You are on page 1of 11

European Journal of Operational Research 269 (2018) 782–792

Contents lists available at ScienceDirect

European Journal of Operational Research


journal homepage: www.elsevier.com/locate/ejor

Innovative Applications of O.R.

Effects of multipurpose shopping trips on retail store location in a


duopoly
Vladimir Marianov a,∗, H. A. Eiselt b, Armin Lüer-Villagra c
a
Department of Electrical Engineering, Pontificia Universidad Católica de Chile, Av. Vicuña Mackenna 4860, Santiago, Chile
b
Faculty of Business Administration, University of New Brunswick Fredericton, NB, Canada
c
Department of Engineering Sciences, Universidad Andres Bello Santiago, Chile

a r t i c l e i n f o a b s t r a c t

Article history: Models for retail store location typically assume that consumers make single purpose trips. However, fre-
Received 11 May 2017 quently, consumers purchase more than one product during the same trip. This is the first attempt to
Accepted 8 February 2018
propose including multipurpose trips in models locating commercial facilities selling non-essential prod-
Available online 15 February 2018
ucts. We model and analyze the case in which two firms sell one product or type of products each.
Keywords: Consumers buy one, both, or none, depending on the utility they obtain. We solve 10 0 0 instances and
Facility location report the results, which indicate that firms significantly benefit from taking into account multipurpose
Multipurpose shopping trips trips when locating their stores. Based on the difficulty to solve one version of the problem, we also
Store clustering derive an efficient heuristic algorithm for the solution and test its performance.
Leader-follower planning © 2018 Elsevier B.V. All rights reserved.

1. Introduction as Drezner and Eiselt (2002). An empirical research by O’Kelly


(1981) for Hamilton, Ontario, indicates that in that area, 63% of
Starting with Weber’s (1909) famous treatise, geographers grocery purchases are made on multipurpose trips, while 74%
and economists (joined later by business administrators, traffic of non-grocery trips are of the multipurpose type. Other re-
engineers, operations researchers, and others) have discussed the searchers have also found significant percentages of multipurpose
location of facilities. Many of the location models have practical trips among consumers. Popkowski Leszczyc, Sinha, and Sahgal
applications, a selection of which can be found in the recent book (2004) cite a number of empirical studies reporting that 30–50% of
by Eiselt and Marianov (2015). This paper addresses the location all shopping trips are multipurpose trips. All these results confirm
of non-competitive retail facilities or stores, which customers the prevalence of multipurpose trips in practice.
patronize when they purchase their products. In contrast, nearly all facility location models, including the
Planners take into account many factors when seeking locations widely used, as the p-median (see, e.g., Hakimi, 1964, 1965;
of one or more facilities. Among them are customers’ geographi- ReVelle & Swain, 1970; Blanco, Puerto, & El-Haj Ben-Ali, 2016) and
cal distribution, their demographics, buying power and behavior. In the simple plant location problem (Balinski, 1965; Correia et al.
customer choice models, i.e., settings in which customers have the 2016; Verter, 2011) explicitly assume that customers make single-
choice about which facility to patronize, their behavior is probably purpose trips i.e., the consumers travel to the facility, purchase
the most important factor in finding optimal locations, particularly a product or receive a service and return back to their origin. A
those of retail outlets. consumer purchasing two different products will make two trips.
One aspect of customer behavior that can frequently be ob- This assumption is also implicitly present in other popular models
served is multipurpose shopping. This behavioral pattern has as the Location Set Covering Problem and its extensions (Toregas,
been studied by economists and marketing researchers (see, e.g., Swain, ReVelle, & Bergman, 1971) and the Maximal Covering
Eaton & Lipsey, 1975, 1979, 1982; McLafferty and Gosh 1986; Location Problem (Berman et al. 2016; Church & ReVelle, 1974;
Ghosh & McLafferty, 1984: Arentze, Oppewal, & Timmermans, Colombo, Cordone, & Lulli, 2016). Almost all other facility location
2005; Mulligan, 1983; Thill 1992, among others); transportation models assume single purpose trips (Eiselt & Marianov, 2011;
researchers (O’Kelly, 1981, 1983), and operations researchers such Ljubić & Moreno, 2018; Marianov & Eiselt, 2016; Gentile, Pessoa,
Poss, & Roboredo, 2018).
∗ Diverging from the assumption of single purpose trips, Hodgson
Corresponding author.
E-mail addresses: marianov@ing.puc.cl (V. Marianov), haeiselt@unb.ca (H.A. (1990), Hodgson and Rosing (1992), and Berman, Krass, and Xu
Eiselt), armin.luer@unab.cl (A. Lüer-Villagra). (1995) study a different trip pattern. They formulated what they

https://doi.org/10.1016/j.ejor.2018.02.024
0377-2217/© 2018 Elsevier B.V. All rights reserved.
V. Marianov et al. / European Journal of Operational Research 269 (2018) 782–792 783

