The Hubbing
Phenomenon
According to Phillips (1987),
LTL hubbing provides the opportunity for
similar consolidation, only in terms of loads instead of passengers. By using breakbulk terminals (hubs), LTL
truckers can achieve similar economies of scale that truckload truckers enjoy
by definition. By travelling along well
established spokes, LTL drivers return home more often than truckload
drivers. Consequently, driver turnover
rates are considerably lower in LTL trucking.
According to Mele (1989), truckload trucking companies experience as
much as 85 to 110% driver turnover each year. Additional information regarding
LTL load movement can be found in Braklow et al. (1992). Other references that
are insightful for LTL hubbing include Winston (1981), Morrison and Winston
(1985), and Barker and
Hubbing in truckload trucking is
motivated differently. Truckload
truckers cannot benefit from increasing the number of city pairs served, nor
can they use hubbing to consolidate loads. Because most truckload truckers
carry loads point to point, hubbing would likely increase average load
circuity. The primary benefit of hubbing
in truckload trucking is to reduce tour length.
The challenge for researchers is to quantify the trade-offs between
savings associated with tour length and driver turnover relative to costs in
other areas.
The HUBNET
Simulator
A key tool designed to help in the
evaluation of H&S transportation networks is the HUBNET simulator. HUBNET is a knowledge-based, user-friendly
simulation system written primarily in the SIMNET II simulation language with a
C-shell user interface. The development
of HUBNET has been funded by J.B. Hunt Transport, Inc., a large truckload
carrier in the
The conceptual development of HUBNET and
system architecture are described in detail in Taha and Taylor (1994). Figure 1 provides an overview of the system
architecture of the HUBNET system. Those
readers interested in detailed information regarding HUBNET are referred to
Taha (1993). Those readers interested in
detailed information regarding SIMNET II simulation are referred to Taha
(1988).
*****Insert
Figure 1 Here*****
II.
Experimental Concerns
In this section, the authors describe
all aspects of the experimental design considered within this paper. Specifically, the authors describe the
differences in operational strategies between current and proposed
methods. They also describe the
experimental factors considered within this research to support H&S
implementation. Furthermore, this
section includes brief discussions related to performance measures and
verification and validation methods.
Current
Operational Setting
Typically, the truckload trucking
industry offers door to door service, taking the most direct route between
customer locations. Thus, any excess
circuity that may be associated with H&S networks is eliminated. Obviously, this procedure can carry a driver
far from his or her home base.
Dispatchers must assign available drivers to new loads in an effort to
minimize deadhead miles (empty miles from a driver's location to a pickup
point) while simultaneously satisfying driver and equipment requests. As pointed out by Powell et al. (1988),
dispatchers attempt to position drivers to reduce deadhead (first dispatch
empty miles), yet the complexity of this task in large dynamic systems makes it
near impossible.
Because truckload trucking organizations
know that many of the advantages of hubbing are not possible in their
environment, they are understandably reluctant to undertake the massive change
in operations that hubbing would require.
Instead, these organizations prefer smaller pilot studies to determine
the feasibility of H&S implementation in general terms and the efficacy of
its use in economic terms.
Currently, truckload truckers are
considering at least two limited implementation experiments. The first experiment is with "zone"
drivers. In this methodology, drivers
are dispatched only within a certain zone and are guaranteed to be home at
least one day each week. The second
experiment is with very limited H&S implementation consisting of regularly
scheduled trips between two hub cities along one lane or spoke. It is conjectured that both of these
strategies present viable options for truckload trucking companies if
implemented on a partial basis. In other
words, the system would likely fail if all drivers are converted to either zone
or lane drivers, yet the optimal percentages of drivers that should operate
using alternative methods is unknown. As
will be shown subsequently, the results of this research strongly support the
idea of limited implementation of H&S networks.
Alternative
H&S Settings
Several issues of experimental concern
arise when considering the implementation of H&S networks in truckload
trucking. Among these issues are the
problems associated with allocating drivers among hubs and among types (lane
drivers, local drivers, and non-network drivers), based upon freight density
analysis and other factors. Also among
these issues are the layout of the H&S network in terms of the number of
hubs, the hub locations, the spoke or lane locations, and service area
allocation. Furthermore, driver usage
rules must be considered as a part of experimental design.
