Journal of Retailing 92 (3, 2016) 333–351

Cross-Channel Effects of Price Promotions: An Empirical Analysis of the

Multi-Channel Grocery Retail Sector
Els Breugelmans ∗,1 , Katia Campo 1,2
KU Leuven, Faculty of Economics and Business, Korte Nieuwstraat 33, 2000 Antwerp, Belgium
Available online 26 February 2016

Abstract
An important consideration for multi-channel grocery retailers is which promotion strategy to adopt across channels. The tendency of grocery
buyers to keep visiting the offline store after they start to buy at the chain’s online store implies that promotions in one channel can have substantial
negative effects on buying behavior in the other channel, especially if the promotions differ across channels. This article examines the cross-channel
effects of price promotions on category purchase decisions, taking into account both contemporaneous effects (during the promotion period) and
cross-period effects (after the promotion period). For cross-period promotion effects, a distinction is made between lagged promotion effects on
subsequent category purchases, and promotion frequency effects on future promotion effectiveness. The results show that (1) promotions in one
channel can have negative effects on category purchases in the other channel during the promotion period, (2) these cross-channel effects are
asymmetric, (3) high promotion frequency can have negative effects on future promotion effectiveness in the other channel, and (4) cross-channel
effects are more negative for more loyal customers of the chain.
© 2016 New York University. Published by Elsevier Inc. All rights reserved.

Keywords: Multi-channel grocery shopping; Price promotions; Cross-channel promotion effects; Promotion effectiveness

“The future of grocery shopping is multi-channel.”
—Joanne Denney-Finch, chief executive (IGD 2009)
Online grocery sales are expected to rise from 3.3 percent
of total U.S. grocery spending to 11 percent by 2023 (Steiman
2014), and are expected to double its value in the U.K. by 2020
to £17.2 bn (IGD 2015). Most online grocery shoppers regularly
visit both online and offline grocery stores, most often using the
online and offline channel of the same (primary) chain (Melis
et al. 2015). These multi-channel shoppers constitute a rapidly
growing segment that dramatically changes the grocery mar-
ket (IGD 2011). As an interesting and consequential aspect of
this change in market conditions, multi-channel shoppers are
exposed to marketing actions of a chain’s online and offline
channel.
The trend of grocery shoppers using multiple channels rather
than migrating to an online channel (as is more likely in
∗ Corresponding author. Tel.: +32 16326481.
E-mail addresses: els.breugelmans@kuleuven.be (E. Breugelmans),
katia.campo@kuleuven.be (K. Campo).
1 Both the authors should be regarded as joint first authors.
2 Tel.: +32 16376260.
non-grocery categories; Ansari, Mela, and Neslin 2008) there-
fore challenges grocery retailers to construct operations models
more strategically and implement marketing mix strategies care-
fully (Retail Systems Research 2013). In this research, the focus
is on promotional strategies, and more particularly on the cross-
channel promotion effects they entail. When implementing
different promotional actions across channels, multi-channel
retailers must wary about whether and how promotions in one
channel influence buying behavior of their customers in the other
channel. Customers of a multi-channel grocery chain can take
advantage of special offers available in the chain’s online and
offline channel. This may lead to subsidization of loyal chain
customers rather than enhancing store traffic and overall sales.
A survey conducted by IGD (2011) indeed confirms that a more
extensive promotion offer is an important motivation for many
shoppers to visit multiple channels.
The decision whether and how to differentiate promotions
across channels is therefore an important managerial issue. Pre-
vious research, however, provided only partial insights into
promotion effects in a multi-channel context. In this research,
we aim to provide a comprehensive analysis and examine the
effect of price promotions on household category purchase

http://dx.doi.org/10.1016/j.jretai.2016.02.003
0022-4359/© 2016 New York University. Published by Elsevier Inc. All rights reserved.
334 E. Breugelmans, K. Campo / Journal of Retailing 92 (3, 2016) 333–351
decisions (i.e. category purchase incidence and quantity
decisions) in a multi-channel grocery retail context, where con-
sumers visit the offline and online channel of the same chain.
We take into account both within and cross-channel promotion
effects (i.e. effects on the own and other channel), as well as con-
temporaneous and cross-period promotion effects (i.e. effects
during and after the promotion period). For cross-period promo-
tion effects, we distinguish between short term lagged promotion
effects on subsequent category purchases, and longer term pro-
motion frequency effects on future promotion effectiveness.
Focusing on cross-channel and cross-period promotion effects
is particularly relevant for repeatedly purchased categories like
groceries, where multi-channel shoppers regularly visit (and are
thus exposed to promotions in) online and offline channels. The
estimation results shed light on several important research ques-
tions: whether and how promotions in one channel reinforce
category sales in the own channel, cannibalize category sales in
the other channel, lead to lower category sales in post-promotion
periods, and/or reduce promotion effectiveness by making con-
sumers less responsive to promotions. Based on the estimation
results, we use simulations to detail different payoffs, in the form
of sales revenue earned from different promotion strategies.
Our unique data set for this study consists of a representative
U.K. household panel from Kantar Worldpanel, obtained from
AiMark. The data cover online and offline household grocery
purchases at Tesco, for a 78-week period. We estimate category
purchase decision models (incidence and quantity) for two fre-
quently purchased categories, milk and cereals, across Tesco’s
online and offline stores. Tesco is the U.K. market leader in the
grocery sector in the offline as well as the online channel.
From an academic point of view, this study provides impor-
tant contributions to promotion and multi-channel retailing
literature. To the best of our knowledge, this is the first study
to examine explicitly the effects of promotions in one channel
on buying behavior in the other channel of the same chain. We
also extend promotion literature by providing a comprehensive
analysis of own- and cross-channel promotion effects, taking
both contemporaneous and cross-period promotion effects into
account. That is, most previous literature focuses on cross-
competitive promotion effects on brand or store choice within
an offline channel (Ailawadi and Neslin 1998; Bell, Chiang,
and Padmanabhan 1999; Blattberg, Briesch, and Fox 1995;
Bucklin and Gupta 1992). Studies that examine promotion
effects in different channels mainly note differences in promo-
tion effectiveness between channels, without examining whether
promotions in one channel also affect purchase decisions in the
chain’s other channel (Chu, Chintagunta, and Cebollada 2008;
Zhang and Wedel 2009). By using estimation results to sim-
ulate the effect of different promotion scenarios, we provide
interesting insights into the sales revenue effects of different
promotion strategies. Such an approach may help multi-channel
retailers improve their overall business strategy and optimize
cross-channel synergies through more effective promotion plan-
ning.
In the next section, we present a brief overview of promo-
tion effects examined in an offline context, and indicate how
these findings can be translated to expected promotion effects
in a multi-channel grocery shopping context. Next, we describe
our modeling approach and the data we used for the empiri-
cal analysis. After an overview and discussion of the estimation
and simulation results, we discuss our main findings and the
resulting managerial implications. We end with limitations and
extensions for future research.
Price Promotion Effects in a Multi-Channel Context
Price promotions, and the different effects they may have
on offline consumer buying behavior, have been examined in
several previous studies (for a review, see van Heerde and
Neslin 2008). We seek to analyze these effects in a multi-
channel shopping environment, from the retailer’s perspective.
For this reason, we focus on category-level (purchase incidence
and quantity) effects, rather than brand-level (choice/switching)
promotion effects (Ailawadi et al. 2007; Bell, Chiang, and
Padmanabhan 1999; van Heerde and Neslin 2008).3 As indicated
in Table 1, we provide a comprehensive analysis, by distin-
guishing between own-and cross-channel promotion effects
(i.e. effects in the same channel or in the other channel,
respectively), and between contemporaneous and cross-period
promotion effects (i.e. effects during or after the promotion
period, respectively). For cross-period promotion effects, we fur-
ther differentiate between lagged promotion effects (i.e. effects
on category purchase decisions in the post-promotion period,
resulting from consumers’ stockpiling behavior; Ailawadi et al.
2007; van Heerde and Neslin 2008) and promotion frequency
effects (i.e. changes in promotion effectiveness due to [high]
promotion frequency in the past; Foekens, Leeflang, and Wittink
1999; Kopalle, Mela, and Marsh 1999).
Own-Channel Promotion Effects
Own-channel promotion effects (contemporaneous, cross-
period lagged and cross-period promotion frequency effects)
have been examined in multiple past papers, and in most cases
provided consistent results (Bucklin and Gupta 1992; Degeratu,
Rangaswamy, and Wu 2000; Foekens, Leeflang, and Wittink
1999; Kopalle, Mela, and Marsh 1999; van Heerde and Neslin
2008; Zhang and Wedel 2009). Based on these results and the
underlying mechanisms discussed in previous studies, we for-
mulate the following hypotheses. First, given the findings that
online and offline promotions can induce consumers to buy
a category earlier than planned or in larger quantities, at the
expense of either competitive chains (category growth at the
chain level; van Heerde and Neslin 2008) or future purchases
(loyal acceleration in the own channel; Ailawadi et al. 2007), we
expect a strong, positive contemporaneous effect of price promo-
tions on consumers’ category purchase decisions (H1). Second,
3 In the following discussions, we do not distinguish explicitly between
category purchase incidence and quantity, because previous research has demon-
strated that promotions have effects of similar direction on both decisions (Bell
and Hilber 2006). We do, however, allow the magnitude of effects to differ (see
the Model section).
 E. Breugelmans, K. Campo / Journal of Retailing 92 (3, 2016) 333–351 335

Table 1
Overview promotion effects.
Own-channel Cross-channel

Contemporaneous price promotion effects H1 H4

H5: Asymmetry H8: Heterogeneity
Cross-period price promotion effects Lagged H2 H6
Promotion frequency H3 H7
H1. Offline (online) price promotions have a positive effect on category purchase decisions in the own offline (online) channel.
H2. Offline (online) price promotions have a negative effect on category purchase decisions in the own offline (online) channel in post-promotion periods.
H3. A high frequency of offline (online) price promotions has a negative effect on promotion effectiveness in the own offline (online) channel in subsequent
periods.
H4. Online (offline) price promotions have a negative effect on category purchase decisions in the other offline (online) channel.
H5. A similar price promotion in the online channel has a stronger negative effect on category purchase behavior in the offline channel than vice versa.
H6. Online (offline) price promotions have a negative effect on category purchase decisions in the other offline (online) channel in post-promotion periods.
H7.A high frequency of online (offline) price promotions has a negative effect on promotion effectiveness in the other offline (online) channel in subsequent
periods.
H8. Own- and cross-channel (contemporaneous and cross-period) promotion effects differ between loyal and non-loyal customers of the chain.