referred to as flow-capturing and flow-intercepting models. In these • Both firms are unaware that customers may engage in multi-
models, consumers travel between two known points, e.g., home purpose shopping, while customers use multipurpose shopping,
and work, following a known path, making their purchase along if their utility indicates that this is the most preferred course of
the way. This scenario could be considered as a two-stop trip in action.
which the origin, the second stop, and the consumers’ routes are • One firm (the leader) has already located its facilities un-
known. To the best of our knowledge, true multipurpose trips have aware that customers may use multipurpose shopping, while
been included in location models only by Araghi, Berman, and the other firm (the follower), locates its branches fully aware
Averbakh (2014), Li and Tong (2017), Berman and Huang (2007), that customers may employ multipurpose shopping.
and Suzuki and Hogdson (2005). Suzuki and Hodgson locate two • There is full cooperation of the two firms (or, equivalently, both
types of public facilities on a discrete space offering two different belong to the same owner or holding company) and they si-
services, A and B. Different percentages of the customers, known multaneously locate their branches fully aware that customers
ex ante, require A, B, or both services. Customers requiring both may engage in multipurpose shopping.
services obtain them during the same trip. The total distance trav-
eled by the customer is the sum of the distances from customer’s The difference between the results of the first model and the
origin to facility A, from facility A to facility B, and from facility B second and third models indicates the value to the firm(s) the
back to the customer’s origin, or the other way around. Single ser- consideration of multipurpose shopping can contribute. It will
vice consumers travel to the closest facility offering their required also indicate the differences between the locational patterns that
service and then return to their origin. Berman and Huang solve emerge as a result of multipurpose shopping and its inclusion in
the same problem, but they also analyze its properties and inves- the planning.
tigate heuristic solution procedures. Araghi et al. (2014) solve the The remainder of the paper is organized as follows. Section 2
problem on a tree. Finally, Li and Tong (2017) locate an activity to models customer behavior with the help of utility functions and
maximize accessibility when consumers perform several activities then formulates the basic model, in which both firms are assumed
on the same trip. to be unaware of actual customer behavior. Section 3 first formu-
In all models in the literature that involve multipurpose trips, lates a model, in which firms sequentially locate facilities with dif-
services are assumed to be essential, so that all demand must be ferent knowledge about customer behavior, followed by a model,
served, and the facilities are to be located to minimize the sum in which the two firms cooperate.
of consumers’ travel distances. This paper is the first to propose a Section 4 contains computational experience with the models,
model that includes multipurpose trips in facility location models, and Section 5 concludes the paper and proposes potential future
in which firms attempt to maximize their market area. Products extensions.
are assumed to be non-essential, so that consumers are not will-
ing to travel too far to obtain them. Furthermore, there are two 2. Modeling customer behavior and the basic model
firms that locate multiple branches (facilities) in order to maxi-
mize their respective market share. Each firm offers one specific This section will first model customer behavior with the help of
product at all of its branches and in that sense, the firms are not utility functions, followed by a slight adaptation of the well-known
competitors. maximum covering model (Church & ReVelle, 1974), which will
The central piece of our models is a feature that attempts to de- refer to as the basic model, and we use as a benchmark.
scribe customer behavior. In particular, we assume that each cus- In order to set the stage, consider first the space in which cus-
tomer has a utility function for each of the products that the stores tomers and facilities are and are to be located. In this paper, we
offer. Given these utility functions, customers will decide whether assume that all locations are in a discrete space, e.g., a network
or not to purchase either of the two products (or both), and, if so, with locations restricted to the nodes. Note that it is not required
which branch(es) to purchase them from. More specifically, cus- for the triangular inequality to hold. We also assume that there
tomers have reservation prices that indicate their willingness to pay exist paths between all pairs of points or nodes i and k, whose
for a product. Based on the relation between the full price for the attribute is total or generalized travel that includes, among other
product under consideration, i.e., the purchase price at the facil- components, the cost of parking, gas, and the value of time.
ity (the mill price) plus the travel costs on the one hand and the Customers are generally interested in two nonessential prod-
reservation price on the other, customers will decide whether or ucts S and P (e.g., shirts and pants), which are offered by a variety
not to purchase the product: if the full price is higher than the of branches of two firms, which, if confusion can arise, are also
reservation price, the purchase will not be made; is it lower, the referred to as S and P.
customer will make a purchase. The magnitude of the demand is Let N be the set of demand points and potential locations of
fixed, so that customers purchase either the full amount of their stores. We assume that customers belonging to the same demand
demand, or nothing at all. In case of multipurpose shopping, the node are homogeneous in their valuation of goods, i.e., all poten-
scenario is similar; here, the joint reservation price is either the tial customers at a node i have the same reservation (or indiffer-
sum of the individual reservation prices or a different value that ence) prices riS , i.e., the maximal price a customer at node i (or,
takes into account interaction effects. Consumers will engage in a for brevity, customer i) is willing to pay for one unit of product S
single-purpose or a multiple-purpose trip, whatever results in the and riP for the product P, generally with riS = riP . There is no loss of
largest utility to them. If neither of these have positive utility, they generality, as each node could be represented by as many copies
will not make any purchase. These model features are in line with as groups of consumers, segmented by their reservation prices.
the usual assumed customer rationality, except that, as opposed to The shirts are identical in all branches S and the pants are the
the usual location models, we do not assume that customers pa- same in all branches of P. Furthermore, all branches are equal,
tronize the facility closest to them, but the one that provides the so that the only distinguishing features are their distances to the
highest positive utility to them. In that sense, this is a generaliza- customers.
tion of the usual models. See, e.g., Espejo, Marín, and Rodríguez- We assume that a consumer i’s choices are guided by the util-
Chía (2016). ity he obtains for purchasing these products. More specifically,
Our contribution analyzes how multipurpose shopping behav- customer i’s utility uPi for purchasing a single unit of product P
ior affects the firms’ decisions. In order to do so, we consider the on a single purpose trip depends on three factors, viz., the cus-
following three models: tomer’s reservation price, the purchasing costs at the facility, and
784 V. Marianov et al. / European Journal of Operational Research 269 (2018) 782–792

the travel costs from the customer to the facility (Sedghi, Shavandi, occurs, if both firms simultaneously, while in the latter case, one
& Abouee-Mehrizib, 2017). Here, we assume that prices are fixed firm, the leader, will locate its branches first, before the second
(in the short run) and consumers have complete information on firm, the follower, will locate its branches. The leader – follower
the prices at every store before they set out on the trip. The latter concept was first introduced by the economist von Stackelberg
assumption may be justified via online searches, flyers, etc. (1943). Note the asymmetry between the two firms: while the
Formally, define the utility and reservation price for product S leader will have to guess how the follower will react to its actions,
as uSi and riS , respectively. Suppose now that firm P has one of its the follower only has to observe the actions of the leader, which
branches located at site k, at which it charges a price πkP for one have already occurred. This paper will consider both simultaneous
unit of its product. Given generalized transportation costs gab be- and sequential locations. Without loss of generality, we assume
tween points a and b, the utility of the customer located at site i that firm P is the leader and firm S is the follower in case of
purchasing one unit of product P, is sequential location.
The remainder of this section describes a benchmark problem,
uPik = riP − πkP − (gik + gki ), (1)
in which both firms assume that customers only engage in single-
Such as shortest paths, these utilities can be computed in purpose trips. In this case, it is irrelevant whether the firms locate
polynomial time for all demand point - candidate location pairs simultaneously or sequentially, as neither firm’s actions will affect
before solving any location optimization problem. Both calcula- the other firm. The formulation will follow Church and ReVelle’s
tions are assumed to be done in advance, as is, at least for the (1974) maximum capture problem, except that, instead of distances
case of shortest paths, customary in all location problems. Note as a coverage criterion, it uses positive consumer utility. We will
that once the price πkP at the store at k is known, a consumer at i formulate the problem for firm P below, the formulation for firm
is willing to travel and make the purchase if the utility he obtains S’s problem is equivalent. First define np as the number of stores
is positive, i.e., if (gik + gki ) < (riP − πkP ). In other words, the size to be located by firm P, and assume that price for the product is
of the market of a store at site k is given by the purchase price at the same everywhere, i.e., πkP = π P ∀k. Define ai as the number
that point and the reservation prices of the consumers at point i. of customers at site i, who use their utility functions to determine
The market size of facility S is computed in the same manner. their purchases, and let the parameters
Now, if a consumer were to purchase products from firms S 
1 uPik = riP − π P − (gik + gki ) > 0
and P on a multipurpose trip, and assuming that the cost matrix P
cik = ∀i, j (3)
is symmetric1 , the utility is: 0 uPik = riP − π P − (gik + gki ) ≤ 0
 
