Professional Documents
Culture Documents
Loacation Based Ad - Interface
Loacation Based Ad - Interface
To cite this article: Dominik Molitor , Martin Spann , Anindya Ghose & Philipp Reichhart (2020)
Effectiveness of Location-Based Advertising and the Impact of Interface Design, Journal of
Management Information Systems, 37:2, 431-456, DOI: 10.1080/07421222.2020.1759922
Article views: 12
ABSTRACT KEYWORDS
Offline retailers increasingly use location-based coupons to target Randomized field
consumers in their vicinity in real-time. The rationale for the use of experiment; mobile pull;
location-based coupons is that geographic proximity increases the mobile push; mobile
analytics; behavioral effects;
relevance for consumers and, thus, the effectiveness of these cam-
interface design;
paigns. Two key interface-specific aspects of location-based coupon m-commerce; location-
applications that influence their effectiveness are 1) the provision of based advertising; mobile
distance information and 2) the distance-based ranking of coupons. advertising
The aim of this paper is to study and quantify the impact of these
two key aspects of interface design on the effectiveness of location-
based coupons. We conduct a randomized field experiment with
399,913 observations, including 3,499 different coupon promotions
offered by 3,930 different stores located in 2,392 ZIP code areas in
a large Western European country. Our results show that the most
effective interface design for location-based coupons is based on
a distance-based ranking. Furthermore, we find significant differences
in the impact of distance and display rank based on the interface
design and the actual geographic location of users. Our results thus
contribute to the understanding of consumers’ behavioral responses
to location-based advertising and provide important implications for
the interface-design of location-based advertising applications.
Introduction
Mobile advertising is the most important digital advertising channel based on ad spend-
ing, accounting for more than 75 percent of digital ad spending in the United States [32].
One of its key applications is location-based coupons. Functionally, location-based cou-
pons can be either sent as a push notification to users without explicit request (i.e., mobile
push) or provided based on explicit request (i.e., mobile pull) within specific applications
in which users can intentionally browse through available coupons [47].
Previous research has almost exclusively focused on investigating the users’ responses
to mobile push coupons. For example, existing studies have analyzed the users’ intention
to redeem coupons [15], the users’ probability of usage [33], and the effectiveness of
mobile short message service (SMS) based [5, 16, 18] and app-based [24] push targeting
strategies.1 Unlike the existing studies analyzing coupons that are pushed on mobile
devices, our study analyzes a location-based pull coupon application.
CONTACT Martin Spann spann@spann.de Institute of Electronic Commerce and Digital Markets, Ludwig-
Maximilians-Universität München, Geschwister-Scholl-Platz 1, D-80539 Munich, Germany
Supplemental data for this article can be accessed at here
© 2020 Taylor & Francis Group, LLC
432 MOLITOR ET AL.
and distance-based rankings. For firms, it is also important to learn about the behavioral
aspects necessary to optimize the effectiveness of location-based coupon campaigns via the
device’s interface.
The aim of this paper is to study and quantify the impact of an app’s interface-specific
choice architecture — via distance-based ranking and the provision of distance informa-
tion — on the effectiveness of location-based coupons, by answering the following
research questions:
Research Question 1 (RQ1): What is the most effective interface design for location-based
pull coupons and how do users’ responses depend on their geographic location?
Research Question 2 (RQ2): How does the magnitude of distance and ranking effects differ
depending on the interface design?
Literature Review
Mobile couponing, as a subset of mobile advertising, is still a fairly new topic covered only
by a few studies in previous literature. Table 1 shows a summary of the relevant prior
studies.
The early studies on mobile coupons focused on survey-based approaches to analyze the
consumers’ intention to redeem coupons (e.g., [15]), their usage probability (e.g., [33]), and
the perceived effectiveness of different targeting methods [44]. More recent literature has
almost exclusively focused on mobile push approaches, often with a location component. For
example, Andrews et al. [5] investigated how commuters in subways responded to push-based
434 MOLITOR ET AL.
mobile ads. In a similar study, Ghose et al. [24] analyzed the differences between commuters
and non-commuters in responses to mobile coupons. Luo et al. [36] used a field experiment to
estimate the impact of temporal and geographical distance — induced by mobile push
notifications — on the response to discounted movie tickets. In a related study, Fong et al.
[20] investigated the response to mobile promotions for movie tickets targeted at consumers
in the vicinity of a competing cinema, a strategy referred to as geo-conquesting. Dubé et al.
[16] replicated this study and found that the returns to geo-conquesting were reduced when
a competitor also implements a mobile targeting campaign. Fang et al. [18] analyzed whether
mobile promotions that are triggered by geo-fences — promotions sent when a consumer
enters a predefined area around the promoting store — were more effective than those
received without triggering by geo-fences. Danaher et al. [13] used observational data to
analyze how mobile coupons — sent out to visitors of a mall — influence the purchase
behavior of these visitors and found temporal and in-mall location effects on coupon
response. Li et al. [34] investigated the impact of weather and found that sunny weather
increased the consumers’ response likelihood to push notifications, while rainy weather
decreased consumers’ response.
In contrast to previous studies, we measure the effectiveness of a different coupon
delivery mechanism — location-based mobile pull — by using a large and novel rando-
mized field experiment that includes a great variety of product categories (13) and a great
variety of stores and geographies (3,930 stores in 2,392 ZIP code areas). Unlike the
previous literature, this approach allows us to investigate and vary the mechanism’s
interface design. More specifically, the interface design reflects the presence of multiple
coupons, enabling us to implement rankings that are informed by distance. This approach
contrasts with mobile push coupons that are usually standardized by the operating
system’s message design and that rarely include more than one coupon.
Theoretical Background
The implementation of mobile pull-based choice interfaces requires the consideration of
specific design mechanisms that govern the presentation of coupons on smartphones.
