Sage Publications, Inc. Journal of Marketing: This Content Downloaded From 2.36.90.50 On Thu, 21 Nov 2019 18:57:04 UTC

Adding Bricks to Clicks: Predicting the Patterns of Cross-Channel Elasticities Over Time
Author(s): Jill Avery, Thomas J. Steenburgh, John Deighton and Mary Caravella
Source: Journal of Marketing, Vol. 76, No. 3 (May 2012), pp. 96-111
Published by: Sage Publications, Inc.
Stable URL: https://www.jstor.org/stable/41714491
Accessed: 21-11-2019 18:57 UTC
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide
range of content in a trusted digital archive. We use information technology and tools to increase productivity and
facilitate new forms of scholarship. For more information about JSTOR, please contact support@jstor.org.
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at
https://about.jstor.org/terms
Sage Publications, Inc. is collaborating with JSTOR to digitize, preserve and extend access to
Journal of Marketing
This content downloaded from 2.36.90.50 on Thu, 21 Nov 2019 18:57:04 UTC
All use subject to https://about.jstor.org/terms
Jill Avery, Thomas J. Steenburgh, John Deighton, & Mary Caravella
Adding Bricks to Clicks: Predicting

the Patterns of Cross-Channel
Elasticities Over Time
The authors propose a conceptual framework to explain whether and when the introduction of a new retail store
channel helps or hurts sales in existing direct channels. A conceptual framework separates short- and long-term
effects by analyzing the capabilities of a channel that help consumers accomplish their shopping goals. To test the
theory, the authors analyze a unique data set from a high-end retailer using matching methods. The authors study
the introduction of a retail store and find evidence of cross-channel cannibalization and synergy. The presence of
a retail store decreases sales in the catalog but not the Internet channel in the short run but increases sales in both
direct channels over time. Following the opening of the store, more first-time customers begin purchasing in the
direct channels. These results suggest that adding a retail store to direct channels yields different results from
adding an Internet channel to a retail store channel, as previous research has indicated.
Keywords : multichannel retailing, channels of distribution, channel management, channel migration, direct
marketing, e-commerce, retail stores
the introduction of the Internet channel (Ansari, Mela, and

practice of multichannel retailing greatly expanded. Neslin 2008; Biyalogorsky and Naik 2003; Deleersnyder et
As practice RetailersRetailers such as Wal-Mart
new technologies of suchopened e-commerce
multichannel as opened Wal-Mart retailing new opened paths greatly to e-commerce market, expanded. the al. 2002; Geyskens, Gielens, and Dekimpe 2002; Ward
websites to supplement their brick-and-mortar stores, and 2001), by testing our theory on a unique data set collected
retailers such as Dell began moving into the shopping mall. from a retailer of high-end apparel, accessories, and home
At present, there is an increasing sentiment among retailers furnishings, which added four new retail stores in areas pre-
that a multichannel presence creates synergy, with stores viously served by only catalog and Internet channels.
acting as billboards for the brand, catalogs providing entic- Our results show that adding a retail store has a differ-
ing reminders to buy, and the Internet providing an ever- ent impact across channels, across time, and between first-
present storefront. Despite the explosion of multichannel time and repeat customers. We show that opening a retail
retailing in practice, the academic literature has yet to store cannibalizes demand in the catalog channel right
develop a broad theory of how channels work together, and away. However, over time, both the direct channels benefit
empirical evidence of cross-channel synergy has yet to be from store presence, with the Internet channel experiencing
documented. a greater boost, illuminating cross-channel synergy left
In this article, we propose a conceptual framework to undiscovered by previous empirical studies. In addition,
predict the patterns of cross-channel elasticities that occur although the store initially cannibalizes repeat customers in
over time following the introduction of a new channel. The the direct channels, over time it brings in new customers at
framework separates short- and long-term effects by ana- a faster rate.
lyzing the channels' underlying capabilities. We predict that Our results show when and how multichannel retailing
the order of entry should matter when new channels are can produce synergistic effects across channels. Although
added to the system, such that adding a retail store to an the results are based on data collected from a single retailer,
Internet channel should produce different effects than our framework suggests other contexts in which they would
adding an Internet channel to a retail store. In addition, we apply and explains why they differ from previous work that
extend the current published literature, which focuses on has focused on the introduction of an Internet channel.
Jill Avery is Assistant Professor of Marketing, Simmons School of Manage-

Conceptual Framework
ment (e-mail: jill.avery@simmons.edu). Thomas J. Steenburgh is Associate Before proposing a set of testable hypotheses for our par-
Professor (e-mail: tsteenburgh@hbs.edu), and John Deighton is Harold ticular research setting, we begin by developing a concep-
M. Brierley Professor of Business Administration (e-mail: jdeighton@ tual framework that helps explain the effects of channel
hbs.edu), Harvard Business School, Harvard University. Mary Caravella is expansion. To predict the patterns of cross-channel elas-
Assistant Professor in Residence, School of Business, University of Con-
ticities that will occur over time, our analysis abstracts
necticut (e-mail: mary.caravella@business.uconn.edu). Ajay Kohli served
as area editor for this article. away from specific channels by focusing on their capabili-
ties. We define a "capability" as an enabling characteristic
© 201 2, American Marketing Association Journal of Marketing