i jk = ri − π j + ri − πk − gi j + g jk + gki
uSP S S P P
(2) indicate the potential facility locations k, at which a customer at
site i can purchase a good with positive utility. Define now the
where uSP i jk
is the utility obtained by the customer for purchas- capture variables yPi , which assume a value of one, if a customer
ing products from S and P at sites j and k, respectively, and the at site i is captured, and define location variables xPk , which equal
term in parentheses is the cost incurred by the customer to visit 1, if firm P locates one of its stores at site k, and 0 otherwise. We
stores j and k and return to his origin. This implies that consumers can then formulate the maximum covering problem for firm P as
at i are willing to travel to visit both stores if (gi j + g jk + gki ) < 
SP : Max ai yPi (4)
(riS − π jS ) + (riP − πkP ), i.e., if the round-trip cost they incur to pur-
i
chase both products is short enough to keep positive their utility 
for such an action. Note that the round-trip in this case is the sum s.t. yPi ≤ P P
cik xk ∀i (5)
k
of three costs, one of which is the cost between a store belonging 
to firm S and a firm P’s store (an S and a P store, for short). If this xPk ≤ nP (6)
cost decreases, the utility for the customer visiting both stores in- k
creases, and multipurpose shopping becomes more attractive. The xPk , yPi ∈ {0, 1} ∀i, k. (7)
firms’ reaction is getting closer to each other, unless precluded by
other considerations, e.g., cannibalization of other stores belonging In the above formulation, the objective (4) maximizes the sales
to the same chain. (or capture) of firm P. Constraints (5) define the values of the
In all cases, consumers will travel and purchase only if the util- capture variables: A customer at site i is considered captured by
ity of an action is positive. Assuming that reservation prices and firm P, only if at least one if its branches is considered to provide a
purchase prices are fixed, there are maximum costs the customer positive utility to the customer. Note that customers at a demand
is willing to incur to obtain either one or two products. A natu- point could be within the market of more than one store of firm
ral assumption is that consumers are in a “purchasing mood”, i.e., P. Whichever store of firm P they choose is irrelevant to the firm.
whenever there is a purchase that results in a positive utility, they Constraint (6) restricts the number of facilities selling product P to
will make that purchase. It is worth mentioning that it is possible np , and constraints (7) define the domain of the variables. Using
that even though the utility of the purchase of a product at one this model, both firms can independently compute the locations
of, say, firm S’s branches k is negative (the purchase of product S of their branches and can thus determine their expected revenues,
does not justify an individual trip), a customer may still purchase by individually counting those consumers that will be captured by
the product at that site, if the utility uSP is positive, i.e., if it is making single purpose trips to one of their stores.
i jk
In what follows, we define the demand ai as consumers who
worthwhile to pick up a unit of S “along the way.”
will purchase nothing, one of the two products, or both of them,
Given this customer behavior, the firms’ interest is now to lo-
depending on whatever action has for them the maximum utility.
cate a number of branches, so as to maximize its sales. Given the
Once the firms have located their branches, customers can then
above results, this means that each firm will attempt to maxi-
determine at which facilities they satisfy their demand, by simply
mize the number of customers (demand) that obtain positive util-
using their utility functions.
ity when purchasing at its stores.
In location theory with multiple firms, we distinguish between
3. The models
simultaneous location and sequential location. The former case
This section will formulate two models: first, we derive and
1
For a non-symmetric matrix, the shortest path visiting stores j and k is chosen formulate a leader – follower model, in which the leader’s facilities
V. Marianov et al. / European Journal of Operational Research 269 (2018) 782–792 785

have already been located without considering the possibility of straints (15)–(17) ensure that a customer chooses to patronize the
customers using multipurpose shopping, while the follower now store that provides the highest utility to him, be it derived from
attempts to site its own facilities, considering the possibility that single or multipurpose shopping. Take, for instance, constraint
some consumers will make multipurpose shopping trips whenever (15), formulated for all P stores. If cik P has a value of one, meaning

it is convenient for them. This means that we must explicitly that a customer at i obtains a positive utility by visiting store k
incorporate the decision in the model. In order to formulate the and purchasing product P, and assuming for the time being that
follower’s decision model, we will need again to define location the second term on the right hand side is zero, yPi will be forced
variables xSj , which assume a value of one, if the follower locates to one, i.e., i will be captured by (within the market of) k for
one of his “S stores” at site j, and zero otherwise. We also need product P. Now, if the second term is not zero, i will not be forced
to define capture variables ySi , yPi , andySP , which assume a value to be captured by k. This second term is non-zero when there is
i
at least one cirt SP = 1 corresponding to a multipurpose trip i-r-t-i
of one, if a branch of the follower’s stores captures customers
(customer demand) at site i, if a branch of the leader captures the with a higher utility for i than the single purpose trip to t, where
demand at customer site i, and if leader and follower capture the r is an S store and t is a P store. Note that there are two condi-
demand at customer site i, respectively. The variables will assume tions: uSP
irt
≥ uPik and uSP
irt
≥ uSi j + uPik . The first condition makes the
a value of zero otherwise. In addition, we require parameters summation non-zero if the utility of a multipurpose trip is equal
 or greater than the utility of a single-purpose trip to k. The second
1 uSi j > 0
ciSj = ∀i, j (8) condition guarantees that the constraint works even if the triangu-
0 uSi j ≤ 0 lar inequality does not hold (which could make two single purpose

1 uPik > 0 trips have a higher utility than a multipurpose trip). This inequal-
P
cik = ∀i, k (9) ity will never choose capture by a single purpose trip if there is an
0 uPik ≤ 0
 equal or higher utility multipurpose trip. Conversely, if there is no
1 uSP >0 multipurpose trip that provides equal or higher utility to a con-
ciSPjk = i jk
∀i, j, k (10)
0 uSP
i jk
≤0 sumer, he will be captured by a single purpose trip to a store of
firm P. Constraints (16) performs the same task with stores of firm
Finally, we define Pˆ as the set of locations k at which P stores S. Constraint (17) forces capture by a multipurpose trip, whenever:
are located. The follower’s location problem can then be defined
as follows. (a) There is a j-k pair that provides a positive utility when vis-
  ited (first term on the right-hand side), and
MP F : Max ai ySi + ai ySP
i (11)
(b) There are no single purpose trips to an S or a P store (second
i i
 and third term on the right-hand side, respectively) with a
s.t. ySi ≤ ciSj xSj ∀i (12) higher utility than the multipurpose trip i-j-k-i (whenever
j a single and a multipurpose trip have the same utility, the

yPi ≤ P
cik ∀i (13) multipurpose trip wins); and
(c) Single trips to the two stores in the pair do not provide a
k∈Pˆ
 higher utility than the multipurpose trip, which could hap-
ySP
i ≤ ciSPjk xSj ∀i (14) pen if the triangular inequality does not hold (fourth term).
j∈N,k∈Pˆ
 Note that constraints (15)–(17) are analogous to the closest
yPi ≥ cik
P
− SP S
cirt xr ∀i, ∀k ∈ Pˆ (15) assignment constraints (Espejo et al., 2016), but instead of assigning
r∈N,t∈Pˆ∧ a demand to its closest facility, they assign it to the trip that
uSP
irt
≥uPik ∧
uSP ≥uSir +uPit has the highest utility. We call these constraints “highest utility
irt
 assignment” constraints. Constraints (18) and (19) ensure that no
ySi ≥ ciSj xSj − SP S
cirt xr ∀i, j (16) capture is counted more than once. Note that a demand node i
r∈N,t∈Pˆ∧ can be simultaneously captured by a P and an S store in single
uSP ≥uSi j ∧
irt purpose trips, but not by a single and a multipurpose trip at the
uSP
irt
≥uSir +uPit
  same time. Constraint (21) defines all variables as binary.
ySP SP S
i ≥ ci jk x j − cirS xSr − citP Having formulated a problem, in which the follower firm
r:uSir >uSP
i jk t∈Pˆ∧ locates its branches by considering single- and multi-purpose
uPit >uSP
i jk shopping behavior by its potential customers, consider now a
 situation in which the two firms fully cooperate as a single firm,
− ciSj cik
P S
xj ∀i, j, ∀k ∈ Pˆ (17)
so as to maximize aggregated capture.in what concerns location
uSi j +uPik >uSP
i jk
of their stores. Although total cooperation between two firms
ySP S
i + yi ≤ 1 ∀i (18) could be considered extreme, firms can collude by streamlining
their location decisions to maximize capture and avoid competing
ySP P
i + yi ≤ 1 ∀i (19) for customers’ purchase power. This strategy makes explicit what
mall owners (knowingly or otherwise) have done for a long time.
 In order to formulate a pertinent model, we now need to define
xSj ≤ nS (20)
additional variables xPk that assume a value of one if a P store is
j
located at p and 0 otherwise, as well as variables xSP , which equal
i ∈ {0, 1}
xSj , ySi , yPi , ySP ∀i, j. (21) jk
1, if there are S and P stores at j and k respectively. The model
Objective (11) maximizes the capture by the follower firm MPB for this situation is as follows:
S that results from single and multi-purpose trips. Constraints      
(12) allow variable ySi to take the value of one only if there is
MP B : Max ai ySi + ySP
i + ai yPi + ySP
i (22)
i i
some store j such that ciSj > 0, i.e., a store that provides customers