Specifically, a key aspect of mobile pull coupons is that users can proactively search and
JOURNAL OF MANAGEMENT INFORMATION SYSTEMS 435
choose between a variety of available campaigns that are presented by mobile coupon
providers [22]. As a result, the users’ choices strongly depend on the manner in which
coupon-specific information is being presented [31]. Thus, the app’s choice architecture
can influence the users’ decisions. This observation is why knowledge about the actual
design of the interface-specific choice architecture is also crucial for location-based pull
coupon applications. The two main factors that inform the underlying choice architecture
via the interface design are the availability of a ranking mechanism in combination with
distance (location) information. We therefore discuss the theoretical background of both
aspects based on previous studies on ranking and distance effects.
Ranking Effects
Previous studies in e-commerce have shown that consumers discover products mainly
through the use of search engines (e.g., [23]). Most search engines use rankings to present
their results to users [45]. Relevance-based rankings aim to reduce the search costs for
consumers by allowing them to find a product more quickly. Previous studies on the
consumers’ search and choice behavior in the context of search engines have revealed
strong ranking effects (e.g., [23]), which are manifested by a preference towards easily
accessible (i.e., highly ranked) choice options. In particular, the literature on sponsored
search advertising indicates that highly sorted options provide an economic benefit
evidenced by the interplay between top-positioned search results and an increasing
numbers of clicks and conversions [2, 25]. Higher-sorted coupons receive more attention
by consumers because these coupons are often assumed to be more relevant [45],
a phenomenon that is also denoted as the primacy effect [37].
It has been shown that the ranking effect is stronger for mobile users than for PC users,
ostensibly due to the device-specific display size [22]; to view information on smaller
screens requires the user to exert comparably higher cognitive efforts [6], sometimes also
referred to as search costs. However, the effect of rankings on choice can be endogenous if,
consistent with a standard practice of the search engines’ ranking algorithms, the con-
sumers’ prior behavior is used to inform the ranking [25]. Therefore, it is important to use
a randomized experiment to test the ranking effects.
In a location-based context, we contribute to these studies by using various ranking
mechanisms to analyze the ranking effects. We treat the ranking mechanisms (in combi-
nation with distance) as a design option that informs our experimental design. We test the
effectiveness of all ranking mechanisms by comparing the mean responses between
different treatment groups in a large-scale randomized field experiment. The ranking
effects, as an empirical outcome of our estimation model, are applicable to all treatment
groups.
Distance Effects
The distance between the user and the coupon-related point of sale is a natural candidate
to inform rankings for location-based applications. When combining distance and rank-
ings, coupons offered from nearby stores should be perceived as more relevant and,
accordingly, should be ranked higher. The impact of distance, for example, spatial
proximity based on the consumers’ home location, on consumer behavior has been
436 MOLITOR ET AL.
highlighted in existing studies involving the consumers’ choices related to offline retail
stores [7]. The results are unambiguous: increasing distances between consumers and
offline stores generally decrease the likelihood of a purchase. For example, Bell et al. [7]
showed that the distance between households and stores negatively influences the con-
sumers’ shopping decisions.
Recent literature has studied the role of location and distance in mobile Internet usage.
Ghose et al. [22] found that distance matters more for mobile Internet usage than it
matters for Internet usage via PCs. Molitor et al. [38] investigated the impact of contextual
factors (e.g., distance, geographical surroundings, weather) related to mobile device usage
and found that distance has a negative impact on the consumers’ choice behavior. These
results indicate that location matters for mobile shoppers [26]. The reason that distance
might matter more for mobile device usage (compared with PC usage) is that mobile
coupon applications are used as a ubiquitous information medium with the intention to
bring consumers to physical stores.
The availability of distance information itself may also have different implications.
Information concerning distance to stores can be assumed to be a signal of the underlying
transportation costs related to different coupons or store options [21]. Increasing distance
thus indicates higher transportation costs that decrease a store’s attractiveness (all else
being equal). However, distance information can also help consumers reduce uncertainties
about where to find stores or products. When considering the relationship between
saliency and transparency, the same coupon without distance information (low salience
and transparency) might be preferred to the one with distance information (high salience
and transparency), assuming that consumers have some knowledge of the location of the
coupon-related store. This expectation can be explained by the fact that in the absence of
distance information, consumers tend to miscalculate the actual distance to stores. The
underlying theoretical explanation can be found in overconfidence [14].
Overconfident consumers tend to overestimate their performance on particular tasks
that require abilities, such as knowledge about store distances in their local environment,
and, thus, underestimate how long it takes to reach the store (i.e., they are more optimistic
about the required time). This underestimation is even valid for tasks that they were
confronted with in the past. Similarly, individuals tend to underestimate the probability of
negative events, such as the required time to finish a project or to travel to a store [10].
Therefore, the same coupon for the same store at the same location might be perceived as
relatively less attractive if exact distance information is provided because the saliency of
actual distances and thus the transparency of transportation costs reduce the risk of
overconfidence. However, this phenomenon is only assumed to be valid in cases where
consumers have some knowledge about the location of stores.
because the average distance to the next store offering coupons is relatively high. However,
the average store travel distance can be perceived as a reference point. In case this
reference point is already relatively high (e.g., larger distances due to a more rural home
location), a marginal increase in distance is perceived to be less negative than is a similar
marginal increase based on a relatively low reference point (e.g., smaller distances due to
a home location close to urban areas). More remotely located — and geographically
isolated — consumers can thus be assumed to be less sensitive to distances. This argument
is supported by the results from prior research. Choi and Bell [12] found that consumers
with high offline shopping costs — due to geographical isolation in more rural areas — are
less price sensitive. Hence, the geographic location presents a contextual reference point
that has to be taken into account as it can lead to heterogeneous responses.
option for displaying content and ads with an explicit offline component (e.g., Yelp,
Foursquare, TripAdvisor, or Groupon).
Experimental Design
In the field experiment, we experimentally manipulated the two key aspects of location-
based pull coupons: the provision of distance information and the ranking mechanism.
For a period of 14 weeks, the collaborating coupon provider agreed to randomly assign
new users (who downloaded the focal application for the first time) to four experimental
groups. This approach ensured that only new users without prior experience with the
design and mechanism of the application were included in the experiment.5 These users
were exclusively assigned to one treatment group (between-subjects experimental design).6
The users also had no information about the existence of other treatment groups.