ISSN: 0022-2429 (print), 1547-7185 (electronic) 96 Volume 76 (May 2012), 96-111
of a channel that allows consumers to accomplish their ties should be observed in the long run. For example, it is
shopping goals. The framework does not depend on being fairly obvious to consumers that a retail store provides the
able to develop an exhaustive list of channel capabilities, opportunity to talk to salespeople, touch merchandise, and
which would seem to be an impossible task because there is get advice about what to buy. If these capabilities matter to
no end to the needs of consumers. Rather, it depends on consumers, we should observe their effects in the short run.
being able to classify each capability of the new channel on Conversely, it takes repeated exposure to the retail store's
two dimensions: "living billboard" to build awareness and positive brand
associations. If this capability matters, we should expect to
1 . Does a given capability of the new channel substitute for or
observe its effect over time. The matrix in Table 1 can be
complement the capabilities of the preexisting channels?
used to predict the net effects of introducing a new channel.
2. Is a given capability quickly apparent to the consumer, or
We should observe short-term effects if the new channel
must it be learned through experience?
has conspicuous capabilities and long-term effects if the
Weighing the answers to the first question helps determine new channel has experiential capabilities. We can predict
whether the new channel will cannibalize demand in the whether the net short-term effect is positive or negative by
preexisting channels or generate incremental demand for judging whether the new channel's conspicuous capabilities
them.1 predominantly complement or substitute for those of the
Cannibalization of customers and sales will result if a preexisting channels.3 (The net effect would be zero if com-
new channel too closely duplicates existing capabilities plementary and substitutive capabilities have equal weight.)
(Deleersnyder et al. 2002; Moriarty and Moran 1990) or Similarly, we can predict the net long-term effects by
offers superior capabilities (Alba et al. 1997). Incremental assessing the experiential capabilities of the new channel.
demand in the existing channels will be generated if a new
channel provides complementary capabilities that attract
new customers to the existing channels or cause existing
Research Hypotheses
customers to purchase more.2 Complementary capabilities We use our conceptual framework to develop a set of
offset weaknesses in the existing channels and/or offertestable hypotheses for our particular research setting.
compatible functionality that encourages cross-channelUnlike prior published research, which involves studying
buying. Because each channel has different capabilities the effect of adding an Internet channel (Ansari, Mela, and
(Baker et al. 2002; Verhoef, Neslin, and Vroomen 2007), Neslin 2008; Biyalogorsky and Naik 2003; Deleersnyder et
the pattern of cross-channel elasticities that occur should al. 2002; Geyskens, Gielens, and Dekimpe 2002; Ward
2001), our study focuses on studying the effects of intro-
depend on the type of channel being added and the existing
ducing brick-and-mortar stores to preexisting direct chan-
channels already in the retailer's channel mix. This suggests
nels. Studying this type of channel addition seems particu-
that the order of entry matters and that different patterns
should occur depending on whether firms add bricks tolarly important now because many retailers born on the
clicks or vice versa. Internet or in direct mail catalogs, such as Athleta, J. Jill,
and Garnet Hill, are establishing a retail store presence to
The answer to the second question determines the time
complement their direct businesses. There is growing senti-
frame in which we should expect to observe a given effect.
ment that having a brick-and-mortar presence can provide a
Existing empirical studies do not separate the short-term
competitive advantage to the direct channels, as suggested
effects from the long-term effects of opening a channel,
by Raul Vasquez, Walmart.com chief executive (Bustillo
thereby leaving the dynamic nature of cross-channel effects
and Fowler 2009, p. Bl): "There was a time when the
unknown. Some channel capabilities are conspicuous: They
online and offline businesses were viewed as being differ-
are quickly apparent to consumers and should change their
ent. Now we are realizing that we actually have a physical
behavior in the short run, thereby affecting sales in the short
advantage thanks to our thousands of stores, and we can use
run as well. Alternatively, experiential capabilities are
learned through experience or accrued over time before
they affect shopping behavior; the impact of these capabili- 3In hypoth
examined ac
negative sy
1 Cannibalization
tomer is
prefa
either partial
mostor prude
full,
2Consumers will
weight pref
capa
already shopping,
outline asopr
inferior capabilities.
tomers and
TABLE 1
Impact of Adding a New Channel
Substitutes for Complements
Conspicuous capab
Experiential capab
Patterns
it to become No. 1 online." Little empirical evidence exists the opening of a retail store, while Internet sales are not. In
to support this idea in the literature. Our framework helps their setting, catalog orders decreased by 14%, and existing
predict when a physical store provides an advantage to the customers made purchases less frequently following the
direct channels and when it does not. store opening.
We provide a general assessment of each channel's
Hj: Following the opening of a store, sales in the catalog and
capabilities for a set of consumer shopping goals (Dorsey Internet channels from customers living within the retail
2006) relevant to our setting in Table 2. The table shows trading area surrounding the store decrease in the short
that the channels differ in their capabilities in our context. run.
This list of shopping goals and channel capabilities will not H2: Following the opening of a store, sales in the catalog
apply in all contexts, but the process of assessing capability channel from customers living within the retail trading
area surrounding the store decrease more than sales in the
differences across channels and matching them to shopping
Internet channel in the short run.
goals can be used in other settings. Enumerating the con-
spicuous and experiential capabilities of a store channel
helps us develop hypotheses for how a store opening should The Long-Term Effects of Experiential
affect the two direct channels over time. We discuss the Capabilities
store's capabilities relative to the Internet channel and to theThere are several experiential capabilities that suggest the
catalog channel next. retail store will complement the direct channels in the long
run. Managers increasingly view stores as living advertise-
The Short-Term Effects of Conspicuous
ments that generate reach and frequency for the brand mes-
Capabilities
sage. Consider the following (Chang 2009):
There are several conspicuous capabilities that favor a typi-
Stores act as the brand's billboard.... Best Buy Mobile
cal retail store over the direct channels. A store allows cus-
stores, located in high-end shopping malls, are attracting
tomers to touch and feel merchandise before purchase, customers who are new to its brand. (Scott Moore, Vice
eliminating some of the risk of purchasing through direct President of Marketing, Best Buy)
channels. A store also gives consumers a chance to talk to
In addition, valuable brand associations attributed to the
salespeople face-to-face when deciding whether to buy and
store channel may transfer to the retailer's other channels
provides an immediate sense of gratification when they do,
(Jacoby and Mazursky 1984; Keller 1993), and positive
two capabilities the direct channels do not offer. By switch-
associations formed through the knowledge and/or patron-
ing their purchases to the retail store, shoppers can avoid
age of the store can transfer to the other channels as a halo
paying shipping and handling charges associated with buy-
effect (Kwon and Lennon 2009). Effects stemming from
ing in the direct channels, thereby lowering their transaction
increases in brand awareness and the creation of positive
costs. These capabilities are fairly obvious to consumers,
brand associations are likely to transfer to the other chan-
given that retail stores are well known to consumers and
nels only after repeated exposure to the living billboard of
most will already be familiar with this type of shopping
the store, as branding effects generally accrue only after a
experience. Because neither of the direct channels provides
minimum threshold level of exposure is achieved and then
these opportunities, they should be substitutive capabilities show an additive effect over time (Simon and Arndt 1980).
that drive short-term cannibalization of both the catalog and
Because repeated exposure to the retail store should
Internet channels.
strengthen brand awareness and deepen brand associations,
Two other substitutive capabilities affect only the cata- we expect the resulting effects of this complementary capa-
log channel. In our setting, the retail store provides a much bility to increase sales in the direct channels over time.
greater assortment of merchandise than the catalog channel Although the branding capabilities of the retail store
does. Because greater assortment is a substitutive capabil- should complement both the Internet and catalog channels,
ity, this suggests that the opening a store channel should the Internet should benefit more than the catalog channel
hurt the catalog more than the Internet. The retail store also does because it is a less intrusive medium. The Internet is
offers more information-search capabilities than the catalog an inbound marketing communications channel (Halligan
channel but fewer information-search capabilities than the and Shah 2009) because consumers, particularly first-time
Internet channel (Balasubramanian, Raghunathan, and customers, must actively search out websites before making
Mahajan 2005). a purchase. In contrast, a catalog intrudes into a consumer's
In total, the store's conspicuous capabilities are largely consciousness and is an outbound marketing communica-
substitutes for those of the preexisting channels. This sug- tions channel. Catalogs serve as highly effective brand
gests that the introduction of a store channel should canni- communications (Frazier 1999), making it more likely that
balize sales from both direct channels in the short run and they will elicit brand awareness and strong brand associa-
that cannibalization effects should be felt more strongly in tions for the retailer, as well as serve as a call to action.
the catalog channel than in the Internet channel, due to the Search advertisements or e-mail marketing used to drive
store's assortment advantage and greater information- demand to the website are less effective brand communica-
search capabilities over the catalog channel. An asymmetri- tions because many consumers ignore them and because
cal response between the two direct channels is consistent they do not offer the strong visual presentation of the cata-
with the finding reported in Pau weis and Neslin (2009), log, making it less likely that they will elicit the same level
who empirically show that catalog sales are cannibalized by of brand awareness and strong brand associations for the
98 1 Journal of Marketing, May 2012
(0 i ryj i JL
n C i ryj c C - O) m E 9 C/) „ ^
1 <c « § C 8,2 c 8 o - g .e g
O
(0 E ~ e 2
2 o Ïe
£ ©o
=9 .cÏco « §"S ©
a) .9? .c §0 ë~
ó "co £
- o E to
c o-£ ce 0 ô
.® (0 ^ S« •££> Öc®Ss§3_°g "co - o c ® "o óo '§.
•tí ^ c 8$£ ^ Š 0 +- ^ -o £ Z3 (/) C0 0 ^ S Ê £5° Q._c 2 2
5 •tí g si c I" 8$£ Hi ^ Š 0 5 +- Isl ^ -o S| £ ?8 Z3 (/) g
I " 1 § p^° ® "i ü = = 1 ^ 2 ^ E 5ew offl.cS38 E £ « 11 =i ã
Õ « ¡| ®® Ic-o = ® 22™ ®2 1 c^ ê| 2|«"§O5|||° ®Í8-8§
a>a>©m^Ssr25«
£ © o. c - ®®
© co•*•§§
<= §5=-°- #rr
= §5Eo
g#3 o-°W0=6 §3 ce
« ^ co Eo©ê^|»íg,Sc: 3 W
° -n E c5 05 « ^ *3Zgl
•* 8
S 5 m CO 9? C» "g < 2?« O) íí£ © co r Sc CÖ 2>®> E Ol E 05 ? m "Č0 ^ ~ 05 © 2 2 -n C C £ -C m
* 1 s ř.; ?i šlí & O) ïs» r s. a li® CÖ 2>®> ss i ¡ 2 g-s-g m US ~ = ! 8 m
1 * 1 Is! P ?i -1" 8 "Iœ-_S i Ss «i f! ä§ll5".|«® 2 Illi|=?Si =
2 ee S^cù-C?
®
E 0c čš
|si s-sO 188
- "O "O čš ^- 00
§„r sCM
a ^cTB
ScSsi."
+* 0s B0 ft
?S2®Q_Í=
B 3ČO^P§®«
®Í8
<M Q_Í= OÊ-Q
ÍÍ8=S.<«®|§
2?S2®
S "£-g®
m i "£
8.i 8.
(U S®5»rS2
§Ž5f38®.8
í®-cImW£o
O C ^ÔScO^c?} ^
C ^ 0 "O r- "O Cr1 + CO c Ü2 „r o 0 c: -7-1 ■+- O -Çz ~3 ZZ- F P 03 m P TJCOütO ^ "D w fll m ^ c
4 * i fisji i|ï h tiUllilSs -7-1 USUI IISÜI l!#|lfi| fl

*Imm
sfilicm
cm ££OûOûc/)c/)
£ MHz
£ SS 00coco" Z"coZIpilptl! Z.EcOcŽi-íSoŠlíJcocl llillfilplf
co Z.EcOcŽi-íSoŠlíJcocl OidE^jdo§8.S5 0 0q.(Tq.(T
OidE^jdo§8.S5 i jO^jãÔto |!i ill!
jO^jãÔto
+ +II II 1+ + +II 11 +
_L
c _L ^ ^ o© co
co C i
. - CD »L
(/3 . ©
„ 0 co c/d
_ c/d 0 "Tg "Tg o
^ o 2o 03 ,203 c
c
3.tí
.tí 3
I si-- ^5
^5 Its
^ -- É-li CO^r^ C
CO^r^ I CXI
ili XI
u .^„ ^1| C
u ^G)
Il ^ .,
'*§§CÖ
C G)2 lit
"mCÖ i5
"m ÃT
ÃT «¿I û)^-" ||_
û)^-" WÈmWÈm
.tí .- -■ ^ _q 3 - cd cd ^ - 0 c-û c o C XI o _. c u c-° 03 §"îa o C ., CÖ 2 "m S
.-
<0
0)
conQ.
co ^-■
rn ^ ^_q
^ Um1-03rn
œ ti 030 S(U
Sil
CD-
œ ¡2
0 Sil
Ju -=p^r-O Ju(U 0¡5E -=p^r-
CD Lf)
c-û ^ c CD
03 SZ ^
q. ^ ^ ^ m q.
o ¥ 0(003-
"S ^0(003-
_.
ü fö !-¥ O
c 03
^ 0)c
sz c~ ™
0"o -Q-jCO
!- "o
WD
C-Q-jCO
™m ^ c: S x:m
03 "S ^ 0
0
¡H
JQ
« Q. I filli Um1- £°?sS! ^ IS|I||8 m "S ^ fö O pi! 0)c s:j WD "S m
? S
ř §£mi-S,
lisi? ig?iSp ?5fs£S| isn.®-oc
eSoo iB|
.®-oc §is||5
iB| lis 1
(0
a
CU 2
C0 p 0
£®"-«3&
"O c
eSoo
^ cE (-2
- |o©
-Cû r-" ^ .y ^ ^ "O O co -c 0"cc: ^"5(0 -
O
k.
ü tr c "i S ? o ^ ^ > 03 _; 3 c Ç0 . ^ g 03 3«^ Q-OCOÔqC c COÔO _ Ç D CO
CM '3
SÉ
cû £
1üi lini
© cse«
"i ? oo,!
ïii3iï£íg¿c°.E"r
^ > _; ups 3 Ç0 . Sc
^ îi g3".
fill 3".
SÏ|li ci-g.
Q-OCOÔqC lilí i-urti Ç
«-S®^s<gl
gHHa>-
3= 5 i 03Zpst|fep^|?r
03 .Í5ñ: 5-c
.Í5ñ: 5-c s 0s 0
ZicofecOo^§^.i5irZcö °2°2 O
^llfll^ O o Z.çm2h3£
os 03¿r
°2 03¿r
C řřlis!
0 0o EE0 |2if§MgPi!lSt
0 ^C
03¿r g-co^ ôg-co^
ř Û.OOJCÛ ^ ?§o ^^^^cosEos
?§ o^^^cosEos
0 0 §£mÔ£Ec3_
¡5 I I I I I I I + lili
i2
a>
a^. o 5 •§ -s S g â i ®s§
c
■S SS i o ? ; E g <,^ £©£ «fe_íg

