricing, transportation infrastructure, and
transit. For some examples, see Bucklew ( 2007 ), Eliasson ( 2009 ), Gorman
( 2008 ), Oster, Rubin, and Strong ( 1997 ), Prud’homme and Bocarejo ( 2005 ),
Tang and Lo ( 2008 ), and Tsamboulas, Vrenken, and Lekka ( 2007 ). There is
not a similar body of literature for inland waterway transportation, but
one related work is Fellin et al. ( 2001 ) that develops a quadratic program-
ming model to assess grain fl ows and estimate the benefi ts to grain ship-
ments from possible lock expansions. Discrete choice random utility
models have been used in numerous transportation analyses to model indi-
vidual agents’ choices among transportation alternatives (e.g., Ben-Akiva
and Lerman 1985 ; Bhat 2012 ; Rich, Holmbad, and Hansen 2009 ; Swait and
Ben-Akiva 1987 ; Train 1980; Vovsha 1997 ; Yai, Iwakura, and Morichi 1997 ).
Studies that have developed discrete choice random utility models for
freight shippers’ choices include Baindur and Veigas ( 2011 ), de Jong and
Ben-Akiva ( 2007 ), Jiang, Johnson, and Calzada ( 1999 ), Mitra ( 2013 ), Swait,
Louviere, and Williams ( 1994 ), and Train and Wilson ( 2007 , 2008 ). Swait,
Louviere, and Williams ( 1994 ) found that incorporating stated preference
(SP) and revealed preference (RP) data provided better predictive power in a
logit model than relying on revealed preference data alone. Jiang, Johnson,
and Calzada ( 1999 ) used disaggregated revealed preference data to pre-
dict whether long distance shipments would use road transportation or
rail. Train and Wilson ( 2007 , 2008 ) further extended the usage of SP and
RP data for freight destination and modal choice models by conducting
experiments on stated preferences constructed from revealed preferences
TJ 53.2_03_Sweeney-Campbell-Sweeney.indd 145TJ 53.2_03_Sweeney-Campbell-Sweeney.indd 145 27/02/14 2:04 AM27/02/14 2:04 AM
146 / TRANSPORTATION JOURNAL™
of bulk freight shippers, illustrating this methodology on the Columbia/
Snake River System. De Jong and Ben-Akiva ( 2007 ) created logistic models
for the countries of Norway and Sweden using disaggregate shipment data,
while Baindur and Veigas ( 2011 ) used an agent based random utility model
to analyze freight market competition in the Italy–France transport corri-
dor. Finally, Mitra ( 2013 ) used RP data to construct a disaggregated model of
modal choice for agricultural freight in North Dakota.
These studies utilized several different functional forms for the dis-
crete random utility models in their analyses, including logit models
(Baindur and Veigas 2011 ; Jiang, Johnson, and Calzada 1999 ; Rich, Holmbad,
and Hansen 2009 ; Swait, Louviere, and Williams 1994 ; Train 1980; Train
and Wilson 2007 , 2008 ; and Vovsha 1997 ) and probit models (Mitra 2013 ;
Yai, Iwakura, and Morichi 1997 ). Probit models place more emphasis on the
normalcy of the underlying error distributions, and for that reason logit
models have seen more usage in discrete choice random utility models.
While the discrete choice random utility model used in our research is a
multinomial logit model (as in Train and Wilson 2007 ), our method for
integrating a discrete choice random utility model into a simulation model
easily accommodates other functional forms provided that the underlying
assumptions of the discrete choice model have been satisfi ed.
There is much research on simulation models of congested freight
transportation systems, with recent examples including Palma, Kilani,
and Lindsey ( 2005 ) for road transport and congestion pricing, Tumer and
Agogino ( 2007 ) for airport traffi c fl ow management, and Kidokoro ( 2006 )
for urban railway management. While inland waterway transportation has
not received the analytical research attention of other transport modes, dis-
crete event simulation for waterway systems dates back to the early 1970s
(Carroll and Bronzini 1973 ). Recent studies that have developed waterway
simulation models for analyzing operational changes and infrastructure
investments for inland waterways include Biles, Sasso, and Bilbrey ( 2004 ),
Dai and Schonfeld ( 1998 ), Franzese et al. ( 2004 ), Golkar, Shekhar, and
Buddhavarapu ( 1998 ), Martinelli and Schonfeld ( 1995 ), Ting and Schonfeld
( 1998 , 2001a , 2001b ), and Wang and Schonfeld ( 2007 ).
There is an ongoing effort to develop detailed simulation models of
UMR river operations, starting with Sweeney ( 2004 ). Smith, Sweeney, and
Campbell ( 2009 ) developed a more detailed simulation model that incor-
porates time-varying dynamics of traffi c and lockage operations on the
UMR. Simulation modeling is especially important for the UMR since the
strong annual, weekly, and daily seasonality of operations render steady
TJ 53.2_03_Sweeney-Campbell-Sweeney.indd 146TJ 53.2_03_Sweeney-Campbell-Sweeney.indd 146 27/02/14 2:04 AM27/02/14 2:04 AM
Sweeney, Campbell, and Sweeney: Integrating Random Utility with Simulation \ 147
state queuing approximations and analytical methods unsuitable for these
locks (Smith et al. 2011 ; Smith, Sweeney, and Campbell 2009 ). However,
none of these previous UMR models, or any other waterway system simula-
tion model, explicitly captures shippers’ behaviors that may divert ship-
ments from the river system to alternate modes (e.g., rail to New Orleans
for export) or destinations (e.g., local ethanol plants) in response to con-
gestion related delays and costs. This failure to include users’ responses to
congestion limits the usefulness of these previous models for evaluating
the effects of systemic congestion mitigation measures.
Some researchers have expressed interest in merging discrete choice
random utility models with simulation models of passenger transporta-
tion (Li et al. 2006), but no research we have found has attempted a similar
merger in freight transportation modeling. In this research, we integrate
a choice model with a detailed, dynamic, freight transportation simula-
tion model to incorporate the decision processes of individual shippers in
deciding whether or not to use the system (as opposed to adopting broad
aggregate generalities of shippers’ behavior). This union of models allows
utility maximizing shippers to respond dynamically and intelligently to
the performance of the system as it varies over time. By combining discrete
choice random utility and dynamic simulation in an integrated model, we
hope to provide a more realistic and accurate tool that policymakers can
leverage to better understand the behavior of a complex transportation
system.
The Upper Mississippi River Navigation System
The Mississippi River navigation system extends over 1,800 miles from the
Gulf of Mexico to Minneapolis–St. Paul, Minnesota. It is a key link in global
supply chains for a variety of US agricultural products. The Mississippi
River transported over 498 million tons of freight in 2011 (the most recent
year for which data is available), of which over 70 million tons were agricul-
tural products (US Army Corps of Engineers 2011 ). Other cargoes include
bulk commodities such as coal, cement, and chemical and petroleum prod-
ucts. Products are carried in barges (typically 195–200 feet long and 35 feet
wide) that can hold over 1,500 tons each (or 53,500 bushels or 450,000 gal-
lons). Barges are joined together into tows pushed by a single towboat. Our
research focus is on the UMR, which extends from the confl uence of the
Ohio River to just north of Minneapolis–St. Paul, Minnesota. The UMR
navigation system includes 29 lock-and-dam facilities that provide reli-
able commercial navigation by maintaining a minimum channel depth of
TJ 53.2_03_Sweeney-Campbell-Sweeney.indd 147TJ 53.2_03_Sweeney-Campbell-Sweeney.indd 147 27/02/14 2:04 AM27/02/14 2:04 AM
148 / TRANSPORTATION JOURNAL™
nine feet, permitting vessels to navigate the UMR’s approximately 300-foot
elevation difference. On the UMR, tow sizes are generally limited to 15 or 16
barges due to navigation channel conditions. The UMR is open year round,
though there is almost no traffi c in winter due to extremely diffi cult operat-
ing conditions.
Each dam on the UMR includes one or two locks that allow vessels to
pass through the dams. A lockage operation consists of a vessel entering
the lock chamber, having the water level raised or lowered as needed, and
then exiting the chamber. Commercial vessels are nearly always locked
individually, while groups of recreational vessels and towboats not pushing
barges are frequently locked together. Locks are 600 feet or 1,200 feet long
and at each lock, congestion occurs periodically as vessels form queues
when arriving while the lock is occupied. The river channels between the
locks and dams, termed lock pools, are not a source of congestion as they
possess nearly unlimited traffi c fl ow capacities during normal operating
conditions (US Army Corps of Engineers 2004 , Engineering Appendix,
A-53). Economies of scale have led to the increased use of large barge tows
(nearly 1,200 feet long) on the UMR that must be decoupled into two sec-
tions for passage through 600-foot-long locks, and then recoupled follow-
ing lockage. These “double lockages” require almost two hours on average
and have been cited as a “key contributor to chronic delays” at locks on
the UMR (National Research Council 2004) . Our research focuses on the
most congested segment of the UMR: locks 20, 21, 22, 24, and 25 (there is no
Lock 23), which are the fi ve southernmost 600-foot-long locks (bracketed
to the north and south by 1,200-foot-long locks) and the four connecting
river pools, covering a total of about 100 river miles just north of St. Louis,
Missouri. The traffi c in the study region includes large tows of agricul-
tural commodities headed downstream for export; upstream backhauls of
empty barges being positioned for future loads; smaller tows of one or a
few barges traveling between local terminals; towboats traveling without
barges for repositioning; and private recreational craft. The annual total
tonnage shipped on the UMR has ranged from 60.7 million to 84.1 million
tons over the period 2001–2011 (US Army Corps of Engineers 2011 ), with
approximately 70 percent of the tonnage in the study section in 2011 being
agricultural products destined downstream.
From an analytical perspective, the congested segment of the UMR can
be viewed as a complex stochastic system consisting of a sequence of fi ve
capacitated servers (locks), each with four separate queues (for upstream
and downstream recreational and commercial traffi c) and nonnegligible
TJ 53.2_03_Sweeney-Campbell-Sweeney.indd 148TJ 53.2_03_Sweeney-Campbell-Sweeney.indd 148 27/02/14 2:04 AM27/02/14 2:04 AM
Sweeney, Campbell, and Sweeney: Integrating Random Utility with Simulation \ 149
stochastic transport times between servers, where vessel routes (sequence
of locks) and arrival frequencies are dynamic, seasonal and interdependent
and lock service times are vessel, season and sequence dependent. This
characterization includes aspects of sequential queuing networks in trans-
portation with two-way traffi c (Hall 2003 ), though the large variation in
service times for different vessels and the stochastic and dynamic nature
of the traffi c distinguish this from other transportation systems. The UMR
navigation system also includes some aspects of dynamic job shop sched-
uling with sequence dependent setup times (Ramesesh 1990 ; Ruiz, Maroto,
and Alcaraz 2005 ).