Focal hypotheses in bold.

buying earlier or more in the category, in reaction to promotions,
can result in larger household stocks and a reduced need to buy
the category on subsequent shopping trips, unless consumers
also increase their consumption rate (Ailawadi et al. 2007; van
Heerde and Neslin 2008) or shift purchases from one store to
the other (Guyt and Gijsbrechts 2014). In line with most previ-
ous research, we expect that promotions have a negative effect
on category purchase decisions in post-promotion periods in the
same channel (H2). Third, previous research has shown that fre-
quent exposures to promotions can make future promotions less
effective (Foekens, Leeflang, and Wittink 1999; Kalyanaram and
Winer 1995; Kopalle, Mela, and Marsh 1999; van Heerde and
Neslin 2008). When promotions become endemic, consumers
change their expectations about the value and frequency of future
deals (Liu and Balachander 2014), such that they even start to
regard promotional deals as normal (Blattberg, Briesch, and Fox
1995). Because shopping frequency is lower in online grocery
stores, it is worthwhile to investigate whether a similar nega-
tive effect of a high promotion frequency on future promotion
sensitivity can be found in the online channel (H3).
Cross-Channel Contemporaneous Effects
Our main contribution lies in the investigation of cross-
channel promotion effects (focal hypotheses, highlighted in bold
in Table 1). To predict whether price promotions in one channel
cannibalize category purchases in the other channel, we focus on
promotion-driven switching effects across channels. Promotion-
driven switching effects have been observed for (offline) store
choice and purchase allocation decisions (Bucklin and Lattin
1992; Guyt and Gijsbrechts 2014; Kumar and Leone 1988;
MacKenzie and Walters 1988). The underlying reason is that
promotions can increase the overall attractiveness of a store and
attract customers that would otherwise not have visited the store
(direct store switch). In addition, promotions can influence the
category allocation patterns of consumers who regularly visit
multiple stores, and stimulate shifts in purchases from one store
to the other (indirect store switch; Bucklin and Lattin 1992).
Based on a similar logic, we expect that multi-channel shoppers
may adjust their allocation decisions (indirect channel switch)
to take maximum advantage of promotion opportunities (H4).
Asymmetry of Cross-Channel Effects
Research on the brand switching effects of promotions has
demonstrated that these effects tend to be asymmetric for brands
that belong to different price/quality tiers (Bronnenberg and
Wathieu 1996; Sivakumar 2004; Sivakumar and Raj 1997). The
general finding is that consumers more easily “switch up” from
a low to a high tier brand in response to a promotion for the high
tier brand, than they “switch down” from a high to a low tier pro-
moted brand. One of the major reasons is that consumers rather
give in on price than on quality (Sivakumar 2004). Cross-channel
contemporaneous promotion effects can for that reason also be
asymmetric, in view of the channel differences in price and con-
venience (Gupta and Kim 2009). The online channel excels in
offering convenience advantages and allows consumers to save
time and effort but often charges a higher price for this extra
service (e.g. Avery et al. 2012; Gupta and Kim 2009). Using
reasoning similar to that applied to explain asymmetry in brand
switching effects (e.g. Sivakumar 2004), we expect that online
price promotions can trigger consumers to take advantage of
online shopping convenience (higher benefits) without having
to suffer substantial increases in shopping costs (price discount).
This may lead to asymmetry in cross-channel promotion effects
to the advantage of the online channel, such that online price
promotions have a stronger negative effect on offline category
purchase behavior than vice versa (H5).
Cross-Channel Cross-Period Effects
Similar to cross-period lagged promotion effects in the
own channel, consumer stockpiling triggered by promotions
may negatively affect subsequent category purchase decisions
in the other channel (Ailawadi et al. 2007; Dawes 2012).
We explore these preemptive channel switching effects, by
336 E. Breugelmans, K. Campo / Journal of Retailing 92 (3, 2016) 333–351
examining whether price promotions in one channel lead to a
reduction of category purchases in the other channel during post-
promotion periods (H6). What is more, multi-channel shoppers
who interchangeably shop in online and offline channels of the
same grocery chain are exposed to price promotions in both
channels and may use this information to form their expecta-
tions about future promotion activities. We therefore expect that
a high frequency of price promotions in one or both channels
can lead to an expectation that price promotions will be widely
available in the future (in offline and/or online channels), and
consequently may reduce the effectiveness of price promotions
in both channels (H7).
Heterogeneity in Promotion Reactions
Finally, cross-channel effects can differ across households
(van Heerde and Neslin 2008; H8). Promotion effects maybe
stronger for loyal chain customers, who can save more by tak-
ing advantage of (different) price reductions in the chain’s online
and offline store. But less loyal, multiple-store shoppers have
more opportunities to shift category purchases across chains
(rather than channels), which may trigger stronger positive con-
temporaneous promotion effects in a given channel (shift from
competitive to the focal chain). The lower visit frequency at the
focal chain, however, also may result in weaker cross-channel
and cross-period promotion effects.
Empirical Analysis: Model
To analyze multi-channel promotion effects, we focus on con-
sumers’ category purchase decisions for a single, multi-channel
grocery chain for two reasons. First, examining customers’
buying behavior at the same chain provides insight into
promotion-driven purchase shifts across channels that can sup-
port the development of profitable multi-channel promotion
strategies. Second, focusing on a single chain allows to isolate
“pure” channel allocation effects at the category level and elim-
inate possible confounding effects of chain differences (e.g. in
price or quality positioning, retail store image), while keeping
the analysis tractable.
Channel choice (i.e. the decision to visit the online or offline
channel) and chain choice (i.e. the decision to visit the exam-
ined chain) are considered as exogenous decisions. In line with
previous studies on category-level promotion effects, we dis-
tinguish between the purchase incidence decision (given the
channel and chain visit decision), and the purchase quantity
decision (given purchase incidence), using a log-normal hurdle
model (McDowell 2003; Wooldridge 2010). This model allows
the magnitude of the promotion effects and the set of relevant
variables to differ between purchase incidence and quantity deci-
sions. We use maximum likelihood procedures and Latent Gold®
software to estimate the model. In this way, we can define latent
classes over both hurdle model components (see Vermunt and
Magidson 2005). The indices i, t and c refer to the household,
purchase occasion,4 and channel (c = 1 for online and c = 2 for
offline channels), respectively. For parsimony, we exclude cat-
egory and segment indices. To distinguish between own- and
cross-channel effects, we use the indicators c and c̄ to denote
the own and other channel, respectively.
The first component of the hurdle model determines whether
the key dependent variable (purchase quantity) is equal to zero
(no purchase in the category) or positive (purchase in the cate-
gory) (McDowell 2003). This purchase incidence decision—that
is, whether household i makes a category purchase in channel c
on purchase occasion t (bit,c )—is described by a binomial Probit
model:
 ∗ >0
1, if bit,c
bit,c = , (1)
0, otherwise
∗ is a latent variable capturing the attractiveness for
where bit,c
household i to make a purchase in the category in channel c on
purchase occasion t, modeled by the following specification:
∗
bit,c = [α0 + α1 ∗ onvstit + α2 ∗ log(CRi ) + α3 ∗ INVRATIOit
+ α4 ∗ log(INVit ) + α5c ∗ loyalit,c ]
+ [μ1c ∗ promoit,c + μ2c ∗ promoit,c̄ ]
+ [ϕ1c ∗ promoi,t−1,c + ϕ2c ∗ promoi,t−1,c̄ ]. (2)
The operationalization of the variables is detailed in the Data
section and Table 3. The expression within the first set of square
brackets measures the household’s baseline category purchase
tendency in the absence of promotions. The average tendency
to buy in the category (captured by the intercept α0 ) is adjusted
for possible channel differences using an online channel dummy
variable (onvstit ). The overall tendency to buy the category in
the online channel is then equal to α0 + α1 . We expect a positive
online correction parameter α1 , given the consumers’ tendency
to buy larger baskets online (Chintagunta, Chu, and Cebollada
2012; see also Table 2). To get an accurate estimate of household
consumption needs and inventory effects, we use the procedure
suggested by Bell and Boztuğ (2007). The consumption rate
parameter α2 should be positive: higher consumption needs lead
to a stronger tendency to make a purchase in the category. In
contrast, the inventory ratio parameter α3 should be negative:
higher than usual household stocks reduce the need to purchase
the category. Bell and Boztuğ (2007) suggest that in addition
to a traditional inventory variable that controls for heterogene-
ity across different households, there can also be a potential
positive effect caused by inventory estimation bias and inven-
tory effects on consumption. To control for this, we include a
log inventory variable that we expect to be positive (captured
4 In line with previous studies on category purchase decisions (e.g. van Heerde
and Neslin 2008), and because we treat the visit to the channel as given, we
model decisions during a purchase occasion (in our case, periods with a visit
to the channel) rather than at absolute points in time (e.g. every week). Hence,
we consider only weeks in which households have the opportunity to buy the
category in a given channel.
 E. Breugelmans, K. Campo / Journal of Retailing 92 (3, 2016) 333–351 337

by α4 ). In addition, we include channel-specific loyalty vari-

2,034 (25.96%)
48,711 (75.3%)

61,178 (25.0%)
33,400 (93.9%)
.425 (s.e.: .35)
.762 (s.e.: .64)
9,862 (15.2%)
Multi-channel

.29 (s.e.: .19)

.25 (s.e.: .26)
.41 (s.e.: .22)
.62 (s.e.: .29)
6,120 (9.5%)
ables (loyalit,c ) (Guadagni and Little 2008), which capture a
household’s tendency to repurchase a category in the same chan-
nel of a given chain (state dependence; van Heerde and Neslin
2008). The related parameters (α5c ) should be positive: A higher
category-specific loyalty to channel c of the focal chain in the
past leads to a higher tendency to repurchase the category in that

Single-channel

.796 (s.e.: .55)

.42 (s.e.: .22)

.68 (s.e.: .25)

2,178 (6.1%)
channel.

42 (.53%)
The expression in the second set of brackets captures own-

(online)
and cross-channel contemporaneous promotion effects. The

870
own-channel promotion variables (promoit,c ) are household and
channel specific, measured as weighted percentage price dis-
counts (Grewal et al. 2011) offered in the category and channel

184,002 (75.0%)
5,760 (73.51%)
on a given purchase occasion. These variables are household

Single-channel

.413 (s.e.: .33)

.31 (s.e.: .20)

.41 (s.e.: .22)

specific, because we focus on visits to a channel during which

(offline)

166,991
the household has the opportunity to buy in the category, and
we use household preferences for specific items as weights. The
parameters (μ11 , μ12 ) should be positive, because promotions
often have an immediate, positive impact on consumers’ cate-
gory purchase incidence decisions in the own channel (H1). The

.416 (s.e.: .33)

.785 (s.e.: .60)
.30 (s.e.: .20)
.28 (s.e.: .21)
.41 (s.e.: .22)
.64 (s.e.: .26)
cross-channel promotion variables (promoit,c̄ ) are constructed

232,554

245,180
Cereals
similarly, but only taking into account those purchase occasions

10,732

35,578
7,836

6,120
Total
in which household i visited the other channel (c̄) during the
same week and was exposed to promotions in that channel.
The parameters (μ21 , μ22 ) should be negative, because promo-
tions in one channel are expected to have an immediate negative

3.14 (s.e.: 2.29)

3.11 (s.e.: 3.24)
2,175 (23.51%)
57,350 (74.3%)
13,028 (16.9%)

98,589 (25.0%)
36,985 (100%)
Multi-channel

.54 (s.e.: .27)

.36 (s.e.: .27)
.75 (s.e.: .21)
.85 (s.e.: .20)
6,750 (8.8%)
impact on consumers’ category purchase incidence in the other
channel (H4). The expression within the third set of brackets
captures own- and cross-channel, cross-period, lagged promo-
tion effects (promoi,t−1,c and promoi,t−1,c̄ respectively). The
parameters (ϕ1c , ϕ2c ) are expected to be negative, to reflect the
negative effects of consumer stockpiling on subsequent purchase

298,212 (75.0%)
2.84 (s.e.: 2.17)
7,076 (76.49%)

incidence decisions in the own and other channel (H2 and H6).
Single-channel

.62 (s.e.: .28)

.78 (s.e.: .23)

To incorporate past promotion frequency effects on promo-
(offline)

194,457

tion effectiveness, we use a model with varying parameters

(Foekens, Leeflang, and Wittink 1999; Kopalle, Mela, and Marsh
1999). The parameters of the own-channel promotion variables
(μ1c ) are a function of: (i) a constant capturing sensitivity to cur-

Expressed in standardized units: liters for milk, kilograms for cereals.

rent own-channel promotions, (ii) a past promotion frequency
2.92 (s.e.: 2.21)
3.11 (s.e.: 3.24)
.60 (s.e.: .28)
.36 (s.e.: .27)
.77 (s.e.: .22)
.85 (s.e.: .20)

variable in the own channel c (Lpromoit,c ) capturing the possi- Percentage of total in the category, across shopper segments.
ble decrease in promotion effectiveness due to a high promotion
251,807

396,801
13,028

36,985

Percentage of total in the shopper segment per category.