To ensure experimental consistency in
this study, the number of drivers supporting the H&S network is equal to
the number of drivers required to support a non-network solution. In an effort to reduce excess circuity,
approximately 75% of all loads considered (those 75% that incur the least
circuity through network travel) are moved on the H&S delivery network.
Therefore, 25% of all drivers are designated as non-network drivers. These drivers carry 25% of the total freight
load using traditional point-to-point delivery methods. Because the percentage of network loads is
heavily dependent upon the network configuration, a number of pilot runs are
needed to determine the allowable circuity required to fix the percentage of
network loads near 75% for all scenarios examined in this paper. Furthermore, the 75% figure is somewhat
arbitrarily chosen. Therefore,
sensitivity analysis on this initial figure is included in the experimental
design for the extended testing presented in Harit et al. (1994). The concept of allowable circuity and
corresponding H&S network loading is also discussed in detail in Harit et
al. (1994).
Local and lane drivers are allocated to
individual hubs in a manner that is directly proportional to the average number
of loads using a given hub as a transshipment point, or originating or
destinating within the service area for the hub. Lane drivers are those that drive on
connecting spokes between hubs. Local
drivers are those that pick up and drop off loads within the service area for a
given hub. Local drivers are assigned to
a particular hub and do not leave the service area for that hub. Non-network drivers are also assigned to home
hub locations as a function of load density.
From an experimental viewpoint, this
research is classified as a three-factor experiment. The first factor is hub
location methodology consisting of three levels; distance-based hubbing, flow-based
hubbing, and hybrid hubbing. The second
factor is the number of hubs with two levels; 24 hubs and 32 hubs. The third factor is driver usage with two
levels; 1-hub tours and 2-hub tours. We
will now explain each of these factors and levels in greater detail.
The hub location methodology is probably
the most complex experimental consideration. The distance-based hubbing method
is heuristic driven. In distance-based
hubbing, hubs are placed in locations that are convenient in terms of being
one-day of driving time from one another.
Additional hubs are allocated within this structure in areas
characterized by high freight density.
Flow-based hubbing techniques attempt to place hubs in regions
characterized by low imbalance between originating and destinating loads, while
restricting feasible solutions to those that satisfy some degree of closeness
between hubs. To allocate hubs in this
manner, we have divided the Continental United States, Northern Mexico, and
Minimize:
|
|
subject to:
|
|
|
|
|
|
In the above formulation, │Cj│ is the absolute value of the load imbalance for
lat./long. grid j; i.e., the absolute value of (loads in - loads out). Xj is a binary decision variable that is 1 if a hub is
assigned to grid j and 0 otherwise. Aij
is 1 if grid i is contiguous to grid j and 0 otherwise. The objective function (1) seeks to locate
hubs in areas of low freight imbalance. The first constraint (2) ensures that
while a hub cannot be in each grid location, each grid has at least one hub in
a contiguous grid. The second constraint
(3) ensures that the correct number of hubs (24 or 32) are assigned. Finally, the last constraint (4) ensures that
each Xj is a binary integer.
The final hubbing methodology considered is that of hybrid hubbing. This methodology is largely based upon
heuristics, expert judgement, and the location of existing terminal locations
for J.B. Hunt Transport, Inc., who graciously provided supporting case study
data. It is called a hybrid method
because it includes some aspects of flow-based or density-based hubbing, and
includes some aspects of distance-based hubbing.
The second major experimental factor is
the number of hubs in the H&S network.
In this study, we have used two levels for this factor; 24 and 32. The 24-hub scenario is representative of a
small network that minimally covers the study region in
The third major experimental factor is
that of driver usage rules. Two levels
exist for this factor; 1-hub rules and 2-hub rules. Under 1-hub rules, lane drivers are not
allowed to travel more than 1 hub from home before transferring his or her
load. The next load must bring the
driver to his or her home hub. Under
2-hub rules, drivers are allowed to travel 2 hubs from home. Obviously, this factor attempts to examine
the trade-off between efficient load movement and driver tour length. Other scenarios involving N-hub rules with
increasing penalty functions
to force the
driver home did not perform well relative to 1-hub and 2-hub rules and have
been dropped from the factorial design.
