Professional Documents
Culture Documents
Teleshopping Versus Shopping: A Multicriteria Network Equilibrium Framework
Teleshopping Versus Shopping: A Multicriteria Network Equilibrium Framework
COMPUTER
MODELLING
PERGAMON Mathematical and Computer Modelling 34 (2001) 783-798
www.elsevier.nl/locate/mcm
Abstract-This paper proposes a network equilibrium framework for the conceptualization, mod-
eling, and analysis of consumers’ selection of teleshopping versus shopping for a frequently purchased
product. The shoppers or consumers are assumed to be multicriteria decision-makers with each class
of consumer characterized by a finite number of criteria which may include, for example, time, cost, op-
portunity cost, as well as security/safety, among other criteria. We construct the network structure of
the consumers’ choices with an explicit identification of the nodes, links, paths, and origin/destination
pairs, and allow each class of consumer or shopper to weight his criteria in an individual manner on
each link of the shopping network. The framework is the first to capture teleshopping versus shopping
alternatives with predictive equilibrium flows. @ 2001 Elsevier Science Ltd. All rights reserved.
1. INTRODUCTION
The advent of telecommunications has transformed the manner in which people work and con-
duct their daily activities. In particular, the relationship between telecommunications and trans-
portation through such activities as telecommuting and teleshopping has been a topic of much
discussion and study by researchers (cf. [l] and the references therein) as well as practitioners
(cf. PI).
Although there is now a growing body of transportation literature on the topic of telecommut-
ing (see, e.g., [3]), the topic of teleshoppin g, a relatively newer concept, has received less attention
0895-7177/01/$ - see front matter @ 2001 Elsevier Science Ltd. All rights reserved. Typeset by dM-Tj$
PII: SO895-7177(01)00099-l
784 A. NAGURNEYet al.
to date. Shopping denotes a set of activities in which consumers seek and obtain information
about products and/or services, conduct a transaction transferring ownership or right to use,
and spatially relocate the product or service to the new owner [l]. Teleshopping can refer to a
case in which one or more of those activities is conducted through the use of telecommunications
technologies. Attention today is focused on the internet as the technology of interest, and in-
deed internet-based shopping is increasing phenomenally (e.g., [4]). In this setting, teleshopping
represents the consumer’s role in “business-to-consumer” (B2C) electronic commerce. Although
we place this paper in the context of internet-based shopping, the framework presented here can
apply more broadly, such as to using the telephone to order from a catalog.
Several researchers have explored the plethora of forms in which shopping is available today
(cf. [5]), the direct and indirect functions fulfilled by shopping (e.g., [6]), and conceptual and em-
pirical models of the choice between store shopping and teleshopping (e.g., [7,8]). However, there
has been virtually no study of the transportation impacts of teleshopping beyond speculation
(e.g., [VW
In this paper, we take on the challenge of developing a conceptual framework for the modeling
and analysis of teleshopping versus shopping. This framework differs from previous models of
“shopping mode” choice by being network-based, essentially modeling shopping mode and desti-
nation (in a simplified sense) simultaneously. By relating the shopping choice to a representation
of the transportation/communication network, we obtain information about the spatial charac-
teristics of consumers’ choices, information that is crucial to improving our understanding of the
transportation and other spatial impacts of teleshopping.
The framework is network-based and has the following features.
equilibrium models (cf. [13-17]), dating to the earlier work of Dial [18] and Dafermos [.19], who
introduced congestion into this problem setting. However, such an approach has not, hereto-
fore, been used in the setting of the examination of teleshopping versus shopping. Nevertheless,
our approach does build on the recent work of Nagurney, Dong and Mokhtarian (201 who devel-
oped integrated multicriteria network equilibrium models to formalize the conceptualisation of
telecommuting versus commuting (see also [21,22]).
The model developed here may be viewed as being complementary to the model of Nagurney,
Dong and Mokhtarian [20] in that both examine human activities surrounding transportation
and telecommunication tradeoffs with the former focusing on the activity of shopping and the
latter--on commuting. The model proposed in this paper, however, has the following original
features.