Although there is very little publically available data with details on
river traffi c and lock operations for the UMR, we are fortunate to have
access to a unique set of data acquired for a previous study of the UMR
(Sweeney 2004 ). These data, compiled from the Corps’ OMNI lock perfor-
mance database for calendar years 2000 through 2003, are the most com-
plete set of operations data available for the UMR and form the basis for
our analyses. This database contains detailed timing information for each
phase of every UMR lockage. The data includes 70,180 individual lockages
in our study section and shows a large variability in the distribution of
lockage times. Analysis of the data reveals that there are three main groups
of vessels that utilize the UMR system (Sweeney 2004 ): (a) “Double Tows”
(typically 1,200 feet long pushing 15 barges) that require two “cuts” to fi t
through the 600-foot-long locks, (b) “Single Tows” that are small enough
to not require a double lockage, and (c) “Recreation&Light Tows,” which
include recreational craft and towboats without barges that can be locked
most quickly, and often simultaneously.
Smith, Sweeney, and Campbell ( 2009 ) identify calendar year 2000 as
the most representative year of normal UMR operations in our lockage
data because that year was relatively unaffected by extraordinary system
events such as extended lock closures, and it experienced relatively typi-
cal weather, water fl ow, and other operating conditions. Consequently we
also adopt calendar year 2000 as the baseline in our analysis here. (Note that
tonnage on the UMR has actually declined about 30 percent from 2000 to
2011.) Table 1 provides the number, percentage of total, and the mean and
standard deviation of lockage times for each of the three vessel groups for
UMR locks 20 through 25 in our baseline data. The mean lockage times
range from less than 20 minutes for Recreation and Light Tows to almost
two hours for double tows. This table also shows the differential usage of
each lock, with a general trend of fewer lockages farther upstream. Table 2
TJ 53.2_03_Sweeney-Campbell-Sweeney.indd 149TJ 53.2_03_Sweeney-Campbell-Sweeney.indd 149 27/02/14 2:04 AM27/02/14 2:04 AM
150 / TRANSPORTATION JOURNAL™
provides more details on the baseline traffi c at locks 20–25 and shows that
50–70 percent of the lockages are for commercial traffi c (i.e. tows) and 59–67
percent of the lockages at each lock are double tows. While the majority of
tonnage is downbound, the numbers of upbound and downbound barges
and tows at each lock each year are about equal.
Currently, the Corps operates each lock under a modifi ed fi rst-
in, fi rst-out local operating policy called RECPRIO that gives locking
Lock Vessel Group Lockages Lockages
Mean
Lockage
Time (min.)
S.D. of
Lockage
Time
20 Double 2266 12.1% 110.6 44.1
Recreation and Light Tows 677 3.6% 18.1 33.5
Single 585 3.1% 41.7 33.1
Total 3528 18.9% 81.4 56.8
21 Double 2340 12.5% 114.8 28.8
Recreation and Light Tows 749 4.0% 15.8 9.7
Single 615 3.3% 40.1 24.7
Total 3704 19.8% 82.4 50.0
22 Double 2360 12.6% 127.7 39.1
Recreation and Light Tows 633 3.4% 19.1 13.6
Single 524 2.8% 52.7 57.8
Total 3517 18.8% 97.0 59.8
24 Double 2468 13.2% 116.8 35.3
Recreation and Light Tows 795 4.3% 16.4 8.4
Single 537 2.9% 41.2 22.4
Total 3800 20.3% 85.1 53.0
25 Double 2485 13.3% 116.4 36.5
Recreation and Light Tows 1069 5.7% 16.5 9.3
Single 597 3.2% 39.5 29.3
Total 4151 22.2% 79.6 54.8
Totals Double 11919 63.7% 117.3 37.4
Recreation and Light Tows 3923 21.0% 17.1 16.7
Single 2858 15.3% 42.8 35.6
Total 18700 100.0% 84.9 55.2
Table 1/ Annual Lockages, Percent of Total, Mean, and Standard Deviation of
Lockage Times at UMR Locks 20–25
Source : Compiled from the Corps’ OMNI baseline data by vessel group.
TJ 53.2_03_Sweeney-Campbell-Sweeney.indd 150TJ 53.2_03_Sweeney-Campbell-Sweeney.indd 150 27/02/14 2:04 AM27/02/14 2:04 AM
Sweeney, Campbell, and Sweeney: Integrating Random Utility with Simulation \ 151
priority to recreational and governmental vessels by requiring that these
noncommercial vessels wait for no more than three commercial ves-
sel lockages before moving to the front of the queue. In this article, we
model this policy along with three policy alternatives proposed for
reducing congestion on the UMR (National Research Council 2004 ):
improved scheduling of barge tows by resequencing local lock queues,
deployment of helper towboats to expedite the double-lockage process,
and construction of new larger locks that completely eliminate the need
for double lockages. The implementation costs for these alternatives
range from very low for tow resequencing policies to very high (several
billion dollars) for constructing new locks.
The UMR Simulation Model
Our discrete event simulation model extends previous UMR simula-
tion models by integrating a random utility model that incorporates the
responses of potential UMR shippers to the dynamic expected utilities of
waterway versus nonwaterway transportation. By incorporating shippers’
responses to variable levels of system congestion (and other economic or
policy incentives), we produce estimates of congestion consistent with
users’ preferences and alternatives and, consequently, better measure the
direct economic benefi ts to users from operating policy and infrastructure
changes. This section describes our integrated simulation model, which
is constructed using Micro Saint Sharp 3.6 software (Alion Science and
Technology 2011 ).
Integrating shippers’ preferences and alternatives entails two criti-
cal enhancements to earlier UMR simulation models (Smith, Sweeney,
and Campbell 2009 ; Sweeney 2004 ) that (a) allow self-interested shippers
to “opt out” of sending a potential waterborne shipment on the UMR,
Lock
Double
Lockages
Number of
Vessels
Number of
Commercial
Vessels
Tons
(×1000)
Downbound
Tons Barges
Average
Delay
(hrs)
20 64.3% 4,134 2,818 35,015 72.1% 34,066 2.23
21 63.3% 4,155 2,913 36,449 72.0% 35,760 1.81
22 67.5% 4,125 2,866 36,813 72.5% 36,000 3.03
24 64.3% 4,542 2,918 38,698 72.0% 37,909 2.12
25 59.1% 5,714 2,924 39,162 71.7% 38,064 2.36
Table 2/ Selected Lockage and Vessel Statistics at UMR Locks 20–25
Source : Compiled from the Corps’ OMNI baseline data.
TJ 53.2_03_Sweeney-Campbell-Sweeney.indd 151TJ 53.2_03_Sweeney-Campbell-Sweeney.indd 151 27/02/14 2:04 AM27/02/14 2:04 AM
152 / TRANSPORTATION JOURNAL™
and (b) properly model the consequences of such behavior. For the fi rst
enhancement we need to generate each potential UMR water shipment
with a specifi c water-related origin, destination and shipment (vessel) size,
which supplants the earlier models’ use of probabilistic decisions at each
lock to determine whether or not a vessel proceeds to the next lock. The
second enhancement embeds a dynamic shipper’s choice decision to deter-
mine whether each potential water shipment (at the time it is generated)
will be transported either via the UMR in a barge tow (“opt in”) or via a non-
water-transportation alternative (“opt out”). Figure 1 presents a fl owchart
of the logic of the integrated simulation model and the following sections
provide details.
UMR Shipment Generation
To model shippers’ opt-in or -out choices in a simulation of the UMR, we
need to know the specifi c origin, intended destination, and size for each
potential shipment that may use the UMR. Unfortunately, a comprehen-
sive data set of all potential UMR shipments is not available. Furthermore,
UMR barge transportation does not operate on published or set schedules;
nor is there detailed vessel tracking information available to create such
data. (While individual waterway carriers track their own vessels, those
data have competitive value and are not publicly available.) Fortunately, the
Corps does collect some data on the vessel lockages at each lock. These data
are primarily for monitoring lock performance and do not track vessels at
Figure 1 Flowchart of the UMR Micro Saint Sharp 3.6
Simulation Model.
TJ 53.2_03_Sweeney-Campbell-Sweeney.indd 152TJ 53.2_03_Sweeney-Campbell-Sweeney.indd 152 27/02/14 2:04 AM27/02/14 2:04 AM
Sweeney, Campbell, and Sweeney: Integrating Random Utility with Simulation \ 153
any other locations in the system or provide detailed data on the specifi c
freight carried, the originations and destinations of vessels, or on any ves-
sel activities that occur solely in the pools between the locks. As stated on
the Corps’ Lock Performance Monitoring System website: “Detailed infor-
mation on specifi c companies or commodities is considered privileged and
is not included in the Corps Locks website” (US Army Corps of Engineers
2013 ). However, by comparing the traffi c at adjacent locks, we can ascer-
tain the origin and destination pools for individual commercial tows.
Therefore, to generate potential water shipments in our model we assume
that each shipment originates or terminates at the midpoints of pools 21,
22, 24 and 25 or, for shipments that pass through the entire study section, at
points determined by the average distance tows travel external to the study
section. This provides 30 origin-destination pairs for shipments through
the locks in our study section (we are not interested in shipments that stay
within a single pool, as they do not impact lock congestion).
From careful analysis of the detailed lockage data for the baseline year,
we identifi ed 4,970 itineraries for double and single tows, with the origin
and destination pattern shown in table 3 . Ignoring the last row and last
column of percentages aggregated by destination pool and origin pool, the
other percentage values in table 3 show the relative observed frequencies
for the 30 shipment origin-destination pairs. These data show that most
tows originate or terminate outside pools 21–25, with nearly a third of all
tow traffi c stopping nowhere within the boundaries of the fi ve-lock system.
Destination Pool
Origin Pool
20 and
below 21 22 24 25
26 and
above Total
20 and
below
- 6.4% 2.3% 1.8% 0.4% 17.8% 28.7%
21 7.2% - 3.1% 1.0% 0.1% 3.6% 15.0%
22 2.5% 2.8% - 1.8% 0.1% 2.7% 9.9%
24 3.4% 1.2% 2.0% - 0.7% 5.3% 12.5%
25 0.4% 0.1% 0.1% 0.7% - 2.0% 3.3%
26 and
above
15.2% 3.9% 2.6% 6.9% 2.1% - 30.6%
Total 28.6% 14.4% 10.1% 12.1% 3.3% 31.4% 100.0%
Table 3/ Observed Frequencies of Tow Itineraries
Source : Compiled from the Corps’ OMNI baseline data.