Although not a part of the experimental
design presented in this paper, extended testing and sensitivity analysis on
the percent of non-network versus network loads have been performed. Additional testing has been completed
regarding the allocation of drivers to support network solutions. The results of this testing are presented in
Harit et al. (1994).
For convenience, a coding scheme has
been developed to aid in identifying scenarios within the factorial
design. Let (HLM/#H/TL) represent a scenario
with hub location methodology (HLM) which is equal to D for distance-based, F
for flow-based, and H for hybrid; number of hubs (#H) equal to 24 or 32, and
tour length (TL) equal to 1 or 2 hubs from home. For example (H/32/1) represents a scenario in
which 32 hubs are placed according to the hybrid method. Drivers are permitted to travel one hub from
home.
A final experimental consideration is
that of service area allocation to hubs.
Using the HUBNET user interface, 2E x 2E lat./long grids are assigned to the nearest hub. As a tie-breaker, grids are assigned to
minimize freight density imbalance.
Measures of
Performance
Five primary and many secondary measures
of performance are used to determine the efficacy of H&S networks in this
study. The five primary performance
measures include lane driver tour length, local driver tour length, average
miles per driver per day, first dispatch empty miles as a function of trip miles,
and average circuity as a function of trip miles. Obviously, these measures are
selected to demonstrate the trade-offs existing between tour length and
traditional measures of efficiency in truckload trucking when using an H&S
transportation network.
Verification and
Validation
The validity of supporting load density
data is guaranteed by using actual historical data supplied by J.B. Hunt
Transport, Inc. The HUBNET code has been
verified by a number of test runs designed to heavily stress the system in areas
of critical concern. Additionally,
HUBNET is installed and in use by J.B. Hunt Transport, Inc. This should help to ensure the validity of
the code and the effectiveness of technology transfer. Because J.B. Hunt Transport, Inc. has not
implemented a complete H&S network at this time, the validity of results is
even more important because actual comparisons with field data are not possible
at this time. Regarding the validity of the simulation results themselves, a
replication design is used to ensure the independence of data between runs.
III.
Analysis of Results
In this section, we discuss the results
of the experimentation described above.
We begin with a discussion of HUBNET use and computational
requirements. Subsequently, we present
the results of the experimentation, which includes statistical analysis of
validated HUBNET output via Analysis of Variance (ANOVA) methods.
HUBNET
Performance
The HUBNET simulator has proven to be an
effective and user-friendly tool for analyzing H&S networks in truckload
trucking operations. Of particular value
is the graphical network builder which allows for simple "what-if"
analyses in terms of network configuration. Figure 2 shows a typical input
screen that enables the user to graphically build the network in terms of hub
locations, spoke locations, and service area allocation. The screen shown in the figure involves the
placement of hubs.
*****Insert
Figure 2 Here*****
The computational needs of HUBNET, on
the other hand, are considerable for the experiments conducted in the course of
this study. Each scenario consists of
five independent replications of 5000 loads each. Each replication requires more than one hour
of CPU time on a Sun 690 mini-computer.
Performance
Evaluation
The performance of each of the twelve
experimental conditions of the factorial design are now discussed relative to
one another and relative to the current methods employed in the truckload
trucking industry. Figures 3 through 7
present the performance of the experimental conditions in terms of the five
performance criteria. Because the first
dispatch empty miles and the average circuity miles are considered proprietary
by J.B. Hunt Transport, Inc., these performance measures are presented in terms
of their percentage relative to trip miles.
The average miles driven per driver per day is also considered
proprietary. Therefore, this performance
measure is presented as a percentage, with the 100% baseline for this
calculation being the miles per driver per day for the traditional
point-to-point delivery system.
****Insert
Figures 3 Through 7 Here*****
Examination of Figure 3 reveals that the
most dominant experimental feature affecting network lane driver tour length is
that of driver usage rules. Obviously,
the 2-hub TL rule would result in longer tour lengths than the 1-hub rule. The surprising element is that the tour
length is increased much more than the factor of two that would seem
intuitive. It would appear that it is
much more difficult than expected to obtain a load going in the right direction
when a driver is more than one hub from home.