The ability to disentangle and test the impact of distance-based rankings and the
provision of distance information was provided by a full factorial 2 x 2 design that yielded
four experimental groups: “coupons sorted by distance, with distance information”
(Group 1; this is the provider’s usual design); “randomly sorted coupons, with distance
information” (Group 2); “coupons sorted by distance, without distance information”
(Group 3); and “randomly sorted coupons, without distance information” (Group 4).
Randomly sorted coupons hereby indicate that coupons in Groups 2 and 4 were randomly
sorted within an interval of 50 km of a user’s location. In contrast, coupons in Groups 1
and 3 were sorted by distance in ascending order with the closest store displayed first.
Omitting distance information (Groups 3 and 4) induces uncertainty about actual dis-
tances but also makes transportation costs less salient; randomly sorted coupons instead of
coupons sorted by distance decreases the relevancy of the ranking mechanism. Users may
have basic knowledge about the approximate location of local stores. Therefore, random
sorting (Group 4) can exhibit less relevance than can distance-based sorting (Group 3)
even without distance information in both groups. Figure 1 depicts screenshots of all four
treatment groups.
In total, 4,364 users participated in the field experiment. Each participant was randomly
assigned to one of the four treatment groups: 1,317 to Group 1, 1,011 to Group 2, 1,036 to
Group 3, and 1,000 to Group 4.7 The underlying randomization process was solved by
a server-side allocation mechanism (i.e., participants did not have the ability to select into
a particular treatment group for themselves). Furthermore, the server-side manipulation
ensured that users remained in the same experimental group throughout the entire
experiment.
Data Description
Our dataset consists of 399,913 impressions (observations) based on the visits of 4,364
unique users in 30,779 sessions. Similar to impressions in online advertising, mobile
impressions are based on the number of coupons displayed on the participants’ smart-
phones. Table 2 reports the summary statistics of all variables based on first 25 coupons
(ranks). Our focal variables are the distance to the offline redemption point (measured in
kilometers), the display rank, and the discount depth (measured as a relative discount).
The variable coupon type denotes the coupons’ classification: the coupons are either
JOURNAL OF MANAGEMENT INFORMATION SYSTEMS 439
21.8
Impressions Clicks
20.1
20
N = 399,913 N = 5,399
16.8 17.1
Frequency in percent
15 15.1
14.3
12.7
10.1
10
9
8.1
6.6 6.4 6.3
5.3 5.7
5
s
n
ry
er
n
er
e
ar
ty
es
ts
t
ia
or
ar
io
io
om
au
an
ed
ce
th
rb
C
af
Sp
sh
at
_B
O
Ba
ro
ur
tim
C
Be
_H
uc
Fa
e_
bs
G
ta
Ed
ng
ul
ur
es
lu
M
vi
is
C
R
Li
Le
as the number of previous clicks (before the current session) as an approximation of the
users’ previous experience.
The dataset includes 3,499 different coupon promotions. The coupon promotions were
offered by 3,930 different stores shops located in 2,392 ZIP code areas. Figure 2 depicts all
coupon-related product categories that were available and clicked; this pattern indicates
some differences in preferences revealed by the users clicking on particular product
categories.9 However, as the relative distribution of categories is similar to that in other
studies that make use of Groupon data from the United States [46], it permits us to make
similar arguments about the categories’ generalizability.
1.53
1.42
1.27
1.16
Group 1
Group 2
Group 3
Group 4
Group 1 Group 2 Group 3 Group 4
sorted by distance & randomly sorted deals & sorted by distance & randomly sorted deals &
With distance information with distance information without distance information without distance information
Figure 3. Comparison of click rates between treatment groups. Notes: N = 399,913. Margin of error:
Group 1: 0.0014, Group 2: 0.0014, Group 3: 0.0016, Group 4: 0.0014.
4
3.5
3
Click Rate in percent 2.8
2.3
2.1
2
1.9 1.9 1.9 1.9
2
1.8 1.8
1.7 1.7
1.6 1.6 1.6
1.5 1.5
1.4 1.4 1.4
1.3 1.3 1.3 1.3 1.3 1.3
1.2 1.2 1.2 1.2 1.2
1 1 1
.9
1
.8
.7
0
1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10
Figure 4. Click rate by display rank and group. Notes: The x-axis indicates the first ten display ranks for
Groups 1-4.
2.5
Group 1 Group 2
2.3
2.2 Group 3 Group 4
Click Rate in percent
2
.8 .8 .8 .8 .8 .8 .8 .8 .8 .8
.7 .7 .7 .7
.6 .6
.5 .5 .5 .5
.5
.4 .4
0
s
*
ar
e*
er
n*
s
es
ty
ry
ia
e
ar
er
t
io
an
rb
C
ur
om
io
au
ed
ce
sh
_B
th
af
ct
Ba
is
ur
O
m
ro
Be
_H
Fa
ua
Le
bs
ta
ti
G
es
ul
ng
Ed
lu
M
C
R
vi
Li
Figure 5. Click rates by group and product category. Notes: A star on the respective product category
indicates significant differences between groups within categories at 0.05 (p < 0.05). Specifically, the
differences in click rates between groups are significant for the following product categories: grocery,
multimedia, cafes, beauty & wellness, leisure & sport, and living & home, as well as education & culture.
The click rates rounded to one decimal place.
order itself does not seem to be the users’ dominating choice heuristic [45]. This result
indicates that the importance of distance and display rank depends on the choice context.
We also analyze the differences in click rates across groups and product categories,
shown in Figure 5. The observed click pattern, indicating a higher click rate for Groups 1
and 3, also holds for many product categories. The differences between treatment groups
are significant for the following categories: grocery, multimedia, cafes, beauty and well-
ness, leisure and sport, living and home as well as education and culture.
within specific geographic locations, such as city centers, as stores are mainly located in
urban areas with higher store densities. Users in a trade area — as part of a specific retail
market agglomeration — are also assumed to shop at centrally located stores within that
area. To this end, we divide the ordered distribution of mean distances to stores within the
top 25 display ranks per session into three tertiles (with approximately 1/3 of all observa-
tions each) to approximate users in urban, suburban and rural areas, similar to the
classification used in previous literature [39].