TJ 53.2_03_Sweeney-Campbell-Sweeney.indd 153TJ 53.2_03_Sweeney-Campbell-Sweeney.indd 153 27/02/14 2:04 AM27/02/14 2:04 AM
154 / TRANSPORTATION JOURNAL™
In the simulation model, potential UMR shipments are generated for
the 30 origin-destination pairs using nonstationary Poisson processes in
accordance with historical norms that encompass monthly seasonality in
shipment generation rates and sizes. To capture the monthly seasonality
for double tows and single tows, we employ 720 shipment generation func-
tions (corresponding to 30 origin-destination pairs, 12 months, and single
or double tows). For the recreational traffi c and light tows, we generate
individual vessels upbound and downbound at each lock for each month
using 120 vessel generation functions (5 locks, 2 directions, and 12 months).
Therefore, we have in our model a total of 840 Poisson processes that gener-
ate potential traffi c for the system.
Modeling UMR Shipment Choices Using Random Utility
For each potential water shipment generated in the simulation model, a
decision is made whether or not it will travel via the UMR as a barge tow.
The alternatives to using the UMR are to send the shipment via some non-
water-transportation mode, or to not send the shipment at all. This opt-in
or -out decision is based on the shipper’s expected utility associated with
each specifi c alternative, which is represented as the sum of a determinis-
tic component involving observable variables incorporating the expected
performance of the segment of the UMR system that the potential tow will
utilize and a randomly determined component. We similarly represent the
utility of nonwater-shipment alternatives for each specifi c potential water
shipment as the sum of a deterministic and a random component. Formally,
let I denote the set of shipment origin, destination and size triples, J
i denote
the set of shipment alternatives for each i∈I and T denote the time-span of
the simulation. The utility of shipment i and alternative j at time t is:
UV iIJtT
jt
ijt
ijt
ii
ε
=+ ∀∈ ∈,,
,,, (1)
where Vjt
i, is the deterministic (observable) component of utility, and jt
i
ε
,
is the random (unobservable) component of utility (assumed i.i.d. Extreme
Value Type I random variables). The alternative j selected for shipment i at
time t is the alternative that provides the greatest utility to the shipper for
that shipment:
argmax
JJ U
ijt
i
∈{}
, (2)
TJ 53.2_03_Sweeney-Campbell-Sweeney.indd 154TJ 53.2_03_Sweeney-Campbell-Sweeney.indd 154 27/02/14 2:04 AM27/02/14 2:04 AM
Sweeney, Campbell, and Sweeney: Integrating Random Utility with Simulation \ 155
We let K denote the set of variables that affect a shipment’s observable util-
ity, so that the deterministic component of utility for each alternative is:
,,,
,,,
1
VXiIjJtTkK
jt
ikjtk
i
k
n
i
∑
β
=∀∈∈∈∈
= (3)
where Xjtk
i,, is a vector of observable shipment attributes associated with
alternative j at time t for shipment i and β
k is a vector of utility coeffi cients.
Variables that affect the utility of shipping on the UMR, such as the trans-
portation cost and the time required to complete the shipment, are there-
fore related to the simulated performance of the UMR system, which is itself
a function of traffi c and resulting congestion. Consequently, individual
potential shippers’ realized utility values (and incentives to use the system)
are linked in a consistent manner to the performance of the system which
is itself determined by the interaction of all shippers’ opt-in decisions and
the operational characteristics of the system. With this logit random utility
model, the probability of selecting alternative for shipment i at time t is
then given by (Train 2009 ):
Unfortunately, detailed information regarding the full set of alternatives
available for each specifi c potential UMR tow shipment does not exist and
acquiring such data is costly and beyond the scope of our study. However,
Train and Wilson ( 2007 ) provide summary information compiled from
a survey of agricultural shippers regarding shipper’s self-reported next-
best alternative to shipment on the UMR. Agricultural shipments com-
prise the majority of traffi c in our study area, accounting for over 70
percent of all tonnage shipped in or through our study area in the base-
line calendar year 2000. Therefore, we model the shipper’s choice for each
potential shipment as a binary opt-in (use the UMR) or -out (use the next-
best nonwater alternative) decision by comparing the expected random
utilities of using the UMR system and using the nonwater alternative at
the time each shipment is generated. The shipment is routed to the option
with the greatest expected utility. Because the expected utility of a water
shipment is related to the expected future performance of the UMR sys-
tem, the decision to use the UMR system is dynamic and responsive to
anticipated congestion.
V
jJ Vjt
i
jJ jt
i
i
∑
)
(
)
(
==
∈
Pr( ) exp
exp
,
,
(4)
J
TJ 53.2_03_Sweeney-Campbell-Sweeney.indd 155TJ 53.2_03_Sweeney-Campbell-Sweeney.indd 155 27/02/14 6:18 PM27/02/14 6:18 PM
156 / TRANSPORTATION JOURNAL™
Implementation of Shippers’ Choice for the UMR System Model
To illustrate the importance of including shippers’ choice among
alternatives in simulation models of congested transportation systems,
we need to calculate the expected deterministic portion of utility in equa-
tion (3) for potential UMR shippers for both shipment on the UMR and
for shipment via the nonwater alternative. For shipments on the UMR,
the simulation model provides an accurate and dynamic way to estimate
tow travel times, so we set the expected value of the deterministic potion
of utility for water shipment equal to the realized deterministic com-
ponent of utility of the last tow to complete the same itinerary. (Other
mechanisms could be employed for this, such as a weighted average of
the realized utilities on recent identical itineraries, or a forward scan to
predict the utility as a function of the anticipated state of the system as
the tow completes its itinerary.)
To calculate the expected deterministic portion of utility for shipments
opting out of traveling on the UMR, we employ an econometric model and
summary statistics compiled from a revealed preference (RP) and stated
preference (SP) survey of US grain shippers reported in Train and Wilson
( 2007 ). Specifi cally, the expected deterministic level of utility associated
with the nonwater-shipment alternative is modeled as the mean determin-
istic utility of the next best nonwater shipping alternative as self-reported
by grain shippers in Train and Wilson ( 2007 , 17). We recognize this may be a
limiting assumption as individual shippers, differentiated by location and
product, may have multiple alternatives (or even no alternative) to using
the UMR, and also that while grain is the majority commodity shipped on
the UMR, other products are shipped as well. For illustrative purposes we
use this single nonwater alternative for all shippers in our model, while
acknowledging that detailed economic evaluation of the real benefi ts asso-
ciated with congestion management measures for the UMR requires more
specifi c information regarding each shipper’s true set of alternatives to
using the system.
Train and Wilson ( 2007 ) report three different econometric model
specifi cations: a fi xed coeffi cient logit model fi t to RP data only; a fi xed
coeffi cient logit model fi t to both RP and SP data; and a mixed logit model
with variable coeffi cients fi t to RP and SP data. While our simulation
model framework can integrate any of these specifi cations (or others,
such as probit or nested logit), we elect to incorporate their fi xed coef-
fi cient model fi t to RP and SP data in our simulation because of its full
TJ 53.2_03_Sweeney-Campbell-Sweeney.indd 156TJ 53.2_03_Sweeney-Campbell-Sweeney.indd 156 27/02/14 2:04 AM27/02/14 2:04 AM
Sweeney, Campbell, and Sweeney: Integrating Random Utility with Simulation \ 157
use of their survey data and its ease of implementation. Therefore, our
model uses the following explanatory variables and shipper utility coef-
fi cients identifi ed by Train and Wilson ( 2007 ) to estimate the determin-
istic portion of utility for each shipment being transported on the UMR
and its next best nonwater alternative: (1) the expected transportation
rate, (2) the expected transportation time, (3) the expected percentage of
on-time delivery, (4) the expected commodity price at the destination, (5)
the shipment distance, (6) a binary variable (0,1) to denote if the alterna-
tive involves movement on rail, (7) the rail loading capacity of the ship-
per if the alternative involves movement on rail, and (8) a binary variable
(0,1) to denote if the alternative involves water transportation. The utility
coeffi cient estimates are taken from table 18 of Train and Wilson’s fi xed
coeffi cient logit model and the mean values of the eight variables are
taken from table 9 of Train and Wilson ( 2007 , 32, 17).
Table 4 summarizes the Train and Wilson ( 2007 ) data as utilized in
our model. The fi rst column in table 4 shows the observable or simulated
variables that impact shipment utility. Column 2 displays the marginal
contribution to the utility of shipping on the UMR of a unit change in each
observable or simulated variable. For example, a one-dollar increase in the
barge transportation rate decreases the utility of shipping via the UMR by
0.0885 utils/ton. Column 3 displays the mean values of the observable vari-
ables for UMR barge grain shippers that reported that they did use the UMR
for shipment. Column 4 shows the mean values of the observable variables
for the self-reported next-best nonwater alternative to UMR shipment iden-
tifi ed by the UMR grain shippers. The fi nal two columns highlight the utility
contribution of the observable variables at their mean values to the utility
of using the UMR for shipment compared to the next best nonwater alterna-
tive. It is the expected utility at mean values displayed in the fi nal column
of table 4 that we employ to represent the expected deterministic utility of
the nonwater shipment alternative for all potential UMR barge shipments.
These utils/ton may be converted to dollars of direct economic value
per ton by rescaling the utility functions by the marginal utility of the
transportation rate (utils/$) yielding shipment utilities measured in $s/ton.
The fi nal row of table 4 demonstrates the relative strength of the prefer-
ence for shipping on the UMR for shipments reported utilizing the UMR
as the expected deterministic component of utility for shipping on the
UMR ($34.56/ton) is about 30 percent larger than the utility expected for
sending that shipment to its next-best nonwater alternative ($26.66/ton).
TJ 53.2_03_Sweeney-Campbell-Sweeney.indd 157TJ 53.2_03_Sweeney-Campbell-Sweeney.indd 157 27/02/14 2:04 AM27/02/14 2:04 AM
158 / TRANSPORTATION JOURNAL™
This higher expected utility for reported UMR shipments results from the
UMR providing a better combination of transportation characteristics
(e.g., rate, transit time, reliability) for those shipments in this region.
The realized utilities from the simulation model for shipments on the
UMR are affected by the performance of the UMR system because these real-
ized utilities depend on the time that it takes a tow to complete its itinerary.