Figure 4 is also very interesting.
With respect to local driver tour lengths, the 2-hub rule generally
provides improvement over the 1-hub rule.
Upon first inspection, it may appear that the 32-hub networks perform
better than similar 24-hub networks.
Intuitively, one may think that 32-hub scenarios would have better
performance relative to local driver tour length, because service areas are
smaller. However, this anticipated
result is subsequently shown, using ANOVA, to be insignificant for this
experimental design. The observed system
behavior under the 2-hub rule is less intuitive, but is a result of the hubs
being fed more efficiently by lane drivers under the 2-hub TL rule.
Interestingly, this efficient feeding by lane drivers is accomplished with
longer lane driver tour lengths. The
most significant factor relative to local driver tour length, however, seems to
be that of network design configuration.
This is confirmed by ANOVA results later in this paper. The hybrid
configuration provides much better performance than either the flow-based or
distance-based hub layout designs. This
result is also somewhat intuitive, because the hybrid system provides smaller
service-areas than distance-based methods for those areas with dense freight
traffic. While the flow-based system
does not necessarily result in large service areas in dense freight regions, it
also does not explicitly use freight density as a criterion for service area
allocation. The hybrid method, on the
other hand, uses freight density as a primary consideration for hub location
and consequently for service area allocation.
The miles per driver per day for the
hybrid scenarios are quite low compared to distance-based or flow-based hub
scenarios, as noted in Figure 5. Furthermore, the 32-hub scenarios result in
fewer miles than their 24-hub counterparts.
With 32 hubs in the system, drivers spend less time driving and more
time waiting for loads. This is not
necessarily a bad result if 24-hub scenarios result in more circuity and first
dispatch empty miles. It is shown
subsequently in this paper that this is indeed the case. Figures 6 and 7 indicate that distance-based
hub layouts perform poorly in terms of first dispatch empty miles and in terms
of circuity. This somewhat negates the
outstanding performance of the distance-based hub location methodology in terms
of miles per driver per day. The hybrid
layout, on the other hand, results in relatively strong performance in terms of
circuity and first dispatch empty miles.
As mentioned earlier, the 32-hub layouts perform better than 24-hub
layouts in terms of first dispatch empty miles and circuity, making their
performance in terms of average miles per driver per day more attractive. Circuity, in point-to-point methods, is
brought about by the desire to get drivers to their home base as often as
possible.
Comparative
Performance
The objective of this section of the
paper is to identify the "best" of the H&S network designs
according to the five primary performance criteria used. To aid in this selection, Table 1 is
presented. In Table 1, each scenario is
rated according to five ratios that indicate performance relative to the best
observed performance among all twelve primary scenarios. For example, the
(H/32/2) scenario provides the best performance among all candidates in terms
of local driver tour length and first dispatch empty miles (indicated by 1.00
values in Table 1). Similarly, it is
only 1% worse than the (F/32/1) and (F/32/2) scenarios in terms of average
circuity. However, it is 555% worse than
(H/32/1) in terms of lane driver tours and achieves only 75% of the average
miles per driver per day achieved by the (F/24/1) scenario.
****Insert
Table 1 Here*****
Based on the information presented in
Table 1, the (H/32/1) scenario is selected as the best scenario. It is the best performer in terms of lane
driver tour length and ties for first in first dispatch empty miles. It is only 1% behind the best scenario for
circuity. The performance for local
driver tour length is 28% lower than the best scenario, yet much better than
all 2-hub driver rule scenarios.
Furthermore, the local driver tour length is less than one day on the
average, which is acceptable. The only
disappointing feature of the (H/32/1) scenario is that it achieves only 77% of
the miles per driver per day that the best scenario achieves. This is not a large concern when comparisons
are made to other H&S scenarios, because many of these other scenarios are
higher in terms of miles per day at the expense of first dispatch empty miles
or circuity. No revenue is generated for
this wasteful driving, so the (H/32/1) scenario is selected as the best H&S
network configuration based on the factorial design presented earlier.