Figure 6 shows the click rates for each tertile and treatment group. The differences between
groups are significant (Urban: χ2 = 14.06, p < 0.01; Suburban: χ2 = 31.18, p < 0.01; Rural: χ2 =
50.76, p < 0.01). Tertile 1 includes all mean store distances between 0.01 and 2.71 km, and we
refer to the users’ respective locations in this interval as “Urban” because proximate stores can
mostly be found in densely populated areas (as is the case in most European cities). Tertile 2
includes all mean store distances between 2.72 and 8.68 km, and we refer to the users’
locations as “Suburban.” Tertile 3 includes all mean store distances between 8.69 and
50 km, and we refer to the users’ locations as “Rural.” We find that treatment-specific click
rates differ depending on the users’ geographical locations as indicated by the tertiles. Group 1,
the default group, performs particularly ineffectively in urban areas (Tertile 1, see Figure 6).
Distance information in combination with distance-based rankings seems to be less informa-
tive if the differences between distances are rather small. This may be explained by better
informed users as well as by more competition based on higher store densities in Tertile 1
(“Urban”). In contrast, close distances seem to be more informative if the coupons are sorted
randomly (both in Groups 2 and 4) and/or if distance information is omitted (Groups 3
and 4). Figure 6 also shows that coupons in suburban (Tertile 2) and rural (Tertile 3) areas are
more effective in treatment groups, such as Groups 1 and 3, which make use of distance-based
rankings. The users’ sensitivity to distance thus seems to be context dependent. The results are
consistent with the assumption that due to their higher average store travel distances, users in
1.52
1.5
1.45
1.39
Click Rate in percent
1.32 1.31
1.19 1.22 1.21
1.09 1.07
1
G1 G3 G2 G4 G1 G3 G2 G4 G1 G3 G2 G4
.5
0
Figure 6. Click rates by tertile (Top 25 ranks). Notes: N = 399,913. The order of the depicted groups is
based on the experimental design – the adjacent Groups 1 and 3 are sorted by distance; conversely,
Groups 2 and 4 are randomly sorted.
444 MOLITOR ET AL.
more remote locations tend to be more accepting of higher transportation costs. Users in more
remote locations (rural vs. urban areas) thus appear to be comparably more inclined to
respond to location-based coupons.
Econometric Analysis
This section outlines our econometric model and describes the underlying estimation process.
Model
We use a logit model to analyze the users’ coupon choices. Specifically, the use of
a hierarchical Bayesian logit model with mixed effects including Markov chain Monte
Carlo (MCMC) methods enables us to control for the users’ individual-level heterogeneity
[3]. The logit model in the hierarchical Bayes framework is based on the panel data
approach introduced by Chib and Carlin [11]. In our model, a user is assumed to choose
the utility-maximizing coupon option. Users reveal their coupon-specific preferences by
clicking on coupons. We model user i’s utility U as follows [43]:
Uijt ¼ vijt þ εijt (1)
The deterministic part of equation (1) is represented by vijt, whereas the stochastic
component is represented by εijt. The stochastic part contains nonsystematic or random
factors that affect the utility U. If users are browsing within the application, they choose
a coupon only if the underlying utility U is greater than 0. Specifically, the users’ direct
responses are represented by the latent response variable yijt = 1 if user i decides to choose
coupon j at time t; otherwise, yijt = 0.
We model the users’ coupon choice probability as a function of user-specific, coupon-
specific, location-specific, and time-specific variables. We thereby aim to measure the
impact of distance, display rank, and their treatment-specific differences, based on the
interactions between distance, rank and treatment groups on the users’ responses to
coupons. We also account for the treatment groups, discount, coupon type (discount vs.
promotion coupon), number of previous clicks, search depth (the number of impressions
per session), product category, time of day, weekday or weekend, week, and coupon
expiration. The number of previous clicks and the search depth are included in the
model to control for the users’ experience as well as the usage intensity of the application.
Controlling for the available product categories also helps us to account for category-
specific coupon preferences. Furthermore, the time of day, weekday, and week dummy
variables allow us to control for time-specific effects.13 In our model, the probability of
user i choosing coupon j at time t is given by the following mean function:
exp vijt
Prijt ¼ (4)
1 þ exp vijt
JOURNAL OF MANAGEMENT INFORMATION SYSTEMS 445
with the underlying utility of user i based on a Bernoulli distributed logit link function [11]
Uijt ¼ Xijt β þ Wijt bi þ εijt with vijt ¼ Xijt β þ Wijt bi (5)
bi , Nq ð0; DÞ (6)
where Xijt is a matrix of right-hand side covariates, and β is a vector of fixed effects.
Similarly, Wijt is another matrix of right-hand side covariates, and bi is a vector of
individual-specific random-effects: bi ~ Nq(0,D) and εijt are error vectors that are inde-
pendent and identically distributed based on a type I extreme value distribution.14 The
hierarchical specification of the mixed model in Equation (5) is completely denoted by the
following prior distributions D−1 ~ W(v0−1R0,v0), σ−2 ~ G(v0/2,δ0/2) and β ~ Np(β0,B0),
where W denotes the Wishart distribution of the random effects-specific precision matrix,
G denotes the Gamma distribution of the conditional error vector, and N denotes the
normal distribution of the fixed-effects parameters. The mean of the Wishart prior is R0,
whereas the Gamma prior mean is v0/δ0 [11].
Based on the general formulation of our model in Equation (5), we can specify the
random and fixed-effects matrices by adding the full set of explanatory right-hand side
covariates for Xijt and Wijt. We apply fixed and random effects for distance, display
rank, and discount depth, as we assume individual-level heterogeneity based on the
users’ idiosyncratic preferences related to their responses to the experimental variation
as well as discount depth. All the other variables are only estimated as fixed effects.
Estimation
We estimate our mixed-effects hierarchical Bayesian model by using Markov chain Monte
Carlo methods with a Gibbs sampling algorithm from the posterior distribution. We run
the Markov Chain for 25,000 iterations and use the last 20,000 iterations to calculate the
posterior means and standard deviations and report the 95 percent posterior interval. To
increase the estimation efficiency, we use a thinning parameter of 20 (i.e., keeping every
20th draw for the posterior distribution).