For example, if a tow takes longer than anticipated to complete its jour-
ney, then the realized utility of the water shipment is decreased through
both the direct effect increased shipment time has on the realized utility
(the utility coeffi cient - 0.000426 in table 4 ) and also the indirect effect that
Mean
Variable Value
Utility Contribution at
Mean Variable Value
Variable
Utility
Contribution
Coeffi cient
Shipment
via UMR
Shipment
via the
Nonwater
Alternative
Shipment via
UMR
Shipment via
the Nonwater
Alternative
Transportation
rate ($s/ton)
-0.088500 26.5 18.5 -2.3453 -1.6373
Transportation
time (hours)
-0.000426 236.0 109.0 -0.1005 -0.0464
Reliability (per-
cent on time)
0.008770 87.1 84.2 0.7639 0.7384
Price at desti-
nation ($s/ton)
0.007730 116.6 102.3 0.9013 0.7908
Distance
(miles)
0.001630 1,061.0 405.0 1.7294 0.6602
Rail indicator (0
= no, 1 = yes)
-0.899000 0.0 1.0 0.0000 -0.8990
Rail*Ln(1+Rail
car loading
capacity)
0.719000 1.0 46.0 0.0000 2.7528
Barge indicator
(0 = no, 1 = yes)
2.110000 1.0 0.0 2.1100 0.0000
Expected utility
at mean values
(utils/ton)
3.0588 2.3595
Scaled utility
at mean values
(2006 $'s/ton)
$34.56 $26.66
Table 4/ Observable and Simulated Variables that Contribute to Deterministic
Utility, Utility Contribution Coeffi cients, Mean Variable Values for UMR Shipment
and Nonwater Alternative Shipment, and Mean Contributions to Utility by
Alternative
Source : Compiled from Train and Wilson (2007).
TJ 53.2_03_Sweeney-Campbell-Sweeney.indd 158TJ 53.2_03_Sweeney-Campbell-Sweeney.indd 158 27/02/14 2:04 AM27/02/14 2:04 AM
Sweeney, Campbell, and Sweeney: Integrating Random Utility with Simulation \ 159
increased time has on realized transportation cost (the utility coeffi cient
- 0.088500 in table 4 ).
To estimate this indirect effect of UMR transportation time on real-
ized transportation costs, we employ the results in Train and Wilson ( 2004 ,
34) relating barge transportation rates to total barge transportation times;
Train and Wilson ( 2004 ) report that the natural log of the rate for barge
transportation changes by a factor of 0.3 for each unit change in the natural
log of total barge transportation time. Thus, we can write
ln( Transportation Rate ) = 0.3* ln( Total Transportation Time )
+ ln( constant ). (5)
Using the mean values of total barge transportation time (236 hours for
UMR grain shipments from table 4 ) and barge transportation rate ($26.50
per ton for UMR grain shipments from table 4 ) we solve equation (5) to
determine the constant value for UMR grain shippers as 5.1448. Then we
can rewrite equation (5) as
Transportation Rate = 5.1448 × Total Transportation Time 0.3 (6)
We note that the transportation rates estimated by equation (6) increase
less than linearly with increases in total transportation time. For example,
a 25 percent increase in the total transportation time produces only a 6.9
percent increase (1.25
0.3 = 1.069) in the transportation rate from equation (6),
thereby providing an attenuated indirect decrease in shipment utility.
We are not able to estimate the impact on realized utilities of changes
in the reliability of performance of water shipments as information regard-
ing expected or promised shipment times of individual barges or railcars
is not publically available. Therefore, we assume that all UMR shipments
are delivered with the mean reliability for UMR shipment of 87.1 percent
on-time delivery reported in table 4 . See Wiegmans ( 2010 ) for a discussion
of cost-reliability tradeoffs in freight transportation.
Finally, we note that the travel time through our 100-mile study section
is relatively short, averaging approximately 21.6 hours in our baseline data,
relative to the mean total transportation time of 236 hours indicated in
table 4 . Thus, the performance of the fi ve-lock study section will have only
a minor impact on the realized utility of most shipments unless system per-
formance degrades signifi cantly and substantially increases the time spent
in the study section.
TJ 53.2_03_Sweeney-Campbell-Sweeney.indd 159TJ 53.2_03_Sweeney-Campbell-Sweeney.indd 159 27/02/14 2:04 AM27/02/14 2:04 AM
160 / TRANSPORTATION JOURNAL™
Integrated UMR Simulation Model Details
The simulation model tracks the activities of the vessels that utilize the
UMR system as they travel through pools, wait in queues at locks, and pass
though locks. The vessel-processing times at the individual locks and tran-
sit times through the pools are distributed log-normally in the model, with
means and variances varying monthly by vessel type, operational lock-
age type, and direction of movement. As in earlier work (Smith, Sweeney,
and Campbell 2009 ; Sweeney 2004 ), three operational types of lockages
are differentiated: an “exchange” lockage occurs when a vessel enters the
chamber after waiting for completion of lockage by a vessel traveling in
the opposite direction; a “turnback” lockage occurs when a vessel enters
the chamber after waiting for completion of lockage by a vessel traveling
in the same direction; and a “fl y” lockage occurs when a vessel arrives to
an empty lock chamber. Thus, there are 1,080 different locking time dis-
tributions (one each for 12 months, 5 locks, 3 vessel types, 3 lockage types,
and 2 directions of travel) and 288 different pool transit time distributions
(one each for 12 months, 4 lock pools, 3 vessel types, and 2 directions of
travel). Periods of impaired lock operating conditions (caused by scheduled
maintenance, adverse river conditions, or lock malfunctions) are imposed
randomly and independently at each lock at seasonally varying rates as in
earlier models (Smith, Sweeney, and Campbell 2009 ; Sweeney 2004 ). As our
focus is waterway congestion resulting from typical operations, we do not
model extraordinary events such as extended droughts, rare fl oods, or cata-
strophic infrastructure failures.
The shipment utilities calculated in equation (1) are measured in utils/
ton scaled by the assumed i.i.d. Extreme Value Type 1 distributions of the
random error terms. These utils/ton may be converted to dollars of direct
economic value per shipment by dividing the utils/ton by the marginal util-
ity of the transportation rate (utils/$) yielding utility measured in $/ton and
then multiplying that result by the mean shipment sizes of 14,245 tons for
double tows and 2,171 tons for single tows derived from the baseline data.
The simulation model tracks the total economic value of all generated ship-
ments (including the random component of utility), whether or not they
travel on the UMR, to provide a consistent estimate of the total direct eco-
nomic value to all shippers of the performance of the UMR system and their
shipment alternatives. By simulating alternative congestion mitigation
measures with increasing numbers of potential shipments, we can system-
atically explore the benefi ts (i.e., total direct economic value to all shippers)
produced by a congestion mitigation measure for a wide range of demand
TJ 53.2_03_Sweeney-Campbell-Sweeney.indd 160TJ 53.2_03_Sweeney-Campbell-Sweeney.indd 160 27/02/14 2:04 AM27/02/14 2:04 AM
Sweeney, Campbell, and Sweeney: Integrating Random Utility with Simulation \ 161
levels, all while accounting for the dynamic self-interested decisions of the
shippers.
Each individual UMR system simulation encompasses a period of one
calendar year, from January through December. This fi nite horizon simu-
lation strategy is appropriate as the study section of the UMR system is
essentially unused from mid-December through mid-March due to adverse
operating conditions (Sweeney 2004 ). Starting the simulation at a point
with little traffi c and no lock queues allows us to closely examine the effects
of the choice function and the different congestion mitigation measures
as traffi c builds from negligible winter levels to its peaks in July and thus
allows for an accurate representation of the true dynamics of the traffi c sys-
tem over the yearlong simulation.
Model Validation and Calibration
The operation of the integrated simulation model was verifi ed by tracing
the movement of individual entities (shipments and vessels) through the
network and by visually inspecting the model animation. For the integrated
discrete choice model, we verifi ed that the expected and realized random
utilities of modeled alternatives were computed correctly and that the
model appropriately sent each potential shipment (tow) to the alternative
with the greatest expected total utility. For detailed validation of individual
activity times and basic system operations (e.g., the number of lockages,
lockage times, pool transit times) we refer the reader to Smith, Sweeney,
and Campbell ( 2009 ). Here we present highlights of the validation of the
integrated model that incorporates shippers’ choice.
To validate our shipment generation procedure we fi rst executed the
simulation model 100 times (each representing one year of operation) with
the existing lock operating policy RECPRIO while requiring all generated
shipments to enter the UMR system as tows. We compared the number
and frequencies of generated tow itineraries to the observed itineraries
in the baseline data and a chi-square goodness of fi t test indicated (at the
5 percent signifi cance level) that the model itinerary counts and frequen-
cies were consistent with those evident in the baseline traffi c. We also
compared the modeled number of lockages, lock utilizations, and itinerary
durations to baseline data and again found a very good fi t. (For example, the
mean number of lockages from the 100 simulations of the system ranged
from a low of 99.7 percent of observed lockages at Lock 25 to a high of 100.1
percent of observed lockages at Lock 20.) Therefore, we conclude that the
itinerary generation mechanism in the simulation model corresponds well
TJ 53.2_03_Sweeney-Campbell-Sweeney.indd 161TJ 53.2_03_Sweeney-Campbell-Sweeney.indd 161 27/02/14 2:04 AM27/02/14 2:04 AM
162 / TRANSPORTATION JOURNAL™
to the real system. Because there is no data on specifi c alternatives to UMR
transport for individual UMR shippers, we could not directly validate the
discrete choice model for our study section of the UMR. Consequently, we
later explore the sensitivity of our model and its results to a signifi cantly
less attractive (relative to water shipment) specifi cation of the alternative
available to potential water shipments.
When we enable the dynamic discrete shippers’ choice model in the
integrated simulation model, shippers will opt UMR shipment only when
provides a greater utility than the nonwater alternative. For example, if we
repeat the 100 simulations with the RECPRIO lock operating policy, but now
with the integrated random utility model used to dynamically allocate gen-
erated shipments to the UMR or to the nonwater alternative as described
earlier, then only 67.5 percent of the potential shipments are actually sent
via the UMR. For the other 32.5 percent, the nonwater alternative provides
a greater utility, given the expected levels of deterministic utilities and
random variation exhibited in table 4 . Because many generated shipments
elect to not use the UMR system in the model that integrates a nonwater
alternative, it is necessary to increase potential shipment generation rates
so that the number and itineraries of shipments that use the UMR system
approximate observed baseline levels. In general, this is no simple exercise
as there are 720 interdependent shipment generation rates for tows over
the year that must be simultaneously altered in a manner that yields sys-
tem activity levels that approximate observed activity levels on the UMR.