c
(0
-C
O c- S fďiř S> ®Ji s.® i #5 g § S-ď s-®«55
iš2-a
2c-g g 5 g o. g.a« 1^1 s -fc®. s "§^§<1
2 §>o> 5¡§g |<| ©-g s O ¡Ctg |gg g««g|
® 2® » ¿i íE<N pITÖ) ©3 © «8 ac^ì Eg;E Î5®®®0)
"S
Q. -a
m®T3
T3 c» rn
c " C0
c „ ^
h?W©«¿-s0-
^ „ h?£=
£=©3
O .<3 © «8 £Qc0-Cô1-==-Coa-
t; "CC S Ž§Ígw-oo2iH-®«S§;« £Qc0-Cô1-=
Q. C0 m _. T3 rn C0 ^ W O ^ =3 O) h? íc,Oog £= O t; "CC 0 0
" ! _. Si loi? ^«1 ® Il sicŠM^c i®"--2
I 0 S O r Sa- é-ál^-a S ^ 2 Si S tl-i.8SïS.ii&t2S
0 ^ O -Q r O'S S c w S c c -n co w ö)c 03 w 5 et; c
w § ©s E«-s'Šfe§8« 2 8s¿j£1^ >Í§ S-S^ S ® «¿"-s ® %°J= 2_
1 I Sc §^SëÎSc5 0-D3§^"®0 SS5?ifl5 S^SlaS
I w
Œ Œ2"C
I ac^
Sc o)«2 çB|
f0"5 ac^ ^-||<
® oÍS^c^^cd I2S88¡8
3® 'O13 «®
2 w 10 10 ©"iôÍÑ-E2ÍS_=E
© c to c 'O
-2 w îS WîSiSs o) f0"5 ® £8ffl.Írí8.E oÍS^c^^cd S? E?c 3® S ¡t5Sc w E ©"iô = © E c
Ô w
'p §03-C^w
^ _-C0 w^-^^^0
hJ JÇ CL
W ~ o Xm0 00 -O
^ JÇ - CO
o 0«en
2-> 2 0
*5co£'wC^^^cq^w
¿ O .En" 50c S o o ~ m -O Ï o ~.-o3oo2<co
0Û _J hb Z Q. O S C^CC _l CL < Q.T- Ï co CO £ CC 0 C Q_0 CLO. C O .E O 0 1- ^
+ I + + + ++ + + + + +
- w
5
® ®
*- .2
"o w
, =
0) i2
CO ^
i2 Q-
^ 0
i=5"0 „V 0 , "C Smo 0 0 (fi 'co
O) 13 o C „V 0 OO E - ~ 0 o Ecqco
E O) B ¿ (0 C _ © J3 59-C J3>-^ - m£ .2^® Ecqco
■5.
« o«O
9-(15
o Om>-
0 _ m r0 Ifl
~ m -Q O
Í2DÕ)i2
■ - Ifl oR-0 O D -5
r~ O^^^-C/D ■ »-
- r~
W ®-9n®
u ii' O^^^-C/D 2
g-O"^ >c 03 O W ~ C0 c 03 ¡5 0 - Q.
£ww2§«°cü
WD0rm^ ~C 0 ü? £ 0 «S®F
O) -g 0 M ü? ^mO
^ O H © i L 0 ° ~C 00 0 o co co P c O) co N i_ 5 °- 0 rr> r
£ 0 o
Ê
iisifs
■- O H0©tr
i L g?
oEco
0 0Eco
° -ÜB 00
EcoEco Sï^-ëg
oj 8o co
ö)P §
c l'E®¿-
oï'ôj§ cooN o.
i_ 5-Q
«.ç rr>
£^®^
¡£r
E w
W D Oî
■- w Q. 0r >
« ££ tr COV^CDhz
« O o o"0)0m0ï a
c CD
g Cg
S čg oj ScO.EO-g 8
8 -5 ss ££ 0"s£2£2
§00(2(2|2(2 Sr Em
E2(2|2
Em COV^CDhz
qj(2
g-x«^
(2
0) C/)Î=C0
m
(2
a «« Cg o5E
Patterns of Cr
retailer as a catalog, particularly for first-time customers. channel capabilities varies according to consumers' levels
Ansari, Mela, and Neslin (2008) find a positive association of familiarity and perceived expertise in choosing (Balasub-
between sending catalogs and subsequent catalog shopping ramanian, Raghunathan, and Mahajan 2005); therefore, we
but a diminishing marginal return for sending e-mails to drive expect different weightings across first-time and repeat cus-
Internet purchase. Therefore, the Internet channel appears tomers. We summarize the consumer shopping goals and
to require greater outbound marketing programs such as the channel capability requirements of first-time customers
living billboard that the store offers, whereas the catalog relative to repeat customers in Table 3 and discuss them
channel is largely self-sufficient at driving its own demand. next to inform our hypothesis development.
In addition, the store's physical presence reduces uncer- It is likely that the retail store will serve as an acquisi-
tainty about buying through the direct channels by provid- tion engine for the direct channels. Given the retail store's
ing the consumer with a physical place to go if trouble superiority on capabilities that are important to prospective
ensues. This capability is likely to be more valuable for cus- customers, such as sales assistance and the opportunity to
tomers shopping in the Internet channel, which was a fairly touch and feel the merchandise, many consumers who are
new channel during the time period of our data collection, not currently purchasing in the direct channels are likely to
as previous research has shown that consumers are nervous try shopping in the store. First-time customers are more
about purchasing online and that the presence of a physical likely than repeat customers to require a multisensory,
store in their local market relieves their concern (Tang and experiential shopping experience, such as that offered by a
Xing 2001). store, to help them assess their options (Balasubramanian
Over time, the addition of a store could increase loyalty 1998). Given the prevalence of multichannel shopping
between the retailer and all its consumers, including those behavior (Reda 2002), some of these new store shoppers are
purchasing in direct channels. Because loyalty is created both likely to begin purchasing across channels over the longer
with the retailer overall and with a particular channel (Ansari, run, bringing them to the website and catalog for some por-
Mela, and Neslin 2008; Reynolds and Beatty 1999), it would tion of their purchases. This is consistent with Coldwater
both complement and substitute for the direct channels' Creek's strategy, a company selling women's apparel
capabilities. Loyalty established with the retailer would be through direct channels that first opened retail stores in
complementary because consumers would become more 1999 to capitalize on complementary effects:
likely to shop with the retailer across all channels. Loyalty
Believing that the ability to occasionally "touch and feel"
to a particular channel would attract new customers to the
merchandise will remain a coveted aspect of the American
retailer who favor shopping in stores but make them less woman's shopping experience and to provide another
likely to patronize direct channels. Finally, the store channel means by which to introduce current and prospective cus-
offers consumers a shared social experience while they tomers to our catalogs and e-commerce web site, we have
shop, increasing the value they derive from their purchase. also embarked on a program of selectively establishing
Because shoppers in the direct channel shop alone, the social for the first time full-line retail stores in highly-trafficked
urban areas. (Coldwater Creek Inc.)
interaction that the store offers is a substitutive capability.
In total, the store's experiential capabilities mostly com- Given that this is a two-step process, we would expect
plement those of the direct channels. This suggests that the increases in the number of first-time customers shopping in
introduction of a store should increase sales in both direct
the direct channels to occur in the long run rather than
channels over time. Given that consumers shopping in theimmediately.
Internet channel require more outbound marketing and Beyond this direct approach, the retail store should also
more reassurance about their purchases, the complementary serve an important branding function for first-time cus-
effects should be felt more strongly in the Internet channel tomers even if they do not purchase in the store before com-
versus the catalog channel. Thus: ing to the direct channels. First-time customers lack brand
H3: Following the opening of a store, sales in the catalog and awareness and need channels to generate outbound marketing
Internet channels from customers living within the retailimpressions, while repeat customers are more aware of the
trading area surrounding the store increase in the long run.brand and need less marketing to spur purchase. The store's
H4: Following the opening of a store, sales in the Internet experiential capabilities such as its living billboard raise
channel from customers living within the retail tradingbrand awareness for the retailer and contribute strong brand
area surrounding the store increase more than sales in the
associations that are required before customers will make
catalog channel over the long run.
the choice to buy from the retailer online or in the catalog.
First-Time Versus Repeat Customer Effects Therefore, the number of first-time customers purchas-
ing in the direct channels should increase over time due to
Finally, our theory allows us to make a distinction in the
the greater outbound marketing presence of the store and its
buying behavior of first-time and repeat customers. Prior
brand awareness and association building capability. Thus:
literature has suggested that consumers use different chan-
nels during their life cycle (Neslin and Shankar 2009). H5: Following the opening of a store, the number of first-time
When assessing which type of customer (first time vs. customers purchasing (as measured by a count of new
customer accounts making at least one purchase per
repeat) will drive cannibalizing and complementary effects,
month) in the direct channels increases in the long run.
it is imperative to understand whether and how the two seg-
ments differentially value the conspicuous and experientialIt is unlikely that opening a retail store will slow down new
capabilities of the new channel. The utility of different customer acquisition in the direct channels in the short run,
100 / Journal of Marketing, May 2012
TABLE 3
Customers Differ in How They Value Capabilities
Consumer Shopping Goal
Conspicuous Capabilities
To have access to broad N
assortment associated with novice choice (Iyengar and one progresses from novice to expert (Iyengar
Lepper 2000) and Lepper 2000)
To shop whenever and May be willing to sacrifice convenience t
wherever wanted personalized attention from a salesperson
To minimize tangible May be willing to pay more to feel good about May be more likely to shop around to look for
transaction costs the purchase because of higher perceived risk lowest price due to lower perceived risk
due to lack of knowledge
To minimize intangible May be more willing to invest time in shopping to May be searching for more time-efficient way
transaction costs reduce risk stemming from lack of familiarity and to purchase given familiarity and expertise
expertise (Alba and Hutchinson 1987) (Alba and Hutchinson 1987)
May be searching for more information-search
material to inform consideration set formation
(Balasubramanian, Raghunathan, and Mahajan
2005)
To have access to face- Necessary for novices who lack familiarity and Less needed by experts who have familiarity
to-face sales support expertise (Alba and Hutchinson 1987) and expertise (Alba and Hutchinson 1987)
during transaction
To be confident in Need to touch and feel merchandise to learn More comfortable purchasing without physical
purchasing product differences due to lack of familiarity and interaction with merchandise due to familiarity
expertise (Alba and Hutchinson 1987; Balasub- and expertise (Alba and Hutchinson 1987;
ramanian, Raghunathan, and Mahajan 2005) Balasubramanian, Raghunathan, and Mahajan
Need to experience physical presence of the 2005)
retailer to reduce uncertainty about purchasing Retailer has already proven itself; no need for
online or by telephone physical presence
Experiential Capabilities
To recognize and/or recall Low levels of brand awareness makes living High levels of
a retailer for a particular billboard of store highly effective previous purchasing
type of purchase Living billboard of store serves as reminder to
purchase
To enjoy a pleasurable May require rich, multisensory brand experience Strong, positive brand associ
shopping experience to create strong, positive brand associations exist from previous purchase ex
because none exist Need to affirm subjectively perceived expertise
by demonstrating choice skills as expert may
require public display of choosing
(Balasubramanian, Raghunathan, and
Mahajan 2005)
To establish a relationship Salespeople reduce risk of purchase and help Less sales assistance needed
with the retailer that close the sale Previous experience with the retailer provides
makes shopping easier other customers validate the new customer's validation for the choice
choice (Balasubramanian, Raghunathan, and
Mahajan 2005)
given that customers who prefer shopping in the Internet and handling costs and they do not need to wait for their
and catalog channels will be largely unaffected by the purchases to arrive by mail. Other existing direct channel
store's opening until its additive effects have time to mani- customers who previously searched and bought online may
fest. Thus, we predict a null effect, that the store opening instead search online and then shop in the store when they
will not affect the number of first-time customers purchas- want to touch and feel the product or discuss the purchase
ing in the direct channels in the short run and, therefore, do
with a salesperson (Verhoef, Neslin, and Vroomen 2007). This
not present a formal hypothesis for this case.
will lead to cannibalization and reduce the number of repeat
We predict that repeat customers will follow the pat-
customers shopping in the direct channels in the short run.
terns outlined in Hj and H3. Some existing direct channel
customers may quickly switch all or some of their shopping H6: Following the opening of a store, the number of repeat
to the store because of the store's conspicuous capabilities, customers purchasing (as measured by a count of repeat
such as its broader assortment (vs. the catalog) and reduced customer accounts making at least one purchase per
tangible and intangible costs because there are no shipping month) in the direct channels decreases in the short run.
Patterns of Cross-Channel Elasticities 1 101
Nevertheless, the experiential capabilities offered by the regarding a reasonable drive time from which a shopping
store opening are predominantly complementary for repeat mall would draw. This resulted in assigning a total of 550
customers, which should increase the number of repeat cus- zip codes to the four retail trading areas shown in Table 4;
tomers buying in the direct channels over time. The con- these geographical areas represent our treatment regions.
stant billboard effect of the store's presence should remind We were interested in understanding how the store
existing customers of the retailer so that when they are at openings affected both sales and the number of customers
home and ready to buy, the retailer's brand should be more purchasing in the direct channels in the surrounding zip
salient in their minds. The retailer's brand associations
codes over time. To answer these questions, we were able to
should be strengthened for direct channel customers who collect data on the net catalog sales, the net Internet sales,
experience shopping in the store, which should make pur- the number (i.e., count) of first-time customers making a
chasing online or by catalog from the retailer more likely.purchase in the direct channels, and the number of repeat
The opportunity to shop across channels should increase thecustomers making a purchase in the direct channels.5 Our
loyalty of direct channel customers to the retailer as a whole
monthly observations begin 36 months before each of the
and make purchasing online or by catalog from the retailer four store openings (34 months in Store A due to data con-
more likely. Thus: straints) and end in December 2006.
H7: Following the opening of a store, the number of repeat
Quasi-Experimental
customers purchasing (as measured by the count of repeat Design with Matching
customer accounts making at least one purchase per
We use a quasi-experimental design to draw inferences
month) in the direct channels increases over the long run.
about how the retail store openings affected direct channel
sales. We compared observations in the four treatment
Data and Research Methodology regions with observations in control regions before and
after stores opened. Like any research design, this approach
Our study is based on data collected from a multichannel
has its advantages and weaknesses. The advantage is that it
retailer of high-end apparel, accessories, and home furnish-
can control for outside events that coincide with the occur-
ings. This retailer operates stores in shopping malls in lim-
rence of an intervention and interfere with its effects. For
ited regions of the country and sells directly to consumers
through catalogs and the Internet. Overall, sales from the example, a general economic recession occurring at the
retail stores have been significantly higher than sales from of a store opening would attenuate the positive brand-
time
the direct channels, but growth in the direct channels has ing effects that a store opening might have on sales. This
been dramatic over the past decade. The retail stores and the of outside event does not pose a problem in a quasi-
type
direct channels carry the same merchandise and use the experimental design because both the treatment and control
same price points for regular ticket pricing. Nevertheless,groups would suffer and its confounding influence would
the day-to-day operations of these two units are largely controlled for. The use of a treatment/control design
be
enables us to control for seasonal variations and annual
independent, with each unit running separate merchandising
and pricing promotions and making separate advertisingfluctuations in sales driven by external factors, such as the
growth of Internet penetration.
and communications decisions. It is important to note that,
during the time of our study, the catalog and Internet chan- The challenge in implementing a quasi-experimental