9,251

6,750

frequency in the own channel (μ11,c < 0; H3), and (iii) a past pro-
Total
Milk

motion frequency variable in the other channel c̄ (Lpromoit,c̄ )

Household purchase behavior, descriptive statistics.

capturing the possible decrease in promotion effectiveness due

to high promotion frequency in the other channel (μ12,c < 0; H7).
Off only
On only

Thus,
Both
Off

Off

Off

Off
On

On

On

On

μ1c = μ10,c + μ11,c ∗ Lpromoit,c + μ12,c ∗ Lpromoit,c̄ . (3)

(standardized units per trip)c

For the second component of the hurdle model, which cap-

Average purchase frequency

Average purchase incidence

(% of visits with purchase)
(nr of purchases per week)
Number of observationsb

tures category purchase quantity decisions, we use a log-normal

Shopping trip quantity
Number of shoppersa

specification that provided a much better model fit than the

linear truncated model (see the “Robustness Checks” section).
Sales (£/year)a

We operationalize the category purchase quantity variable as

the logarithm of the total quantity household i bought in chan-
Table 2

nel c on purchase occasion t. We incorporate all explanatory

a
b
c

variables—with the exception of channel loyalty, which is not

338 E. Breugelmans, K. Campo / Journal of Retailing 92 (3, 2016) 333–351
relevant for quantity decisions—in Eqs. (4) and (5) in the same
way as before. We have similar expectations for the direction of
the effects, but allow the magnitude of the effects to be different.
Thus,
ln(PQ∗it,c ) = [β0 + β1 ∗ onvstit + β2 ∗ log(CRi )
+ β3 ∗ INVRATIOit + β4 ∗ log(INVit )]
+ [δ1c ∗ promoit,c + δ2c ∗ promoit,c̄ ]
+ [γ1c ∗ promoi,t−1,c + γ2c ∗ promoi,t,1,c̄ ], and
(4)
δ1c = δ10,c + δ11,c ∗ Lpromoit,c + σ12,c ∗ Lpromoit,c̄ . (5)
To account for household differences in promotion effects,
we estimate the log-normal hurdle model using a latent-class
analysis. Our dataset includes multi- and single-channel shop-
pers (see “Data” section). Because we are especially interested
in promotion effects for multi-channel shoppers, we only include
heterogeneity in promotion reactions for these shoppers. For the
single-channel shoppers, we estimate the ‘mean’ effects using
the fixed-class member option in Latent Gold® . This approach
significantly simplifies our analyses and results, but does not
bias the insights in cross-channel promotion effects for multi-
channel shoppers (see “Robustness Checks” section). For similar
reasons, we introduce heterogeneity in the effects for the promo-
tion variables but not for the control variables, like consumption
rate and inventory level. To explain latent-class membership,
we include the household’s chain loyalty (i.e. share of category
purchases at the focal chain, CatSharei ) as concomitant vari-
able. Similar to the formulation used by Gupta and Chintagunta
(1994), we specify the segment membership probability pig as:
exp(λ0g + λ1g ∗ CatSharei )
pig = G−1 . (6)
1 + m=1 exp(λ0m + λ1m ∗ CatSharei )
The log-likelihood function for the entire sample is given by:
⎛ ⎞
 I G 2 Tic
LL = log ⎝ pig Lit,c|g ⎠ , (7)
i=1 g=1 c=1tc =1
where pig is household i’s probability of belonging to segment
g (g = 1, . . ., G), Tic is the number of purchase occasions during
which household i made a visit to channel c of the chain, and
I is the number of households in the data. Furthermore, Lit,c|g
is household i’s simultaneous likelihood of purchase incidence
and quantity decisions on purchase occasion t in channel c (van
Heerde and Neslin 2008), which is conditional on i belonging
to segment g:
Lit,c|g = (Pg (bit,c = 1))yit,c (1 − Pg (bit,c = 1))1−yit,c
× (Pg (PQit,c = PQ∗it,c |bict = 1))Zit,c (8)
with yit,c as the category purchase indicator, equal to 1 if house-
hold i purchased the category in channel c on purchase occasion
t, and 0 otherwise; and Zit,c as the purchase quantity indicator,
equal to 1 if household i purchased PQ∗it,c units in channel c on
purchase occasion t, given purchase incidence, and 0 otherwise.
Empirical Analysis: Data
We have data from a representative U.K. household panel
from Kantar Worldpanel, obtained from AiMark for the period
from July 2006 to December 2007. The U.K. online grocery
market is among the most developed of all major online grocery
markets (Kearney 2012), and in our analysis period, already
reached much higher levels of multi-channel grocery shopping
than is observed in many other western countries nowadays. Our
data cover 78 weeks of online and offline household grocery pur-
chases at Tesco, the market leader in the U.K. grocery sector for
both online and offline markets. In our estimation period, the
chain only offered a home delivery service. Data for the first 26
weeks served to initialize the household-specific control vari-
ables; data for the next 52 weeks are used for model estimation.
We also have data about grocery purchases at other chains dur-
ing the same period, which we used to operationalize the control
variables (consumption rate, inventory level) and concomitant
category-specific chain loyalty variable.
We estimate category purchase decision models for two fre-
quently purchased categories, milk and cereals. We include
all households that made at least one store visit to Tesco in
the estimation period (irrespective of the channel) and that are
regular users of the category (i.e. consumption rate equal to
at least the 25 percent level of the consumption rate distribu-
tion across households). The sample includes 9,251 households
(2,175 multi-channel and 7,076 single-channel offline shop-
pers) for milk and 7,836 households (2,034 multi-channel, 5,760
single-channel offline, and 42 single-channel online shoppers)
for cereals.
Table 2 provides descriptive statistics about household pur-
chase behavior in each category. For multi-channel shoppers,
the average category purchase frequency is higher offline, but
the percentage of visits with a purchase and the quantity pur-
chased per shopping trip (for cereals) are substantially higher
online. As previous research indicates, major shopping trips
more likely occur in the online channel, whereas fill-in trips
happen in the offline channel (e.g. Chintagunta, Chu, and
Cebollada 2012). Comparing offline purchases by multi- and
single-channel shoppers, we find that the former visit offline
store less frequently than the latter and tend to buy larger
quantities in each shopping trip. This finding is in line with
observations that time-constrained shoppers are more likely to
take advantage of the online shopping convenience benefits
(Degeratu, Rangaswamy, and Wu 2000; Perkins 2014).
We highlight some key features of the variable operational-
ization here and provide more details in Table 3. We use the
initialization period of 26 weeks and data from competitive
chains to compute the consumption rate and provide a starting
value for the inventory level. To account for household differ-
ences in the levels of safety stock and make inventory effects
comparable across households, we divide the inventory level
by the household’s average inventory level (INVRATIOit ). The
 E. Breugelmans, K. Campo / Journal of Retailing 92 (3, 2016) 333–351
Table 3
Variable operationalizations.
Notation Name
bit,c Category purchase incidence for
household i in channel c on t
ln(PQit,c ) Logarithm of category quantity for
household i in channel c on t
onvstit Visit to the online channel for
household i on purchase occasion t
both vstit Visit to both channels for household i
on purchase occasion t
CRi Weekly consumption rate for
household i
INVit Inventory rate for household i on
purchase occasion t
INVRATIOit Mean-centered inventory rate for
household i on purchase occasion t
loyalit,c Loyalty variable for household i in
channel c on purchase occasion t
WPctDisctis,c Average percentage price reduction
in category for household i in
channel c in week s
Promoit,c Current sales promotions for
household i in channel c on purchase
occasion t
339
Description Specification
Indicator variable, equal to 1 if
household i makes a category purchase
in channel c on purchase occasion t (i.e.
week with a shopping trip), 0 otherwise.
Logarithm of total quantity bought in a
category (in standardized units:
milliliters, milk; grams, cereals) for
household i in channel c on purchase
occasion t.
Indicator variable, equal to 1 if
household i visits the online channel on
purchase occasion t, 0 otherwise.
Indicator variable, equal to 1 if
household i visits both channels on
purchase occasion t, 0 otherwise.
26
Overall total quantity of the category CRi = s=1
PQis /26
bought by household i in all channels
and chains in the initialization period
(PQis ; s = 1, . . ., 26; s = week indicator),
divided by the number of weeks in
initialization period (26).
Inventory in the previous week INVit = max[0, (INVi,s−1 + POi,s−1 − CRi )]
(INVi,s−1 ), plus the amount bought in the
previous week across all chains (PQi,s−1 )
minus the weekly consumption rate
(CRi ). If this inventory level is less than
0, we use 0. The inventory level at the
starting point of the initialization period
(INVi0 ) is set equal to 0.
INVit
Mean-centered version of the inventory INVRATIOit = AVG INVi
rate, using the average inventory for
household i (AVG INVi ).
Weighted average of past loyalty for the loyalit,c = θ * loyali,t−1,c + (1 − θ) * bi,t−1,c
category in channel c at the focal chain
(loyali,t−1,c ) and channel choice at the
focal chain on the last purchase occasion
(bi,t−1,c ) for household i, with weight
θ = .7 to capture fading effects.

N
The average percentage of price discount WPctDisctis,c = win ∗ PctDisctns,c
(PctDisctnsc ) of a category across all n=1
SKUs (n = 1, . . ., N) in channel c in week
s, weighted by spending shares for SKU
n in each household i’s long-term
shopping basket in the focal chain (win ).
Promotions are determined on the basis
of price deviations from the average
price level (equal to 1 if unit price of
SKU n in week s < [average price of
SKU n − 1 standard deviation]). The
price discount percentage is calculated as
the difference between unit price of SKU
n minus average price of SKU n in week
s, divided by average price of SKU n.
Weighted percentage price reduction for Promoit,c = WPctDisctit,c * Vstit,c
household i in channel c, given that
household i visits the channel in
purchase occasion t (Vstit,c ).
340 E. Breugelmans, K. Campo / Journal of Retailing 92 (3, 2016) 333–351

Table 3 (Continued)
Notation Name Description Specification

Promoit,c̄ Current sales promotions for
household i in channel c̄ on purchase
occasion t
Promoi,t−1,c Lagged promotion for household i in
channel c on purchase occasion t
26 
LPromoit,c Past promotion frequency for
household i in channel c on purchase
occasion t
CatSharei Share of category purchases at the
focal chain for household i
Weighted percentage price reduction for Promoit,c̄ = WPctDisctit,c ∗ Vstit,c ∗ both vstit
household i in channel c̄, given that
household i visits both channels in
purchase occasion t.
Weighted percentage price reduction for
household i in channel c of the previous
purchase occasion t − 1.
 N
Weighted sum of previous promoted Lpromoit,c = λs ∗ n=1
Disctn,t−s,c ∗ Vstit,c
items of the category (number of s=1
promoted items, Nrpromoi,t-s,c ) on
purchase occasion t, given the visits of
household i in channel c in past 26 weeks
(s = 1, . . ., 26), with weight λ = .75 to
capture fading effects.
PQi Tesco
Overall total quantity of the category CatSharei = PQi All
bought by household i in the estimation
period in all channels at the focal chain
Tesco (PQi Tesco ) relative to overall total
quantity of the category bought by
household i in all channels at all chains
(PQi All )