(a) We no longer limit the number of criteria facing a decision-maker to two or three and
consider, instead, the more general situation of a finite number of criteria, where the
number is as large (or small) 8s necessary. Of course, each class of consumer can weight
the criteria differently. Also, for the sake of generality and modeling flexibility, we allow
the consumers’ weights to be both class-dependent and link-dependent.
(b) We demonstrate how human, that is, consumer choices can be mapped into purchases of
goods and, ultimately, their delivery, within a consistent network equilibrium framework
with predictive capabilities.
(c) We incorporate information into the model explicitly.
The paper is organized as follows. In Section 2, the multicriteria network equilibrium model
of teleshopping versus shopping is developed. In Section 3, we derive the variational inequality
formulation and provide qualitative properties of the equilibrium pattern. In Section 4, we
propose a computational procedure, along with convergence results. We then apply the algorithm
in Section 5 to several illustrative numerical examples. Our summary and conclusions are given
in Section 6.
...
Physical
Transaction Links
Transportation Links
of consumer locations given by m and the number of information locations given by N, where N
also corresponds to the number of shopping sites. We denote the consumer location nodes (before
the shopping experience) at the top level of nodes by: 1,. . . , m, with a typical such node denoted
by j. We emphasize that each location may have many consumers. The second level of nodes,
in turn, corresponds to the information locations (and where the transactions also take place),
with nodes: m + l,... , m + n representing the virtual or internet-based locations and nodes:
m+n+ l,... , m + N denoting the physical locations of information corresponding to stores,
for example. Such a typical node is denoted by K. The third level of nodes corresponds to the
completion of the transaction with nodes: m + n + 1,. . . , m + N + n corresponding to internet
sites where the product could have been purchased (and where we have assumed information has
also been made available in the previous level of nodes) and nodes: m + N + n. + 1,. . . , m + 2N
corresponding to the completion of the transaction at the physical stores. A typical such node is
denoted by 1. The bottom level of the nodes are enumerated as: m + 2N + 1,. . . ,2m + 2N and
denote the locations of the consumers following the completion of the shopping experience (here
assumed to be the same as their original locations before beginning the shopping experience).
We now discuss the links connecting the nodes in the network in Figure 1. There are four sets
of links in the network. A typical link (j, K) connecting a top level node (consumers’ location) j
to an information node K at the second level corresponds to an access link for information. The
links terminating in nodes: m + 1, . . . , m + n of the second level correspond to telecommunication
access links and the links terminating in nodes: m + n + 1, . . . , m + N correspond to (aggregated)
transportation links.
As can be seen from Figure 1, from each second tier node K, there emanates a link to a node 1,
which corresponds to a completion of a transaction node. The first mn such links correspond to
virtual orders, whereas the subsequent links denote physical orders/purchases. Finally, we have
links emanating from the transaction nodes to the consumers’ (final) destination nodes, with the
TeleshoppingversusShopping 787
Having fixed the above ideas, we are now ready to present the notation which will allow us to
clarify the costs, demands, and flows on the network. In addition, we introduce the behavior of
the shoppers, who are assumed to be multicriteria decision-makers. Henceforth, we refer to the
network in Figure 1 as the shopping network.
We consider the network depicted in Figure 1, which we denote by G = [JV, L], where N
denotes the set of nodes in the network and L: the set of directed links. The origin nodes in the
network are the top level nodes corresponding to the consumers’ locations and the destination
nodes are the bottom level nodes. An origin/destination pair in this network corresponds to a
pair (j, m + 2N + j), for j = 1,. . . ,m. Thus, we assume that the shopper originates from the
same location to which the product is delivered. Let a denote a link of the network connecting
a pair of nodes and let p denote a path, assumed to be acyclic, consisting of a sequence of links
connecting an origin/destination (O/D) p air of nodes. There are nL links in the network and np
paths. Let W denote the set of m O/D pairs. The set of paths connecting the O/D pair w is
denoted by P, and the entire set of paths in the network by P.