As we have no data regarding individual shippers’ seasonal rates of gener-
ating shipments or their real sets of seasonally available alternatives and
acquiring such data is diffi cult, costly and beyond the scope of our study
(see Casavant et al. 2008 ; Train and Wilson 2004 , 2007 , 2008 for discussions
of the diffi culty of acquiring suitable survey data), we adopt an iterative
procedure to uniformly increase shipment generation rates in the model
to levels that approximate the observed baseline activity levels. We employ
an iterative nonparametric Mann-Whitney U test (Mann and Whitney 1947 )
of the equality of two independent distributions (at the 5 percent confi -
dence level) by increasing the shipment generation rates by the reciprocal
of the percentage of traffi c opting in until we accept the hypothesis that the
number of opting-in UMR shipments generated by the 100 model runs that
include the nonwater alternative is sampled from the same distribution of
the number of shipments from the model without a nonwater alternative to
shipment on the UMR.
The last column in table 5 presents the result of the calibration process.
Table 5 indicates that 33.4 percent of all potential shipments generated in the
TJ 53.2_03_Sweeney-Campbell-Sweeney.indd 162TJ 53.2_03_Sweeney-Campbell-Sweeney.indd 162 27/02/14 2:04 AM27/02/14 2:04 AM
Sweeney, Campbell, and Sweeney: Integrating Random Utility with Simulation \ 163
model opt out of using the UMR system, thereby requiring a mean of 7,470
generated potential shipments (vs. 4,980 shipments when there is no alter-
native to the UMR) to produce the level of traffi c in the baseline data. The
shipments that opt out of using the UMR may be viewed as representative of
the universe of shipments that could utilize the UMR system, but elect not
to do so because they receive greater utility sending their products to des-
tinations not involving UMR transportation. Stated differently, the poten-
tial number of shipments that could move on the UMR when an alternative
exists is some 50 percent greater (7,470 vs. 4,980 generated shipments) than
the realized number of shipments that actually opt to use the UMR, with
the baseline UMR performance. Comparison of the realized utility per ship-
ment in columns 2 and 3 of table 5 shows that when the alternative is avail-
able, the utilities for shipping via the UMR and via the alternative are both
greater than the utility via the UMR with no alternative. This illustrates how
including shippers’ alternatives in the model allows the UMR to be better
utilized by having it serve those shipments for which it is best suited.
Without a
Nonwater Alternative
Calibrated with a
Nonwater Alternative
Shipments generated
Mean 4,980 7,470
Standard deviation 74 88
Realized UMR tows
Mean 4,981 4,978
Standard deviation 74 68
Percent choosing the alternative
Mean 0.0% 33.4%
Standard deviation 0.0% 0.8%
Realized utility per shipment via the UMR
Mean $490,399 $566,421
Standard deviation $3,435 $4,435
Realized utility per shipment via the alternative
Mean $0 $499,553
Standard deviation $0 $5,412
Total utility to all shippers
Mean $2,442,678,774 $4,064,464,124
Standard deviation $39,972,971 $49,921,172
Table 5/ Comparisons of Selected Summary Statistics Compiled from 100 Simula-
tions Each of the UMR System with and without a Nonwater Alternative with the
Existing RECPRIO Lock Operating Policy
TJ 53.2_03_Sweeney-Campbell-Sweeney.indd 163TJ 53.2_03_Sweeney-Campbell-Sweeney.indd 163 27/02/14 2:04 AM27/02/14 2:04 AM
164 / TRANSPORTATION JOURNAL™
Analysis, Results, and Discussion
Our integrated simulation model incorporating shippers’ choice allows us
to analyze more accurately changes in important metrics of system perfor-
mance, including the economic value produced by the system for shippers,
that result from changes in demand, operating policies, infrastructure,
availability of alternatives to UMR transport, and other factors that impact
on the operation of the system or the incentives of shippers to use the sys-
tem. As the key UMR policy issues concern changes resulting from poten-
tially increasing congestion levels (National Research Council 2004 ), we
examine system performance metrics with existing and increased levels of
shipment demands. Demand increases of up to 40 percent are modeled by
increasing the arrival rates of the nonstationary Poisson shipment genera-
tion functions by factors of 1.1, 1.2, 1.3, and 1.4. An increase in demand of 40
percent over the baseline level observed in 2000 represents a near doubling
from traffi c levels in recent years due to the decline in usage of the UMR
over the last decade (US Army Corps of Engineers 2011 ). However, we note
that demand levels for UMR traffi c are forecast by the US Army Corps of
Engineers to increase by between 64 and 98 percent from 2004 through 2060
(US Army Corps of Engineers 2008 , 3–19), thereby positioning our demand
increase factors well within the range of offi cial Corps traffi c estimates. We
note further that the Corps traffi c forecasts for the UMR have been sharply
criticized as overly optimistic (e.g., National Resource Council 2004, 6, 34).
Our goals are not to validate or endorse specifi c traffi c forecasts, but rather
to use the increasing levels of demand to highlight the value of including
shippers’ alternatives in modeling a transportation system when traffi c
increases to levels that might create signifi cant congestion.
In all our analytical cases, the nonwater alternative is presumed to have
enough capacity to accommodate all shipments that elect to use it without
altering the expected deterministic portion of the utility of the nonwater
alternative. All results we present in the following sections are compiled
from 100-year-long simulations with each of the specifi ed levels of demand.
Results with Existing Lock Operating Policies (RECPRIO)
To explore the sensitivity of the performance to the characteristics of the
nonwater alternative we consider a less attractive nonwater alternative to
UMR shipment than that derived from using the mean values of agricul-
tural shippers as identifi ed in the Train and Wilson ( 2007 ) survey. As an
extreme case, we also consider performance when there is no alternative to
UMR transportation. This permits us to examine the sensitivity of system
TJ 53.2_03_Sweeney-Campbell-Sweeney.indd 164TJ 53.2_03_Sweeney-Campbell-Sweeney.indd 164 27/02/14 2:04 AM27/02/14 2:04 AM
Sweeney, Campbell, and Sweeney: Integrating Random Utility with Simulation \ 165
performance (and the value afforded system users) to the presumed deter-
ministic component of the utility of the nonwater alternative. Having less
attractive alternatives, or no alternative, to using the UMR will increase the
probability of shippers using the UMR and may heighten congestion.
We create a less-attractive alternative to the UMR by doubling the
transport cost of the nonwater alternative, thereby decreasing the expected
deterministic component of utility of the nonwater alternative to $8.16 per
ton from our original estimate of $26.66 per ton displayed in table 4 . We then
recalibrate the model using the existing lock operating policy (RECPRIO)
with the expected deterministic utility of the alternative set equal to $8.16
per ton using the calibration procedure described earlier. This doubling of
transportation costs for the alternative requires a 9.9 percent increase in
shipment generation rates relative to the baseline to yield the same under-
lying baseline distribution of the total number of UMR shipments. In con-
trast, the original nonwater alternative (with a utility of $26.66 per ton)
required a 50 percent increase in shipment generation rates from the base-
line rates to generate the appropriate number of UMR shipments.
We also consider the situation where there is no viable alternative to
UMR transportation using the baseline shipment generation rates. Table
6 provides results of these experiments with shipment demands increased
up to 40 percent above the baseline level (Demand Level = 1.4 in the table)
using the existing lock operating policies (RECPRIO). The fi rst fi ve rows of
table 6 show the effect of increasing demand with the original $26.66 per
ton deterministic utility for the alternative. The next fi ve rows of table 6
show the effect of increasing demand with the doubled alternative trans-
portation cost, and the last fi ve rows provide results where there is no non-
water alternative to UMR transportation. Each row shows the number of
shipments and the associated utilities (measured in dollars) for the UMR
and for the nonwater alternative.
The effect of the availability and relative attractiveness of a nonwater
shipping alternative is clearly apparent at all levels of demand in table 6 .
The more attractive the alternative to UMR shipment, the greater the per-
centage of shippers that opt out of using the UMR at all demand levels:
33–40 percent of shipments opt out of using the UMR with the baseline
alternative expected deterministic utility value ($26.66/ton), but only
9–14 percent opt out of the using the UMR when the expected transport
cost of the nonwater alternative is doubled. As expected, as the shipment
demand increases, the congestion increases (column 8 of table 6 ), and
more shippers opt for the alternative to UMR transportation (column 7
TJ 53.2_03_Sweeney-Campbell-Sweeney.indd 165TJ 53.2_03_Sweeney-Campbell-Sweeney.indd 165 27/02/14 2:04 AM27/02/14 2:04 AM
166 / TRANSPORTATION JOURNAL™
of table 6 ). However, the last column of table 6 shows that the increasing
congestion in response to high levels of demand (a 30 percent or more
increase from the baseline) is signifi cantly lower when shippers have
an alternative to shipping on the UMR. The less attractive the nonwater
alternative to UMR shipment, the less the dampening effect on system
congestion. Thus, with a 40 percent increase in traffi c from the baseline,
the mean system time for tows increases from 118 hours when there is an
alternative with the baseline utility values to 200 hours for an alternative
with a doubled transport cost, and to 360 hours when there is no viable
alternative to using the UMR.
We observe further that as demand increases, it is only when ship-
pers are modeled as having no alternative to the UMR and demand levels
increase signifi cantly (by 30 percent or more above the baseline) that sys-
tem performance degrades suffi ciently to decrease the total utility received
Via the UMR Via the Alternative
Utility of
Nonwater
Alternative
($/ton)
Demand
Level
Number of
Shipments
Total
Utility of
Shipments
( millions)
Number
of Ship-
ments
Total
Utility of
Shipments
(millions)
Percent
of
Ship-
ments
Sent
via the
UMR
System
Time
per
Tow
(hours)
$26.66/ton 1.0 4.978 $2,820 2,491 $1,245 66.6% 24.0
1.1 5,430 $3,057 2,788 $1,394 66.1% 31.2
1.2 5,799 $3,227 3,163 $1,578 64.7% 51.0
1.3 6.073 $3,304 3,651 $1,815 62.5% 83.3
1.4 6,286 $3,345 4,174 $2,077 60.1% 118.1
$8.16/ton 1.0 4,979 $2,602 493 $ 148 91.0% 22.3
1.1 5,468 $2,845 550 $ 166 90.9% 28.2
1.2 5,920 $3,024 638 $ 192 90.3% 49.4
1.3 6,279 $3,077 823 $ 247 88.4% 114.0
1.4 6,585 $3,054 1,073 $323 86.0% 200.3
No alternative 1.0 4,980 $2,443 0 $ 0 100.0% 21.6
1.1 5,467 $2,668 0 $ 0 100.0% 26.8
1.2 5,964 $2,845 0 $ 0 100.0% 48.5
1.3 6,474 $2,824 0 $ 0 100.0% 158.3
1.4 6,979 $2,634 0 $ 0 100.0% 359.5
Table 6/ Mean Results Compiled from 100 Simulations Each of Selected Modeling
Assumptions with RECPRIO
TJ 53.2_03_Sweeney-Campbell-Sweeney.indd 166TJ 53.2_03_Sweeney-Campbell-Sweeney.indd 166 27/02/14 2:04 AM27/02/14 2:04 AM
Sweeney, Campbell, and Sweeney: Integrating Random Utility with Simulation \ 167
by shippers on the UMR (column 4 of table 6 ). When there are alternatives
to the UMR, then as demand levels increase, and congestion increases, the
relative total utility of UMR shippers becomes greater as the alternative to
water shipment becomes increasingly attractive. This greater attractive-
ness of the alternative draws traffi c off the UMR system thereby affording
higher utility for the shipments that still opt to use the UMR. Figures 2 and
3 respectively present the relationship between shipment demand levels
and the mean utilities and the total utilities for shipments on the UMR.