design lies in identifying a control group that is comparable
nels did not locally customize their marketing policies in
regions in which stores were opened but rather pursuedto a the treatment group. In our setting, it is reasonable to
national marketing plan that was consistent across believe all that the retailer is more likely to open stores in trad-
regions. This is important because it provides a clean test ing
of areas with favorable geodemographic and shopping
the impact of opening a retail store, in that our retailer did
not adjust its marketing strategies in the direct channels to 5The company defines these data as sales net of returns respec-
anticipate or reflect changes in sales because of the launch
tively generated by catalog mailings and online purchases.
of the stores.
The retailer opened four retail stores in one U.S. state TABLE 4
during our observation period. Two of the stores were Quasi-Experimental Treatment Regions
opened in retail trading areas previously served by only the Retail Number of
direct channels, and two were opened in areas neighboring Trading Year Neighboring Zip Monthly
Area Opened Store Codes3 Observations
a preexisting retail store. (The preexisting stores were open
for more than five years before our observation period.) We Store A Fall 2000 Yes 61 108
assigned zip codes to retail trading areas according to theStore B Fall 2001 No 209 100
resident consumers' driving time to the nearest store. We Store C Fall 2002 Yes 97 87
used a maximum time of 60 minutes to find the boundary Storeof D Fall 2002 No 183 87
each retail trading area.4 This cutoff was based on discus- aThese numbers reflect the elimination o
sions with retailing experts and shopping mall managers of data.
Notes: Because the Internet channel op
were limited to 1 2 months of data in
4For Stores A and C, where there were already existing stores data in Store B, 36 months of data in Store C, and 35
within a 60-minute drive, we assigned to each store region only months of data in Store D in the period before the store
those zip codes for which the new store was the closest store. openings.
behavior characteristics. Thus, it is conceivable that the effect on net web sales, Model 3 estimates the effect on the
population of consumers that resides near stores is funda- number of first-time customers purchasing in the direct
mentally different from the population of consumers that channels (as measured by a count of first-time customer
does not. In an ideal test, we would address this concern by accounts making at least one purchase in a month), and
randomly assigning store openings to different retail trading Model 4 estimates the effect on the number of repeat cus-
areas to create balance between the two groups. This solu- tomers purchasing in the direct channels (as measured by a
tion is not feasible in the real world, so we used matching to count of repeat customer accounts making at least one pur-
make the treatment and control groups comparable. chase in a month). The variable treatment - controlt mea-
The basic idea of matching is to transform observational sures the difference between the treatment and control
data so that they more closely resemble data that would groups for the outcome variable of interest (e.g., the differ-
have resulted from random experimentation. These proce- ence in monthly catalog sales in Model 1) at time t.
dures work by making the control group as similar- and Our analysis focuses on the variables store .opent and
therefore as comparable- as possible to the treatment group post .open .monthst, which identify the nature of the
across covariates believed to affect the outcome variable of response to the store openings over time. The variable
interest. Although matching techniques are just beginningstore .opent is a step-function that represents the store open-
to be used in marketing (e.g., Von Wangenheim and Bayoning intervention. It takes the value 0 before the store open-
2007), they have already gained wide acceptance in other ing and 1 afterward. A negative coefficient indicates that
fields in the social sciences.6 The matching literature is opening a store has a short-term detrimental impact on the
broad and theoretically sophisticated, which can make it direct channels that persists over time. The variable post.
difficult for an applied researcher to develop a coherent open.monthst represents the number of months from the
strategy to implement these methods. We addressed this store opening to month t, taking the values 0 to 73 for Store
problem using the strategy that Ho et al. (2007) suggest, A, 0 to 63 for Store B, and 0 to 50 for Stores C and D. (It
which enabled us to test most of the commonly used proce-takes the value 0 before the store openings and in the month
dures and to choose the particular procedure that providedthat the store opens.) A positive coefficient indicates that
the best match for our data. (Details of our matching proce-opening a store is increasingly beneficial to the direct chan-
dure appear in the Appendix.) nel over time. Prior theory does not necessarily suggest that
In summary, to isolate the effects of the store opening this needs to be the case. It might also be possible for the
and rule out alternative explanations for sales changes duestore to increasingly cannibalize sales from the direct chan-
to confounding variables, we matched four treatment andnel, which would be suggested by a negative coefficient.
We included store dummies to account for cross-sectional
control regions across a broad set of geographic, demo-
differences across stores.
graphic, behavioral, sales, marketing activity, and competi-
tive variables. We were then ready to analyze how the To test for robustness, we tried several differently
physical store openings affected direct channel sales. shaped response curves after the store openings. For exam-
ple, we added a squared post. open .monthst term to the
Model Specification model to test whether the response was nonlinear over time.
We
Our interest was in understanding how opening a store found these terms to be nonsignificant, which allowed
would affect the direct channel sales and the number of cus- us to keep the simple linear specification. This had the
tomers making purchases in both the short and long run. added benefit of making our analysis relatively easy to
Thus, to operationalize the short-term sales, we included a interpret.
term in the model to capture an immediate shift in direct
channel sales in the month that the store opened, and to Findings
operationalize the long-term sales, we included a term to For each outcome variable of interest, we estimated a full
capture a trend in sales afterward. Thus, we specified the model that included data from all four stores. Then, to
model as follows:
investigate potential differences between stores that opened
in virgin retail trading areas and those that opened in areas
treatment - controlt = ß0 + ßi store .opent + ß2 post .open .monthst
neighboring a preexisting store, we ran separate models,
+ ß3 store. dummy t + £t.
one that combined Stores B and D (virgin retail trading
We estimated four sets of models to understand the effect of areas) and one that combined Stores A and C (retail trading
the store openings on the direct channels7: Model 1 esti- areas neighboring a preexisting store).
mates the effect on net catalog sales, Model 2 estimates the
The Effect of the Store Openings on the Catalog
Sales Channel
6Examples include Winship and Morgan (1999) in sociology;
Table sci-
Lee and Wahal (2004) in finance; Ho et al. (2007) in political 5, Model 1, reports the results for catalog channe
ence; Jaffe, Trajtenberg, and Henderson (1993), Meyer (1995),
sales. Beginning with the full model, the store .opent coeffi
and Heckman et al. (1998) in economics; and Hansen (2004)
cientinis negative and significant (ßj = -12,924 ,p < .001)
education.
indicating that catalog channel sales dropped in the short
7Reported results are based on ordinary least squares regression.
run after the brick-and-mortar stores opened, in support o
In a robustness check, we tested the errors for autocorrelation and
reestimated all models with time series models when appropriate. Hj. The post .open .monthst coefficient is positive and sig
Our results were robust to these additional tests. nificant (ß2 = 164,/7 < .05), indicating that the catalog chan
TABLE 5 due to the store openings. In the long run, however, catalog
Sales in the Direct Channels channel managers should be supportive of store openings
because catalog sales recover to the point that they would
A: Model 1 : Sales in the Catalog Channel
have been had the stores never opened, and the benefits of
Regions Regions having a brick-and-mortar store in a retail trading area con-
Without with tinue to increase over time.
Full Preexisting Preexisting Turning to the submodels, we observe a similar pattern
Model Stores Stores
of cannibalization and complementarity. The store .opent
Intercept -8,278** -6,862* 14,382*** coefficient is negative and significant regardless of whether
(2,764) (3,403) (2,972) the store opened in virgin territory (ßj = -14,157 ,p < .01)
Store.open -12,924*** -14,157** -11,350*
or in a region neighboring a preexisting store (ßj = -1 1 ,350,
(3,416) (5,287) (4,430)
Post.open. months 164* 133 1 81 1 p < .05). The post .open .monthst coefficient is insignificant
(81) (134) (99) (ß2 = 133, p > .10) if the store opened in virgin territory but
Store A dummy 33,094*** - 8,929** is positive and directionally significant (ß2 = 181, p < .10) if
(3,185) (3,165) it opened in a region neighboring a preexisting store. We
Store C dummy 23,832*** - -
then tested whether there was a statistically significant dif-
(3,366)
Store D dummy 8,837*** 8,600* - ference between the store.opent (ßj) and post .open .monthst
(3,366) (3,627) (ß2) coefficients between the virgin and preexisting store
Adjusted R2 .27 .06 .07 regions for catalog sales. We found no statistical differences
between the respective coefficients (ßji t = -.407, p >.10;
B: Model 2: Sales in the Online Channel
ß2: t = -.288,/?> .10).
Intercept -20,235*** -5108t 17,926***
(4178) (2969) (5306) The Effect of the Store Openings on the Internet
Store.open 35 -4857 12,586 Channel
(4676) (4069) (7783)
Post.open. months 823*** 334*** 1165*** Table 5, Model 2 reports the results for the Internet channel
(105) (99) (162) sales. Beginning with the full model, the store .opent coeffi-
Store A dummy 94 - -59,596*** cient is insignificant (ßj =35 ,p> .10), indicating no short-
(4566) (5710)
term drop in sales following the store opening; thus, Hi was
Store C dummy 50,816*** - -
(4586) not supported in the Internet channel. The post.open.
Store D dummy 13,204** 8242** - monthst coefficient is positive and significant (ß2 = 823, p <
(4576) (2838) .001), in support of H3, indicating that the Internet channel
Adjusted R2 .39 .08 .47 sales continuously grew over time. Although there was no
tpc.10. immediate cannibalization, the Internet channel increas-
*p < .05. ingly benefited from the presence of the brick-and-mortar
**p< .01.
stores over time.
***p< .001.
Notes: Standard deviations appear in parentheses. Our results suggest that the store openings have a
greater positive impact on the Internet channel than on the
nel sales continuously grew over time after the initial catalog channel. Although the catalog channel suffered can-
decline, in support of H3. These results suggest that catalog nibalization immediately following the introduction of
channel sales were cannibalized shortly after the brick-and- brick-and-mortar stores, the Internet channel did not, in
mortar stores opened but also that the catalog channel support of H2. We found a significant difference between
increasingly benefited from the store's presence over time. the store .opent (ßj) coefficients of the catalog and Internet
To give economic meaning to these results, we calcu- channels (t = -2.317,/? < .05). This result is consistent with
lated the percentage change in sales relative to the monthly Pauwels and Neslin (2009), who also find that the introduc-
average in the pre-store opening period. Catalog channel tion of brick-and-mortar stores more adversely affects the
sales dropped by 1 1 .9% shortly after the brick-and-mortar catalog channel than the Internet channel. Furthermore,
stores opened. Nevertheless, these sales would recover to although both the catalog and the Internet channels increas-
the level that would have been expected had the store never ingly benefited from the presence of brick-and-mortar
opened in 79 months and would subsequently continue stores, we found a statistically significant difference
growing. The short-term decrease in sales we uncovered is between the post .open .monthst (ß2) coefficients of the cata-
consistent with the cannibalization effect that Pauwels and log and Internet channels (t = -4.969, /7 < .001). The com-
Neslin (2009) find. However, the complementary effect we plementary effect was approximately five times greater in
found has not been previously detected, leaving long-term the Internet channel than it was in the catalog channel, in
benefits to the catalog channel undiscovered. support of H4. These results are consistent with the idea that
From a managerial perspective, these results suggest the capabilities of the store channel, such as its ability to
that adding stores creates conflict with the catalog channel. generate greater brand awareness and associations, comple-
In the short run, managers in the catalog channel might be ment the Internet channel more than they complement the
entitled to some temporary relief in their revenue targets catalog channel.
The submodels paint the same picture. Regardless of TABLE 6
where the store opened, the store.opent coefficient is insig- Customer Counts in the Direct Channels
nificant (ßi = -4857,/? > .10 in virgin areas, and ßi = 12,586,
A: New Customer Households
p > .10 in areas neighboring preexisting stores), reflecting
the notion that web sales are resistant to cannibalization
Regions Regions
from retail stores. Given that the store .opent (ßj) coefficient Without with
was insignificant in both types of regions, we did not run a Full Preexisting Preexisting
Model Stores Stores
comparison test between the two coefficients in this case.
The post. open. monthst coefficient is positive and significant Intercept -3.85 6.86 63.7***
(ß2 = 334, p < .001 in virgin areas, and ß2 = 1 ,165,/? < .001 (6.39) (7.53) (7.0)
in areas neighboring preexisting stores). We tested whetherStore.open -10.2 -31.8** 11.8
there was a statistically significant difference between the (7.9) (11.7) (10.5)
Post.open. months 1.13*** 1.29*** .98***
post .open .monthst (ß2) coefficients between the virgin and
(-19) (.30) (.23)
preexisting store regions for Internet sales. This was indeed Store A dummy 27.3*** - -51.3***
the case (t = -4.377,/? < .001), indicating that the comple- (7.4) (7.5)
mentary effect was significantly greater when there were Store C dummy 78.1*** - -
preexisting stores in the area. This suggests that the store's (7.8)
billboard effect was additive and built on existing aware- Store D dummy 19.4* 19.1* -
ness in the market. The store's billboard effect on brand (7.8) (8.0)
Adjusted R2 .28 .09 .28
awareness appears to be stronger for customers living in a
region with a preexisting store who have already built up B: Repeat Customer Househ
some level of brand awareness and brand associations, due
Intercept -22. 5t 72 ~2
to the additive nature of branding effects. (11.9) (14.2) (12.5)
As Stern, El-Ansary, and Coughlan (1996, p. 340) note, Store.open -103*** -162*** -43.1*
"one of the key elements of channel management is decid- (15) (22) (18.6)
ing how many sales outlets should be established in a given Post.open. months 2.35*** 2.75*** 1.93***
geographic area." Although store placement decisions are (.35) (.56) (.42)
Store A dummy 174*** _ 120***
primarily made according to whether the area can generate
(14) (13.3)
incremental sales in the store to cover its costs, our resultsStore C dummy 51 .9*** - -
suggest that having a denser store population in an area can (14.5)
offer value for sales from the Internet channel as well. Hav- Store D dummy 55.8*** 54.8*** -
ing more than one store in an area may be beneficial for the (14.5) (15.1)
Internet channel due to the branding effects stores have on Adjusted R2 .39 .25 .40
website sales. Adding a store to a region that already had tp< .10.
*p < .05.
one accelerated the complementary effects observed.
**p < .01 .
***p< .001.
The Effect of the Store Openings on First-Time Notes: Standard deviations appear in parentheses.
Direct Channel Customers
We used the same procedures to analyze the customertime,