In the shopping network framework, a path consists of a sequence of choices made by a con-
sumer. Forexample, thepathconsistingofthelinks: (1,m+l),(m+l,m+N+1),(m+N+1,m+
2N + 1) would correspond to consumers located at location 1 accessing virtual location m + 1
through telecommunications, placing an order at the site for the product, and having it shipped to
them. The path consisting of the links: (m, m + N), (m + N, m + 2N), and (m + 2N, 2m + 2N),
on the other hand, would reflect that consumers at location m drove to the store at location
m + N, obtained the information there concerning the product, completed the transaction, and
drove home. Note that a path represents a sequence of possible options for the consumers. We
will utilize flows to depict how many consumers actually select the particular paths with a zero
flow on a path corresponding to the situation that no consumer elects to choose that particular
sequence of links.
We assume that there are k classes of consumers in the network with a typical class denoted
by i. Let ft denote the flow of class i on link a and let xb denote the nonnegative flow of class i
on path p. The relationship between the link loads by class and the path flows is
where 6,, = 1, if link a is contained in path p, and 0, otherwise. Hence, the load of a class of
consumer on a link is equal to the sum of the flows of the class on the paths that contain that
link.
In addition, let fa denote the total flow on link a, where
La=&:, i=l
VaEC. (2)
788 A. NAGURNEY et al.
Thus, the total load on a link is equal to the sum of the loads of all classes on that link. Group the
class link loads into the knL-dimensional column vector f with components: {f,‘, . . . , f:, . . . , fi,“,
. . . , f$} and the total link loads: {fa,. . . , fn} into the nL-dimensional column vector f. Also,
group the class path flows into the Icnp-dimensional column vector 2 with components: {#,, , . . . ,
Xi,,> 1.
Note that in this network framework, we seek to determine the number of individuals who
select shopping options in the case of a specific product. We assume that the demand for the
product for each class of consumer is fixed and known a priori. We let the demand for the
product associated with origin/destination (O/D)p air w and class i be denoted by db and let
P, be the set of paths connecting the origin/destination pair w.
Clearly, the demands must satisfy the following conservation of flow equations:
that is, the demand by a consumer class at a location must be equal to the sum of the flows of
the class of consumer over all the available shopping options.
We are now ready to describe the functions associated with the links. In particular, we assume
that there are H criteria which the consumers may utilize in their shopping decision-making with
a typical criterion denoted by h. We assume that Cha denotes criterion h associated with link a,
where we have that
&a = Cha(f), Vu E 13, (4
where ta(f) denotes the time associated with traversing link a. In the case of an access link,
we would expect the function to be low for telecommunications and higher for transportation.
Another relevant criterion for shopping would be cost, that is, we may have
which would reflect (depending on the link a) an access cost, transportation or shipment cost, or
product price. In addition, another relevant criterion in evaluating teleshopping versus shopping
is opportunity cost since one may expect that this cost would be high in the case of teleshop-
ping (since one cannot physically experience and evaluate the product) and lower in the case of
shopping. Hence, the third criterion may be opportunity cost where
with oa(f) denoting the opportunity cost associated with link a. Finally, a shopper may wish to
associate a security cost with teleshopping, in which case the fourth criterion may be
and can incorporate specific link-dependent factors which could include, for a particular class,
factors such as safety, convenience, and sociability.
We then construct the generalized cost of class i associated with link a and denoted by ui as
._
u; = u; f 0 ) Vi, Va E C, (10)
and group the generalized link costs into the knL-dimensional column vector u with components:
{UA,..‘) UX,,)‘..) U: ,..., Ui,,}.
Link-dependent weights were introduced by Nagurney and Dong [22] to provide a greater degree
of generality and flexibility in the modeling of travel decision-making. Nagurney, Dong and
Mokhtarian [20], in turn, used link-dependent weights but assumed three criteria, in particular,
travel time, travel cost, and opportunity cost, in their integrated multicriteria network equilibrium
models for telecommuting versus commuting.
