do market characteristics impact the relationship … of retailing 83 (3, 2007) 309–324 do market...

A

itaii©

K

ftabptsit

k

2to

0d

Journal of Retailing 83 (3, 2007) 309–324

Do market characteristics impact the relationshipbetween retailer characteristics and online prices?

Rajkumar Venkatesan a,∗, Kumar Mehta b,1, Ravi Bapna c,2

a 100 Darden Blvd., Darden Graduate School of Business, Charlottesville, VA 22903, United Statesb Management Information Systems, 4400 University Drive, MSN 5F4, School of Management,

George Mason University, Fairfax, VA 22030, United Statesc Information Systems, Indian School of Business, Gachibowli, Hyderabad, 500 032, India

Received in revised form 30 April 2006; accepted 30 April 2006

bstract

We propose that market characteristics interact with retailer characteristics to determine online prices. The retailer characteristics examinednclude—service quality of a retailer, channels of transaction provided by a retailer and the size of a retailer. The market characteristics capturehe level and nature of competition, and the price level of a product. We utilize a Hierarchical Linear Model (HLM) framework for capturing
nd testing the proposed interactions. The better fit between the model and the online market structure is reflected by a twenty-five percentncrease in explainable price dispersion over results from comparable studies. Our study demonstrates that while retailer characteristics dompact online prices, this influence is significantly enhanced or diminished by the accompanying market characteristics.
2007 New York University. Published by Elsevier Inc. All rights reserved.

ity; Com

ppiiIipAr

eywords: Internet markets; Retailer pricing strategy; Retailer service qual

Introduction

Industry surveys show that online firms have transitionedrom the initial phase of investment and brand-building tohe profit maximization phase. The published profitabilitynd sales reports from 2001 to 20033 indicate that Internet-ased retailing is now a firmly established and an increasinglyrofitable market. In light of this maturation, it is imperativeo ask whether the extant understanding of retailer pricing
trategies comprehensively explains the online retailers’ pric-ng behavior. Shankar and Bolton (2004) have suggestedhat descriptive research on the determinants of retailer
∗ Corresponding author. Tel.: +1 434 924 6916; fax: +1 434 243 7676.E-mail addresses: [email protected] (R. Venkatesan),

[email protected] (K. Mehta), ravi [email protected] (R. Bapna).1 Tel.: +1 703 993 9412; fax: +1 703 993 1809.2 Tel.: +1 860 486 8398; fax: +1 860 486 4839.3 eMarketer (2004) indicates 79 percent online retailers were profitable in003. Further, a Mullaney (2003) indicates that 40 percent of the publiclyraded Internet companies reported profits in 2002, with the key areas beingnline retailing and finance.

ltkrpad

ic2g

022-4359/$ – see front matter © 2007 New York University. Published by Elsevieoi:10.1016/j.jretai.2006.04.002

petition

ricing strategies aids managers in profiling their pricingractices and predicting competitors’ behavior. However, its important to note that the Internet as a retailing channels significantly different from the traditional marketplace.n contrast to the traditional channels, Internet is character-zed by vastly increased availability of information regardingroduct attributes, prices and retailer service quality metrics.

key challenge has been to obtain an understanding of theetailers’ continued ability to price differentiate in the face ofowered search costs in Internet markets. The main findingso date have been that—(a) price dispersion in Internet mar-ets is persistent with time, (b) in some, but not all, instancesetailer service quality explains some of the variance in onlinerices, and (c) the transaction channels provided by a retailernd level of competition matter (see Pan et al. 2004 for aetailed review).

Economic theories of consumer demand and retailer pric-
ng suggest that product prices are a function of both—theharacteristics of the retailers and the competition (Pakes003; Betancourt and Gautschi 1993). Studies of pricing byrocery retailers in offline markets have found that retail-
r Inc. All rights reserved.

mailto:[email protected]



dx.doi.org/10.1016/j.jretai.2006.04.002

3 l of Ret

edtiawts2t

ntkec

p

TC

R

A

E

P

10 R. Venkatesan et al. / Journa

rs customize their prices at the brand store level. Theyo so based on their intimate knowledge about their cus-omers, product categories, store characteristics, and mostmportantly in response to market competition (Shankarnd Bolton 2004). Empirical evidence also indicates thathen setting prices the retailers consider multiple objec-

ives such as—reaction to competitor actions, maximizing
tore brand share, and manufacturer incentives (Chintagunta002). These observations from offline markets establishhat retailer’s pricing decisions not only reflect their inter-
Wmp

able 1omparing proposed framework with existing studies

epresentative research Model includes characteristics of

Shoppers Product Retailer Competition Comreta

nalytical modelsLal and Sarvary (1999) Ya Y N N N

Chen and Hitt (2003) Y N Y N N

Smith (2001) Y N Y N N

mpirical modelsTang and Xing (2001) N H Y N N

Ancarani and Shankar(2004)