at i with a positive utility when visited. Constraints (13) and s.t. ySi ≤ ciSj xSj ∀i (23)
(14) accomplish the same for variables yPi , ySP i
, respectively. Con- j
786 V. Marianov et al. / European Journal of Operational Research 269 (2018) 782–792

 goods from both firms. This soft competition is characterized by the


yPi ≤ P P
cik xk ∀i (24) fact that it is not directly adversarial, as, depending on their util-
k ity function, customers may purchase from both firms. Given that

ySP
i ≤ ciSPjk xSP
jk ∀i (25) multipurpose trips reduce a customer’s distance to visit both firms,
j,k it will make it more attractive to patronize both stores and thus re-
 duce the competitive pressure. Applying the MPB model, in fact, is
yPi ≥ cik
P P
xk − SP SP
cirt xrt ∀i, k (26) how this soft competition is reduced the most.
r,t:uSPirt
≥uPik ∧
uSP ≥uSir +uPit
irt 4. Computational experiments with exact and heuristic

ySi ≥ ciSj xSj − SP SP
cirt xrt ∀i, j (27) methods
r,t:uSP
irt
≥uSi j ∧
uSP ≥uSir +uPit
This section will report the results of our computational tests.
irt
  The tests on integer programing models were run on an INTEL®
ySP SP SP
i ≥ ci jk x jk − cirS xSr − citP xtP Core TM i7-4790 CPU @ 3.60 gigahertz, with a 32 gigabytes RAM.
r:uSir >uSP
i jk
t:uPit >uSP
i jk
The operating system is Ubuntu 16.04.2 LTS, and the modeling
 language, AMPL version 20170111, with CPLEX 12.7 running using
− ciSj cik
P SP
x jk ∀i, j, k (28) 1 thread.
uSi j +uPik >uSP
i jk

4.1. Example instance


xSP S
jk ≤ x j ∀ j, k (29)
In order to obtain some insight into the workings of the
xSP P
jk ≤ xk ∀ j, k (30) models and the differences that the assumptions make, we first
generate some problems and assess the effects of the different
xSP S P
jk ≥ x j +xk − 1 ∀ j, k (31) assumptions. In particular, we randomly generate 100 demand
points on a 100 × 100 square with demands being uniformly
ySP S
i + yi ≤ 1 ∀i (32) distributed between 0 and 100. Distances are calculated according
to the Euclidean metric.
ySP P
∀i Our tests included different numbers of stores for the firms, and
i + yi ≤ 1 (33)
different values of what we call the initial consumer’s surplus for
 a product, i.e., the difference (riX − π X ) between the reservation
xSj ≤ nS (34)
price riX and the actual price π X for consumer i and product X. For
j
 these tests, we assumed that all consumers have the same reser-
xPk ≤ nP (35) vation price for a product, although the models allow distinctions
k between demand nodes. Furthermore, if a node houses different
types of consumers, we simply create as many co-located copies
jk , yi ∈ {0, 1}
xSj , ySi , xPk , yPi , xSP SP
∀i, j, k. (36)
of the node as consumer types, treating each copy as an indepen-
The objective (22) maximizes the total capture of the two dent node.
stores. Constraints (23)–(28) are equivalent to those of the previ- In order to streamline our discussion, we will present one ex-
ous model, except that now there are no stores located a priori. ample of our tests in detail. Fig. 1 consists of four parts: part (a)
Constraints (29)–(31) define the location of pairs of stores, through shows the locations of the “unaware” leader, follower and con-
variables xSP jk
, constraints (32) and (33) preclude double capture, sumers, i.e., none of them considers multipurpose shopping. Part
constraints (34) and (35) fix the number of stores of each type, (b) shows the facility locations in case leader and follower are un-
and constraints (36) define the domains of the variables. aware of the customers’ potential use of multipurpose shopping,
In this situation, there is no leader or follower as both types of while the customers will engage in such practice, in case their util-
stores locate simultaneously. While the model is completely sym- ity functions indicate that they should. In part (c), the leader P is
metric with respect to both players, the solution could show differ- unaware and locates as before, but the follower S and consumers
ent market captures of the two firms, depending on the reservation are aware of multipurpose shopping. Finally, part (d) represents si-
and purchase prices of their products and the number of stores of multaneous location of both leader and follower, both aware that
each firm. customers may use multipurpose shopping, and consumers who
It is important to note that the inclusion of consumers who de- may or may not use multipurpose shopping. Customers are shown
mand only one of both products is trivial. In such a case, it suffices as disks in case their demand is not captured by either facility, as
to set cirtSP = 0 ∀r, t and cP = 0 ∀r for consumers i who demand rectangles, if their demand is captured by the leader P, and as tri-
ir
only product S. The same applies to consumers who demand only angles, if they are captured by the follower S. Pentagons indicate
product P. These consumers will not make any multipurpose trips. that a customer is captured by firms P and S. The latter case can
Then, the demand could be divided in demand for product P only, occur, if either markets overlap when firms are not aware of mul-
for product S only, and consumers that decide whether purchasing tipurpose trips (Fig. 1(a)), or because there are consumers who do
one product, the other, or both. In all cases, the purchases will be engage in multipurpose trips (Fig. 1(b)–(d)). The size of the respec-
made only if the utility of such an action is positive. tive shape indicates the magnitude of the demand of the customer,
It is also important to note that our model is not competitive in with a minimum given by the size of the number inside the circle.
the strong sense, meaning that firms do not compete directly with A shape with a bold edge shows that the leader P (rectangle), a
each other in order to capture customers’ demand. However, even follower S (triangle) or both firms are co-located with a customer.
though the two firms offer different products, they still compete In our chosen example, the leader faces a consumer surplus
for the utility function of their customers. As an example, if the of (riP − π P ) = 30, while the follower’s product has a consumer
follower were to offer an attractive product by locating close to a surplus of (riS − π S ) = 50. As Fig. 1(a) shows, the unaware leader
customer, that customer may purchase only the follower’s product and unaware follower are not co-located, although there are a few
rather than that offered by the leader, or, alternatively, purchase demands that patronize both types of stores in single trips (e.g.,
V. Marianov et al. / European Journal of Operational Research 269 (2018) 782–792 787

Fig. 1. Both firms locate two stores each. Leader’s consumer surplus 30, follower’s consumer surplus 50. (a) Unaware leader, follower and consumers (SP). (b) Smart con-
sumers (MP). (c) Smart follower and consumers (MPF). (d) Smart leader, follower and consumers (MPB).