Let vi denote the generalized cost of class i associated with path p in the shopping network
where
vi = Cu6 (j) hap, Vi, Vp. (11)
aEL
Hence, the generalized cost associated with a class and a path is that class’s weighted combination
of the various criteria on the links that comprise the path.
Note, from the structure of the criteria on the links as expressed by (4) and the generalized
cost structure assumed for the different classes on the links according to (9) and (lo), that we
are explicitly assuming that the relevant criteria are functions of the flows on the links where
we recall that the flows correspond to the number of shoppers of a particular class that selects a
particular link. Indeed, it is not unreasonable to assume that criteria which would be relevant,
for example, to the access links (cf. Figure 1) would be flow-dependent and would be a function of
t,he number of shoppers traveling and/or accessing the information sites. Indeed, we can expect
more congestion on the telecommunication links as more shoppers seek information via this route;
similarly, we can expect more traffic congestion as more shoppers elect to travel to the shop sites.
Moreover, one can expect a similar kind of flow dependence (as well as congestion) as shoppers
search either virtually or physically for information and the products. Furthermore, one can
expect that the criteria associated with the transaction links will also be flow-dependent, as in
the case: for example, of the price of the product depending on the number of consumers that
seek to purchase the product at the particular site. Finally, clearly, the criteria associated with
the shipment,/transportation links can be expected to be flow-dependent as well.
Since we have assumed repetitive shopping for the product, the behavioral assumption that we
propose is that shoppers choose their shopping paths so that their generalized costs are minimized.
Specifically. the behavioral assumption utilized is similar to that underlying traffic assignment
models (see also. [23]) in that we assume that each cl~sss of shopper in the network selects a path
so as to minimize the generalized cost on the path , given that all other shoppers have made their
choices.
In particular, we have the following network equilibrium conditions for the problem outlined
‘above.
790 A.NAGURNEYet al.
For each class i, for all O/D pairs w E W, and for all paths p E P,,,, the flow pattern 2* is said
to be in equilibrium if the following conditions hold:
=
( )i
f-* XL, if 5: > 0,
%
2 XL, if xi* = 0.
(12)
In other words, all utilized shopping paths by a class connecting an O/D pair have equal and
minimal generalized costs. The term XL is simply an indicator whose value is not known a priori.
THEOREM 1.VARIATIONAL
INEQUALITY
FORMULATION.Thevariationalinequalityformulation
of the multicriteria network model satisfying equilibrium conditions (12) is given by: determine
$* E K, satisfying
u; = Wz;,& + w;,C& + . . . + wtff+ C~H-I), + (I- gda) Cfh, Vi, Va EC, (14)
where
Cha = %X(f) + c/La, VaEC, Vh, (15)
that is, each criterion differs from any other on a link solely by the fixed cost term cha. Each of
the weights preceding the respective criterion is assumed to be nonnegative.
Assume now that g(f), where g(f) is the column vector with components: {ga, . . . ,gn,,}, is
strictly monotone, that is,
=
Wfa (% (f ‘) + Cla) + f-da (Sa (f ‘) + Cza)
/ H-l \ 1
- wZ;a
(ga(f2) + Cla)+ & (ga(f”) + a) + **. + (ga (f2) + CHa)
]
But, (22) must be greater than or equal to zero, under assumption (20), and hence, the con-
clusion follows. I
Also, we have the following result.
THEOREM 5. LIPSCHITZ CONTINUITY. If the generalized link cost functions have bounded
first-order derivatives, then the function F that enters variational inequality (13b) is Lipschitz
continuous, that is, there exists a constant L > 0, such that
4. THE ALGORITHM
In this section, an algorithm is presented which can be applied to solve any variational inequal-
ity problem in standard form (see (13b)), that is, determine X’ E Ic, satisfying
The algorithm is guaranteed to converge provided that the function F that enters the variational
inequality is monotone and Lipschitz continuous (and that a solution exists). The algorithm is
the modified projection method of Korpelevich [26].