Results
In this section, we present our estimation results and predict the click rates based on
alternative coupon specifications.
Estimation Results
Table 3 presents the posterior means of our estimation results (based on top 25 display ranks).15
Geographical distance has a significant negative impact on the probability of choosing
coupons (β = −0.183; p < 0.05). Increasing distances between stores and the users’ real-
time location are assumed to increase the underlying transportation costs, which has
a negative impact on the users’ preferences toward these stores [7], as they have to travel
to a store to make use of a chosen coupon.
446 MOLITOR ET AL.
Similarly, we find that display rank has a negative impact on the users’ coupon response
behavior (β = −0.354; p < 0.05). The further down a coupon is displayed (numerically
increased display rank), the lower the probability of users choosing that coupon. Ghose
et al. [22] found a comparable result, which was explained by an increased cognitive
load — sometimes referred to as search costs — related to the screen size and requiring
more scrolling to discover lower-ranked coupons.
The interactions between distance and treatment groups show that users’ sensitivity to
distances becomes more negative when coupons are sorted randomly (i.e., Groups 2
and 4) than when the default distance-based ranking for coupons is employed (i.e.,
JOURNAL OF MANAGEMENT INFORMATION SYSTEMS 447
Group 1). This result suggests that distances to stores are an important driver of coupon
response: randomly sorted coupons increase the negative impact of geographical distance.
More precisely, for randomly sorted coupons, the distance effect is slightly more negative
if the distance information is provided (Distance×Group2: β = −0.330; p < 0.05) but still
significant without distance information (Distance×Group4: β = −0.276; p < 0.05). These
results reflect the descriptive results from Figure 4; the distance-based click rates in
Groups 1 and 3 indicate substantial differences between ranks, whereas the random
sorting-based click rates in Groups 2 and 4 are almost uniformly distributed, demonstrat-
ing less rank-specific differences in click rates. One cautious explanation for this finding is
that users are not just selecting top-ranked coupons; as reflected in the estimation results,
the users care about the transportation costs based on the (more or less salient) distances
to stores. For Group 2, this effect might be increased by the presence of distance-based
contrast effects [40] between adjacent and randomly sorted coupons (i.e., 100 m vs. 1 km
is a larger contrast than 100 m vs. 200 m).16 Comparably closer store locations seem to be
more attractive in this scenario. For Group 4, we also assume that users respond more
positively to stores they recognize from their local environment once these stores appear
in their coupon feed.
Further, the significant interactions between display rank and groups can be considered
as the flipside of the distance-based interactions. Ranking effects are found to be weaker
not only when coupons are not sorted by distance and distance information is available
(Rank×Group2: β = 0.239; p < 0.05), but also when coupons are not sorted by distance and
distance information is not available (Rank×Group4: β = 0.225, p < 0.05). Hence, users are
willing to choose lower ranked coupons (and incur higher search costs) if they stem from
stores that are geographically close.
The group-specific effects are not significant, given they are time invariant for indivi-
dual users (i.e., each user remains in the same treatment group). However, the differences
in variation between groups in terms of distance and display rank effects are captured by
the interaction terms based on distance, display rank and treatment groups, as previously
described.
Additionally, the discount depth has a positive and significant impact on the choice of
discount coupons. However, promotion coupons without direct monetary impact seem to
be generally preferred to discount coupons, as indicated by the negative effect of the
coupon type variable. This finding may be explained by the presence of the coupons’
advertising effect [17], which informs users about the availability of (potentially) attractive
products and stores.
We also control for the search depth via the number of impressions per session,
number of previous clicks, time of day, weekday, and number of days before a coupon
expires. The number of impressions per session has a significantly negative impact on the
probability of choosing coupons. Similarly, the number of previous clicks — a proxy for
usage experience — has a significantly negative impact on the users’ coupon choice
probability.
Location-based coupons seem to be slightly more effective during the nighttime than
during the morning, afternoon and evening. There are also significantly fewer coupon
choices on weekdays versus weekends. Additionally, the fewer the number of days until
a coupon expires is, the higher the probability that users will decide to choose that
448 MOLITOR ET AL.
coupon.17 This effect is consistent with the results of the previous studies on traditional
coupon promotions [30] and can be explained by the fear of missing out.
Finally, the category-specific differences in Table 3 indicate that users prefer coupons
for the categories comprising beauty and wellness, cafes, clubs and bars, grocery, fashion
and accessories, multimedia, restaurant and others than for the baseline category barber-
shops. Compared to the barbershop coupons, the education and culture category coupons
significantly decrease the observed choice probability.
Please refer to the Online Supplemental Appendix for additional analyses as well as for
robustness checks. The additional analyses focus on users’ heterogeneity regarding their
usage intensity and differentiate between the coefficients of the top 15 percent of users and
the rest of the user base. The robustness checks include analyses based on each user’s very
first login, treatment group-specific results and controls for the appearance of offers on the
first screen/page of the app. Overall, our main variables — distance and display rank —
remain qualitatively similar based on the sign and the statistical significance.
Managerial Implications
This study provides several implications for firms. First, our results indicate that loca-
tion-based pull coupon providers can think about careful modifications to a strict
distance-based ranking. For example, providers could think about grouping coupons
in distance buckets, such as 0–1 km and 1–2 km. Within a specific bucket, providers can
450 MOLITOR ET AL.
auction off the specific ranking slot, as done in search engine advertising campaigns, and
achieve a higher click-through or impression-based price. Please note that such ranking
mechanisms are very different from the ranking currently applied by most coupon
providers. Within distance buckets, a ranking that is not strictly based on distance
should not substantially increase the disparity in the distances of the adjacent coupons.
Thus, coupon providers might implement an auction mechanism to optimize the
allocation of top-ranked promotions within distance buckets. Additionally, prices
could differ by the particular geographic area (e.g., urban areas vs. suburbs). In this
case, real-time location information about actual usage frequencies can increase the
value of area-specific and ranking slots. The bidding competition would then not be
based on particular coupons but instead would be based on specific (highly frequented)
areas.