customers 64, 86, 90, 94). The leader P’s market is 1311, while the model, there is co-location. These results suggest using co-location
follower S’s market is 2297. This is not unexpected, since the fol- for the joint location problem.
lower faces the higher consumer surplus. When consumers con- We first show the situations in which co-location is optimal,
sider the possibility of multipurpose trips, they are willing to travel with the help of Fig. 2.
farther away and both leader and follower’s market increase to Suppose there is an optimal solution to the joint location prob-
1563 and 2810 respectively, i.e., by 19% and 22%, respectively. This lem in which there are at least two non-co-located stores p and s,
effect is mostly visible in the “southern” nodes, close to one of the belonging respectively to P and S, located at points k and j. The
positions of the leader. If the follower includes multipurpose trips prices at respective stores are π P and π S . Naturally, riS − π S > 0
in his planning, he co-locates with the leader and both markets be- and riP − π P > 0
come 3270, with gains of 149% and 42% over the base model (case
a), respectively. Note that the leader benefits the most, because the Proposition 1. Suppose p and s have either a single purpose shop-
follower surplus added to the leader’s market radius is more signif- pers’ market only, or share multipurpose shoppers with each other,
icant than the contribution of the leader to the follower, which has but not with other stores, as depicted in Fig. 2(a). A is the single
a larger surplus to begin with. This situation is shown in Fig. 1(c). shoppers’ market of p, C the single shoppers’ market of s and B the
When leader and follower cooperate and all parties include mul- multiple shoppers’ market shared by both, and their expressions are
tipurpose trips in their planning, the leader moves away from the as follows:
borders of the market towards its center, and both markets reach
  
A = i|(gik + gki ) < riP − π P ∧ gi j + g jk + gki ≤ riS − π S + riP − π P
a peak of 4312 each.
  
4.2. Heuristic for the both firms joint location problem B = i| gi j + g jk + gki < riS − π S + riP − π P .

    
Preliminary tests with the SP, MP and MPF models on 100-node C = i| gi j + g ji < riS − π S ∧ gi j + g jk + gki ≤ riS − π S + riP − π P
instances run in less than one second per instance. The model MPB,
however (cooperation of the two firms) took between 800 and The total market of p is A ∪ B, and the total market of s is C ∪ B.
137,0 0 0 seconds (≈ 38 hours) for the few tests we conducted, de- Then, if |A| ≥ |C |, co-location at k is optimal.
pending on the number of stores located of each type. The longest
Proof. The objective value is z∗ = K + |A ∪ B| + |C ∪ B|, where K is
time was for an instance with three stores each for the leader and
the term that includes all demand captured by other stores belong-
the follower. We consequently propose a heuristic for solving the
ing to both types. Then, if s co-locates with p, both stores’ markets
joint location problem, inspired by the preliminary tests.
are identical, composed entirely by multipurpose shoppers belong-
In the preliminary tests we observe that P and S stores ap-
ing to the set M:
pear co-located in the solutions in all instances of the MPB model, 
whenever it is possible. Furthermore, in most solutions to the MPF M = i|(gik + gki ) < riS − π jS + riP − πkP .
788 V. Marianov et al. / European Journal of Operational Research 269 (2018) 782–792

Fig. 2. Different cases for three or more non-co-located stores. Store s (point j) is at the center of the circle marked “C”. Store p (point k) is at the center of the circle marked
“A”.

Table 1
Optimal solution, heuristic solution and upper bound for some instances of the joint location problem.

CPLEX LB (Co-located stores) Upper bound (relaxation of Total heuristic Opt gap
max utility constraints) time (s)

nP nS P S Total Time P S Market Total Time Total market Time(s) (%)


Market Market market (s) Market market (s)

1 1 2728 2728 5456 840 2728 2728 5456 8 5456 2 10 0.00


1 2 2728 3553 6281 1877 2728 3553 6281 4 6281 3 7 0.00
2 1 3368 2728 6096 1348 3368 2728 6096 2 6096 1 3 0.00
2 2 4312 4312 8624 38564 4312 4312 8624 2 8624 1 3 0.00
1 3 2728 4232 6960 1184 2728 4232 6960 8 6960 3 11 0.00
3 1 3912 2728 6640 1200 3912 2728 6640 2 6640 1 3 0.00
2 3 4312 4627 8939 28883 4312 4627 8939 20 8939 11 31 0.00
3 2 4540 4312 8852 21825 4540 4312 8852 2 8852 2 4 0.00
3 3 4996 4996 9992 137309 4996 4996 9992 2 9992 1 3 0.00

Since A ∪ B ⊆ M, the market of p after co-location is at least the As the Table shows, the heuristic not only solves the joint loca-
same size as before. The market of s is also M, and since |A| ≥ |C |, tion problem in a short time, but it also finds optimal solutions for
its new market M ⊃ C ∪ B, i.e., is larger than its previous market. all the instances we were able to solve using CPLEX. Furthermore,
Actually, the joint market increases in |A| − |C | plus the possible the upper bound exactly equals the optimal solution in these cases.
demands around A, captured due to the added utility riS − π S . As
a consequence, in the case of Fig. 2(a), the co-located solution is
optimal.  4.3. Runs with 1000 instances

Proposition 2. Let p and s share multipurpose shoppers with each With this background, we ran tests on 10 0 0 instances, using
other, and p share market with other S stores, as depicted in Fig. 2(b). CPLEX for models SP, MP, MPF, and the heuristic for the joint
If |A| ≥ |C |, co-location at k is optimal. location problem. We generated the 10 0 0 instances exactly in the
same form as the example instance in a square of 100 × 100 units.
Proof. The proof is equivalent to that of the previous Proposition. The horizontal and vertical coordinates of the locations of nodes,
The market increases in |A| − |C | plus the possible extra market as well as their demand volumes, were randomly determined using
due to the added utility riS − π S .  a uniform distribution between 0 and 100. The metric is Euclidean.
The tests were run on the same computer, operating system and
The only case remaining is that of Fig. 2(c), in which |A| < |C |. software as the example instance. We used a range of values of
In this situation, store s could be co-located with p, in which case, leader surplus going from 2.5 to 100, in steps of 2.5, while the fol-
co-location is optimal if the extra market around A compensates lower surplus equaled 30. We report results for different numbers
the loss of |C | − |A|. Conversely, store p could be co-located with of stores opened by the leader and follower. In particular, (1, 1), (1,
s, and the solution would be optimal if |C | − |A| plus the extra 2), (2, 1), (2, 2). Note that for each combination of surpluses and
market around C would compensate the loss of part of D. store numbers, we run the 10 0 0 instances. The running time never
Since co-location seems to be optimal in many situations and exceeded 1 second for any instance of the SP, MP, MPF problems
provides a good solution in general, we propose a heuristic that solved using CPLEX. The joint location problem was solved using
assumes, for the joint location problem, that there are always the heuristic and the solution time never exceeded 30 seconds.
min{nP , nS } co-located stores. Then, a feasible solution (lower Table 2, along with Fig. 3, display the results of our tests
bound) for the joint location problem can be found by solving the for a large variety of the leader’s surpluses. For each problem,
model MPB in which capture of multipurpose shoppers is forced Table 2 shows averages for 10 0 0 runs and different values of the
to be possible only by co-located P and S stores. This is achieved leader’s surplus, given that the follower’s surplus is fixed at 30.
by solving model MPB with j = k in variable xSP jk
. In order to assess For these instances, the heuristic always found the optimal values
the quality of the solutions, we also include the computation (the gap between the upper and lower bounds was zero).
of an upper bound, consisting of relaxing the maximum utility Fig. 3 shows graphically the average values in Table 2. Note
constraints (25)–(28). that curves showing the leader’s (P) market, start from a small
For the joint location problem, Table 1 shows the objective val- value: for a surplus of 2.5, the leader has a very small market. The
ues, upper bounds, run times and gaps using CPLEX, the heuristic follower’s market, on the other hand, starts with a market around
and the upper bound computation. 1400, because of his consumer’s surplus of 30.
V. Marianov et al. / European Journal of Operational Research 269 (2018) 782–792 789