The statement of the modified projection method is as follows, where 7 denotes an iteration
counter.
STEP 0. INITIALIZATION. Set, X0 E K. Let 7 = 1 and let y be a scalar such t,hat 0 < y < l/L,
where L is the Lipschitz continuity constant (cf. [26]; see also (23)).
STEP 1. COMPUTATION. Compute _%7 bv_ solving the variational inequality subproblem
x7) 2
0 x7 + yF (X7-1)T - X7-1)T ,X - 0, VXEK.
((X’+yF(X7)T-X7-1)T,X-X7)>0. VXcK.
STEP 3. CONVERGENCE VERIFICATION. If max IXF - XT-‘1 5 E, for all 1, with E > 0, a
prespecified tolerance, then stop; else, set 7 =: 7 + 1, and go to Step 1.
We now give an explicit statement of the modified projection method for the solution of vari-
ational inequality problem (13a) for the multicriteria network equilibrium model.
Teleshopping versus Shopping 793
STEP 0. INITIALIZATION. Set j” E Ic. Let 7 = 1 and set y such that 0 < y < l/L, where L is
the Lipschitz constant for the problem.
2c
i=l c4E.c
pii7 + y (?A:
(p-1)) - p) x (fi - f(7) 2 0, V’JEK. (27)
STEP 3. CONVERGENCE VERIFICATION. If Ifi’ - ft7-‘I 5 E, for all i = 1,. . . , k, and all a E C,
with E > 0, a prespecified tolerance, then stop; otherwise, set 7 := 7 + 1, and go to Step 1.
We now state the convergence result for the modified projection method for this model.
THEOREM 6. CONVERGENCE. If the generalized link cost functions are monotone and have
bounded first-order derivatives, then the modified projection method described above converges
to a solution of variational inequality (13a).
PROOF. According to Korpelevich [26], the modified projection method converges to a solution
of the variational inequality problem of form (13b) provided that the function F that enters the
variational inequality (see (13b)) is monotone and Lipschitz continuous and a solution exists.
Existence of a solution for the model has been established in Theorem 2 and Lipschitz continuity
in Theorem 5. I
Hence, it follows from Theorems 4 and 5, that if the generalized link cost functions are of
form (14) and (15) with g being strictly monotone, and have bounded first-order derivatives,
then the modified projection method is guaranteed to converge to a solution of variational in-
equality (13a).
This algorithm was proposed by Nagurney [21] and applied for the computation of the mul-
ticriteria traffic network equilibrium in the case of a model with two criteria and fixed travel
demands. Here, however, we consider the more general situation in which there can be a finite
number of criteria associated with each class of consumer and each link. Moreover, here we allow
the weights to be not only class-dependent but also link-dependent and, consequently, the model
in this paper directly extends the model of Nagurney [21] to the case of a finite number of criteria
and to the setting of shopping versus teleshopping.
5. NUMERICAL EXAMPLE
In this section, we present a numerical example for which we then compute the equilibrium flow
pattern using the modified projection method described in the preceding section. Specifically, we
consider a situation in which there are consumers located at two locations with the possibility of
shopping virtually through telecommunications at two sites and physically at two other sites. We
assume that there are two classes of shoppers. The shopping network for the problem is given
in Figure 2. The network consists of two origin nodes at the top; two destination nodes at the
bottom, with two O/D pairs given by: wi = (1,11) and wg = (2,12).
There is a total of 20 links in the network, where eight links (in the first set) are access links,
four links (in the second set) are transaction links, and eight links (in the final set of links) are
shipment/transportation links. In the first set of links, four links represent access links to the
virtual sites through telecommunications and the remaining four links correspond to access links
794 A. NAGIJRNEY
et al.
to the physical sites through transportation. The links have been enumerated in Figure 2 for
data presentation purposes.