The selection of this baseline is
important, because additional testing and sensitivity analysis is desired. Adding to the factorial design of experiments
would have been cost prohibitive and time prohibitive in terms of the CPU
requirements to examine a full factorial experimental design. By selecting a strong baseline and fixing
some of the H&S system parameters, additional sensitivity analysis is
possible through factorial experiments.
This reduces the computational requirements for the extended testing by
a factor of twelve. This sensitivity
analysis and additional testing is presented in Harit et al. (1994), along with
discussion concerning H&S implementation and business strategy.
The performance of the (H/32/1) baseline
compared to current point-to-point delivery methods of operation is also
interesting and important. The largest
improvement associated with H&S operations is in the area of driver tour
length. Using the (H/32/1) scenario,
average tour length for over-the-road drivers is less than two days. This represents almost a 90% improvement over
existing dispatching methods with point-to-point deliveries. Local drivers get home even more
frequently. Total miles per driver per
day are reduced by 14.6% in (H/32/1) compared to the point-to-point delivery
system. This is a significant reduction.
Related problems are noted in terms of circuity and first dispatch empty miles
as a percent of total trip miles.
Circuity increases from 3.5% of trip miles to 12.1% of trip miles. First dispatch empty miles increase from 5.6%
of trip miles to 8.2% of trip miles.
It is not immediately clear whether the
H&S network approach is a viable option based on this performance
evaluation. This is largely dependent
upon the cost of high driver turnover. Would drivers be willing to drive fewer
miles per day and accept the inevitable lower wages in return for more time at
home with their families? It is,
however, immediately clear that additional H&S evaluation is needed to
determine network efficacy in truckload trucking in general terms. It is also clear that it would be prudent to
test H&S systems in limited implementation plans as opposed to making a
basic change in operational techniques.
Both of these points are addressed in greater detail in Harit et al.
(1994).
Results of ANOVA
Testing
In Tables 2 through 6, the results of
ANOVA testing are presented for each of the five primary performance
criteria. These results are important
because they statistically validate the findings discussed in the Performance
Evaluation section of this paper. For
the lane driver tour length criterion, as shown in Table 2, only the tour
length (TL) is a significant factor at an alpha level of 0.05. TL is in fact highly significant at the 0.01
alpha level. Throughout the remainder of
this paper, a result is said to be statistically significant if found to be
significant at the α = 0.05 level and highly
significant if found to be significant at the α
= 0.01 level. The ANOVA
test presented in
Table 2 supports the previous discussion regarding the 2-hub versus 1-hub
driver rules.
****Insert
Tables 2 through 6 Here*****
For the local driver tour length
criterion, as shown in Table 3, both TL and the hub location methodology (HLM)
are statistically significant. As
discussed previously, the number of hubs (#H) criterion affects local driver
tour length due to the service area size.
Surprisingly, this result is not statistically significant as a main
effect. We also discussed the fact that
the 2-hub TL rule seems to result in more efficient hub feeding by lane
drivers. The 2-hub rule seems to be a
strong alternative strategy in terms of freight movement, but seems to be a
poor alternative relative to driver tour length. These results are supported by the ANOVA
testing, which shows the TL main effect to be statistically significant. The HLM is highly significant for this measure
of performance, with the best performance indicated for those scenarios which
tend to place hubs with small service areas in regions of high freight
density. The hybrid scenario is best in
this regard.
Table 4 presents the results of ANOVA
testing for the average miles per driver per day criterion. In this case, the main effects HLM and #H are
statistically significant with HLM once again attaining a highly significant
level. Obviously, distance-based hub
locations would result in more miles per driver per day. The #H result is also intuitive, because
32-hub scenarios have fewer inter-hub miles than 24-hub scenarios. The reader should use caution, however, in
using these results without considering circuity and first dispatch empty
miles.
Tables 5 and 6 present the results of
ANOVA testing for the first dispatch empty miles and average circuity criteria,
respectively. Because the circuity and
first dispatch empty miles are calculated in the HUBNET simulation pre-run,
these criteria are not affected by replication effects. For these criteria, differences are brought
about only when the scenarios are changed in terms of #H or HLM. Therefore, TL is not included in the ANOVA
testing for these criteria. In Tables 5
and 6, it is clear that HLM and #H are highly significant as main affects.