Table 2
Market sizes (Average) for 10 0 0 runs of each instance. SP, MP, MPF, refer to the models in Sections 3–5. Follower’s market marked gray. Follower surplus = 30, leader surplus
variable.
P’s surplus SP MP MPF Joint location lower bound Joint location upper Total heur.
bound time (s)
Market P Market S Market P Market S Market P Market S Market P Market S Total Time (s) Total Time (s)
market market
2.5 238 1339 314 1343 354 1349 1475 1475 2951 5 2951 2 7
5.0 318 1339 425 1352 548 1374 1613 1613 3226 5 3226 2 7
7.5 388 1339 530 1364 773 1419 1759 1759 3518 5 3518 2 8
10.0 473 1339 682 1390 1005 1484 1907 1907 3813 6 3813 2 8
12.5 560 1339 853 1428 1253 1579 2059 2059 4119 6 4119 3 9
15.0 652 1339 1021 1477 1467 1690 2216 2216 4431 6 4431 3 9
17.5 752 1339 1188 1524 1646 1792 2377 2377 4754 7 4754 3 10
20.0 855 1339 1372 1606 1826 1917 2537 2537 5074 7 5074 3 11
22.5 970 1339 1569 1719 2005 2061 2697 2697 5394 8 5394 3 12
25.0 1087 1339 1751 1845 2158 2194 2855 2855 5711 9 5711 4 13
27.5 1208 1339 2011 2046 2337 2355 3014 3014 6027 10 6027 4 14
30.0 1339 1339 2502 2502 2499 2505 3171 3171 6342 11 6342 4 15
32.5 1475 1339 2287 2252 2662 2663 3321 3321 6642 12 6642 5 17
35.0 1613 1339 2316 2234 2805 2808 3463 3463 6926 13 6926 5 19
37.5 1759 1339 2429 2291 2960 2960 3594 3594 7189 15 7189 6 21
40.0 1907 1339 2540 2334 3111 3111 3724 3724 7449 17 7449 6 22
42.5 2059 1339 2660 2393 3257 3257 3846 3846 7692 18 7692 6 24
45.0 2216 1339 2789 2443 3403 3404 3962 3962 7925 19 7925 7 26
47.5 2377 1339 2918 2500 3539 3539 4075 4075 8149 20 8149 7 27
50.0 2537 1339 3043 2530 3666 3666 4181 4181 8362 21 8362 6 27
52.5 2697 1339 3175 2554 3794 3794 4281 4281 8563 21 8563 7 28
55.0 2855 1339 3317 2597 3920 3920 4375 4375 8749 22 8749 6 28
57.5 3014 1339 3447 2619 4031 4031 4468 4468 8936 21 8936 6 27
60.0 3171 1339 3580 2636 4146 4146 4555 4555 9111 21 9111 5 26
62.5 3321 1339 3708 2632 4237 4237 4643 4643 9287 20 9287 5 24
65.0 3463 1339 3828 2627 4321 4321 4724 4724 9448 19 9448 4 23
67.5 3594 1339 3939 2660 4403 4403 4799 4799 9598 17 9598 4 21
70.0 3724 1339 4048 2697 4486 4486 4870 4870 9740 16 9740 3 20
72.5 3846 1339 4147 2678 4559 4559 4925 4925 9850 16 9850 3 19
75.0 3962 1339 4247 2713 4622 4622 4964 4964 9929 16 9929 2 18
77.5 4075 1339 4343 2747 4686 4686 4987 4987 9974 16 9974 2 18
80.0 4181 1339 4432 2772 4750 4750 4997 4997 9994 16 9994 2 18
82.5 4281 1339 4520 2808 4804 4804 4999 4999 9999 17 9999 2 19
85.0 4375 1339 4596 2831 4855 4855 50 0 0 50 0 0 9999 18 9999 2 20
87.5 4468 1339 4681 2850 4908 4908 50 0 0 50 0 0 9999 18 9999 2 20
90.0 4555 1339 4745 2863 4940 4940 50 0 0 50 0 0 9999 19 9999 2 21
92.5 4643 1339 4810 2893 4968 4968 50 0 0 50 0 0 9999 19 9999 2 21
95.0 4724 1339 4862 2937 4983 4983 50 0 0 50 0 0 9999 20 9999 2 22
97.5 4799 1339 4910 2945 4993 4993 50 0 0 50 0 0 9999 20 9999 2 23
100.0 4870 1339 4947 2951 4997 4997 50 0 0 50 0 0 9999 21 9999 2 23

Fig. 3. Both firms locate two stores each. Follower’s consumer surplus 30, leader’s surplus between 0 and 100.

As expected, for model SP, as the leader’s surplus increases, his addition to the growing surplus, there are consumers making mul-
market increases almost linearly. The follower’s market stays the tipurpose trips and being captured by P. This growth is almost lin-
same no matter what the leader’s surplus is, because, as per as- ear, except for the neighborhood of the singular point at which the
sumption, there are no multipurpose trips. consumer surpluses of both firms are the same (30). At that point,
When consumers engage in multipurpose trips (MP), Fig. in most cases, both firms solving SP co-locate because they are
3 shows that the leader P’s market as a function of his surplus solving exactly the same instance of the same problem, and conse-
is consistently higher than in the previous case (SP), because in quently, their markets are the same. As they co-locate, the multi-
790 V. Marianov et al. / European Journal of Operational Research 269 (2018) 782–792

Table 3
Markets for different settings, averaged over 10 0 0 instances. Also shown is the run time of the heuristic.
Surplus P,S (nP , nS ) SP MP MPF Joint location heuristic - lower bound Joint location Full heuristic
heuristic - upper
bound
Market P Market S Market P Market S Market P Market S Market P Market S Total Time Total Time Gap (%) Time
market market
(30,30)
(1,1) 724 724 1425 1425 1425 1425 1820 1820 3641 9 3641 3 0.00 12
(1,2) 724 1339 1379 1960 1424 2020 1816 2417 4233 9 4234 6 0.02 15
(2,1) 1339 724 1960 1379 2069 1477 2417 1816 4233 4 4234 3 0.02 7
(2,2) 1339 1339 2503 2503 2498 2504 3171 3171 6342 6 6342 4 0.00 11
(30,50)
(1,1) 724 1408 1396 1719 2047 2101 2803 2803 5606 10 5606 3 0.00 13
(1,2) 724 2537 1494 2840 2025 3163 2767 3640 6407 10 6437 4 0.47 15
(2,1) 1339 1408 2156 1838 2665 2219 3318 2795 6113 4 6117 3 0.06 6
(2,2) 1339 2537 2530 3043 3130 3456 4181 4181 8362 6 8362 4 0.00 9
(50,30)
(1,1) 1408 724 1719 1397 2295 2295 2803 2803 5606 13 5606 3 0.00 17
(1,2) 1408 1339 1837 2156 2290 2869 2795 3317 6113 20 6117 8 0.06 28
(2,1) 2537 724 2840 1494 3288 2205 3640 2767 6407 6 6437 4 0.47 10
(2,2) 2537 1339 3043 2530 3666 3666 4181 4181 8362 14 8362 7 0.00 21

Table 4
Market increase (Average, in percentages) in 10 0 0 runs of each instance.