Figure 2 shows that as system congestion increases with increasing
demand, there is a corresponding decrease in the utility per UMR shipment
in all three cases. However the decrease in utility per shipment increases
as the attractiveness of the alternative decreases. Figure 3 depicts a slightly
different outcome for the total utilities received by UMR shippers as a func-
tion of demand level. When there is a relatively attractive alternative to
shipping on the UMR, even with dramatically increased levels of demand,
the system continues to deliver increasing levels of total utility to its users.
Even the presence of only a modestly attractive alternative keeps the total
level of utility delivered to system users from signifi cantly degrading with
Figure 2 Mean Estimated Utility of Shipments via the UMR with Existing Lock
Operating Policy (RECPRIO) at Selected Forecast Demand Levels
Source : Compiled from simulation model results.
TJ 53.2_03_Sweeney-Campbell-Sweeney.indd 167TJ 53.2_03_Sweeney-Campbell-Sweeney.indd 167 27/02/14 2:04 AM27/02/14 2:04 AM
168 / TRANSPORTATION JOURNAL™
much larger demands (at least to an increase in demand of 40 percent
relative to the baseline). In summary, including alternatives and shippers’
choice is important because even a moderately attractive alternative to sys-
tem use prevents (or delays) the degradation of the system, and only when
all potential shippers are forced to use the system are there extreme levels
of congestion.
Results with Proposed Congestion Mitigation Measures
To further explore the impact of including shippers’ choice in the model,
we examine system performance under increasing traffi c demands in
conjunction with the implementation of three proposed UMR congestion
mitigation measures (National Research Council 2004 ): (1) amended vessel
scheduling at locks via a “fastest expected lockage time” policy (FLT), (2)
providing helper boats at locks in conjunction with FLT scheduling (HLP),
and (3) construction of larger new locks (NEW). These three congestion
mitigation measures have widely varying estimated implementation costs
ranging from approximately $100,000 annually for implementing amended
vessel scheduling (Mundy et al. 2005 ), to $5–$10 million annually for imple-
menting helper boats, and to over $1.5 billion in construction costs for new
locks (US Army Corps of Engineers 2004 ).
Figure 3 Total Estimated Utility of Shipments via the UMR with Existing Lock
Operations (RECPRIO) at Selected Forecast Demand Levels
Source : Compiled from simulation model results.
TJ 53.2_03_Sweeney-Campbell-Sweeney.indd 168TJ 53.2_03_Sweeney-Campbell-Sweeney.indd 168 27/02/14 2:04 AM27/02/14 2:04 AM
Sweeney, Campbell, and Sweeney: Integrating Random Utility with Simulation \ 169
Under an FLT scheduling mechanism, which has been demonstrated
to be effective for clearing lock queues in this system (Smith, Sweeney, and
Campbell 2009) , tows are placed into local lock queues in accordance with
the expected time required to complete their lockage, with lower expected
times placed in the front of the queue. Thus, FLT gives priority to Single
Tows over Double Tows. This is a dynamic scheduling mechanism; the
queue processing order at each lock changes depending on the composition
of the queue at any point in time. Measure HLP employs FLT scheduling
along with dedicated “helper” towboats at each lock to assist the lockages
of Double Tows. These helper boats decrease the lockage time for each dou-
ble tow lockage by approximately 15 minutes (US Army Corps of Engineers
2004 ).
The fi nal congestion mitigation measure we analyze, denoted NEW,
is the construction of new 1,200-foot-long locks at Locks 20 through 25,
as proposed by the Corps (US Army Corps of Engineers 2004 ). These new
locks are already authorized for construction by Congress in the Water
Resource Development Act of 2007 (Public Law 110-114) and are sup-
ported for priority construction funding by a joint task force of the Army
Corps of Engineers and the Inland Waterways Users Board (Inland Marine
Transportation System Capital Investment Strategy Team 2010 ). With
new 1,200-foot locks, all lockages become single cut lockages and lockage
times drop dramatically for tows that are currently processed in a double
cut lockage. For our simulations, we adjusted the lockage times of Double
Tows to refl ect the Corps’ estimated mean lockage times for the new locks,
which range from 50 to 64 minutes and refl ect a 42–50 percent reduction
in mean lockage times at these locks (US Army Corps of Engineers 2004 ).
Because we have no data on the likely composition of tows after lock expan-
sion, we retain the baseline mix of vessel traffi c in our evaluation of the
new locks.
We fi rst simulate the system without the inclusion of a non-water
alternative for demand levels of 1.0, 1.1, 1.2, 1.3 and 1.4 (relative to baseline
demand) with the existing RECPRIO policy and for FLT, HLP, and NEW.
These 2,000 simulations (100 simulations each with 5 demand levels, and
3 congestion mitigation measures plus RECPRIO) refl ect circumstances
where all traffi c is forced to use the UMR, regardless of congestion. Selected
statistics compiled from these runs—including mean number of realized
UMR shipments (tows), mean total utility ($millions) for all UMR ship-
ments, and the mean time spent in the system per tow (hours)—are sum-
marized by demand level and congestion management measure in table 7 .
TJ 53.2_03_Sweeney-Campbell-Sweeney.indd 169TJ 53.2_03_Sweeney-Campbell-Sweeney.indd 169 27/02/14 2:04 AM27/02/14 2:04 AM
170 / TRANSPORTATION JOURNAL™
Table 8 summarizes a second set of 2,000 simulations that include a
nonwater alternative with the expected deterministic portion of the ran-
dom utility for that alternative set equal to the baseline utility of $26.66
per ton. Selected statistics compiled from these simulations include:
mean number of shipments via the alternative, mean number of UMR
shipments (tows), mean total utility ($millions) for all shipments via the
alternative, mean total utility (millions) for all UMR shipments, mean
total utility for all shipments, and the mean time spent in the system per
tow (hours).
The importance of including the nonwater alternative in the simula-
tions is clearly evident in the striking differences between the results pre-
sented in tables 7 and 8 . When shippers have an alternative to using the
Via the UMR
Demand Level
Congestion Miti-
gation Measure
Number of
Shipments
Total Utility
of Shipments
( millions)
System Time per
Tow (hours)
1.0 FLT 4,981 $2,444 20.7
HLP 4,994 $2,461 17.0
NEW 4,972 $2,475 12.3
RECPRIO 4,980 $2,443 21.6
1.1 FLT 5,484 $2,676 25.5
HLP 5,478 $2,696 18.7
NEW 5,490 $2,735 12.5
RECPRIO 5,467 $2,668 26.8
1.2 FLT 5,967 $2,859 41.0
HLP 5,988 $2,933 21.3
NEW 5,979 $2,965 12.7
RECPRIO 5,964 $2,845 48.5
1.3 FLT 6,473 $2,904 119.5
HLP 6,493 $3,159 27.5
NEW 6,481 $3,224 13.0
RECPRIO 6,474 $2,824 158.3
1.4 FLT 6,978 $2,870 242.3
HLP 6,972 $3,303 52.8
NEW 6,966 $3,459 13.3
RECPRIO 6,979 $2,634 359.5
Table 7/ Mean Results Compiled from 100 Simulations Each for Congestion Miti-
gation Measures Modeled without a Nonwater Alternative to the UMR
TJ 53.2_03_Sweeney-Campbell-Sweeney.indd 170TJ 53.2_03_Sweeney-Campbell-Sweeney.indd 170 27/02/14 2:04 AM27/02/14 2:04 AM
Sweeney, Campbell, and Sweeney: Integrating Random Utility with Simulation \ 171
UMR, then even at signifi cantly elevated demand levels, system congestion
(as measured by the mean hours per tow spent in the study section) is signif-
icantly dampened as lower utility shippers opt out of using the congested
UMR system. For example, under the FLT operating policy with demand
40 percent greater than the baseline, excluding shippers’ choice from the
model (the fourth row from the bottom of table 7 ) results in signifi cantly
degraded system performance with 242.3 mean hours per tow spent in the
study section and a mean total utility for all UMR shipments of $2,870 mil-
lion. The similar situation modeled with a nonwater alternative (the fourth
row from the bottom of table 8 ) shows a much lower mean time per tow of
only 90.8 hours in the study section, and a much higher mean total utility
Number of Shipments
Total Utility of
Shipments
Demand
Level
Congestion
Mitigation
Measure
Via the
Alternative
Via the
UMR
Via the
Alternative
(millions)
Via the
UMR
(millions)
Total
Utility All
Shipments
(millions)
System
Time
per Tow
(hours)
1.0 FLT 2,473 5,006 $1,244 $2,829 $4,074 23.0
HLP 2,446 5,036 $1,228 $2,861 $4,090 18.5
NEW 2,411 5,058 $1,203 $2,900 $4,103 12.9
RECPRIO 2,491 4,978 $1,245 $2,820 $4,064 24.0
1.1 FLT 2,743 5,477 $1,383 $3,074 $4,457 29.1
HLP 2,704 5,518 $1,359 $3,129 $4,488 20.5
NEW 2,643 5,569 $1,320 $3,195 $4,515 13.1
RECPRIO 2,788 5,430 $1,394 $3,057 $4,451 31.2
1.2 FLT 3,097 5,864 $1,567 $3,245 $4,812 41.9
HLP 2,995 5,993 $1,508 $3,383 $4,890 24.5
NEW 2,906 6,068 $1,448 $3,486 $4,934 13.4
RECPRIO 3,163 5,799 $1,578 $3,227 $4,805 51.0
1.3 FLT 3,520 6,183 $1,797 $3,349 $5,146 66.0
HLP 3,279 6,450 $1,654 $3,617 $5,271 31.9
NEW 3,149 6,571 $1,573 $3,766 $5,339 13.7
RECPRIO 3,651 6,073 $1,815 $3,304 $5,120 83.3
1.4 FLT 3,986 6,463 $2,049 $3,411 $5,460 90.8
HLP 3,648 6,806 $1,852 $3,751 $5,604 47.8
NEW 3,387 7,072 $1,689 $4,057 $5,747 14.1
RECPRIO 4,174 6,286 $2,077 $3,345 $5,422 118.1
Table 8/ Mean Results Compiled from 100 Simulations Each for Congestion Miti-
gation Measures Modeled with a Nonwater Alternative to the UMR
TJ 53.2_03_Sweeney-Campbell-Sweeney.indd 171TJ 53.2_03_Sweeney-Campbell-Sweeney.indd 171 27/02/14 2:04 AM27/02/14 2:04 AM
172 / TRANSPORTATION JOURNAL™
for all UMR shipments of $3,411 million. Thus, including shippers’ choice
with an alternative to water shipment suggests a much less congested sys-
tem that still produces signifi cant value for its users. Potential users that do
not benefi t suffi ciently from using the congested system opt out of using
the UMR, thereby reducing realized levels of system congestion for remain-
ing users.