Additionally, the interaction affect of HLM/#H is highly significant in both
cases. As #H increases, the distance
from a hub to pick-up and delivery points decreases and circuity also
decreases. Obviously, as #H approaches
infinity, we would optimize performance relative to these two criteria. With a reasonable number of hubs, however, we
cannot achieve the same level of performance that is achievable with the
current point-to-point pick-up and delivery method. The hybrid hub location method once again
performs well due to hub proximity to dense freight regions, with well
positioned transshipment hubs along dense freight paths. The HLM/#H interaction effects are more
subtle but are evident in Figures 6 and 7.
The interaction affect is perhaps most easily described by discussing
the difference in performance between 24-hub and 32-hub scenarios as HLM
varies. In this regard, Table 7 is
helpful. Table 7 presents the difference
between 24-hub and 32-hub scenarios for circuity and first dispatch empty miles
as a percent of total trip miles. For
first dispatch empty miles, Table 7 indicates that this percent difference is
greater for some HLM scenarios than others.
The percent difference is least for flow-based hub location methods,
which is intuitive. The percent
difference is higher for distance-based hub locations, and highest for the
hybrid method of hub location. This
indicates that the hybrid method is more greatly affected by the shift from 24
to 32 hubs for this measure of effectiveness.
For the circuity performance measure, this disparity in percent
difference is also apparent, only the distance-based HLM provides the greatest
difference with the hybrid HLM results indicating the smallest difference. This result is also intuitive in that the
hybrid hubs are well placed relative to freight density. The addition of more hubs does not buy as
much in performance as in the distance-based HLM.
*****Insert
Table 7 Here*****
IV.
Concluding Remarks
Hub-and-spoke networks have been used
successfully in a number of transportation systems. It is still not clear whether or not H&S
systems can be profitable in the truckload trucking industry. This paper represents the first efforts known
to the authors in quantifying the effects of H&S in truckload trucking, and
therefore provides a first and unique contribution in this area. We will now discuss our findings in general
terms.
The simulation architecture presented in
Taha and Taylor (1994) and the HUBNET system developed by Taha (1993) have
proven to be effective for evaluating H&S networks in truckload
trucking. The only disappointment in
this regard is the considerable computational requirements associated with
running the actual simulation scenarios.
The results of experimentation indicate
that H&S networks perform well relative to some performance criteria and
poorly compared to others. The key
improvement is in the area of driver tour length. This improvement, however, is obtained at the
expense of other important criteria such as average miles per driver per day,
circuity, and first dispatch empty miles.
Only a detailed cost comparison using company specific data would reveal
whether or not this is a worthwhile trade-off.
Therefore, the authors are not prepared to make a blanket statement
regarding the efficacy of the H&S system at this point. However, the authors are prepared to make two
statements based on the results of the factorial experimental design evaluated
and presented within this paper.
Firstly, an opportunity seems to exist for limited implementation.
Secondly, additional testing is needed.
Limited implementation of H&S
networks could prove very beneficial in truckload trucking. With limited implementation, it would be
possible to establish regular driving lanes between two high density city pairs
that are well separated in terms of distance.
In this way, drivers would be assured of regular work and would be more
or less guaranteed mileage minimums.
This type of limited network could be extended between many potential
city pairs that have regular and heavy traffic.
As a result, capacity could be sold to customers according to a regular
schedule, potentially contributing to an increase in the market share for freight
between high volume city pairs. The
majority of loads could still be handled according to more traditional methods
until a near-optimal percentage of network loads could be determined. Limited
implementation is rapidly becoming a necessity under current operating
conditions because of the increasing intermodal emphasis in truckload
trucking. With more freight movement via
rail, trucks constantly pick up and drop off loads at fixed railroad terminal
locations. These fixed locations are de
facto hubs, which are fed by a service area on one end with dispersion to
another service area on the other end.