channel loyalty variable (loyalit,c ) is a weighted average of past
channel loyalty (loyali,t−1,c ) and the households’ previous chan-
nel choice for the category at Tesco (bi,t−1,c , equal to 1 if channel
c was selected on the previous purchase occasion t − 1 and 0
otherwise). It captures state dependency effects, including the
tendency of consumers to repurchase the category in the same
channel. The concomitant variable, category-specific chain loy-
alty, is operationalized as the share of category purchases at
Tesco relative to other chains.5
To construct the promotion variables, we first determine the
percentage price discount for each stockkeeping unit (SKU) in
the category—because “consumers process their price savings
in a relative fashion” (Grewal et al. 2011, p. S48)—and each
time period when SKU n was on promotion. In line with previ-
ous price promotion studies, we identify promotion periods as
weeks in which the SKU’s unit price is at least one standard devi-
ation below its regular (average) price level (Geyskens, Gielens,
and Gijsbrechts 2010; Nijs et al. 2001).6 For the selected SKUs
and promotion weeks, we then calculate the percentage price
5 This time-invariant variable captures category-specificchain loyalty and indi-
cates whether household i spends a larger share of category purchases at the
focal chain Tesco. It is not to be confused with the time-specific channel loyalty
variable (loyalit,c ) which reflects a household’s loyalty to the online or offline
channel of the focal chain, at a specific point in time.
6 We used the entire estimation period to determine the regular (average) price
and its standard deviation. For milk, there was a significant (category-wide) price
increase in week 62 (week 36 of the estimation period) in the online and offline
channels. Therefore, we calculated an average price for the period before and an
average price for the period after the price increase. To verify our classification
of (no) promotion SKUs/weeks, we conducted robustness checks using different
operationalizations of the average price level (e.g. based on a moving window)
and different cut-offs (e.g. two standard deviations). Our classification remained
discount as the ratio of (1) the difference between the SKU’s
regular unit price and its actual price and (2) the SKU’s regular
unit price. To make these promotion variables household spe-
cific, we focus on visits to a channel when the household has an
opportunity to buy in the category, and we use SKU spending
shares in each household’s long-term shopping basket (i.e. over
the entire time period and all chains) as weights. These weights
reflect household-specific SKU preferences (Raju 1992; Zhang
and Breugelmans 2012). The category promotion variable thus
is larger when more SKUs of the category are promoted, pro-
motions are offered for more strongly preferred SKUs, and/or
the price discounts provided are greater. This operationalization
allows us to capture that more and deeper price discounts, espe-
cially when offered for the household’s favorite items, generate
stronger positive effects (Nijs et al. 2001; Shankar and Bolton
2004). Lagged promotion variables are constructed in a similar
way, using a one-period lag.7
Past promotion frequency variables (Lpromoit,c , Lpromoit,c̄ )
are operationalized as a weighted sum of category promotions
in a specific channel during the past 26 weeks (s = 1, . . ., 26),
given that household i visited the channel and was exposed
to these promotions (Vstit,c = 1). Because we want to capture
the overall effect of past promotion exposure in the category
(e.g. Foekens, Leeflang, and Wittink 1999; Kopalle, Mela, and
Marsh 1999), we take all SKU promotions in the category into
account
N (i.e. no weighting with household-specific preferences;
n=1 PctDisct n,t−s,c ). In line with previous research, we use
the decay factor λ to control for recency and frequency effects.
The resulting past promotion frequency variable is greater when
7 Note that the one-period lag, t − 1, refers to the previous shopping trip rather

robust. than the previous time period (week).

 E. Breugelmans, K. Campo / Journal of Retailing 92 (3, 2016) 333–351
Table 4
Marketing mix, descriptive statistics.
Milk
Offline
Number of SKUs in the assortment 79
Average paid price (in £)a .75 (s.e.: .26)
Average number of SKUs on promotion/weekb 10.17 (s.e.: 12.03)
Average percentage price discountb 13.84% (s.e.: .0476)
Percentage promotion overlapc 11.59%
a Expressed in standardized units: liters for milk, kilograms for cereals. Note that these are paid prices and therefore include the effect of promotional price
reductions.
b These variables are not weighted with household-specific preference shares.
c The percentage of SKUs that are simultaneously promoted in both channels, relative to the total number of promotions in the channel.
a household has more frequently and recently been exposed to
category promotions.
26  We first discuss the estimation results of the homogeneous
 N
 model. The results for milk and cereals are very similar, and we
Lpromoit,c = λs ∗ PctDisctn,t−s,c ∗ Vstit,c . (9)
s=1 n=1
Table 4 provides descriptive statistics of some marketing
mix variables. As is often observed for online grocery stores,
the online assortment is a subset of the offline assortment and
substantially smaller than the offline assortment for both milk
and cereals (Cheng 2010). The offline channel outperforms the
online channel in terms of the average number of SKUs on pro-
motion, but promotional advantages tend to be greater in the
online channel. For both categories, promotions overlap to some
extent across channels: About 50 percent of SKUs promoted in
a particular week in the online channel are also on promotion in
the offline channel. Hence, for some products there is a common
base for the promotional calendar of both channels (possibly in
response to manufacturers’ trade promotions), yet the two chan-
nels also retain significant independence and responsibility for
day-to-day operations, such as promotion plans.
Empirical Analysis: Results
We estimate the previously described model for each category
separately, as well as for a varying number of latent classes for
the multi-channel shoppers. For both categories, the Bayesian
Information Criterion (BIC) reaches a minimum only with a
large number of multi-channel latent classes (11 for milk, 5 for
cereals), and with negligible improvements in fit thereafter. In
such cases, it is suggested to use an elbow plot to explore the
diminishing gains in model fit (Masyn 2013; Petras and Masyn
2010). BIC plots reveal a clear elbow at three segments for both
milk and cereals, with an improvement in the BIC of only .74
percent for milk and .71 percent for cereals when moving from
three to four multi-channel latent classes. Tables 5 and 6 contain
the estimation results for the homogeneous and heterogeneous
models for milk and cereals, respectively. In addition to the
three multi-channel latent classes, there is one fixed class for
single-channel offline shoppers in the milk category and two
fixed classes for single-channel shoppers (one for offline-only
and one for online-only shoppers) in the cereals category.
341
Cereals
Online Offline Online
23 187 42
.61 (s.e.: .13) 3.38 (s.e.: 1.61) 3.25 (s.e.: 1.56)
2.56 (s.e.: 3.28) 25.12 (s.e.: 15.60) 5.15 (s.e.: 4.25)
17.50% (s.e.: .0774) 22.38% (s.e.: .0514) 27.64% (s.e.: .0825)
46.21% 10.85% 52.81%
Estimation Results
therefore discuss the results for both categories simultaneously,
citing category-specific outcomes only if we observe signifi-
cant differences. We obtain significant and expected effects for
the control variables—consumption rate and inventory—on pur-
chase incidence and quantity. The online correction parameter,
as expected, exerts a significant positive effect on the incidence
and quantity decisions of multi-channel shoppers (α1 = .640 and
β1 = .239 for milk; α1 = .817 and β1 = .098 for cereals). This
finding confirms results from previous studies, in that multi-
channel shoppers tend to order larger baskets online, with a
higher purchase probability for regularly bought categories.
The channel-specific category loyalty variables also have the
expected positive effect on purchase incidence (α5,1 = 1.015
and α5,2 = .962 for milk; α5,1 = .407 and α5,2 = .659 for
cereals).
Focusing on multi-channel shoppers, we observe signifi-
cant, expected, positive own-channel contemporaneous effects
of online and offline promotions on purchase incidence and
quantity (μ10,1 /δ10,1 > 0 and μ10,2 /δ10,2 > 0), except for the con-
temporaneous effect of offline promotions on offline purchase
incidence for milk (μ10,2 not significant). Most cross-channel
contemporaneous promotion effects point to channel canni-
balization, in that the effects are significant and negative
(μ2,1 /δ2,1 < 0 and μ2,2 /δ2,2 < 0), with the exception of the cross-
channel offline promotion effects on online purchase incidence
decisions for milk (μ2,2 not significant) and on online purchase
quantity decisions for cereals (δ2,2 not significant). Own- and
cross-channel lagged promotion effects (ϕ1,1 , ϕ1,2 , ϕ2,1 , ϕ2,2
and γ 1,1 , γ 1,2 , γ 2,1 , γ 2,2 ) are not significant in most cases.
When significant (mostly for milk), they are positive, in con-
trast with our expectations, but in accordance with preference
reinforcement effects observed in previous studies (see e.g.
Guyt and Gijsbrechts 2014). We performed alternative tests
of these lagged promotion effects and obtained largely the
same results (see the “Robustness Checks” section). Finally,
we find significant, negative own-channel promotion frequency
effects on both purchase incidence and quantity for cereals
(μ11,1 /δ11,1 < 0 and μ11,2 /δ11,2 < 0) and on purchase quantity for
milk (δ11,1 and δ11,2 < 0; μ11,1 not significant; μ11,2 significant
342 E. Breugelmans, K. Campo / Journal of Retailing 92 (3, 2016) 333–351

Table 5
Estimation results, milk.a
Homogeneous Model Heterogeneous Model

Offline-only Multi-channel Offline-only Multi-channel Multi-channel Multi-channel

(S1) (S2) (S3)

Category purchase
incidence decision
Intercept Online (α1 ) NR .640** NR .561** .729** .234**
Intercept Online (␣1 ) NR .640** NR .561** .729** .234**
Log cons. rate (α2 ) .336** . 345**
Inventory, MC (α3 ) −.090** −.087**
Log inventory (α4 ) .026** .024**
Loyalty Online (α5,1 ) NR 1.015** NR .933** 1.031** 1.232**
Loyalty Offline (α5,2 ) NR .962** NR .757** .986** .799**
Own-channel Online → Online (μ10,1 ) NR 2.307* NR .857 .952 2.692
contemporaneous
Offline → Offline −.130 .384 −.138 −.649 2.285 −2.748*
(μ10,2 )
Lagged Online → Online (ϕ1,1 ) NR .549 NR .354 −.406 .545
Offline → Offline (ϕ1,2 ) 3.192** 1.277** 3.192** −.461 .301 1.414*
Frequency Online → Online (μ11,1 ) NR −.315 NR .010 .317 −.725
Offline → Offline .217** .095* −.218** .024 −.093 .268**
(μ11,2 )
Cross-channel Online → Offline (μ2,1 ) NR −7.829** NR −9.772** -8.066** −8.271**
contemporaneous Offline → Online (μ2,2 ) NR −2.436 NR −4.337* 9.120** −7.073*
Lagged Online → Offline (ϕ2,1 ) NR .769** NR .532 .994 −.553
Offline → Online (ϕ2,2 ) NR .024 NR −1.235 5.122* −2.696
Frequency Online → Offline NR .342 NR .499 .717 .165
(μ12,1 )
Offline → Online NR −.086 NR −.049 −.172 −.038
(μ12,2 )

Category quantity decision

Intercept Offline (β0 ) .588** .654** .572** 1.019** .292** .523**
Intercept Online (β1 ) NR .239** NR .215** .427** .111**
Log cons. rate (β2 ) .607** .617**
Inventory, MC (β3 ) −.055** −.048**
Log inventory (β4 ) .032** .028**
Own-channel Online → Online (δ10,1 ) NR 1.892** NR 1.521** 1.652 .642
contemporaneous Offline → Offline (δ10,2 ) 2.762** 2.715** 2.750** 1.077* 2.832** 1.364**
Lagged Online → Online (␥1,1 ) NR .012 NR −.312 1.013 −.028
Offline → Offline (␥1,2 ) 1.033** .556** 1.035** −.346 −.093 .521*
Frequency Online → Online (δ11,1 ) NR −.415** NR −.379** −.182 −.203
Offline → Offline (δ11,2 ) −.106** −.111** −.105** −.080** −.065 −.050
Cross-channel Online → Offline (δ2,1 ) NR −2.886** NR −5.151** 2.076 −2.087**
contemporaneous Offline → Online (δ2,2 ) NR −1.939** NR −1.805* −.710 −.970
Lagged Online → Offline (␥2,1 ) NR .352** NR −.006** 1.186** −.029
Offline → Online (␥2,2 ) NR .650* NR .192 .123 .655
Frequency Online → Offline (δ12,1 ) NR −.010 NR −.002 .007 .148
Offline → Online (δ12,2 ) NR −.147** NR −.137** −.122 .062
Variance .302** .284**

Segment membership Intercept .938** −.938** 2.523** −2.564** 1.493** −1.453**

Category-specific chain −.465** .465** −.843** 2.701** −3.422** 1.564**
loyalty
a NR = not relevant.
* Significant at 5% level.
** Significant at 1% level.