Second, dynamic coupon pricing could be applied in combination with geo-targeting
and weighted by insights into the users’ heterogeneous sensitivities to distances, rank-
ings and discounts. Our results help stores that engage in location-based couponing to
quantify the impact of distance, display rank, and discount depth on coupon choice.
Based on predicted click rates, stores who partner with coupon providers have multiple
tools to leverage: the real-time distance between the user and the store, the rank of the
coupon on the screen, and the actual discount. For example, if a store is 1 km away
from a user relative to another competing store that is 0.5 km away, then the former
can still incentivize the user to come to the store by increasing the discount relative to
the latter store. Similarly, if stores displayed in positions 4 and 6 on a mobile screen are
close competitors, then the store with the lower (worse, numerically lower) position can
increase its attractiveness by increasing the discount relative to the store that is closer to
the user (if these activities are relatively close to urban areas). Thus, there is room to
implement dynamic pricing strategies based on the real-time distance between a user
and the store and the rank of the coupon on the screen by additionally considering
users’ sensitivities to those variables.
Third, our results on ranking effects for distance-based rankings are closely related
to the smartphones’ limited screen sizes. Smartphone displays are several inches smaller
than a laptop or PC display. This small size significantly reduces the number of
coupons viewable on the screen. Thus, customizing or targeting advertisements
becomes more critical on a smartphone. The advertisement’s position on the mobile
screen and the distance between a user and the nearest store advertised produces an
interesting set of three-way interactions for marketers to exploit. Thus, although many
firms compete for attention via coupons, a mobile screen has only limited space.
Therefore, the topmost slots of a mobile screen have greater value to firms and can
be priced at a premium. The value of the topmost slots might also inform firms about
potential modifications of the application’s interface design, which could include the
implementation of filter buttons that enable users to sort coupons by certain criteria,
such as product categories, discount depth, or expiration dates, in addition to the
default distance-based ranking criteria.
Finally, our results also deliver more general insights for firms. As mobile marketing
becomes mainstream, firms struggle with the question of how to effectively measure the
impact of their campaigns. Presumably, location information can help stores achieve
higher response rates for their campaigns [1]. Most importantly, managers using location
JOURNAL OF MANAGEMENT INFORMATION SYSTEMS 451
to localize promotions cannot only increase their campaigns’ return on investment (ROI)
but also make those campaigns better tailored and easier to measure to enable a better
customer experience.
Conclusions
Offline retailers increasingly use location-based coupons to target consumers in their
vicinity in real-time. The rationale for the use of location-based coupons is that geographic
proximity increases the relevance for consumers and, thus, the effectiveness of these
campaigns. We focus on quantifying the most effective interface design for the presenta-
tion of location-based coupons. Our results show that the most effective interface design
for location-based coupons is based on a distance-based ranking. We also find significant
differences in the impact of distance and display rank based on the interface design and
the actual geographic location of users. In turn, we extend the literature on ranking and
distance effects by applying behavioral theories on overconfidence, trade-off contrast
heuristics, and context-dependent choices.
To conclude, this paper paves the way for future research on pull coupons in the
growing and important area of location-based advertising and mobile analytics. Although
the specific findings of our randomized field experiment may not generalize to every
possible location-based advertising context, the results provide a critical baseline to inform
firms and managers about the effects of location-based pull coupons. Our results thus
contribute to the understanding of consumers’ behavioral responses to location-based
advertising and provide important implications for the interface design of location-based
advertising applications.
452 MOLITOR ET AL.
Notes
1. Note that mobile push notifications can be transmitted via SMS or app-based notifications.
2. See, for example, http://www.lsoft.com/resources/optinlaws.asp.
3. The devices’ locations are only tracked when the login is initiated. Movements while using
the app are not tracked.
4. The app does not apply any sort of coupon-specific customization/personalization. Previous
usage behavior neither informs the future selection nor the order of coupons. The default
sorting logic is distance-based.
5. Please note that the descriptions in both app stores (Apple and Google Play) only promised
coupons from stores nearby and did not provide any specifics about the precise information
provided (e.g., exact distance) or the type of sorting.
6. It is worth noting that experiments based on individual-level randomization are less prone to
unobserved heterogeneity and endogeneity [28].
7. Non-systematic (server-side) technical issues caused the disparity in group sizes. We tested
the integrity of our randomization procedure by using a logit model reflecting the first login
of each user. The non-experimental control variables were used to explain the users’ alloca-
tion to each of the two treatment variables (i.e., ranking by distance versus a randomly
generated ranking and location information provided versus location information not pro-
vided). We found that none of the non-experimental variables explains this allocation, thus
suggesting a clean randomization procedure.
8. The expiration date of each coupon is shown after clicking on the coupon profile in the
coupon feed.
9. The location-based coupon provider uses its sales force to recruit stores. To this end, the sales
force does not prioritize specific store categories, as the provider is aiming for a highly diverse
selection of coupons.
10. The raw number of clicks and impressions for each group is as follows: in Group 1, there
were 1,632 clicks and 114,668 impressions; in Group 2, there were 1,230 clicks and 96,571
impressions; in Group 3, there were 1,427 clicks and 93,224 impressions; and in Group 4,
there were 1,110 clicks and 95,450 impressions. Of all the clicks, 46.32 percent can be
attributed to the first five coupons (i.e., the first screen).
11. In comparison, the session-level click rate is even higher, based on a mean of 16.38 percent.
12. In Group 3, all available coupons were sorted based on the actual distance between the user
and all participating stores (e.g., if the user lives next to a participating retail store, this store
would be ranked first/very high in Group 3). However, if the same user were assigned to
Group 4 instead, there would have been a high likelihood that this participating store (and
other coupons from proximate stores) would not have shown up in the top results.
13. We log-transformed all non-categorical and non-percentage scaled variables (i.e., distance,
display rank, impressions per session, previous clicks and coupon expiration), given their
variance in scale (see Table 2).
14. The type I extreme value distribution, also known as the Gumbel distribution, closely
resembles a normal distribution, which is often the preferred distribution to characterize
random errors.