Surplus P,S (nP , nS ) MP/SP MPF/MP Joint/MPF Joint/MP

P S P S P S P S

(30,30)
(1,1) 96.9 96.9 0.0 0.0 27.7 27.7 151.4 151.4
(1,2) 90.5 46.4 3.2 3.1 27.5 19.7 150.8 80.6
(2,1) 46.4 90.5 5.6 7.1 16.8 22.9 80.6 150.8
(2,2) 86.9 86.9 −0.2 0.1 26.9 26.6 136.9 136.9
(30,50)
(1,1) 92.9 22.0 46.6 22.3 37.0 33.4 287.1 99.0
(1,2) 106.4 11.9 35.6 11.4 36.6 15.1 282.1 43.5
(2,1) 61.1 30.5 23.6 20.7 24.5 26.0 147.8 98.5
(2,2) 89.0 20.0 23.7 13.6 33.6 21.0 212.3 64.8
(50,30)
(1,1) 22.0 92.9 33.5 64.3 22.2 22.2 99.0 287.1
(1,2) 30.5 61.1 24.7 33.1 22.0 15.6 98.5 147.8
(2,1) 11.9 106.4 15.8 47.6 10.7 25.5 43.5 282.1
(2,2) 19.9 89.0 20.5 44.9 14.1 14.1 64.8 212.3

purpose shoppers’ market reaches a peak. Past that neighborhood, setting Y. The first column shows the surplus of P and S, in that
the instances in which there is co-location decrease as surpluses order. The second column indicates the numbers of stores located
differ and the curve continues its previous trend. Note how the by the two firms. The next two columns show, for the leader
curve for the follower market does not grow much, as in this case, and the follower respectively, the average of the percentage of
the follower is not taking advantage of the multipurpose shoppers. increase in the size of the market obtained by the firms when the
The curves for model MPF start at different points, are always consumers make multipurpose trips (MP), compared to the case in
increasing, and tend to meet when the surpluses are the same. which there are single purpose trips only (SP), which is the usual
Beyond that point, the stores co-locate and the markets are the assumption in location models. These columns suggest that the
same for both firms. It is important to see that the MPF model market forecast by the firms, when multipurpose shopping is not
results in locations with market areas that are consistently higher considered, can be significantly underestimated. The following two
for both firms than those in the MP models. columns show the percentage in which the markets of both firms
The last curve shows the market captured by both firms when increase (or decrease, in one case) by the follower taking into
they locate jointly. Clearly, it is significantly larger than in any account multipurpose trips (MPF), as compared with the case in
other case. which both firms locate unaware of multipurpose trips, but shop-
Table 3 shows the results of the models and the heuristic, av- pers engage in this behavior (MP). Note that as it is the follower
eraged over 10 0 0 instances. The first column shows the surpluses who maximizes its market, the leader’s market can decrease, as it
of the firms. The second column, the number of stores located by does in one case. The next two columns show the percentage in-
each firm. The remaining columns are self-explanatory. The column crease of the markets when both firms locate cooperatively (Joint),
marked “Gap” shows the gap between upper and lower bound in relative to a situation in which only the follower locates thinking
the heuristic. The Table shows how as more actors become aware of multipurpose shoppers (MPF). Next, the two columns that show
of the multipurpose shopping behavior, the larger become the cap- how larger is the market when both firms locate together, as
tured markets. It also shows that the heuristic solves a problem compared to the MP case. This is the difference between the case
that takes a long time using CPLEX, and finds solutions of a very in which the firms know that consumers will engage in multipur-
good quality, with gaps that do not exceed 0.5%. pose trips, and locates accordingly, and the usual location model.
Table 4 shows the percentage of increase in the markets, as The last two columns show the percentage of difference between
actors become aware of multipurpose shopping. The notation X/Y the markets captured by firms that locate using the heuristic,
shows how larger is the market under setting X, compared to as compared to the markets they compute without thinking of
V. Marianov et al. / European Journal of Operational Research 269 (2018) 782–792 791