The results displayed in tables 7 and 8 also show that FLT, HLP and NEW
provide increasingly effective congestion reduction for successively higher
levels of demand. However, we note that for signifi cantly increased levels of
demand (greater than 30 percent above baseline), the decrease in conges-
tion associated with each mitigation measure relative to RECPRIO is sig-
nifi cantly smaller when measured either by the decrease in system time per
tow or by the increased total utility of shipments completed via the UMR
system when a non-water alternative is included in the model. For exam-
ple, with demand 40 percent greater than the baseline and no alternative to
UMR transportation ( table 7 ), the FLT operating policy relative to RECPRIO
results in a reduction of 117.3 mean hours per tow time in the system and a
mean increase in total utility for all UMR shipments of approximately $236
million. The similar situation with a nonwater alternative ( table 8 ) shows a
reduction of only 27.3 mean hours per tow time in the system and a mean
increase in total utility for all UMR shipments of approximately $66 million.
This muted impact of the congestion mitigation measures when including
shippers’ alternatives can be critically important when assessing the direct
economic benefi ts of implementing the congestion mitigation measures.
Figure 4 presents the mean total utility received by all UMR shipments
associated with the three proposed congestion management measures and
also with RECPRIO at selected levels of increased demand with a nonwater
alternative (denoted ALT in the fi gure) and without a nonwater alternative
(denoted NO ALT) enabled in the model. Two key observations are that the
inclusion of a nonwater alternative in the simulation model increases the
estimates of total utilities, and also moderates the differences in total util-
ity levels for all levels of demand relative to the results without a nonwater
alternative. Clearly, if meaningful alternatives to system use are available,
then incorporating these alternatives in the simulation yields dramatically
different estimates of the increased utility levels of water shippers afforded
by implementing the congestion mitigation measures. This result shows
that when shippers have an alternative there is an implicit lower limit on
how far system performance can degrade before many shippers simply just
opt out. Because waiting time is an important component of the shipper’s
TJ 53.2_03_Sweeney-Campbell-Sweeney.indd 172TJ 53.2_03_Sweeney-Campbell-Sweeney.indd 172 27/02/14 2:04 AM27/02/14 2:04 AM
Sweeney, Campbell, and Sweeney: Integrating Random Utility with Simulation \ 173
utility function, as lock wait times lengthen from increased traffi c on the
river, shippers become more and more likely to utilize alternatives to barge
shipment, thereby self-limiting the realized level of system congestion.
The importance of appropriately modeling shippers’ choice to evaluate
the direct benefi cial economic impacts of alternative congestion mitigation
measures is highlighted in fi gure 5 , which displays the mean annual utility
of all generated shipments, whether or not transported on the UMR, with
FLT, HLP, and NEW, measured incrementally to the total utility of all ship-
ment under existing lock operating policy RECPRIO. When shippers are
modeled as having no alternative to using the UMR, we observe very large
incremental benefi ts to all shippers at increased demand levels for all three
congestion mitigation measures (the dashed lines in fi gure 5 ). However,
if water shippers can instead utilize alternative modes or destinations for
their shipments and this choice is explicitly included in the simulation
model, then the incremental direct economic benefi ts of all three conges-
tion mitigation measures to all shippers are dramatically reduced (the solid
Figure 4 Estimated Total Utility of All Shipments via the UMR for Selected
Congestion Mitigation Measures and Forecast Demand Levels with and
without a Nonwater Alternative
Source : Compiled from simulation model results.
TJ 53.2_03_Sweeney-Campbell-Sweeney.indd 173TJ 53.2_03_Sweeney-Campbell-Sweeney.indd 173 27/02/14 2:04 AM27/02/14 2:04 AM
174 / TRANSPORTATION JOURNAL™
lines in fi gure 5 ). For example, the incremental direct benefi ts to all poten-
tial UMR shippers in the study section associated with construction of new
locks is about $500 million less per year at greatly increased demand levels
if shippers have a relatively attractive alternative to shipment via the UMR.
Of course, the level of direct benefi t reduction depends upon the relative
quality of the alternatives to UMR transportation.
Conclusions and Further Research
The fundamental goal of this article is to illustrate the importance of incor-
porating users’ behavior in modeling congested freight transportation sys-
tems. By integrating a random utility model of shippers’ alternatives within
a detailed simulation model of a congested segment of the UMR naviga-
tion system, we generate results that clearly demonstrate the signifi cant
Figure 5 Total Utility (Direct Economic Benefi ts) for All Potential UMR
Shipments Incremental to RECPRIO with Selected Congestion Mitigation
Measures and Forecast Demand Levels with and without a Nonwater
Alternative Compiled from Simulation Model Results
Source : Compiled from simulation model results.
TJ 53.2_03_Sweeney-Campbell-Sweeney.indd 174TJ 53.2_03_Sweeney-Campbell-Sweeney.indd 174 27/02/14 2:04 AM27/02/14 2:04 AM
Sweeney, Campbell, and Sweeney: Integrating Random Utility with Simulation \ 175
impacts of explicitly modeling shippers’ choice when evaluating system
performance. Including users’ preferences and alternatives is an especially
important concern for evaluating the economic impacts of changes to a
transportation system, as the expected net economic benefi ts of structural
or operational improvements to the system cannot be accurately identifi ed
without understanding how users will react with and without the changes.
Our results show how the availability of an alternative to a given trans-
portation system places a limit on how congested the transportation sys-
tem can become. We demonstrate, in the case of the UMR, that estimates
of the expected direct economic benefi ts received by users of the system
are heavily dependent on the behavior of shippers and on the modeling
assumptions regarding the alternatives available to shippers. There is a
large potential gap between the expected economic benefi ts generated
by the system with and without the inclusion of a utility-based shipper’s
choice function. Further, the incremental economic benefi ts of different
congestion mitigation measures are shown to be signifi cantly lower if ship-
pers have access to meaningful alternatives to shipment on the UMR. More
broadly, when evaluating policy decisions in any transportation system, it
is important to include the behavior of those who will utilize the system.
There are numerous opportunities for future research. First, there are
many different mechanisms by which shippers may form their expectations
when deciding whether or not to utilize the system. In our simulation, we
employ a simple mechanism where potential tows expect the system to per-
form as it did for the last tow that used the system with an identical itiner-
ary. However, many other expectation functions could be modeled, such as
forward-scanning mechanisms that evaluate the likely future performance
of the system, weighted averages of past voyages, or risk adverse or seeking
mechanisms. A second avenue of future research is modeling an even more
complex system where the level of alternative deterministic utility is a func-
tion of the volume of shipments that choose that alternative. This would cre-
ate a rich multifaceted model of the complex interactions of the decisions of
many shippers over alternative transport modes and destinations for their
products. A more applied area for future research is to use this type of model
to undertake a more detailed assessment of proposed congestion mitigation
measures for the UMR, using a fi ner level of detail on shippers’ alternatives
to the UMR, as well as current and future river operations and costs.
A fourth vein of potential research is to examine the effects on shippers’
utilities of other congestion management measures such as more complex
system scheduling rules and fi nancial based measures such as congestion
TJ 53.2_03_Sweeney-Campbell-Sweeney.indd 175TJ 53.2_03_Sweeney-Campbell-Sweeney.indd 175 27/02/14 2:04 AM27/02/14 2:04 AM
176 / TRANSPORTATION JOURNAL™
based tolls, time-of-use pricing mechanisms, licensing fees, and lock usage
fees. Additionally, the UMR system consists almost solely of commercial
traffi c, which is in stark contrast from the composition of the traffi c using
most transportation infrastructure. The performance of systems geared
more toward noncommercial use with many different types of users would
also be an interesting area to examine.
Note
The authors thank the editor and three anonymous reviewers for their very insightful
and constructive comments that have helped improve this paper. We also thank Sven
Muller for helpful comments on an earlier version of this paper.
References
Alion Science and Technology. 2011. Micros Saint Sharp User Guide Version 3.6. Alion
Science and Technology Corporation, Boulder, CO.
Baindur, D., and J. M. Viegas. 2011. “An Agent-based Model Concept for Assessing
Modal Share in Interregional Freight Transport Markets.” Journal of Transport
Geography 19 (6): 1093–1105.
Ben-Akiva, M., and S. Lerman. 1985. Discrete Choice Analysis: Theory and Application to
Travel Demand. Cambridge, MA: MIT Press.
Bhat, C. R. 1998. “Accommodating Variations in Responsiveness to Level-of-Service
Measures in Travel Mode Choice Modeling.” Transportation Research Part A 32
(7): 495–507.
—— 2012. “Recent Developments in Discrete Choice Model Formulation, Estimation,
and Inference.” Transportation Research Part B 46 (2): 273–75.
Biles, W., D. Sasso, and J. Bilbrey. 2004. “Integration of Simulation and Geographic
Information Systems: Modeling Traffi c Flow in Inland Waterways.”
Proceedings of the 2004 Winter Simulation Conference, 1392–98.
Bucklew, K. J. 2007. “The Heartland Fast-freight Rail System.” Transportation Journal 46
(4): 36–41.
Carroll, J., and M. Bronzini. 1973. “Waterway Transportation Simulation Models:
Development and Application.” Waterway Resources Research 9:51–63.