but positive). We also observe significant and negative promo- milk, offline promotion frequency negatively influences online
tion frequency effects across channels, but only for two purchase purchase quantity (δ12,2 < 0).
quantity effects. For cereals, online promotion frequency neg- Turning to single-channel shoppers, their promotion reac-
atively influences offline purchase quantity (δ12,1 < 0), and for tions are highly similar to the average (homogeneous) promotion
Table 6
Estimation results, cereals.a
Homogeneous model Heterogeneous model
Offline-only Online-only Multi-channel Offline-only Online-only Multi-channel (S1) Multi-channel (S2) Multi-channel (S3)
Category purchase incidence decision
Intercept Offline (α0 ) .466** NR −.105** .469** NR .251** .404** −.587**
Intercept Online (α1 ) NR 1.142** .817** NR 1.144** .642** .522** .827**
Log cons. rate (α2 ) .444** .449**
Inventory, MC (α3 ) −.196** −.197**
Log inventory (α4 ) .036** .032**
Loyalty Online (α5,1 ) NR NR .407** NR NR .536** .408** .518**
Loyalty Offline (α5,2 ) NR NR .659** NR NR .376** .539** .669**
Own-channel Online → Online (μ10,1 ) NR 1.217 2.393** NR 1.225 10.025** .962 −.240
Offline → Offline (μ10,2 ) .694** 1.438** .668** 2.042**

E. Breugelmans, K. Campo / Journal of Retailing 92 (3, 2016) 333–351

contemporaneous NR NR .508 1.067
Lagged Online → Online (ϕ1,1 ) NR .444 .532 NR .456 1.048 −.420 1.050
Offline → Offline (ϕ1,2 ) .475** NR .047 .472** NR .335 .170 −.582
Frequency Online → Online (μ11,1 ) NR −1.656 −4.652** NR −1.639 −20.302* −.809 −1.458
Offline → Offline (μ11,2 ) −.510 NR −3.340** −.453 NR −2.100 −4.910** −.923
Cross-channel Online → Offline (μ2,1 ) NR NR −6.484** NR NR −4.973** −9.155** −5.016**
contemporaneous Offline → Online (μ2,2 ) NR NR −1.189* NR NR −1.442 −2.284* 1.192
Lagged Online → Offline (ϕ2,1 ) NR NR .834** NR NR 1.286** −.139 1.030**
Offline → Online (ϕ2,2 ) NR NR .274 NR NR 1.413 .568 −.631
Frequency Online → Offline (μ12,1 ) NR NR 2.183 NR NR −.615 .809 .800
Offline → Online (μ12,2 ) NR NR −2.760 NR NR −20.655** −2.215 3.165
Category quantity decision
Intercept Offline (β0 ) .230** NR .251** .219** NR .527** .102** .120**
Intercept Online (β1 ) NR .328** .098** NR .318** .185** .003 .199**
Log cons. rate (β2 ) .302** .296**
Inventory, MC (β3 ) −.092** −.089**
Log inventory (β4 ) .021** .018**
Own-channel Online → Online (δ10,1 ) NR 2.527** 2.014** NR 2.502** 1.679* 1.118* .547
contemporaneous Offline → Offline (δ10,2 ) 1.805** NR 1.575** 1.769** NR 2.059** .818* .780
Lagged Online → Online (␥1,1 ) NR −.538 .078 NR −.538 .608 −.124 −.073
Offline → Offline (␥1,2 ) −.052 NR −.086 −0.055 NR −.186 −.242 −.478
Frequency Online → Online (δ11,1 ) NR −6.730** −6.424** NR −6.653** −8.351** −2.450* −4.000*
Offline → Offline (δ11,2 ) −4.126** NR −3.389** −4.043** NR −4.680** −1.750* −.268
Cross-channel Online → Offline (δ2,1 ) NR NR −2.092** NR NR −3.998** −1.239 −.681
contemporaneous Offline → Online (δ2,2 ) NR NR −.340 NR NR .891 −.240 −1.492*
Lagged Online → Offline (␥2,1 ) NR NR .178 NR NR .225 .146 −.455
Offline → Online (␥2,2 ) NR NR .122 NR NR −.414 −.152 .965*
Frequency Online → Offline (δ12,1 ) NR NR −3.473** NR NR −6.179** −1.180 −4.112
Offline → Online (δ12,2 ) NR NR −1.144 NR NR 1.300 −1.500 3.430
Variance .335** .323**

Segment Membership Intercept 4.647** −7.823** 3.176** 4.467** −8.197** .112 .056 3.561**
Category-specific chain loyalty −2.966** 5.386** −2.420** −2.444** 6.124** .147 .269 −4.097**
a NR = not relevant.
* Significant at 5% level.
** Significant at 1% level.

343
344 E. Breugelmans, K. Campo / Journal of Retailing 92 (3, 2016) 333–351

Table 7
Segment profile.a
Offline- Online-only Multi-channel Multi-channel Multi-channel Multi-channel
only (S1) (S2) (S3) (average)

Panel A: Milk
Segment size .739 .082 .057 .121
Concomitant variable
Average category-specific chain loyalty .77 .89 .53 .85 .80
Purchase behavior characteristics
Average purchase incidence probability .72 .78 .49 .85 .75
Average purchase quantity 2.85 4.89 1.71 3.20 3.46
Control variables
Average consumption rate 3.05 3.16 3.33 3.27 3.52
Average inventory ratio 1.16 1.28 1.11 1.17 1.15
Average online channel loyalty NR .23 .22 .18 .20
Average offline channel loyalty NR .78 .79 .85 .81
Promotion exposure variables
Average weighted percentage online price discount NR .011 .007 .010 .010
Average weighted percentage offline price discount .007 .009 .005 .008 .007
Average percentage of purchase occasions with visits NR .07 .12 .09 .09
to online and offline channel
Panel B: Cereals
Segment size .725 .004 .076 .094 .101
Concomitant variable
Average category-specific chain loyalty .79 .98 .88 .89 .66 .80
Purchase behavior characteristics
Average purchase incidence probability .40 .68 .50 .59 .27 .44
Average purchase quantity .41 .79 .76 .51 .27 .49
Control variables
Average consumption rate .41 .46 .48 .42 .43 .44
Average inventory ratio 1.14 1.15 1.31 1.22 1.04 1.18
Average online channel loyalty NR NR .22 .25 .26 .25
Average offline channel loyalty NR NR .78 .76 .74 .76
Promotion exposure variables
Average weighted percentage online price discount NR .017 .014 .015 .013 .014
Average weighted percentage offline price discount .021 NR .021 .023 .021 .022
Average percentage of purchase occasions with visits NR NR .07 .08 .11 .09
to online and offline channel
a NR = not relevant.

reactions of multi-channel shoppers. Among offline-only shop-
pers, we find significant and positive contemporaneous effects
of promotions on purchase incidence (cereals) and quantity
(milk and cereals), as well as significant and negative promo-
tion frequency effects on quantity (milk and cereals). The lagged
promotion effects again are not significant (purchase quantity,
cereals) or else have an unexpected positive sign (purchase inci-
dence, milk and cereals; purchase quantity, milk). The small
online-only segment for cereals is sensitive to promotions for
purchase quantity decisions, with a significant, positive contem-
poraneous effect of online promotions and a significant, negative
online promotion frequency effect.
To explore differences in the own- and cross-channel pro-
motion reactions across multi-channel shoppers, we focus
on the estimation results from the heterogeneous models in
Tables 5 and 6. For each category, the three segments exhibit
distinctive promotion reactions patterns. To understand these
percent for milk segment 2, 66 percent for cereals segment 3).
This share is lower than the 80 percent cut-off used to clas-
sify consumers as multiple store shoppers in previous research
(Gijsbrechts, Campo, and Nisol 2008; Vroegrijk, Gijsbrechts,
and Campo 2013). Shoppers in these segments thus represent
multiple-chain shoppers.8 They are further characterized by an
above-average percentage of weeks in which they make both an
online and offline visit (12 percent for milk segment 2, 11 per-
cent for cereals segment 3) and lower average purchase incidence
probability and quantity. In both categories, these segments are
the least sensitive to promotions. Except for offline promotions
on offline purchase quantity for milk, none of the own-channel
contemporaneous promotion effects is significant for these seg-
ments. This lack of promotion sensitivity likely arises because
8 Although we use category-specific Tescoshares, chain loyalty is largely con-

segment differences, we provide additional profile characteris-
tics in Table 7.
The segment profiles in Table 7 indicate that, for each cat-
egory, one segment exhibits a below-average Tesco share (53
sistent across categories. The correlation between the category-specific shares
for milk and cereals equals .751. Using the 80 percent cut-off value to clas-
sify multiple-store shoppers, 87.1 percent of single-store shoppers for milk also
would be classified as single-store shoppers for cereals, and 76.9 percent would
be classified as multiple-chain shoppers in both categories.
 E. Breugelmans, K. Campo / Journal of Retailing 92 (3, 2016) 333–351
multiple-chain shoppers visit different stores on a regular basis to
buy each category in the most preferred store (i.e. the one offer-
ing the best price/quality for that category; Gijsbrechts, Campo,
and Nisol 2008). Rather than using promotions to reduce their
grocery spending, these shoppers realize savings by patronizing
multiple (complementary) stores with different pricing formats
(e.g. traditional supermarkets and hard discounters; Vroegrijk,
Gijsbrechts, and Campo 2013).
The other two segments in each category show above-average
Tesco shares (89 percent and 85 percent for milk segments 1
and 3; 88 percent and 89 percent for cereals segments 1 and
2). They mainly differ in their purchase incidence and purchase
quantity patterns. Consumers in milk segment 3 and cereals seg-
ment 2 buy the categories more frequently, while consumers
in milk segment 1 and cereals segment 1 buy in larger quanti-
ties. This is in line with their distinct promotion sensitivities.
The latter groups adjust their purchase quantity, rather than
timing decisions, in reaction to promotions, and they exhibit
significant, positive own-channel and negative cross-channel
contemporaneous promotion effects; significant, negative pro-
motion frequency effects, especially in the own channel and for
purchase quantity decisions; and no significant lagged promo-
tion effects. The former groups are more sensitive to promotions
in their purchase incidence decision, with substantial negative
cross-channel contemporaneous effects (but no significant cross-
channel frequency effects). We therefore describe milk segment
3 and cereals segment 2 shoppers as loyal frequent shoppers,
whereas milk segment 1 and cereals segment 1 are loyal large
basket shoppers.
Simulation Results
To gain better insights into the asymmetry of cross-channel
contemporaneous promotion effects, as well as the overall chain-
level effect of alternative multi-channel promotion strategies, we
conduct simulations, using the estimated parameters and actual
data set. To investigate the magnitude and direction of asym-
metry in cross-channel promotion effects, we compare shifts in
channel sales caused by online and offline promotions, rather
than computing cross-elasticities (Sivakumar 2004):
|
online sales| |
offline sales| contribute relatively little to the changes in total sales, com-
Asymmetry advantage = − , (10)
offline promo
online promo
where a positive (negative) value suggests asymmetry in the
advantage of the offline (online) channel.
We compute baseline values for the expected online/offline
category volume sales for the case in which there are no
promotions in the category in either channel (Promoit,c =
0 and Promoit,c̄ = 0).9 We use observed online promotions
9 We focus on contemporaneous promotion effects and use the estimates of the
significant parameters of the homogeneous models (see Tables 5 and 6), as well
as the observed values of the control variables and own- and cross-channel cross-
period promotion variables, as given. Previous exposure to promotions therefore
is treated as a household characteristic at a given point in time. In line with the
operationalization (Table 3), cross-channel contemporaneous promotion effects
are only observed during the promotion period and for households that visit both
345
(timing and depth) as test scenarios to simulate online and offline
cross-channel promotion effects, because all online SKUs are
also available offline. We compute the effect of the simulated
online/offline promotions on expected offline/online category
spending for household i (E(PQit,c )) as:
E(PQit,c ) = P(bit,c = 1) × E(PQit,c |bit,c = 1), (11)
where P(bit,c = 1) is the purchase incidence probability, and
E(PQit,c |bit,c = 1) is the expected purchase quantity, conditional
on incidence.
We then aggregate households’ expected category spending
in a given channel to derive the expected change in category vol-
ume sales. Because the change in promotion is the same for both
channels (denominator in Eq. (10)), we can derive conclusions
about the asymmetry of cross-channel promotion effects from
the difference in sales change between channels. Table 8, Panel
A, contains an overview of the results, which confirm that cross-
channel contemporaneous promotion effects are asymmetric.
For both milk and cereals, we find that the negative impact of
online promotions on expected offline sales is stronger than the
reverse.
Next, to determine the relative size of own- and cross-channel
effects, we use a similar procedure, but focus on all households
(single- and multi-channel shoppers) and all purchase occasions.
In Table 8, Panel B, we report, for each category, the percentage
change in category volume sales obtained with the current pro-
motion pattern, in the online channel (online sales), the offline
channel (offline sales), and the chain as a whole (total sales). We
also provide results for scenarios with online promotions only
(offline promotions set to zero), and with offline promotions only
(online promotions set to zero), to gain insights into how online
and offline promotions drive the overall sales effect. To obtain
the percentage change in sales, we compare the change in sales
caused by promotions with the baseline sales in the absence of
promotions. For both categories, the overall effect of the current
promotion practice on total sales is positive (+2.09 percent for
milk, +6.21 percent for cereals). An important conclusion there-
fore is that own-channel contemporaneous promotion effects
are tempered, but not completely canceled out by (negative)
cross-channel effects. It is further clear that online promotions
pared with offline promotions (+.06 percent and +28 percent vs.
+2.04 percent and +5.97 percent, for respectively milk and cere-
als). This difference can be explained by the higher offline visit
frequency and the large group of single-channel offline shoppers.
Finally, to gain insights into the different payoffs in sales rev-
enue from different multi-channel promotion strategies (cf. Wolk
and Ebling’s (2010) discussion of different multi-channel pri-
cing strategies), we investigate three scenarios: integration (same
frequency and average percentage price discount for online and
offline promotions), differentiation in the average percentage
price discount (higher for offline or online promotions), and
channels within that period. We therefore compute the change in sales caused by
cross-channel promotion effects for households that visit both online and offline
channels during the promotion period.
346 E. Breugelmans, K. Campo / Journal of Retailing 92 (3, 2016) 333–351