data.in support of H5. This suggests that the store acted as
a billboard for the direct channels by attracting new cus-
We ran separate models for first-time and repeat customers
tomers to the retailer at a faster rate than would have been
to determine whether the short-term drop and longer-term
expected had the store never opened.
growth in direct channel sales varied across the customer
Some differences emerged depending on whether
life cycle. We first explored the effect of the store openings
on first-time direct channel customers. First-time customers
another store was present in a neighboring retail trading
area. The
represent households that have not purchased from either of store .opent coefficient is significant and negative
in virgin areas (ßj = -31.8,/? < .01) but is insignificant in
the direct channels in the past. The company identified
households as new only in the first month that they areas make neighboring
a preexisting stores (ßj = 11.8,/? > .10).
direct channel purchase; subsequent purchases from The the
difference between these coefficients is statistically
household appear in the repeat customer data. Table significant
6, (t = -2.773,/? < .01). This suggests that a store
Model 3, reports the results for the shopping behavior cannibalizes
of some new customers, but in regions where pre-
first-time customers. existing stores existed, this cannibalization had already hap-
In the full model, the store .opent coefficient is insignifi- pened. The post .open .monthst coefficient is positive and
cant (ßj = -10.2, p > .10), indicating that the number of statistically significant in both the regions (ß2 = 1.29,/? <
first-time customers shopping with the direct channels was .001 in virgin areas, and ß2 = 0.98,/? < .001 in areas neigh-
not immediately affected after retail stores were opened. boring preexisting stores). We found no statistically signifi-
The post .open .monthst coefficient is positive and statisti- cant difference between the post .open .monthst (ß2) coeffi-
cally significant (ß2 = 1.13,/? < .001). This shows that the cients between regions (t = .82,/? > .10).
introduction of a physical store increased the number of Our results suggest that some customers who would like
first-time customers shopping in the direct channels over to try the brand (and who would have done so in the direct
channels even if the store had not opened) immediately opening of an additional store had less effect on their
choose to do so through the store instead of the direct chan-behavior than it did for customers in virgin regions. Exist-
nels. This could be because it would be difficult for a cus- ing customers who did not have access to a retail store in
tomer to try the brand in virgin territories except through the past switched some of their demand from the direct
the direct channels before the store opening. Over time, channels when a new store opened in their region.
however, the stores entice new customers to the brand at a The post .open .monthst coefficient is positive and statis-
faster rate in both types of territories. tically significant in both virgin territories (ß2 = 2.75,/? <
.001) and areas neighboring preexisting stores (ß2 = 1.93,/? <
The Effect of the Store Openings on Repeat Direct.001). We found no statistically significant difference between
Channel Customers
the post .open .monthst (ß2) coefficients between regions (t =
Next, we explored the effect of the store openings on 1.171, p > .10). This suggests that the introduction of a
exist-
physical store increased the number of repeat customers
ing direct channel customers. Repeat customers represent
households that have previously purchased frompurchasing
direct from the direct channels regardless of whether
theshop-
channels. Table 6, Model 4, reports the results for the surrounding region already had a preexisting store.
ping behavior of repeat customers. In the full model, the
store .opent coefficient is negative and statistically signifi- General Discussion
cant (ßj = -103, p < .001), in support of H6, indicating that
Our research contributes to the knowledge of multichannel
the number of repeat customers purchasing in the direct
retailing by showing whether, when, for which type of cus-
channels decreased in the short run following the opening
tomer (first-time vs. repeat), and why a new channel helps
of the stores. The post .open .monthst coefficient is positive
and hurts existing channels. First, we propose a conceptual
and statistically significant (ß2 = 2.35, p < .001), in support
framework that predicts the pattern of cross-channel elas-
of H 7, indicating that the number of repeat customers pur-
ticities that occur over time following the introduction of a
chasing in the direct channels increased in the long run fol-
new channel. Our framework, which analyzes underlying
lowing the opening of the stores.
capabilities, suggests that both the type of channel being
The complementary result could be caused by one of
added and the composition of the existing channels already
three consumer behavior processes. First, the incremental
in the retailer's channel mix matter because different chan-
first-time customers who were added to the direct channels
nels have different capabilities. This helps explain why
following the store opening (see Model 3 results) continued
empirical results differ depending on whether the Internet
to purchase from the direct channels in the long run.channel is being added to a retail store channel (the focus of
Because new customers become repeat customers the sec-
most previous studies) or a retail store is being added to an
ond time they purchase, all the purchases beyond their firstInternet channel (the focus of our work). Crucially, our
roll into the repeat customer counts. Second, some of the framework examines how the passage of time, something
repeat customers who switched some of their purchasing tothat has been largely ignored in previous work, influences
the store channel when it opened returned to the direct cross-channel effects. We propose that some channel capa-
channels over the long run and resumed purchasing there.bilities are immediately apparent to consumers while others
Third, some of the repeat customers who did not switch must be learned, resulting in different short- and long-term
some of their purchasing to the store channel when iteffects. We also propose that first-time and repeat cus-
opened shortened their interpurchase cycle in the direct tomers have different shopping goals and therefore differ-
channels as a result of the store opening; these customers entially value channel capabilities.
began shopping in the direct channels more frequently than We then test our framework using sales data. Our
they had before the store opening. Given that we do not empirical results confirm our hypotheses, showing that
have household-level data, it is impossible to discern whichadding a retail store cannibalizes sales in the catalog chan-
of these three processes is occurring. nel, but not the Internet channel, right when it opens but
The same patterns held across regions with and withoutthat its continued presence benefits both the Internet and
preexisting stores. The store .opent coefficient is negativecatalog channels over time. Furthermore, we show that
and significant in both virgin territories (ßj = -162, p <while the store initially reduces the number of repeat cus-
.001) and areas neighboring preexisting stores (ßj = -43 ,/? <tomers purchasing in the direct channels, it brings more
first-time customers in at a faster rate and encourages
.05). We then tested whether there was a statistically signifi-
cant difference between the store .opent (ßj) and post .open. higher numbers of new and repeat customers to purchase in
monthst (ß2) coefficients between the virgin and preexisting the direct channels over time.
store regions for repeat customer counts. We found a statis-
tically significant difference between the store .opent (ßj)Relationship with Prior Empirical Studies
coefficients between regions (t = -4.127,/? > .001), indicat-Although these findings may be particular to our research
ing that the cannibal ization effect was significantly greatersetting, our broad conceptual framework provides a way to
when there were no preexisting stores already in the area. generate hypotheses in other settings, suggests the limita-
This pattern of results suggests that some existing cus- tions of existing empirical findings, and explains why multi-
tomers who were likely to switch some of their demand tochannel retailing systems must be studied from multiple
retail stores already had done so in the regions where retailperspectives. There are only a few previous empirical stud-
stores existed before the new store opening. Therefore, theies on multichannel systems. Rather than being at odds with
our findings, their results help highlight the value of our predict that direct retailers with less established brands may
framework. Because the Internet channel has different capa- benefit more from the billboard effect that the opening of a
bilities than a store channel, we would expect different new store brings. Their direct business may recover more
results if an Internet channel is added to a brick-and-mortar quickly from the initial drop in sales and experience more
system than we would if a retail store is added to an Internet rapid growth afterward. For product moderators, Inman,
channel. Examining the capabilities of each channel helps Shankar, and Ferraro (2004) hypothesize that certain prod-
us predict both the valence of the effect on sales in the uct categories are more highly associated in consumers'
existing channel and the timing in which to expect sales minds and purchasing practices with one particular channel
changes. than others. Research (Alba et al. 1997; Balasubramanian,
In the newspaper (Deleersnyder et al. 2002) and music Raghunathan, and Mahajan 2005) also suggests that the
(Biyalogorsky and Naik 2003) industries studied in previ- type of good being purchased (search/experience/credence
ous work, the Internet channel offered customers a broader goods or utilitarian/experiential goods) can affect the con-
assortment of content or products than the brick-and-mortar sumer's shopping goals and therefore affect whether the
channel, as illuminated in Anderson (2006). With the addi- capabilities of a particular channel are attractive. The tim-
tion of the Internet channel, customers could shop 24/7, ing of the opening of a retail store in relation to the growth
increasing the convenience of purchasing. Given that both of the Internet channel may be a context moderator affect-
news and music are both products that do not need to be ing cross-channel switching behavior.
handled before purchase, shopping for these items on the In the early days of the Internet channel, as studied here,
Internet is less risky than purchasing products such as cloth- Internet sales were not cannibalized by the opening of a
ing or furniture. Thus, it seems likely that in both of these retail store; however, as Internet penetration grows and as
industries, an Internet channel would cannibalize the brick- shopping over the Internet becomes more predominant, the
and-mortar operations, and indeed both sets of researchers opening of retail stores may begin to cannibalize Internet
hypothesize a cannibalizing effect. However, both studies sales to a greater extent. This may reflect that early adopters
yield a null effect, showing neither cannibalization nor of Internet shopping were younger than retail store shop-
complementarity (within 42 months for the newspaper pers, so the opening of a store had little impact on them.
study and 13 months for the music CD study). At the time However, as Internet shopping diffuses into the mainstream,
these studies were run, e-commerce was relatively new, and the demographic differences between Internet shoppers and
perhaps consumers needed to experience reading and shop- retail store shoppers are beginning to disappear, making it
ping before deciding to switch to this channel. A longer more likely that the opening of a retail store will both can-
time frame may have been necessary to detect the cannibal- nibalize Internet sales in the short run and complement
ization, which seems now to have occurred in both these Internet sales in the longer run.
industries (Benilde 2010; Danaher et al. 2008). In summary, our theoretical model is just the beginning
Because marketing communications on the Internet are of the exploration of possible contingencies that affect
more inbound, introducing the Internet channel would not cross-channel elasticities. It is our hope that researchers will
necessarily create greater awareness for newspaper sub- investigate some of these additional contingencies to
scriptions or music retail stores. Therefore, complementary increase collective understanding of the complicated inter-
effects stemming from increased brand awareness and play involved in predicting channel cannibalization and
strengthened brand associations would probably not be complementarity.
observed. Across both existing studies, the authors found no
Limitations
complementary effects when the Internet channel was
added. Because our data set was aggregated at the zip code level,
we were unable to fully observe how customer heterogene-
Opportunities for Further Research
ity affected our results. Although we were able to explore
We recognize that other factors also play a role in determin- some differences between first-time and repeat customers, it
ing customers' likelihood to switch channels in a multichan-would be worthwhile to explore differences among cus-
nel system, such as company and product (Balasubramanian, tomer segments in greater depth. In particular, when apply-
Raghunathan, and Mahajan 2005), customer heterogeneity ing our theory in a real- world setting, retailers could survey
(Inman, Shankar, and Ferraro 2004; Thomas and Sullivancustomers to understand their shopping goals and could
2005), marketing (Ansari, Mela, and Neslin 2008), competi- develop weighting schemes for different capabilities using
tive, and contextual contingencies. These contingenciestools such as conjoint analysis. Our results suggest that this