Additional testing is needed to support
H&S implementation. At this point,
we have identified solid baseline configurations, but have not optimized H&S
configurations. This further testing and
sensitivity analysis is taken up in Harit et al. (1994). Specifically, they address the idea of
limited implementation by performing sensitivity analysis on the percent of
jobs carried by the H&S network.
Furthermore, they perform sensitivity analysis on the number of drivers
available to support the system.
Finally, these authors present a discussion on the implications of
H&S usage for truckload operations.
While the analysis of H&S systems in
truckload trucking is far from complete, this paper provides an important first
step in quantifying the potential costs and benefits. Based on the results presented herein, it
would appear that H&S networks can play at least a limited role in the
truckload trucking industry.
V.
References
Barker, H.H., and
Sharon, S. (1981), "From Freight Flows and Cost Patterns to Greater
Profitability," Interfaces, Vol. 1, pp. 4‑20.
Braklow, J.W.,
Graham, W.W., Hassler, S.M., Peck, K.E., and Powell, W.B. (1992),
"Interactive Optimization Improves Service and Performance for Yellow
Freight System", Interfaces, Vol.
22, No. 1, pp. 147‑172.
Brenner, M.A.,
Leet, J.O., and Schott, E. (1985), Airline Deregulation, Eno Foundation
for Transportation: Westport, Connecticut, USA.
Harit, S.,
Taylor, G.D., and Whicker, G., "Hub and Spoke Networks in Truckload
Trucking: Sensitivity Analysis and Extended Testing," In review for
publication in The Logistics and Transportation Review.
Kanafani, A., and
Ghobrial, A.A. (1985), "Airline Hubbing--Some Implications for Airport
Economics," Transportation Research, Vol. 19A, pp. 15‑27.
Mele, J. (1989),
"Carriers Cope With Driver Shortage," Fleet Owner, Vol. 84,
No. 1 (Jan) pp. 104‑111.
Mele, J. (1989),
"Solving Driver Turnover," Fleet Owner, Vol. 84, No. 9 (Sept.)
pp. 45‑52.
Morrison, S.A.,
and Winston, C. (1985), "Intercity Transportation Route Structures Under
Deregulation: Some Assessments Motivated by the Airline Experience," American
Economic Review, Vol. 75, (May) pp. 57‑61.
Morrison, S.A.,
and Winston, C. (1986), Economic Effects of Airline Deregulation,
Brookings Institute: Washington, D.C., USA.
Phillips, L.T.
(1987), "Air Carrier Activity at Major Hub Airports and Changing Interline
Practices in the United States' Airline Industry," Transportation
Research, Vol. 21A, No. 3, pp. 215‑221.
Powell, W.B.,
Sheffi, Y., Nickerson, K.S., Butterbaugh, and Atherton, S. (1988),
"Maximizing Profits for North American Van Lines Truckload Division: A New
Framework for Pricing and Operations, Interfaces, Vol. 18, No. 1, pp. 21‑41.
Taha, H.A.
(1988), Simulation Modeling and Simnet, Prentice-Hall, Inc.: Englewood
Cliffs, NJ, USA.
Taha, T.T.
(1993), " An Integrated Modeling Framework for Evaluating Hub-and-Spoke
Networks in Truckload Trucking Operations, " Master's Thesis, Department
of Industrial Engineering, University of Arkansas, USA.
Taha, T.T., and
Taylor, G.D., (1994), "An Integrated Modeling Framework for Evaluating
Hub-and-Spoke Networks in Truckload Trucking Operations," The Logistics
and Transportation Review, Vol. 30, No. 2, pp. 141‑166.
Toh, R.S., and
Higgins, R.G. (1985), "The Impact of Hub and Spoke Network Centralization
and Route Monopoly on Domestic Airline Profitability," Transportation
Journal, Vol. 24, pp. 16‑27.
Wheeler, C.F. (1989),
"Strategies for Maximizing the Profitability of Airline Hub-and-Spoke
Networks," Transportation Research Record, No. 1214, pp. 1‑9.
Winston, C.
(1981), "A Disaggregate Model of the Demand for Intercity Freight
Transportation," Econometrica, Vol. 49 (July) pp. 981‑1006.