multipurpose shopping. This difference can be close to three stores, and purchasing it later online) and webrooming (the oppo-
times. site); see, e.g., Wolny and Charoensuksai (2016). For some related
In all cases, if the follower’s surplus is higher than that of the work on competing brick-and-mortar stores and online stores, see,
leader, consumers who consider multipurpose trips make a big e.g., Tang and Lin (2015). These behavioral patterns can radically
difference especially for the leader, who has more to gain when a change the way in which facilities are located, which makes them
very “attractive” follower (with a large market radius) appears in worthy of research.
the market and locates in the leader’s surroundings. However, the
follower’s market increase is smaller than that of the leader. Fi- Acknowledgments
nally, the follower’s gain is high when the leader’s surplus is large.
Finally, Tables 3 and 4 show that by jointly planning their This research was supported by Grant FONDECYT 1160025 and
location, both firms have a significant gain in demand, which by the Complex Engineering Systems Institute through grants ICM
suggests that this is a desirable strategy. P-05-004-F and CONICYT PIA FB0816. This support is gratefully
acknowledged. We also much appreciate the thoughtful comments
5. Conclusions and future research of the anonymous referees, who have enabled us to clarify a
number of issues and produce a more tightly written paper.
With a few exceptions, existing prescriptive location models
assume that consumers make single purpose trips to purchase a References
good. However, it is very common for shoppers to make multipur-
pose trips, in which they either perform more than one activity Araghi, M., Berman, O., & Averbakh, I. (2014). Minisum multipurpose trip location
problem on trees. Networks, 63/2, 154–159.
or purchase more than one item. Very few articles in the location Arentze, T. A., Oppewal, H., & Timmermans, H. J. P. (2005). A multipurpose shopping
literature address the problem of locating two types of stores trip model to assess retail agglomeration effects. Journal of Marketing Research,
when customers make multiple purpose trips. The few existing 42, 109–115.
Balinski, M. (1965). Integer programming: Methods, uses, computation. Management
articles make the fairly restrictive assumption that all demand Science, 12, 253–313.
must be satisfied and there are known proportions of customers Berman, O., & Huang, R. (2007). The minisum multipurpose trip location problem
that make multiple purpose trips. on networks. Transportation Science, 41, 500–515.
Berman, O., Kalcsics, J., & Krass, D. (2016). On covering location problems on net-
This paper presents prescriptive location models that allow works with edge demand. Computers & Operations Research, 74, 214–227.
studying this situation for the first time in a discrete space, in Berman, O., Krass, D., & Xu, C. W. (1995). Locating flow-intercepting facilities: New
which consumers choose what facility to patronize, have reserva- approaches and results. Annals of Operations Research, 60/1, 121–143.
Blanco, V., Puerto, J., & El-Haj Ben-Ali, S. (2016). Continuous multifacility or-
tion prices for each good, and have perfect information about the
dered median location problems. European Journal of Operational Research, 250/1,
prices and the locations of the stores they can visit. Consumers 56–64.
purchase one of the products, both, or none, depending on what Church, R. L., & ReVelle, C. (1974). The maximal covering location problem. Papers
action provides them with the highest utility. We assume that con- of Regional Science, 32/1, 101–118.
Colombo, F., Cordone, R., & Lulli, G. (2016). The multimode covering location prob-
sumers purchase up to two products, but the models are possible lem. Computers & Operations Research, 67, 25–33.
to extend for cases in which shoppers visit one isolated store and a Correia, I., & Melo, T. (2016). Multi-period capacitated facility location under delayed
strip center or a mall, selling multiple products, which is the next demand satisfaction. European Journal of Operational Research, 255/3, 729–746.
Drezner, T., & Eiselt, H. A. (2002). Consumers in competitive location models. In
task in this line of research. Z. Drezner, & H. Hamacher (Eds.), Facility Location: Applications and Theory
This work presents analyses and extensive computational expe- (pp. 151–178). Berlin – Heidelberg – New York: Springer Verlag.
rience. From these analyses, it is possible to confirm a practical Eaton, B. C., & Lipsey, R. G. (1975). The principle of minimum differentiation recon-
sidered: Some new developments in the theory of spatial competition. Review
fact: store location is dependent not only on the distribution of of Economics Studies, 42, 27–49.
the demand, but also on what stores and how many of them are Eaton, B. C., & Lipsey, R. G. (1979). Comparison shopping and the clustering of ho-
already in the region. Naturally, for each product, there are certain mogeneous firms. Journal of Regional Science, 19, 421–435.
Eaton, B. C., & Lipsey, R. G. (1982). An economic theory of central places. The Eco-
other products that exert influence. This influence is due to multi- nomic Journal, 92, 56–72.
purpose shopping trips. Eiselt, HA, & Marianov, V (Eds.) (2011). Foundations of location analysis. In the In-
We show how the inclusion of multipurpose trips in the con- ternational Series in operations research & management science (Vol. 155). New
York: Springer. 509p.
sumer behavior, results in locations that considerably differ from
Eiselt, HA, & Marianov, V (Eds.) (2015). Applications of location analysis. In the In-
those computed assuming single purpose trips only. In particular, ternational series in operations research & management science (Vol. 232). Cham:
the tendency of firms to locate close to each other, or co-locate, Springer-Verlag. 437p.
is much stronger, as this action maximizes their market. proposed Espejo, I., Marín, A., & Rodríguez-Chía, A. M. (2016). Closest assignment constraints
in discrete location problems. European Journal of Operational Research, 219,
models can be used to i) make corrections to the market estima- 49–58.
tions made with traditional models, and ii) find good locations for Gentile, J., Pessoa, A. A., Poss, M., & Roboredo, M. C. (2018). Integer programming
a firm that is entering a market or wants to add stores to their formulations for three sequential discrete competitive location problems with
foresight. European Journal of Operational Research, 265/3, 872–881.
already existing set of stores, in the presence of complementary, Ghosh, A., & McLafferty, S. (1984). A model of consumer propensity for multipur-
non-competing products stores already in the region. pose shopping. Geographical Analysis, 16/3, 244–249.
Given the very long solution times for the MPB problem with Hakimi, S. L. (1964). Optimal location of switching centers and the absolute centers
and medians of a graph. Operations Research, 12/3, 450–459.
an exact algorithm, we devise a heuristic that solves in a short Hakimi, S. L. (1965). Optimum distribution of switching centers in a communication
time the problem of the simultaneous location of both firms, pro- network and some related graph theoretic problems. Operations Research, 13/3,
viding very good solutions and including the computation of an 462–475.
Hodgson, M. J. (1990). A flow capturing location-allocation model. Geographical
upper bound that ensures this quality.
Analysis, 22/3, 270–279.
Future extensions we plan to study include solving the “leader” Hodgson, M. J., & Rosing, K. E. (1992). A network location-allocation model trading
problem, a very difficult bi-level optimization problem. Another off flow capturing and p-median objectives. Annals of Operations Research, 40/1,
247–260.
planned addition is to include demand elasticity, i.e., the fact
Li, R., & Tong, D. (2017). Incorporating activity space and trip chaining into facility
that the magnitude of the demand depends on the customers. A siting for accessibility maximization. Socio-Economic Planning Sciences, 60, 1–14.
further extension is to introduce utility functions with a stochastic Ljubić, I., & Moreno, E. (2018). Outer approximation and submodular cuts for maxi-
element. mum capture facility location problems with random utilities. European Journal
of Operational Research, 266(1), 46–56.
Finally, new behaviors have emerged together with the appear- Marianov, V., & Eiselt, H. A. (2016). On agglomeration in competitive location mod-
ance of the Internet, e.g., showrooming (shopping for a product in els. Annals of Operations Research, 246, 31–55.
792 V. Marianov et al. / European Journal of Operational Research 269 (2018) 782–792

McLafferty, S. L., & Gosh, A. (1986). Multipurpose shopping and the location of retail Suzuki, T., & Hodgson, M. J. (2005). Optimal facility location with multi-purpose trip
firms. Geographical Analysis, 18(3), 215–226. making. IIE Transactions, 37/5, 481–491.
Mulligan, G. (1983). Consumer demand and multipurpose shopping behavior. Geo- Tang, Q., & Lin, M. (2015). Showrooming vs competing. How does brand selection
graphical Analysis, 15, 76–80. matter? http://ink.library.smu.edu.sg/sis_research/2811/, Accessed 2/6/2018.
O’Kelly, M. E. (1981). A model of the demand of retail facilities, incorporating mul- Thill, J-C. (1992). Spatial duopolistic competition with multipurpose and multistop
tistop, multipurpose trips. Geographical Analysis, 13, 134–148. shopping. The Annals of Regional Science, 26, 287–304.
O’Kelly, M. E. (1983). Multipurpose shopping trips and the size of retail facilities. Toregas, C., Swain, R., ReVelle, C., & Bergman, L. (1971). The location of emergency
Annals of the Association of American Geographers, 73/2, 231–239. service facilities. Operations Research, 19, 1363–1373.
Popkowski Leszczyc, P. T. L., Sinha, A., & Sahgal, A. (2004). The effect of multi-pur- Verter, V. (2011). Uncapacitated and capacitated facility location problems. In
pose shopping on pricing and location strategy for grocery stores. Journal of Re- H. A. Eiselt, & V. Marianov (Eds.), Principles of location science (pp. 25–37). Hei-
tailing, 80, 85–99. delberg: Springer.
ReVelle, C. S., & Swain, R. W. (1970). Central facilities location. Geographical Analysis, Weber, A. (1909). Über den Standort der Industrien. 1. Teil: Reine Theorie des Stan-
2, 30–42. dortes. Tübingen. Translated as: On the location of industries p. 1929. Chicago, IL:
Sedghi, N., Shavandi, H., & Abouee-Mehrizib, H. (2017). Joint pricing and location University of Chicago Press.
decisions in a heterogeneous market. European Journal of Operational Research, Wolny, J., & Charoensuksai, N. (2016). Multichannel customer journeys: Mapping the
261/1, 234–246. effects of ZMOT, showrooming and webrooming. Marketing challenges in a tur-
von, Stackelberg H. (1943). Grundlagen der theoretischen Volkswirtschaftslehre (trans- bulent business environment. Part of the series developments in marketing science:
lated as the theory of the market economy). London: W. Hodge & Co. Ltd. Proceedings of the academy of marketing science (2014) (pp. 205–206). Springer
International Publishing Cham.

You might also like