suggest moderating conditions that offer opportunities for should be done separately for first-time and repeat customers.
further research, some of which we highlight next. When a new channel is added to an existing distribution
First, we would expect that the retailer's brand equity, a channel portfolio, the retailer has the opportunity to strate-
company contingency, would moderate our results. In our gically manage cannibalization and complementarity by
framework, we propose that synergistic effects are driven taking marketing actions to mute or encourage each type of
by a greater ability to provide outbound marketing commu- effect (Ansari, Mela, and Neslin 2008). In our research set-
nications: The billboard effect of the retail stores increases
ting, the retailer did not strategically manage the cross-
brand awareness and brand associations for all channels.
channel cannibalizing and complementary effects through
Compared with our analysis of a well-known company,marketing
we activities. Marketing activities were constant
across experimental and control conditions, providing a assess the best ways to manage cross-channel elasticities to
clean test of the effect of the store opening on the direct encourage its customers to use the most profitable channels.
channels without marketing intervention. Just as the retailer Next, we offer strategies for managing sales forecasting,
may strategically manage its cross-channel elasticities promotion planning, and customer acquisition.
through targeted marketing programs, so too may its com-
Sales forecasting. Correctly predicting the drop in direct
petitors try to thwart the growth of a new channel in the
channel sales and their eventual recovery and acceleration
marketplace. Our data set did not include competitive sales
is crucial to managing all channels during retail store
data or marketing activity, so we were unable to observe
expansion if the retailer wants to maintain customers in the
whether our retailer's competitors strategically intervened.
catalog and Internet channels, particularly because most
However, our retailer's managers did not observe a change
retailers use recency, frequency, and monetary value (RFM)
in the marketing activities of their primary competitor dur-
models to drive catalog mailings and e-mail solicitations.
ing the time period of our test. Furthermore, a strong These models calculate whether to send a customer a cata-
competitive response would make it more difficult for us to
log or an e-mail on the basis of their purchase RFM to the
find a significant complementary effect associated with the
firm. Retailers that rely on RFM models may decrease cata-
store opening. Modeling the retailer's and its competitors'
log mailings to customers who have temporarily switched
strategic responses to channel additions as a dynamic and
some of their purchasing to the retail store channel. Particu-
interactive competitive game holds promise for further
larly in the catalog channel, this decrease in marketing sup-
research.
port may intensify the drop in sales and prolong the onset of
Because our data set includes sales and customer data
synergistic effects. Retailers that understand the patterns of
from only one retailer, caution must be used when general-
cross-channel elasticities can adjust the algorithm driving
izing our empirical results to other retail settings. In an
the RFM model to account for the store opening to avoid
effort to generalize our work, we present a conceptual any counterproductive decrease in marketing support.
model that offers theoretical and managerial insight into the
underlying consumer goals that drive cross-channel pur- Promotion planning. Our findings also provide guid-
ance to managers on how to structure promotions when a
chases. This article joins other empirical work in multichan-
new channel is added. Given that stores lead to sales
nel retailing that focuses on analyzing a single retailer
(Ansari, Mela, and Neslin 2008; Biyalogorsky and Naik increases in the direct channels over time, promotions that
2003; Pauwels and Neslin 2009; Venkatesan, Kumar, and encourage customers to shop across channels should be
Ravishanker 2007) or a single product category (Deleersny- implemented. Structurally, the retailer studied here main-
der et al. 2002), moving the field closer toward empirical tained the retail store and direct channels operating units as
generalizations. separate entities, perhaps limiting complementary effects
Another limitation of our data set is that our customer due to a dearth of cross-channel promotion. Cooperative
data are aggregated at the direct channel level, unlike ourcross-channel marketing can improve sales in all channels
sales data, which exist at the individual catalog and Internetor drive sales from less profitable channels to more prof-
channel levels, thus limiting our ability to observe customeritable ones. For example, if catalog cannibalization is unde-
count differences between the catalog and Internet chan- sirable to the retailer, it can offset drops in sales by increas-
nels. Our data set is also limited in that it does not allow us ing direct channel promotions in the area surrounding the
to investigate retail store sales. However, over the time store during the store opening period. This may keep exist-
period of the study, sales from the retail stores were signifi- ing customers in the catalog channel rather than enticing
cantly higher than sales from the direct channels. Given the them to shop in the store. In this case, the store becomes a
strong in-store sales levels, the low levels of cross-channel new customer acquisition engine with its promotional vehi-
cannibalization, and the high levels of cross-channel com- cles targeted toward new, rather than existing, customers,
plementarity uncovered, the addition of the retail stores waswith the hope that these new customers will eventually
highly beneficial to the retailer's sales. Finally, our data setmigrate to the lower-cost catalog and online channels.
is limited in that it does not allow us to investigate the profit Customer acquisition management. Our findings pro-
implications of channel elasticities. Thus, our findingsvide insight into customer relationship management by illu-
report changes in sales revenues, not in profits. minating how the addition of a new channel affects cus-
tomer acquisition in existing channels. The opening of a
Managerial Implications
retail store has a small impact on the rate at which first-time
Our framework and findings offer insight into managingcustomers come into the direct channels in the short run;
multichannel systems during channel expansion. Because
therefore, direct channel managers should continue to
of the differing profitability across channels, managers first
invest in customer acquisition programs during the months
must ascertain which channels are most desirable for cus- surrounding a new store opening if the retailer finds it more
tomers to frequent. In some cases, channel switching is profitable to serve customers in the direct channels than in
beneficial for the retailer- for example, when existing cus- the store channel. Over the long run, the existence of a retail
tomers switch from a high-cost channel, such as using a store increases the rate at which first-time direct channel
bank teller to process a deposit, to a lower-cost channel
customers are acquired. Thus, prospecting materials for
such as using an automated teller machine (Hitt and Frei new direct channel customers should include a retailer's
2002). In others, it is detrimental. Thus, the retailer must
brick-and-mortar store location and should highlight cross-
channel benefits, such as the ability to pick up or return enabled us to rule out potential alternative explanations that
items ordered on the Internet at the store. changes in sales were due to differences in Internet penetra-
In conclusion, our results uncover the previously elu- tion or consumers' familiarity with purchasing from differ-
sive synergy that comes from operating a multichannel sys- ent types of channels. We sourced the demographic and
tem that includes retail stores, catalogs, and Internet chan- shopping behavior data from Esri, and we calculated drive
nels. Given a better understanding of both positive and time to the mall using Maplt software.
negative cross-channel effects, retailers can better anticipate Finally, we matched the treatment and control regions
and respond to changes in sales in existing channels when a on catalog and web sales levels in the period before the
new channel is added, thus strategically managing its chan- store openings. We averaged the catalog and web sales in
nels as a portfolio rather than as separate entities. each zip code in the treatment regions for 36 months before
the store openings. We then added this to the match so that
we could find corresponding zip codes from the control
Appendix regions that best matched these preperiod sales. This
Treatment and Control Group ensured that zip codes of similar direct channel market
Matching Process potential were matched, enabling us to rule out the potential
alternative explanation that the retailer located stores in
We accomplished our matching process in three steps. We
regions that were stronger or weaker in direct channel sales
began by creating a large pool of zip codes from markets in
than the control region.
which stores were not opened. This pool needed to be het-
erogeneous enough in its composition such that reasonable We tested five matching procedures to match zip codes
in the treatment and control regions and chose the one that
matches could be found for each of the zip codes in the four
produced the best fit (Ho et al. 2004). The nearest-neighbor
treatment areas. We accomplished this by collecting data for
matching algorithm works by finding a zip code in the con-
192 zip codes located in three metropolitan areas in other
trol group that is closest to a zip code in the treatment group
parts of the state. We chose each metropolitan area specifi-
on a logistic distance measure. This is a "greedy" algorithm
cally because it contained a shopping mall that was similar
in the sense that the closest match is chosen at each step
to the malls in which the new stores opened. This ensured
without trying to minimize a global distance measure. In
that the direct channels faced a similar competitive environ-
contrast, the optimal matching algorithm (Hansen 2004)
ment in both the treatment and control regions. Further-
finds the matched samples with the smallest average
more, we confirmed with the retailer's managers that there
was no difference in the execution and amount of the
absolute distance across all matched pairs. The subclassifi-
cation matching algorithm forms subclasses of zip codes
retailer's marketing activity in the direct channels betweensuch that the distributions of the covariates for the treatment
the treatment and control group regions. Thus, we qualita-
and control groups are as similar as possible within each
tively matched the treatment and control regions on the
subclass. The full matching algorithm (Hansen 2004;
presence of competition and on marketing activity in the Rosenbaum 2002) is a refined version of subclassification.
direct channels.
It is optimal in the sense of minimizing the weighted dis-
Next, we quantitatively matched zip codes in each treat-
tance between each treatment and control zip code within
ment region with zip codes in the control region on geo-
each subclass. The genetic matching algorithm (Abadie and
demographic, shopping behavior, and direct channel sales
Imbens 2006; Diamond and Sekhon 2005) uses a genetic
characteristics. To begin, we chose six geodemographic search technique to find a set of weights for each covariate
matching variables that affect store location decisions. Our
such that optimal balance is achieved after matching.
discussions with retail managers and previous research sug- Genetic search techniques often provide good solutions for
gest that geodemographics, such as distance to the store difficult search spaces.
(Bell, Ho, and Tang 1998; Forman, Ghose, and Goldfarb The genetic algorithm provided the best match between
2009; Fox, Montgomery, and Lodish 2002; Venkatesan, the treatment and control groups for our data across the 1 1
Kumar, and Ravishankar 2007), age (Ansari, Mela, and geodemographic, shopping behavior, and preperiod sales
Neslin 2008; Ward 2001), and income (Ansari, Mela, and variables. Table Al reports the standardized mean differences
Neslin 2008; Fox, Montgomery, and Lodish 2002; Inman, between the treatment and control groups (the standardized
Shankar, and Ferraro 2004), affect channel choice. Thus, we bias) both before matching and after the genetic match.
included drive time to the mall, average population, the The literature (Ho et al. 2004) suggests that a good
compound annual growth rate (CAGR) of the population, match will result in the means of the treatment and control
average income, the CAGR of income, and average age. groups being less than a quarter of a standard deviation
This enabled us to rule out potential alternative explana- apart, implying that standardized biases will be less than .25
tions that changes in sales were due to differences in the for most matching covariates. The standardized biases were
density of customers, purchasing power, or age of cus- often large before matching, which suggests, as expected,
tomers living near stores. that the zip codes in the treatment groups were fundamen-
We also chose to match on three key multichannel shop- tally different from the entire collection of zip codes from
ping behavior variables to control for alternative explana- which the control groups were drawn. Nevertheless, the
tions of our results, including the market penetration of genetic matching algorithm greatly improved the balance
adults purchasing by catalog, adults purchasing by website, between the groups. After matching, the standardized bias
and the percentage of households with Internet access. This was greater than .25 in only one case across the 1 1 match-
TABLE A1
Matching Assessment for the Genetic Matching Algorithm
Standardized Mean Difference Between Treatment and Control
Before Matching Genetic Matching