Casavant, K., W. Wilson, D. Moore, and K. Miller. 2008. “Ohio River Transportation Needs
Survey.” IWR Report 08-NETS-R-01, US Army Corps of Engineers, Washington,
DC.
Dai, M., and P. Schonfeld. 1998. “Metamodels for Estimating Waterway Delays through a
Series of Queues.” Transportation Research Part B 32 (1): 1–19.
de Jong, G., and M. Ben-Akiva. 2007. “A Micro-simulation Model of Shipment Size and
Transport Chain Choice.” Transportation Research Part B 41:950–65.
TJ 53.2_03_Sweeney-Campbell-Sweeney.indd 176TJ 53.2_03_Sweeney-Campbell-Sweeney.indd 176 27/02/14 2:04 AM27/02/14 2:04 AM
Sweeney, Campbell, and Sweeney: Integrating Random Utility with Simulation \ 177
Eliasson, J. 2009. “A Cost-Benefi t Analysis of the Stockholm Congestion Charging
System.” Transportation Research Part A 43 (4): 468–80.
Fellin, L., S. Fuller, W. Grant, and C. Smotek. 2001. “Measuring Benefi ts from Inland
Waterway Navigation Improvements.” Journal of the Transportation Research
Forum 40 (2): 113–36.
Franzese, L., L. Abdenur, R. Botter, D. Starks, and A. Cano. 2004. “Simulating the Panama
Canal: Present and Future.” Proceedings of the 2004 Winter Simulation
Conference, 1835–38.
Golkar, J., A. Shekhar, and S. Buddhavarapu. 1998. “Panama Canal Simulation Model.”
Proceedings of the 1998 Winter Simulation Conference, 1229–37.
Gorman, M. 2008. “Evaluating the Public Investment Mix in US Freight Transportation
Infrastructure.” Transportation Research Part A 42 (1): 1–14.
Hall, R. W. 2003. “Transportation Queueing.” In Handbook of Transportation Science,
International Series in Operations Research and Management Science 56, edited by
R. W. Hall. New York: Springer.
Inland Marine Transportation System Capital Investment Strategy Team. 2010. Inland
Marine Transportation Systems (IMTS) Capital Projects Business Model Final
Report Revision 1. http://www.iwr.usace.army.mil/Portals/70/docs/Wood_doc/
IMTS_Final_Report_13_April_2010_Rev_1.pdf .
Jiang, F., P. Johnson, and C. Calzada. 1999. “Freight Demand Characteristics and Mode
Choice: An Analysis of the Results of Modeling with Disaggregate Revealed
Preference Data.” Journal of Transportation and Statistics 2 (2): 149–59.
Kidokoro, Y. 2006. “Regulatory Reform and Congestion of Urban Railways.” Transportation
Research Part A 40 (1): 52–73.
Li, T., E. van Heck, P. Vervest, J. Voskuilen, F. Hofker, and F. Jansma. 2006. “Passenger
Travel Behavior Model in a Railway Network Simulation.” Proceedings of the
2006 Winter Simulation Conference, 1380–87.
Mann, H., and D. Whitney. 1947. “On a Test of Whether One of Two Random Variables Is
Stochastically Larger than the Other.” Annuals of Mathematical Statistics 18 (1):
50–60.
Martinelli, D., and P. Schonfeld. 1995. “Approximating Delays at Interdependent Locks.”
Journal of Waterway, Port, Coastal, and Ocean Engineering 121 (6): 300–307.
Mitra, S. 2013. “Discrete Choice Model of Agricultural Shipper’s Mode Choice.”
Transportation Journal 52 (1): 6–25.
Mundy, R., J. Campbell, R. Nauss, D. Rust, L. D. Smith, and D. C. Sweeney II. 2005.
“Management Systems for Inland Waterway Traffi c Control.” Center for
Transportation Studies, University of Missouri–St. Louis.
National Research Council. 2004. “Review of the US Army Corps of Engineers
Restructured Upper Mississippi River–Illinois Waterway Feasibility Study:
Second Report.” National Academies Press, Washington DC.
Nauss, R. 2008. “Optimal Sequencing in the Presence of Setup Times for Tow/Barge
Traffi c through a River Lock.” European Journal of Operations Research 187 (3):
1268–81.
TJ 53.2_03_Sweeney-Campbell-Sweeney.indd 177TJ 53.2_03_Sweeney-Campbell-Sweeney.indd 177 27/02/14 2:04 AM27/02/14 2:04 AM
178 / TRANSPORTATION JOURNAL™
Oster, C. V., Jr., B. M. Rubin, and J. S. Strong. 1997. “Economic Impacts of Transportation
Investments: The Case of Federal Express.” Transportation Journal 37 (2): 34–44.
Palma, A., M. Kilani, and R. Lindsey. 2005. “Congestion Pricing on a Road Network:
A Study Using the Dynamic Equilibrium Simulation METROPOLIS.”
Transportation Research Part A 39 (7–9): 588–611.
Prud’homme, R., and J. Bocarejo. 2005. “The London Congestion Charge: A Tentative
Economic Appraisal.” Transport Policy 12 (3): 279–87.
Ramesesh, R. 1990. “Dynamic Job Shop Scheduling: A Survey of Simulation Research.”
OMEGA 16:43–57.
Rich, J., P. Holmbad, and C. Hansen. 2009. “A Weighted Logit Freight Choice Model.”
Transportation Research Part E 45:1006–19.
Ruiz, R., C. Maroto, and J. Alcaraz. 2005. “Solving the Flowshop Scheduling Problem with
Sequence Dependent Setup Times Using Advanced Metaheuristics.” European
Journal of Operations Research 165 (1): 34–54.
Smith, L. D., D. C. Sweeney II, and J. F. Campbell. 2009. “Simulation of Alternative
Approaches to Relieving Congestion at Locks in a River Transportation
System.” Journal of the Operational Research Society 60:519–33.
Smith, L., R. Nauss, D. C. Mattfeld, J. Li, J. Ehmke, and M. Reindl. 2011. “Scheduling
Operations at System Choke Points with Sequence-dependent Delays and
Processing Times.” Transportation Research Part E 47 (5): 669–80.
Swait, J., and M. Ben-Akiva. 1987. “Incorporating Random Constraints in Discrete Models
of Choice Set Generation.” Transportation Research Part B 21 (2): 91–102.
Swait, J., J. Louviere, and M. Williams. 1994. “A Sequential Approach to Exploiting
the Combined Strengths of SP and RP Data: Application to Freight Shipper
Choice.” Transportation 21 (2): 135–52.
Sweeney, D. C., II. 2004. “A Discrete Event Simulation Model of a Congested Segment
of the Upper Mississippi River Inland Navigation System.” Institute for Water
Resources, US Army Corps of Engineers, Washington, DC.
Tang, S., and H. Lo. 2008. “The Impact of Public Transport Policy on the Viability
and Sustainability of Mass Railway Transit—The Hong Kong Experience.”
Transportation Research Part A 42 (4): 563–76.
Ting, C. J., and P. Schonfeld. 1998. “Optimization through Simulation of Waterway
Transportation Investments.” Transportation Research Record 1620:11–16.
——. 2001a. “Effi ciency versus Fairness in Priority Control: Waterway Lock Case.” Journal
of Waterway, Port, Coastal and Ocean Engineering 127 (2): 82–88.
——. 2001b. “Control Alternatives at a Waterway Lock.” Journal of Waterway, Port, Coastal
and Ocean Engineering 127 (2): 89–96.
Train, K. 1980. “A Structured Logit Model of Auto Ownership and Mode Choice.” Review
of Economic Studies 47 (2): 357–70.
——. 2009. Discrete Choice Methods with Simulation. Cambridge: Cambridge University
Press.
TJ 53.2_03_Sweeney-Campbell-Sweeney.indd 178TJ 53.2_03_Sweeney-Campbell-Sweeney.indd 178 27/02/14 2:04 AM27/02/14 2:04 AM
Sweeney, Campbell, and Sweeney: Integrating Random Utility with Simulation \ 179
Train, K., and W. Wilson. 2004. “Shippers’ Responses to Changes in Transportation Rates
and Times: The Mid-American Grain Study.” IWR Report 04-NETS-R-02, US
Army Corps of Engineers, Washington, DC.
——. 2007. “Transportation Demands for Agricultural Products in the UMR and IL River
Basin.” IWR Report 07-NETS-R-2, US Army Corps of Engineers, Washington,
DC.
——. 2008. “Estimation on Stated Preference Experiments Constructed from Revealed
Choices.” Transportation Research Part B 42:191–203.
Tsamboulas, D., H. Vrenken, and A. Lekka. 2007. “Assessment of a Transport Policy
Potential for Intermodal Mode Shift on a European Scale.” Transportation
Research Part A 41 (8): 715–33.
Tumer, K., and A. Agogino. 2007. “Distributed Agent-based Air Traffi c Flow Management.”
Proceedings of the 6th International Conference on Autonomous Agents and
Multiagent Systems, 342–49.
US Army Corps of Engineers. 2004. “Integrated Feasibility Report and Programmatic
Environmental Impact Statement for the UMR-IWW System Navigation
Feasibility Study.” US Army Corps of Engineers, Rock Island District,
September.
——. 2008. “Final Re-evaluation of the Recommended Plan: UMR-IWW System
Navigation Study.” US Army Corps of Engineers, Rock Island District, March.
——. 2011. “Waterborne Commerce of the United States, Calendar Year 2011, IWR-
WCUS-10-2.” Institute for Water Resources, US Army Corps of Engineers,
Washington, DC.
——. 2013. “Lock Performance Monitoring System.” http://corpslocks.usace.army.mil/
lpwb/f?p=121:1:1980085043898216 .
Vovsha, P. 1997. “Cross-nested Logit Model: An Application to Mode Choice in the Tel-
Aviv Metropolitan Area.” Transportation Research Record 1607:6–15.
Wang, S., and P. Schonfeld. 2007. “Demand Elasticity and Benefi t Measurement in a
Waterway Simulation Model.” CD-ROM. Transportation Research Board
Annual Meeting, 1–11.
Wiegmans, B. W. 2010. “The Freight Transport Portfolio: A New Way to Analyze
Intermodal Freight Transport as Compared to Single-mode Road Transport.”
Transportation Journal 49 (2): 44–52.
Yai, T., S. Iwakura, and S. Morichi. 1997. “Multinomial Probit with Structured Covariance
for Route Choice Behavior.” Transportation Research Part B 31 (3): 195–207.
TJ 53.2_03_Sweeney-Campbell-Sweeney.indd 179TJ 53.2_03_Sweeney-Campbell-Sweeney.indd 179 27/02/14 2:04 AM27/02/14 2:04 AM