15. Note that the correlation between display rank and distance is 0.135. Our experimental design
also allows us to disentangle the impact of display rank from that of distance. More
specifically, the rank and distance correlation varies between 0.054 (Group 4) and 0.220
(Group 3). But even in Groups 1 and 3, which are “sorted by distance,” this correlation is not
straightforward due to the heterogeneous nature of user locations. We additionally estimated
a random-effect logit model based on the same sample to estimate VIFs for distance (6.53)
and display rank (7.21). Both approaches indicate that collinearity does not seem to be
a critical issue in our models.
16. The distance-specific differences between adjacently displayed coupons in the app (based on
the session-level mean and median distances) are significantly higher in both groups with
JOURNAL OF MANAGEMENT INFORMATION SYSTEMS 453
randomly sorted coupons: the median difference in distance is 3.55 times higher in Groups 2
and 4 than it is in Group 1 and 3.3 times higher in Groups 2 and 4 than it is in Group 3.
17. In order to differentiate between the impact of expiration dates and the impact of retargeting, the
possibility that a user was exposed to a specific coupon more than once, we investigate the
correlation between expiration date and retargeting and also estimate a model that jointly accounts
for both variables. We find the correlation coefficient to be small and close to zero (r = −0.010, p >
0.05) and the estimated expiration date coefficient robust to the inclusion of retargeting
(βExpiration Date = −0.112, p < 0.05). The most desired approach to differentiate between the impacts
of both variables would be a randomized field experiment (as future research). The design of this
experiment could then consist of a treatment group that only includes expirable coupons that are
displayed only once, another treatment group that allows for retargeting but only includes
campaigns without expiration dates, and a third treatment group that combines both expirable,
and retargeted campaigns; the latter situation would be similar to our current scenario.
18. The predicted click estimated log odds ^yijkt followed by a non-linear
rates are based on the
transformation, exp ^yijkt =1 þ exp ^yijkt , which yields the predicted click rates.
Acknowledgments
The authors gratefully acknowledge may helpful suggestions from Thomas Otter, the participants of
the Statistical Challenges in eCommerce Research Symposium (SCECR) in Lisbon, the Marketing
Science Conference in Istanbul, and the INFORMS Annual Meeting in Minneapolis. We are
especially thankful to Boris Lücke for his support.
ORCID
Martin Spann http://orcid.org/0000-0003-4645-3913
References
1. AdAge. 3 mini case studies show how location data is moving marketing. http://adage.com/
article/print-edition/case-s/308190/(accessed on June 25, 2019).
2. Agarwal, A; Hosanagar, K; and Smith, M.D. Location, location, location: an analysis of
profitability of position in online advertising markets. Journal of Marketing Research, 48, 6
(2011), 1057–1073.
3. Allenby, G.M.; and Ginter, J.L. Using extremes to design products and segment markets.
Journal of Marketing Research, 32, 4 (1995), 392–403.
4. Anderson, S.P.; and de Palma, A. Spatial price discrimination with heterogeneous products.
The Review of Economic Studies, 55, 4 (1988), 573–592.
5. Andrews, M; Luo, X; Fang, Z; and Ghose, A. Mobile ad effectiveness: hyper-contextual
targeting with crowdedness. Marketing Science, 35, 2 (2016), 218–233.
6. Bang, Y; Lee, D.-J; Han, K; Hwang, M; and Ahn, J.-H. Channel capabilities, product
characteristics, and the impacts of mobile channel introduction. Journal of Management
Information Systems, 30, 2 (2013), 101–126.
7. Bell, D.R; Ho, T.-H; and Tang, C.S. Determining where to shop: fixed and variable costs of
shopping. Journal of Marketing Research, 35, 3 (1998), 352–369.
8. Biocca, F; Owen, C; Tang, A; and Bohil, C. Attention issues in spatial information systems:
directing mobile users’ visual attention using augmented reality. Journal of Management
Information Systems, 23, 4 (2007), 163–184.
9. Bradlow, E.T.; and Schmittlein, D.C. The little engines that could: modeling the performance
of World Wide Web search engines. Marketing Science, 19, 1 (2000), 43–62.
454 MOLITOR ET AL.
10. Buehler, R; Griffin, D; and Ross, M. Exploring the “planning fallacy”: why people under-
estimate their task completion times. Journal of Personality and Social Psychology, 67, 3
(1994), 366–381.
11. Chib, S.; and Carlin, B.P. On MCMC Sampling in hierarchical longitudinal models. Statistics
and Computing, 9, 1 (1999), 17–26.
12. Choi, J.; and Bell, D.R. Preference minorities and the internet. Journal of Marketing Research,
48, 4 (2011), 670–682.
13. Danaher, P.J; Smith, M.S; Ranasinghe, K; and Danaher, T.S. Where, when and how long:
factors that influence the redemption of mobile phone coupons. Journal of Marketing
Research, 52, 5 (2015), 710–725.
14. DellaVigna, S. Psychology and economics: evidence from the field. Journal of Economic
Literature, 47, 2 (2009), 315–372.
15. Dickinger, A.; and Kleijnen, M. Coupons going wireless: determinants of consumer intentions
to redeem mobile coupons. Journal of Interactive Marketing, 22, 3 (2008), 23–39.
16. Dubé, J.-P; Fang, Z; Fong, N; and Luo, X. Competitive price targeting with smartphone
coupons. Marketing Science, 36, 6 (2017), 944–975.
17. Edelman, B; Jaffe, S; and Kominers, S.D. To Groupon or not to Groupon: the profitability of
deep discounts. Marketing Letters, 27, 1 (2016), 39–53.
18. Fang, Z; Gu, B; Luo, X; and Xu, Y. Contemporaneous and delayed sales impact of
location-based mobile promotions. Information Systems Research, 26, 3 (2015), 552–564.
19. Fang, X; Chan, S; Brzezinski, J; and Xu, S. Moderating effects of task type on wireless
technology acceptance. Journal of Management Information Systems, 22, 3 (2006), 123–157.
20. Fong, N.M; Fang, Z; and Luo, X. Geo-conquesting: competitive locational targeting of mobile
promotions. Journal of Marketing Research, 52, 5 (2015), 726–735.