Table 8
Simulation results.
Milk Cereals
Panel A: Asymmetric Cross-Channel Effects
Offline promotions Online promotions Difference Asymmetry advantage Offline promotion Online promotion Difference Asymmetry advantage
|Δonline sales| |Δoffline sales| |Δonline sales| |Δoffline sales|
153.66 477.75 −324.09 ONLINE 13.03 80.21 −67.18 ONLINE
Panel B: Percentage Change in Sales Compared with Baseline Sales (No Promotion) for Current Practice
Total Offline sales generated Total Total sales Online sales Total Offline sales generated Total Total sales
online by offline generated by online by offline
sales Single- Multi- sales Single- Multi- sales Single- Multi- sales
channel channel channel channel channel channel
shoppers shoppers shoppers shoppers shoppers shoppers
Online and offline 2.62% 2.14% 1.84% 2.07% 2.09% 5.33% 7.34% 7.19% 5.83% 7.24% 6.16% 6.21%
promotions
Online promotions 2.99% 0% −0.32% −.07% .06% 5.33% 7.84% 7.66% 0% −.57% −.13% .28%
only
Offline promotions −.33% 2.14% 2.19% 2.15% 2.04% 0% −.40% −.37% 5.83% 7.95% 6.21% 5.97%
only
Panel C: Percentage Change in Sales Compared with Baseline Sales (No Promotion) for Different Scenarios (values in brackets are percentage change in return)
Total Offline sales generated Total Total sales Online sales Total Offline sales generated Total Total sales
online by offline generated by online by offline
sales Single- Multi- sales Single- Multi- sales Single- Multi- sales
channel channel channel channel channel channel
shoppers shoppers shoppers shoppers shoppers shoppers
Integration 2.44% 3.04% 2.67% 2.95% 2.93% 5.33% 7.40% 7.25% 3.51% 4.04% 3.64% 3.83%
(1.11%) (1.70%) (1.36%) (1.61%) (1.59%) (2.74%) (4.63%) (4.50%) (1.84%) (2.21%) (1.93%) (2.06%)
Differentiation in the average percentage price discount
a. Higher for offline 2.41% 3.32% 2.94% 3.23% 3.20% 5.33% 7.37% 7.22% 3.82% 4.45% 3.97% 4.14%
promotion (1.08%) (1.82%) (1.49%) (1.75%) (1.72%) (2.74%) (4.60%) (4.47%) (1.98%) (2.42%) (2.08%) (2.21%)
b. Higher for online 2.69% 3.04% 2.65% 2.95% 2.94% 5.83% 8.14% 7.98% 3.51% 4.01% 3.63% 3.86%
promotions (1.21%) (1.70%) (1.34%) (1.61%) (1.59%) (2.95%) (5.02%) (4.88%) (1.84%) (2.18%) (1.92%) (2.08%)
Differentiation in the promotion frequency
a. More offline 2.21% 4.49% 4.04% 4.39% 4.30% 5.33% 7.32% 7.17% 5.02% 5.80% 5.20% 5.30%
promotions (0.88%) (2.58%) (2.19%) (2.49) (2.43%) (2.74%) (4.55%) (4.42%) (2.74%) (3.33%) (2.85%) (2.94%)
b. More online 3.78% 3.04% 3.09% 3.05% 3.08% 6.68% 9.36% 9.17% 3.51% 3.80% 3.58% 3.87%
promotions (1.88%) (1.70%) (1.73%) (1.70%) (1.71%) (3.53%) (5.97%) (5.80%) (1.84%) (1.96%) (1.87%) (2.08%)