The modified projection method was coded in FORTRAN and the system used was the IBM
SP2 located at the Computer Science Department at the University of Massachusetts at Amherst.
We report both the CPU time (exclusive of input and output) and the number of iterations. We
used the equilibration algorithm of Dafermos and Sparrow (271 to solve the embedded quadratic
programming problems corresponding to variational inequality subproblems (27) and (28) since
the feasible set is that of a network.
The convergence criterion was that the absolute value of the path flows at two successive
iterations was less than or equal to 6 with E set to 10e5. The y parameter in the modified
projection method was set to .Ol. The demand for the product for each class of consumer was
equally distributed among the paths connecting the O/D pair to construct the initial feasible
path flow (and link load) pattern.
Locations of Consumers/Shoppers
Locations of Consumers/Shoppers
Denoting the O/D pairs by WI = (1,11) and 202 = (2,12), the demands for the products by
the two classes were: dLl = 80, d;, = 160, and d& = 150, &, = 75.
There were four paths connecting
each O/D pair. The paths connecting O/D pair ‘~1 were:
p1 = (1,9,13), P2 = 15), P3 = (3,11,17), and p4 = (4,12,19), whereas the paths connecting
(2,107
O/D pair 202 were: p5 = (5,9,14), p6 = (6,10,16), ~7 = (7,11, la), and PS = (8,12,20). we
assumed that th& were three criteria associated with each link and consisting, respectively, of:
time (Criterion I), monetary cost (Criterion 2), and an opportunity cost (Criterion 3).
The weights were constructed as follows: for Class 1, the weights were: wi.1 = .25, UJ.& = .25,
& = 1.0, w;,~ = .25, w;?~ = .25, wiS2 = 1.0, w:.~ = .4, w;,~ = .4, w;>3 = 1.0, w:,q = -5,
Teleshopping versus Shopping 795
For Class 2, we have the following: U& = .5, w& = .5, w& = .5, IL& = .5, ~$2 = .4,
W& = .4, & = .4, 20;s = .3, w;s = .7, w;* = .3, w$‘, = .2 w& = .6, w:a = .5, 20;s = .4,
w;‘s = .5, w:‘s = .7 w;‘s = .6, w& = .7, 2~:‘~ = .4 Wz”‘T= .3: w& = .a, w&J = .3, w&j = .2,
?&s = .6, W&, = (2, ?$ = 3, &;,+ = .9,‘?&, ‘= .i, W$ =' .4, W+ ‘= 3, W& = 4
w+ = .5, W&l = .9, WI,~~ = .5, W2,12 = -5, W&s = .7, Wl,l3 = -4, w2,1s = .6, w&s = .9,
WI,+ = .3, W&d = .4, w&4 = 1.0, w;,Ja = .2, w&s = .3, U&5 = .2, u&s = 1.0, w$ui = 1.0,
w;,,, = 1.0, w& = 1.0, w&7 = 1.0, w&7 = 1.0, U&s = 1.0, W&B = 1.0, W&j = 1.0,
WI,+) = 1.0, w&g = 1.0, w&9 = 1.0, w:,3 = 1.0, t&O = 1.0, W&J = 1.
The generalized link cost functions were constructed according to (9).
The time functions and the cost functions were given by
and
The modified projection method converged in .65 CPU seconds and required 382 iterations for
convergence. It yielded the following equilibrium multiclass link load pattern:
REFERENCES
1. P.L. Mokhtarian and I. Salomon, Emerging travel patterns: Do telecommunications make a difference?, In
HSM Conference Book (to appear).
2. Telecommunications, In The Economist. pp. 5-28, (September 1995).
3. P.L. Mokhtarian, A synthetic approach to estimating the impacts of telecommuting on travel, CIrhnn Studies
35, 215-241, (1998).