21. Forman, C; Ghose, A; and Goldfarb, A. Competition between local and electronic markets:
how the benefit of buying online depends on where you live. Management Science, 55, 1
(2009), 47–57.
22. Ghose, A; Goldfarb, A; and Han, S.P. How is the mobile internet different? Search costs and
local activities. Information Systems Research, 24, 3 (2013), 613–631.
23. Ghose, A; Ipeirotis, P.G; and Li, B. Examining the impact of ranking on consumer behavior
and search engine revenue. Management Science, 60, 7 (2014), 1632–1654.
24. Ghose, A; Kwon, H.E; Lee, D; and Oh, W. Seizing the commuting moment: contextual
targeting based on mobile transportation apps. Information Systems Research, 30, 1 (2019),
3–13.
25. Ghose, A.; and Yang, S. An empirical analysis of search engine advertising: sponsored search
in electronic markets. Management Science, 55, 10 (2009), 1605–1622.
26. Goldfarb, A. The internet killed distance. Mobile computing brought it back. MIT Technology
Review (2013). https://www.technologyreview.com/s/520796/the-internet-killed-distance-
mobile-computing-brought.
27. Goldfarb, A. What is different about online advertising? Review of Industrial Organization, 44,
2 (2014), 115–129.
28. Goldfarb, A. and Tucker, C. Online display advertising: targeting and obtrusiveness.
Marketing Science, 30, 3 (2011), 389–404.
29. Huff, D.L. Defining and estimating a trading area. Journal of Marketing, 28, 3 (1964), 34–38.
30. Inman, J.J.; and McAlister, L. Do coupon expiration dates affect consumer behavior? Journal
of Marketing Research, 31, 3 (1994), 423–428.
31. Johnson, E.J; Shu, S.B; Dellaert, Benedict G. C; Fox, C; Goldstein, D.G; Häubl, G; Larrick, R.P;
Payne, J.W; Peters, E; Schkade, D; Wansink, B; and Weber, E.U. Beyond nudges: tools of
a choice architecture. Marketing Letters, 23, 2 (2012), 487–504.
32. Koetsier, J. Mobile Advertising Will Drive 75% Of All Digital Ad Spend In 2018: here’s
what’s changing. https://www.forbes.com/sites/johnkoetsier/2018/02/23/mobile-advertising-
will-drive-75-of-all-digital-ad-spend-in-2018-heres-whats-changing (accessed on June 25,
2019).
JOURNAL OF MANAGEMENT INFORMATION SYSTEMS 455
33. Kondo, F.N; Uwadaira, Y; and Nakahara, M. Stimulating customer response to promotions:
the case of mobile phone coupons. Journal of Targeting, Measurement and Analysis for
Marketing, 16, 1 (2007), 57–67.
34. Li, C; Luo, X; Zhang, C; and Wang, X. Sunny, rainy, and cloudy with a chance of mobile
promotion effectiveness. Marketing Science, 36, 5 (2017), 762–779.
35. Li, H; Shen, Q; and Bart, Y. Local market characteristics and online-to-offline commerce: an
empirical analysis of Groupon. Management Science, 64, 4 (2018), 1860–1878.
36. Luo, X; Andrews, M; Fang, Z; and Phang, C.W. Mobile targeting. Management Science, 60, 7
(2014), 1738–1756.
37. Mantonakis, A; Rodero, P; Lesschaeve, I; and Hastie, R. Order in choice: effects of serial
position on preferences. Psychological Science, 20, 11 (2009), 1309–1312.
38. Molitor, D; Reichhart, P; and Spann, M. Location-based advertising and contextual mobile
targeting. In Proceedings of the Thirty Seventh International Conference on Information
Systems, 2016, https://aisel.aisnet.org/icis2016/EBusiness/Presentations/14/.
39. Renski, H. New firm entry, survival, and growth in the United States: a comparison of
urban, suburban, and rural areas. Journal of the American Planning Association, 75, 1
(2008), 60–77.
40. Simonson, I.; and Tversky, A. Choice in context: tradeoff contrast and extremeness aversion.
Journal of Marketing Research, 29, 3 (1992), 281–295.
41. Thaler, R.H; Sunstein, C.R; and Balz, J.P. Choice architecture. Working paper. University of
Chicago, 2010. http://ssrn.com/abstract=1583509.
42. The Financial Brand. Statistics & trends that will shape your digital marketing strategy.
https://thefinancialbrand.com/66080/trends-facts-digital-marketing-advertising/(accessed on
June 25, 2019).
43. Train, K. Discrete Choice Methods with Simulation. New York: Cambridge University Press,
2003.
44. Unni, R.; and Harmon, R. Perceived effectiveness of push vs. pull mobile location based
advertising. Journal of Interactive Advertising, 7, 2 (2007), 28–40.
45. Ursu, R.M. The power of rankings: quantifying the effect of rankings on online consumer
search and purchase decisions. Marketing Science, 37, 4 (2018), 530–552.
46. Wu, J; Shi, M; and Hu, M. Threshold effects in online group buying. Management Science, 61,
9 (2015), 2025–2040.
47. Xu, H; Teo, H.-H; Tan, B.C.Y; and Agarwal, R. The role of push-pull technology in privacy
calculus: the case of location-based services. Journal of Management Information Systems, 26,
3 (2009), 135–174.
the Internet and mobile technologies on industries and markets transformed by its shared infra-
structure, (2) the trade-offs between data privacy concerns and the economic value accruing to
consumers, and (3) the business impact of digital advertising and mobile marketing. He has
published papers in Information Systems Research, MIS Quarterly, Management Science,
Marketing Science, Statistical Science, and other journals.
Philipp Reichhart (lmu@philippreichhart.de) is an associated researcher at the Institute of
Electronic Commerce and Digital Markets at the Ludwig-Maximilians-Universität München,
Germany. His research interests include mobile commerce, consumer behavior, word of mouth
and viral marketing. He has published in Journal of Interactive Marketing, Electronic Markets,
International Journal of Electronic Business, and other journals.