Variable Store A Store B Store C Store D Store A Store B Store C Store D
Drive time -.138 .285 -.860 .406 -.003 -.035 -.173 .029
Average population .602 .341 .670 .294 -.107 .025 .056 -.027
Population CAGR -.026 -.329 -.100 .075 -.131 -.004 .002 .140
Average income .550 -.138 .015 -.021 .149 .036 .141 -.011
Income CAGR -.874 .190 -.965 .038 .077 .079 -.180 -.017
Average age .554 .113 -.895 -.315 .013 .054 .431 -.011
Adults buying through catalog -.153 .147 -.423 -.118 -.071 .030 -.106 -.063
Adults buying through website -.230 -.065 -.263 -.019 .014 .101 .123 .103
Households with Internet access .571 -.207 .000 -.052 .153 -.035 .097 -.059
Preperiod catalog sales .343 -.339 .113 -.512 .103 .031 .090 -.001
Preperiod web sales .077 -.034 .255 -.387 .097 .070 .127 -.010
ing covariates in the four retail trading

ing areas:
improved thethe age between the
balance
variable for Store C. Furthermore,before
even in
tothis
.431case,
aftermatch-
matching.
REFERENCES
Abadie, Alberto and Guido W. Imbens (2006), "Large Sample
Chang, Rita (2009), "Boost, Microsoft Market Via Bricks, Mor-
Properties of Matching Estimators for Average Treatment
tar," Advertising Age , 80 (32), 8.
Effects," Econometrica , 74 (1), 235-87. Coldwater Creek Inc. (2002), SEC Filing Form 10-K.
Alba, Joseph A. and J. Wesley Hutchinson (1987), "Dimensions of
Danaher, Brett, Samita Dhanasobhon, Michael D. Smith, and
Consumer Expertise," Journal of Consumer Research Rahul
, 13 (4),
Telang (2008), "Converting Pirates Without Cannibaliz-
411-54.
ing Purchasers: The Impact of Digital Distribution on Physical
Sales and Internet Piracy," Heinz Research Working Paper 57,
Lutz, Alan Sawyer, and Stacy Wood(accessed
(1997), "Interactive
December Home
14, 2011), [available at http://repository.
cmu.edu/heinzworks/57]Incentives
Shopping: Consumer, Retailer, and Manufacturer . to
Participate in Electronic Markets," Journal
Deighton, of "The
John (2005), Marketing, 61in the Informa-
Presentation of Self
(July), 38-53. tion Age," in Inside Consumption: Consumer Motives , Goals,
Anderson, Chris (2006), The Long Tail. New York: Hyperion. and Desires , S. Ratneshwar and D.G. Mick, eds. New York:
Ansari, Asim, Carl F. Mela, and Scott A. Neslin (2008), "Cus- Routledge, 233-51.
tomer Channel Migration," Journal of Marketing Research , 45 Deleersnyder, Barbara, Inge Geyskens, Katrijn Gielens, and
(February), 60-76. Marnik G. Dekimpe (2002), "How Cannibalistic Is the Internet
Baker, Julie, A. Parasuraman, Dhruv Grewal, and Glenn B. Voss Channel? A Study of the Newspaper Industry in the United
(2002), "The Influence of Multiple Store Environment Cues on Kingdom and the Netherlands," International Journal of
Perceived Merchandise Value and Patronage Intentions," Jour- Research in Marketing , 19 (4), 337^4-8.
nal of Marketing, 66 (April), 120-41. Diamond, Alexis and Jasjeet Sekhon (2005), "Genetic Matching
Balasubramanian, Sridhar (1998), "Mail Versus Mall: A Strategic for Estimating Causal Effects: A General Multivariate Match-
Analysis of Competition Between Direct Marketers and Con- ing Method for Achieving Balance in Observational Studies,"
ventional Retailers," Marketing Science , 17 (3), 181-95. working paper, Travers Department of Political Science, Uni-
versity of California at Berkeley, (accessed December 14,
sumers in a Multichannel Environment: 2011), [available
Product at http://sekhon.berkeley.edu/papers/Gen-
Utility,
Process Utility and Channel Choice,"Match.pdf]. Journal of Interactive
Marketing , 19 (2), 12-30. Dorsey, Moira (2006), "Match Channel Capabilities to Customer
Bell, David R., Teck-Hua Ho, and Christopher Goals," Forrester Research , (October 12).
S. Tang (1998),
"Determining Where to Shop: Fixed Forman,and Variable
Chris, Anindya Ghose, and Costs
Avi Goldfarbof
(2009), "Com-
Shopping," Journal of Marketing Research petition Between Local
, 35 and Electronic
(August), Markets: How the
352-69. Benefit of Buying Online Depends on Where You Live," Man-
Benilde, Marie (2010), "The End of Newspapers," The New Yorkagement Science , 55 (1), 47-57.
Times Online , (March 16, 2010), (accessed December 14. 2011),
Fox, Edward J., Alan L. Montgomery, and Leonard M. Lodish
[available at http://www.nytimes.com/2010/03/17/opinion/(2002), "Consumer Shopping and Spending Across Retail For-
1 7 iht-edbenilde .html?page wanted=all] . mats," Journal of Business, 11 (2), S25-S60.
Biyalogorsky, Eyal and Prasad Naik (2003), "Clicks and Mortar:
Frazier, Gary L. (1999), "Organizing and Managing Channels of
The Effect of Online Activities on Offline Sales," MarketingDistribution," Journal of the Academy of Marketing Science,
Letters , 14(1), 21-32. 21 (2), 226-40.
Bustillo, Miguel and Geoffrey A. Fowler (2009), "Wal-Mart Sees
Geyskens, Inge, Katrijn Gielens, and Marnik G. Dekimpe (2002),
Stores as Online Edge," The Wall Street Journal , (December"The Market Valuation of Internet Channel Additions," Journal
15), Bl. of Marketing, 66 (April), 102-119.
110/ Journal of Marketing, May 2012
Halligan, Brian and Dharmesh Shah (2009), Inbound Marketing: Neslin, Scott A. and Venkatesh Shankar (2009), "Key Issues in
Get Found Using Google , Social Media, and Blogs. New York: Multichannel Customer Management: Current Knowledge and
John Wiley & Sons. Future Directions," Journal of Interactive Marketing, 23 (1),
Hansen, Ben B. (2004), "Full Matching in an Observational Study 70.
of Coaching for the SAT," Journal of the American Statistical Pauwels, Koen and Scott A. Neslin (2009), "Building with Bricks
Association , 99 (467), 609-18. and Mortar: The Revenue Impact of Opening Physical Stores
Heckman, James J., Hidehiko Ichimura, Jeffrey Smith, and Petra in a Multichannel Environment," working paper, Tuck School
Todd (1998), "Characterizing Selection Bias Using Experimen- of Business, Dartmouth College.
tal Data," Econometrica , 66 (5), 1017-98. Pitt, Leyland, Pierre Berthon, and Jean-Paul Berthon (1999),
Hitt, Lorin M. and Frances X. Frei (2002), "Do Better Customers "Changing Channels: The Impact of the Internet on Distribu-
Utilize Electronic Distribution Channels? The Case of PC tion Strategy," Business Horizons, 42 (2), 19-28.
Banking," Management Science , 48 (6), 732-48. Read, D. and George F. Loewenstein (1995), "Diversification
Ho, Daniel E., Kosuke Imai, Gary King, and Elizabeth StuartBias: Explaining the Discrepancy in Variety Seeking Between
(2004), Matchit: Matching as Nonpar ametric Preprocessing Combined and Separated Choices," Journal of Experimental
for Parametric Causal Inference , software program, [available Social Psychology, 1(1), 34-49.
at http://gking.harvard.edu/matchit/] . Reda, Susan (2002), "Active Multi-Channel Shoppers May Be a
Liability," Stores, 84, 78-82.
metric Preprocessing
Reynolds, Kristy E. and Sharon E. Beatty (1999), "A Relationship f
metric Causal
Customer Typology," Journal
Inferenc
of Retailing, 75 (4), 509-523.
Inman, J. Jeffrey,
Rosenbaum, Paul R. (2002), Observational Studies, 2d ed. New V
(2004), "The Roles
York: Springer- Verlag. of
Geodemographics
Simon, Julian L. and Johan Arndt (1980), "Thein Shape of the Ch
ing, 68 (April),
Advertising Response Function,"51-71.
Journal of Advertising
Iyengar, Sheena
Research, 20 (4), 767-84. S. and
Demotivating: Can
Stern, Louis, Adel El-Ansary, and Anne On
Coughlan (1996), Market-
Journal of Personal
ing Channels, 5th ed. Upper Saddle River, NJ: Prentice Hall.
995-1006. Tang, Fang-Fang and Xiaolin Xing (2001), "Will the Growth of
Jacoby, Jacob and David Mazursky (1984), "Linking Brand and Multi-Channel Retailing Diminish the Pricing Efficiency of the
Retailer Images- Do the Potential Risks Outweigh the Poten-Web?" Journal of Retailing, 11 (3), 319-33.
tial Benefits?" Journal of Retailing, 60 (2), 105-122. Thomas, Jacquelyn S. and Ursula Y. Sullivan (2005), "Managing
Jaffe, Adam B., Manuel Trajtenberg, and Rebecca Henderson Marketing Communications with Multichannel Customers,"
(1993), "Geographic Localization of Knowledge Spillovers Journal
as of Marketing, 69 (October), 239-51 .
Evidenced by Patent Citations," Quarterly Journal of Econom-
Venkatesan, Rajkumar, V. Kumar, and Nalini Ravishanker (2007),
ics, 108 (3). 577-98. "Multichannel Shopping: Causes and Consequences," Journal
Keller, Kevin Lane (1993), "Conceptualizing, Measuring, and of Marketing,! 1 (April), 114-32.
Managing Customer-Based Brand Equity," Journal of Market-Verhoef, Peter C., Scott A. Neslin, and Bjorn Vroomen (2007),
ing, 57 (January), 1-31 . "Multichannel Customer Management: Understanding the
Research-Shopper Phenomenon," international Journal of
Kwon, Wi-Suk and Sharron J. Lennon (2009), "Reciprocal Effects
Between Multichannel Retailers' Offline and Online Brand Research in Marketing , 24 (2), 129^8.
Images," Journal of Retailing, 85 (3), 376-90. Von Wangenheim, Florian and Tomas Bayon (2007), "Behavioral
Lee, Peggy M. and Sunil Wahal (2004), "Grandstanding, Certifi- Consequences of Overbooking Service Capacity," Journal of
cation, and the Underpricing of Venture Capital Backed IPOs," Marketing,! 1 (October), 36-47.
Journal of Financial Economics , 73 (2), 375-407. Ward, Michael R. (2001), "Will Online Shopping Compete More
Meyer, Bruce D. (1995), "Natural and Quasi-Experiments in Eco- with Traditional Retailing or Catalog Shopping?" Netnomics, 3
nomics," Journal of Business and Economic Statistics, 13 (2), (2), 103-117.
151-61. Winship, Christopher and Stephen L. Morgan (1999), "The Esti-
Moriarty, Roland T. and Ursula Moran (1990), "Managing Hybrid mation of Causal Effects from Observational Data," Annual
Marketing Systems," Harvard Business Review, 90 (6), Review of Sociology, 25, 659-706.
146-55.
Patterns of Cross-Channel Elasticities /111