4. Business and the internet: The net imperative, In The Economist, pp. 5-44, (June 1999).
5. M. Batty, The retail revolution, Environment and Planning B 24. 1-2, (1997).
6. P. Underhill, Why We Buy: The Science oj Shopping. Simon and Schuster, New York, (1999).
7. F. Koppehnan, I. Salomon and K. Proussaloglou, Teleshopping or store shopping? A choice model for
forecasting the use of new telecommunications-based services, Envirvnment and Planning B: Planning and
Design 18. 4733489, (1991).
8. C.F. Manski and I. Salomon, The demand for teleshopping: An application of discrete choice models, Regional
Science and Urban Economics 17, 1099121, (1987).
9. J. Gould, Driven to shop? Role of transportation in future home shopping, 7!ransportation Research Record
1617, 149-156. (1998).
10. S. Handy and T. Yantis, The impacts of telecommunication technologies on non-work travel behavior, In
Research Report SWUTC/97/721927-fF, Southwest Region University Transportation Center, Center for
Transportation Research, The University of Texas at Austin, (1997).
11. R.E. Quandt, A probabilistic abstract mode model, In Studies in 7kavel Demand VIII. pp. 127-149, Mathe-
matica, Princeton, NJ, (1967).
12. M. Schneider, Access and land development, In &ban Development Models, Highway Research Board Special
Report 97, pp. 164-177, (1968).
13. R.B. Dial, Bicriterion traffic assignment: Basic theory and elementary algorithms, Transportation Scrence
30,933111, (199G).
14. R.B. Dial, Network-optimized road pricing: Part I: A parable and a model, Operatzons Research 47, 54-64,
(1999).
15. F. Leurent. Modelling elastic, disaggregate demand, In Proceedings of the First Meetzng of the Euro Working
Grozlp on Urban Trafic and ‘Pransportation, (Edited by J.C. Moreno Banos. B. Friedrich, hl. Papageorgiou
and H. Keller), Technical University of Munich, Munich, Germany, (19Y3).
16. F. Leurent, Cost versus time equilibrium over a network, ErLropean Journal of Operatzons Research 71,
205-221, (1993).
17. F. Leurent, The theory and practice of a dual criteria assignment model with continuously distributed values
of-times. In Transportation and -Pafic Theory, (Edit,ed by J.B. Lesort). pp. 455-477, Pergamon, Exeter,
England, (199(i).
18. R.B. Dial, A model and algorithms for multicriteria route-mode choice, ‘Pansportatzon Research 13B. 3ll-
316, (1979).
19. S. Dafermos, A Multzcnteria Route-Mode Chozce 7!rafic Equilibrzum Model, Lefschetz Center for Dyuamical
Systems, Brown University, Providence, RI, (1981).
798 A. NAGURNEY et al.
20. A. Nagurney, J. Dong and P.L. Mokhtarian, Integrated Multicriteria Network Equilibrium Models for Com-
muting Versus Telecommuting, Isenberg School of Management, University of Massachusetts, Amherst, MA,
(2000).
21. A. Nagurney, A multiclsss, multicriteria traffic network equilibrium model, Mathl. Comput. Modelling 32
(3/4), 393-411, (2000).
22. A. Nagurney and J. Dong, A Multiclass, Multicriteria ‘Rafic Network Equilibrium Model with Elastic De-
mand, Isenberg School of Management, University of Massachusetts, Amherst, MA, (2000).
23. M.J. Beckmann, C.B. McGuire and C.B. Winsten, Studies in the Economics of l’ransportation, Yale Univer-
sity Press, New Haven, CT, (1956).
24. A. Nagurney, Network Economics: A Variational Inequality Approach, Second and Revised Edition, Kluwer
Academic, Dordrecht, (1999).
25. D. Kinderlehrer and G. Stampacchia, Introduction to Variational Inequalities and Their Applications, Pren-
tice Hall, New York, (1980).
26. G.M. Korpelevich, The extragradient method for finding saddle points and other problems, Matekon 13,
35-49, (1977).
27. S.C. Dafermos and F.T. Sparrow, The traffic assignment problem for a general network, Journal of Research
of the National Bureuu of Standards 73B, 91-118, (1969).