differentiation in the promotion frequency (more offline or more
online promotions). In the first scenario, we use online promo-
tions as a source to construct own- and cross-category promotion
effects (cf. the asymmetry calculations). To determine a viable,
higher average percentage price discount for the second sce-
nario, we draw on prior findings (DelVecchio, Krishnan, and
Smith 2007; Kalwani and Yim 1992) and use 7.5 percent as the
average discount increase. This results in an average high price
discount of 25 percent for milk and 35.14 percent for cereals.
For the third scenario, we randomly add promotions (i.e. aver-
age weighted price discount) in 10 percent of the cases. Table 8,
Panel C, offers an overview of the results, including percentage
changes in volume sales and returns (in brackets). For the lat-
ter, we correct the percentage change in quantity sales by the
expected price discount percentage.10
Both differentiation strategies result in higher positive overall
changes in total sales, compared with the integration strat-
egy. In addition, the differentiation strategies exert a stronger
10 Expected change in returns for household i in channel c on purchase occasion
t, due to promotion, can be approximated by the following equation:
E(POit,c |promoit,c ) ∗ [1 − E(WPctDisctit,c )]
− 1,
E(PQit,c |promoit,c )
where E(PQit,c |promoit,c ) is the expected purchase quantity (expected purchase
quantity, conditional on incidence × purchase incidence probability), given
promotions in the category;E(PQit,c |promoit,c ) is the expected purchase quan-
tity without promotions in the category; and E(WPctDisctit,c ) is the expected
(weighted) average percentage price reduction for household i in channel c at
purchase occasion t.
positive effect when the differentiation pertains to offline rather
than online promotions. For both categories, a differentiation in
the promotion frequency dimension prompts a higher percentage
increase in sales than a differentiation in the percentage dis-
count depth. The large group of single-channel offline shoppers,
higher offline visit frequency, and stronger negative cross-
channel online promotion effects together drive stronger positive
effects for offline promotions. The positive percentage changes
in returns show that the increase in sales due to promotions more
than compensates for the decrease in profit margin. Similar con-
clusions on the relative performance of alternative promotion
strategies hold for return increases as for sales increases.
Robustness Checks
We conducted several robustness checks to verify the valid-
ity and specification of our model variables and the consistency
of our findings. First, our model specification outperformed
alternative specifications for the baseline purchase tendency,
using different indicators for category need (e.g. average spend-
ing per trip instead of consumption per period) and inventory
effects (e.g. mean-centered lagged spending). Second, our model
that includes both cross-period promotion variables clearly out-
performed models that included only one of the cross-period
effects. Model fit did also not improve for alternative oper-
ationalizations of the lagged promotion variable, including a
double lag specification (promotion in t − 1 and in t − 2) and a
weighted lagged promotion variable defined over the 26 previous
periods.
 E. Breugelmans, K. Campo / Journal of Retailing 92 (3, 2016) 333–351
Third, we tested alternative operationalizations of the promo-
tion variable, including the number of promoted SKUs which
gave very similar results; and the absolute (rather than percent-
age) price discount which provided clearly inferior results. We
also checked the sensitivity of the weight we used to capture fad-
ing effects for the promotion frequency variables (Eq. (9) and
Table 3), but the results for decay factors in the .5–.9 range
were similar. Unweighted promotion variables lead to simi-
lar results for the within-channel promotion effects but cause
estimation problems for cross-channel promotion effects due
to multi-collinearity. Fourth, we estimated our model with a
larger sample (i.e. no cut-off for consumption rates, such that the
sample includes very light category users) which provided sub-
stantially inferior results, and a smaller sample (multi-channel
shoppers only) which gave very similar results. Consumers who
shop in one channel are a substantial portion of the population
in most grocery markets, and retailers care about overall sales
and overall promotion lift, so we consider it justified to include
single-channel shoppers in our estimation. We also tested a
model that allowed for heterogeneity among these shoppers and
thereby confirmed that the restriction of single-channel latent
classes to one (or two for cereals) does not affect the reported
results.11 Lastly, we found that a linear truncated model pro-
vided robust results, but the log-normal model provided better
model fit.
Discussion and Managerial Implications
The vast growth of multi-channel retailing in recent years
(Neslin et al. 2006; Neslin and Shankar 2009; Shankar et al.
2011) emphasizes the increasing need for marketing managers
to understand how to manage promotions in complex, multi-
channel shopping environments. The major objectives of this
research were to examine the cross-channel effects of price
promotions on category purchase decisions in a multi-channel
context; analyze how they affect the chain’s overall results,
taking contemporaneous and cross-period effects into account;
and investigate the implications for the multi-channel retailer’s
choice of promotion strategy.
Own-Channel Promotion Effects
Our results replicate previous findings: price promotions can
have positive effects on category purchase decisions, both offline
and online, during the promotion period (support H1). Consis-
tent with previous research, we find that a high frequency of
promotions in the offline channel reduces promotion effective-
ness during subsequent visits to the same channel (Foekens,
Leeflang, and Wittink 1999; Kalyanaram and Winer 1995;
11 The results of the model where we allowed for heterogeneity among single-
channel offline shoppers are available on request. Most single-channel offline
shopper segments have significant, positive, contemporaneous promotion effects
(especially for the quantity decision), and some also have significant, negative
promotion frequency effects. Similar to our findings for multi-channel shoppers,
we find no evidence of lagged promotion effects among different single-channel
offline shopper segments.
347
Kopalle, Mela, and Marsh 1999). Although online grocery stores
typically are visited less frequently (Chu et al. 2010; see Table 2),
we find evidence of a similar, own-channel online promotion fre-
quency effect (support H3). In contrast with our expectations, a
positive effect arises for some lagged promotion variables (no
support H2). Past promotions, rather than reducing purchase
tendencies in the category, reinforce own-channel purchase
behavior within the chain. Similar effects have been reported
for offline purchases (e.g. Guyt and Gijsbrechts 2014).
Cross-Channel Contemporaneous Effects
Our results show that promotions can have negative effects on
purchase decisions in the other channel (support H4). By pre-
empting category purchases in the other channel (i.e. reduced
purchase incidence probability and/or reduced quantity in the
other channel), channel differences in promotion attractiveness
can, within the same chain, lead to a shift in category sales from
one channel to the other. These shifts depend on the extent to
which promotions increase the relative attractiveness of a cat-
egory, which is a function of the number of promoted items,
the discount depth, and consumer preferences for promoted
SKUs. Our simulations further indicate that price promotions
are an effective instrument to stimulate category sales in a multi-
channel grocery shopping context, even though they create a risk
of cannibalization: for both categories, the overall effect of the
current promotion practice on total sales continues to be posi-
tive, both in terms of the percent increase in quantity sold and
the percent increase in return. So, a positive own-channel effect
does not get completely canceled out by a negative cross-channel
effect.
We further find that online promotions make relatively small
contributions to the changes in sales (and revenue) compared
with offline promotions. This is the result of the large, dom-
inant single-channel offline shopper segment, and the higher
offline visit frequency present in our sample. Hence, a differen-
tiation strategy that favors offline promotions appears to generate
a greater chain-level effect. Yet, multi-channel grocery retailers
should monitor the size and promotion sensitivity of their differ-
ent shopper segments constantly, considering the fast-increasing
number of multi-channel grocery shoppers.
It is also evident that grocery retailers and researchers
in this field must take both own- and cross-channel promo-
tion effects, and both multi- and single-channel shoppers, into
consideration when assessing the impact of promotions. In a
multi-channel context, promotion effects will be overestimated
and conclusions will be inaccurate if cannibalization triggered
by cross-channel effects is ignored or a channel-by-channel anal-
ysis is used.
Asymmetry of Cross-Channel Effects
Cross-channel, contemporaneous promotion effects appear to
be asymmetric, that is online category price promotions have a
stronger impact on offline category purchase decisions than vice
versa (support H5). For the two grocery categories under con-
sideration (milk and cereals)—which are frequently purchased,
348 E. Breugelmans, K. Campo / Journal of Retailing 92 (3, 2016) 333–351
low risk categories—online shopping convenience thus consti-
tutes an important advantage and triggers stronger reactions to
online promotions, at the expense of offline sales. Multi-channel
retailers could capitalize on these asymmetric promotion effects
by adopting a differentiated promotion strategy. However, to
assess the profitability of such an approach, channel differences
in operational costs and profit margins also have to be taken into
account, information we unfortunately did not have access to.
Cross-Channel Cross-Period Effects
In line with the own-channel lagged promotion effects,
we find few significant—and a limited number of positive
impacts—of previous promotions on category purchases in the
other channel in the post-promotion period (no support H6).
We do find that high promotion frequency in one channel can
decrease future promotion sensitivity in the same but also in
the other channel (support H7). Therefore, exposure to a more
extensive online and offline price promotion offer in a multi-
channel context might reduce future promotion effectiveness.
This reduction in promotion effectiveness is relatively limited
though, and our results indicate that own-channel promotion fre-
quency effects clearly dominate the cross-channel effects. For
the examined retail chain, high promotion frequency is thus not
an important disadvantage associated with a differentiated pro-
motion strategy. Yet, with the growing number of multi-channel
shoppers, multi-channel retailers should be aware that provid-
ing more varied and extensive promotions in a differentiated
multi-channel strategy can entail a risk of reduced promotion
effectiveness.
Heterogeneity in Promotion Reactions
The promotion effects depend on households’ loyalty to the
chain (support H8). We find that multiple-chain shoppers who
buy a substantial amount of their groceries at competitive chains
are less inclined to buy sooner or more at the focal chain in reac-
tion to promotions (negligible own-channel contemporaneous
promotion effects). Although they are more sensitive to pro-
motions in general, the two loyal segments reveal some clear
differences in their promotion reactions. The strongest (own- and
cross-channel) contemporaneous and cross-period (frequency)
promotion effects emerge from the segment that buys less fre-
quently but in larger quantities (without consuming substantially
more). These loyal customers appear to align their category pur-
chases with actual and expected promotion actions, not only
within but also across channels. For the retailer, the high pro-
motion sensitivity of these loyal customers points to a risk of
subsidization, in that price promotions lower the profit margins
earned from the customers who already are likely to make their
category purchases from the chain. Managers therefore need to
be wary of negative cross-channel effects in these segments.
Nevertheless, it is clear that a differentiated promotion strategy
can pay off, especially if it includes more offline promotions and
the differentiation takes place along the frequency dimension.
Limitations and Further Research
Our research has several limitations that also provide inter-
esting indications for further research. First, we concentrate on
price discount promotions and do not distinguish between pro-
motions initiated by the manufacturer versus the retailer, nor
do we have data to differentiate among sales promotion forms
(e.g. price vs. non-price promotions) and communications (e.g.
store flyer vs. in-store promotions). Additional research might
assess the importance of the initiator, the promotion form or its
communication vehicle. It also would be interesting to extend
our models with a price variable (which we partly capture in
the price discount promotion variable) and information about
the chain’s delivery time and fee (data that we unfortunately did
not have). We also lacked any data about unit cost information
(margins) and therefore had to restrict our simulations to sales
revenue figures.
Second, we investigated cross-channel promotion effects for
just two categories. Analyzing a larger set of categories and
explicitly linking cross-channel effects to category character-
istics, such as perceived online purchase risk and shopping
convenience, could provide more specific guidelines for multi-
channel promotion planning. What is more, we focus on an
online grocery shopping context. Replicating the study in other
multi-channel retail settings would be interesting. Our research
focuses on cross-channel promotion effects at the category level,
which are most relevant for a retailer. It also could be inter-
esting to investigate cross-channel effects at the brand level,
with the notion that manufacturers could face a similar multi-
channel promotion planning problem when selling their brands
through multi-channel retailers. Likewise, it would be inter-
esting to investigate the impact of cross-promotions on traffic
building and channel visits, as well as on cross-category pur-
chase behavior.
Third, our research focuses on analyzing cross-channel
promotion effectiveness in a multi-channel, single-chain
context. We consider it a promising avenue for additional
research to investigate cross-channel promotion effectiveness
in a multi-channel, multi-chain context, such that researchers
explicitly model chain and channel choice decisions and take
the activities of competitive chains into account.
Executive summary
More and more offline grocery chains are catching up with
the trend towards multi-channel retailing, where they offer con-
sumers the possibility to purchase via an online store. While
creating interesting opportunities to solidify customer loyalty
through improved service and to reach a larger group of con-
sumers, multi-channel retailing also adds to the complexity of
the marketing strategy. This is especially the case for grocery
retailers, where most grocery buyers keep visiting the offline
store after they start to buy at the chain’s online store. These
multi-channel shoppers are exposed to, and can take advantage
of, potentially different marketing offers. They can, for instance,
adjust their purchase decisions to take maximum advantage
of different price reductions, leading to cannibalization of the
 E. Breugelmans, K. Campo / Journal of Retailing 92 (3, 2016) 333–351