Sage Publications, Inc. Journal of Marketing: This Content Downloaded From 2.36.90.50 On Thu, 21 Nov 2019 18:57:04 UTC

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Sage Publications, Inc. Journal of Marketing: This Content Downloaded From 2.36.90.50 On Thu, 21 Nov 2019 18:57:04 UTC

Uploaded by

Copyright:

Available Formats

Adding Bricks to Clicks: Predicting the Patterns of Cross-Channel Elasticities Over Time

Adding Bricks to Clicks: Predicting

the introduction of the Internet channel (Ansari, Mela, and

Jill Avery is Assistant Professor of Marketing, Simmons School of Manage-

© 201 2, American Marketing Association Journal of Marketing

98 1 Journal of Marketing, May 2012

4 * i fisji i|ï h tiUllilSs -7-1 USUI IISÜI l!#|lfi| fl

■S SS i o ? ; E g <,^ £©£ «fe_íg

100 / Journal of Marketing, May 2012

Consumer Shopping Goal

Patterns of Cross-Channel Elasticities 1 101

102 1 Journal of Marketing, May 2012

Patterns of Cross-Channel Elasticities 1 103

104 / Journal of Marketing, May 2012

We used the same procedures to analyze the customertime,

Patterns of Cross-Channel Elasticities 1 105

106 / Journal of Marketing, May 2012

Patterns of Cross-Channel Elasticities 1 107

108 1 Journal of Marketing, May 2012

Patterns of Cross-Channel Elasticities 1 109

Before Matching Genetic Matching

ing covariates in the four retail trading

110/ Journal of Marketing, May 2012

Patterns of Cross-Channel Elasticities /111

You might also like