chain’s sales in each of the channels. Consequently, when imple-
menting promotional strategies across channels, multi-channel
retailers must wary about whether and how promotions in one
channel influence buying behavior of their customers in the other
channel.
In this paper, we perform a comprehensive analysis and exam-
ine the impact of price promotions on consumers’ category
purchase decisions in a multi-channel grocery retail context,
where consumers visit the offline and online channel of the
same chain. From previous research, it is clear that price promo-
tions not only have a direct effect on the consumers’ purchase
decisions in the same store. Promotion effects can also extend
to later periods, and impact purchase behavior in other stores.
We investigate similar effects in the new multi-channel context,
by takinginto account both within and cross-channel promotion
effects (i.e. effects on the own and other channel), as well as con-
temporaneous and cross-period promotion effects (i.e. effects
during and after the promotion period). For cross-period promo-
tion effects, we distinguish between short term lagged promotion
effects on subsequent category purchases (a larger stock at home
leading to a reduced tendency to buy the product on the next
shopping trip), and longer term promotion frequency effects on
future promotion effectiveness (very frequent promotions lead-
ing to the expectation that there will be another price promotion
very soon, and that it is worthwhile to delay a category purchase
until the next shopping trip).
This study is the first to explicitly examine these multiple
cross-channel promotion effects. We thereby extend the pro-
motion and multi-channel retailing literature by providing a
comprehensive analysis of different promotion effects: whether
and how promotions in one channel reinforce category sales in
the own channel, cannibalize category sales in the other chan-
nel, lead to lower category sales in post-promotion periods,
and/or reduce promotion effectiveness by making consumers
less responsive to promotions.
To analyze these effects, we got access to a very rich data
set of online and offline purchases by a large and representative
U.K. household panel, obtained from AiMark. The data cover
online and offline household grocery purchases at Tesco, for
a 78-week period, and are used to investigate price promotion
effects on the consumers’ decision whether and how much to
buy of a given product category in the chain and channel. We
examine the different promotion effects for two categories, milk
and cereals, which are frequently purchased by a large num-
ber of consumers, and are often promoted by price reductions.
Following the purchase incidence and quantity model estima-
tions, we use simulations to assess the effect of three different
multi-channel promotion strategies: similar price promotions in
both channels, different price promotion depth in the online and
offline channel, and different frequency of price promotions in
the online and offline channel.
The results show that price promotions not only have pos-
itive effects on purchase decisions in the own channel but
also generate negative effects on category purchases in the
other channel during the promotion period. Yet, even though
promotions create a risk of cannibalization, our simulations indi-
cate that positive own-channel effects do not get completely
349
cancelled out by negative cross-channel effects. We further find
that these cross-channel, contemporaneous promotion effects
appear to be asymmetric, that is online category price pro-
motions have a stronger impact on offline category purchase
decisions than vice versa. While we find few significant—and a
limited number of positive—instances of short term lagged pro-
motion effects, we do find that high promotion frequency in one
channel can decrease future promotion sensitivity in the same
but also in the other channel. This reduction in promotion effec-
tiveness is relatively limited though, and our results indicate that
own-channel promotion frequency effects clearly dominate the
cross-channel effects. Lastly, we find that cross-channel effects
are more negative for more loyal customers of the chain.
Our results make clear that a differentiated promotion strat-
egy can pay off, especially if it includes more offline promotions
and the differentiation takes place along the frequency dimen-
sion. It is also evident that grocery retailers and researchers
in this field must take both own- and cross-channel promotion
effects, and both multi- and single-channel shoppers, into con-
sideration when assessing the impact of price promotions. In a
multi-channel context, promotion effects will be overestimated
and conclusions will be inaccurate if cannibalization triggered
by cross-channel effects is ignored or a channel-by-channel anal-
ysis is used.
Acknowledgments
The authors thank AiMark for providing the data and greatly
appreciate the valuable suggestions by the editor and the anony-
mous JR reviewers. Special thanks go to Huiying He and Kim
Goeleven for their help. They also thank Koert van Ittersum,
Siegfried Dewitte, Vera Blazevic, and Lien Lamey for their
helpful suggestions on previous versions of this article.
References
Ailawadi, Kusum L., Karen Gedenk, Christian Lutzky and Scott Neslin (2007),
“Decomposition of the Sales Impact of Promotion-Induced Stockpiling,”
Journal of Marketing Research, 44 (3), 450–67.
Ailawadi, Kusum.L. and Scott A. Neslin (1998), “The Effect of Promotion on
Consumption: Buying More and Consuming It Faster,” Journal of Marketing
Research, 35 (3), 390–8.
Ansari, Asim, Carl Mela and Scott Neslin (2008), “Customer Channel Migra-
tion,” Journal of Marketing Research, 45 (1), 60–76.
Avery, Jill, Thomas J. Steenburgh, John Deighton and Mary Caravella (2012),
“Adding Bricks to Clicks: Predicting the Patterns of Cross-Channel Elastic-
ities Over Time,” Journal of Marketing, 76 (3), 96–111.
Bell, David R. and Yasemin Boztuğ (2007), “The Positive and Negative Effects of
Inventory on Category Purchase: An Empirical Analysis,” Marketing Letters,
18 (1), 1–14.
Bell, David R., Jeongwen Chiang and V. Padmanabhan (1999), “The Decompo-
sition of Promotional Response: An Empirical Generalization,” Marketing
Science, 18 (4), 504–27.
Bell, David R. and Christian A.L. Hilber (2006), “An Empirical Test of the
Theory of Sales: Do Household Storage Constraints Affect Consumer and
Store Behavior,” Quantitative Marketing and Economics, 4 (2), 87–117.
Blattberg, Robert C., Richard Briesch and Edward J. Fox (1995), “How Promo-
tions Work,” Marketing Science, 14 (3), 122–32.
Bronnenberg, Bart J. and Luc Wathieu (1996), “Asymmetric Promotion Effects
and Brand Positioning,” Marketing Science, 15 (4), 379–94.
350 E. Breugelmans, K. Campo / Journal of Retailing 92 (3, 2016) 333–351
Bucklin, Randolph E. and Sunil Gupta (1992), “Brand Choice, Purchase Inci-
dence and Segmentation – An Integrated Modeling Approach,” Journal of
Marketing Research, 29 (2), 201–15.
Bucklin, Randolph E. and James M. Lattin (1992), “A Model of Product Category
Competition among Grocery Retailers,” Journal of Retailing, 68 (3), 271–93.
Cheng, Jacqui (2010), “The New Age of Online Grocery Shopping,”
(accessed April 10 2015), [available at http://arstechnica.com/web/news/
2010/03/the-new-age-of-online-grocery-shopping.ars].
Chintagunta, Pradeep K., Junhong Chu and Javier Cebollada (2012), “Quan-
tifying Transaction Costs in Online/Off-line Grocery Channel Choice,”
Marketing Science, 31 (1), 96–114.
Chu, Junhong, Marta Arce-Urriza, Jose-Javier Cebollada-Calvo and Pradeep K.
Chintagunta (2010), “An Empirical Analysis of Shopping Behavior across
Online and Offline Channels for Grocery Products: The Moderating Effects
of Household and Product Characteristics,” Journal of Interactive Marketing,
24 (4), 251–68.
Chu, Junhong, Pradeep Chintagunta and Javier Cebollada (2008), “A Com-
parison of Within-Household Price Sensitivity Across Online and Offline
Channels,” Marketing Science, 27 (2), 283–99.
Dawes, John.G. (2012), “Brand-Pack Size Cannibalization Arising from Tem-
porary Price Promotions,” Journal of Retailing, 88 (3), 343–55.
Degeratu, Alexandru.M., Arvind Rangaswamy and Jianan Wu (2000), “Con-
sumer Choice Behavior in Online and Traditional Supermarkets: The Effects
of Brand Name, Price, and Other Search Attributes,” International Journal
of Research in Marketing, 17 (1), 55–78.
DelVecchio, Devon H., Shanker Krishnan and Daniel C. Smith (2007), “Cents
or Percent? The Effects of Promotion Framing on Price Expectations and
Choice,” Journal of Marketing, 71 (3), 158–70.
Foekens, Eijte W., Peter S.H. Leeflang and Dick R. Wittink (1999), “Varying
Parameter Models to Accommodate Dynamic Promotion Effects,” Journal
of Econometrics, 89 (1), 249–68.
Geyskens, Inge, Katrijn Gielens and Els Gijsbrechts (2010), “Private Label
Proliferation: How Introducing Economy and Premium Private Labels Influ-
ences Brand Choice,” Journal of Marketing Research, 47 (5), 791–807.
Gijsbrechts, Els, Katia Campo and Patricia Nisol (2008), “Beyond Promotion-
Based Store Switching: Antecedents and Patterns of Systematic Multiple-
Store Shopping,” International Journal of Research in Marketing, 25 (1),
5–21.
Grewal, Dhruv, Kusum L. Ailawadi, Dinesh Gauri, Kevin Hall, Praveen Kopalle
and Jane R. Robertson (2011), “Innovations in Retail Pricing and Promo-
tions,” Journal of Retailing, 87S (1), S43–52.
Guadagni, Peter M. and John D.C. Little (2008), “A Logit Model of Brand
Choice Calibrated on Scanner Data,” Marketing Science, 27 (1), 29–48.
Gupta, Sachin and Pradeep K. Chintagunta (1994), “On Using Demographic
Variables to Determine Segment Membership in Logit Mixture Models,”
Journal of Marketing Research, 31 (1), 128–36.
Gupta, Sumeet and Hee-Woong Kim (2009), “Value-Driven Internet Shopping:
The Mental Accounting Theory Perspective,” Psychology and Marketing,
27 (1), 13–35.
Guyt, Jonne Y. and Els Gijsbrechts (2014), “Take Turns or March in Sync?
The Impact of the National Brand Promotion Calendar on Manufacturer and
Retailer Performance,” Journal of Marketing Research, 51 (6), 753–72.
IGD (2009), “Online Grocery Shopping to Double in Five Years,”
(accessed April 10, 2015), [available at http://www.telegraph.co.uk/finance/
newsbysector/retailandconsumer/7020524/Online-food-shopping-expected-
to-double-in-five-years.html].
IGD (2011), “Hunt for Value Drives Multi-Channel Shopping Across
Europe,” (accessed April 10, 2015), [available at http://www.igd.com/our-
expertise/Shopper-Insight/Channels-and-In-store/4589/European-shoppers-
hunt-around-multiple-channels-for-value/].
IGD (2015). “The Growth Potential of Online Grocery,” (accessed February
1, 2016), [available at http://www.igd.com/Research/Shopper-Insight/The-
growth-potential-of-online-grocery/].
Kalwani, Manohar U. and Chi Kin Yim (1992), “Consumer Price and Promotion
Expectations: An Experimental Study,” Journal of Marketing Research, 29
(1), 90–100.
Kalyanaram, Gurumurthy and Russell S. Winer (1995), “Empirical Generaliza-
tions from Reference Price Research,” Marketing Science, 14 (3), 161–9.
A.T. Kearney (2012), “A Fresh Look at Online Grocery,” (accessed April
10, 2015), [available at http://www.atkearney.com/paper/-/asset publisher/
dVxv4Hz2h8bS/content/a-fresh-look-at-online-grocery/10192].
Kopalle, Praveen K., Carl F. Mela and Lawrence Marsh (1999), “The Dynamic
Effect of Discounting on Sales: Empirical Analysis and Normative Pricing
Implications,” Marketing Science, 18 (3), 317–33.
Kumar, V. and Robert P. Leone (1988), “Measuring the Effect of Retail Store Pro-
motions on Brand and Store Substitution,” Journal of Marketing Research,
25 (2), 178–86.
Liu, Yan and Subramanian Balachander (2014), “How Long Has It Been Since
the Last Deal? Consumer Promotion Timing Expectations and Promotional
Response,” Quantitative Marketing and Economics, 12 (1), 85–126.
MacKenzie, Scott B. and Rockney G. Walters (1988), “A Structural Equations
Analysis of the Impact of Price Promotions on Store Performance,” Journal
of Marketing Research, 25 (1), 51–64.
Masyn, Katherine (2013), “Latent Class Analysis and Finite Mixture Modeling,”
in Oxford Handbook of Quantitative Methods in Psychology, Little Todd ed.
Oxford: Oxford University Press
McDowell, Allen (2003), “From the Help Desk: Hurdle Models,” The Stata
Journal, 3 (2), 178–84.
Melis, Kristina, Katia Campo, Els Breugelmans and Lien Lamey (2015),
“The Impact of the Multi-Channel Retail Mix on Online Store
Choice: Does Online Experience Matter?,” Journal of Retailing, 91 (2),
272–88.
Neslin, Scott A., Dhruv Grewal, Robert Leghorn, Venkatesh Shankar, Marije L.
Teerling, Jacquelyn S. Thomas and Peter C. Verhoef (2006), “Challenges and
Opportunities in Multichannel Customer Management,” Journal of Service
Research, 9 (2), 95–112.
Neslin, Scott A. and Venkatesh Shankar (2009), “Key Issues in Multichannel
Customer Management: Current Knowledge and Future Directions,” Journal
of Interactive Marketing, 23 (1), 70–81.
Nijs, Vincent R., Marnik Dekimpe, Jan-Benedict E.M. Steenkamp and
Dominique M. Hanssens (2001), “The Category Demand Effects of Price
Promotions,” Marketing Science, 20 (1), 1–22.
Petras, Hanno and Katherine Masyn (2010), “General Growth Mixture Analysis
with Antecedents and Consequences of Change,” in Handbook of Quan-
titative Criminology, Piquero Alex and Weisburd David, eds. New York:
Springer-Verlag.
Perkins, Garals (2014), ‘Millennials Take their Growing Buying Power Online,”
(accessed June 23, 2014), [available at http://www.thepacker.com/opinion/
Millennials-take-their-growing-buying-power-online-260424061.html].
Raju, Jagmohan (1992), “The Effect of Price Promotions on Variability in Prod-
uct Category Sales,” Marketing Science, 11 (3), 207–20.
Retail Systems Research (2013), “Omni-Channel 2013: The Long Road to Adop-
tion,” (accessed June 23, 2014), [http://www.rsrresearch.com/2013/06/11/
omni-channel-2013-the-long-road-to-adoption/].
Shankar, Venkatesh and Ruth N. Bolton (2004), “An Empirical Analysis of
Determinants of Retailer Pricing Strategy,” Marketing Science, 23 (1),
28–49.
Shankar, Venkatesh J., Jeffrey Inman, Murali Mantrala, Eileen Kelley and Ross
Rizley (2011), “Innovations in Shopper Marketing: Current Insights and
Future Research Issues,” Journal of Retailing, 87 (1), 29–42.
Sivakumar, K. (2004), “Manifestations and Measurement of Asymmetric Brand
Competition,” Journal of Business Research, 57 (8), 813–20.
Sivakumar, K. and S.P. Raj (1997), “Quality Tier Competition: How Price
Change Influences Brand Choice and Category Choice,” Journal of Mar-
keting, 61 (3), 71–85.
Steiman, H.C. (2014), “Why 2014 Will Finally Be the Year of the Online Gro-
cer,” (accessed November 10, 2014), [available at http://www.businessweek.
com/articles/2014-01-16/why-2014-will-finally-be-the year-of-the-online-
grocer].
van Heerde, Harald J. and Scott A. Neslin (2008), “Sales Promotion Models,” in
Handbook of Marketing Decision Models, Wierenga Berend ed. New York:
Springer
Vermunt, Jeroen K. and Jay Magidson (2005), Latent GOLD 4.0 User Manual,
Belmont, MA: Statistical Innovations Inc.
Vroegrijk, Mark, Els Gijsbrechts and Katia Campo (2013), “Close Encounter
with the Hard Discounter Entry: A Multiple-Store Shopping Perspective on
 E. Breugelmans, K. Campo / Journal of Retailing 92 (3, 2016) 333–351 351

the Impact of Local Hard-Discounter Entry,” Journal of Marketing Research,
50 (5), 606–26.
Wolk, Agnieszka and Christine Ebling (2010), “Multi-Channel Price Differen-
tiation: An Empirical Investigation of Existence and Causes,” International
Journal of Research in Marketing, 27 (2), 142–50.
Wooldridge, Jeffrey M. (2010), Econometric Analysis of Cross Section and Panel
Data, Cambridge, MA: Massachusetts Institute of Technology.
Zhang, Jie and Els Breugelmans (2012), “The Impact of an Item-Based Loyalty
Program,” Journal of Marketing Research, 49 (1), 50–65.
Zhang, Jie and Michel Wedel (2009), “The Effectiveness of Customized Pro-
motions in Online and Offline Stores,” Journal of Marketing Research, 46
(2), 190–206.
Reproduced with permission of the copyright owner. Further reproduction prohibited without
permission.