N H Y N N

Baye and Morgan(2001)

N Y N Y N

Baye and Morgan(2003)

N Y Y Y N

Carlton and Chevalier(2001)

N H Y Y N

Clay et al. (2001, 2002) N H N Y N

Baylis and Perloff(2002)

N H Y N N

Pan et al. (2002, 2003b) N H Y Y N

resent study N H Y Y Y

a Y: yes, N: no, H: homogenous products.

ailing 83 (3, 2007) 309–324

al resources and objectives, but also point to the fact thathe retailers pay close attention to characteristics of the mar-et in which they operate. We are motivated to investigatexistence of a similar effect in online markets where theseharacteristics are easily observable by the retailers.

A distinct feature of this study is that we explicitly incor-orate interactions between retailer and market level factors.
e believe that by examining the moderating influence ofarket level factors, we can provide a more comprehensive
icture of the relationship between prices and retailer char-

Major findings

petition andiler Interaction

Price sensitivity of online shoppers decreases when (1)there is a large pool of online shoppers, (2) nondigitalproduct attributes are important, (3) consumers have afavorable loyalty to the brand they currently own, and(4) the fixed cost of a shopping trip is higher than thecost of visiting an additional store.Better known retailer has higher price on average than alesser known retailer, but lower price on some productsand/or some of the time resulting in price dispersion.Focus of consumer search on few well known retailersleads to interdependence among these and also higherprices. Lesser known retailers however adopt randompricing strategy in equilibrium.

Traditional retailers charge the highest prices, followedby brick-and-click retailers, and pure-play retailers havethe lowest prices.With regard to posted prices, traditional retailers chargethe highest prices, followed by multichannel retailersand pure play retailers. However, when shipping costsare included, multichannel retailers charge the highest,followed by pure-play retailers and traditional retailers.Price dispersion is persistent with time. However,product life cycle leads to changes in number ofcompeting firms and the range of price dispersion.Use “certified” versus noncertified merchants as ameasure of good and poor service quality. Certifiedmerchant enjoys a premium but vanishes with increasednumber of certified competing firms.Manufacturers charge higher prices on their websitesand limit availability to avoid some retailers from freeriding on the sales efforts of other retailers.Price dispersion is persistent with time, and decreaseswith increase in competitionObserved online price dispersion is not due to servicedifferentiation.Significant portion of price dispersion is due tononretailer characteristics. Increase in level ofcompetition reduces prices charged by retailers at adiminishing rate. Influence of service quality is notconsistent across product categories.

Do market characteristics interact with retailercharacteristics to determine price levels?

l of Reta

aasacb(cr

R

etilactte

trut

E

FirtTrtcobtpsm

raiMvecaWw

aaBwvo

ahetifoffd

M

S

epfipei2ectaihlqrt

Ht

C

tpaD

R. Venkatesan et al. / Journa

cteristics. Especially since existing empirical evidence fromnalyses of online prices indicate only a weak influence ofervice quality—an important asset of retailers (Zeithaml etl. 2002). Considering the persistent online price dispersion,oupled with firms’ choice of strategic and tactical decisionsased on the competitive forces that exist in their marketsPorter 2001), we examine whether this supposed anomalyan be explained by considering the interaction between theetailer and market characteristics.

esearch question

The main objective of our paper is to study online retail-rs’ pricing strategies in a single framework that accounts forhe influence of—(a) retailer characteristics - service qual-ty and channels of transaction, (b) market characteristics -evel and nature of competition, and the price of a product,nd (c) interactions among these factors. Table 1 provides aomparison of our study with relevant literature. It is evidenthat the interaction between retailer and market characteris-ics and its effect on online prices has never been explicitlyxamined. This is the main focus of our study.

In remainder of the paper we use retailer in referenceo retailers with a web presence. They can be multichanneletailers, that is they may have a physical presence and/or alsose a catalog channel. Our hypotheses and analyses focus onhe online pricing strategies of these retailers.

xpected contributions

Our study intends to make two primary contributions.irst, we develop an integrated framework for studying the

nteractions between market and retailer characteristics. Foretailers selling nondifferentiable products, market competi-ion can vary widely across products and product categories.he differences in competitive forces make it feasible for the

etailer to make a strategic choice (level of service quality, andransaction channels) across all products, and vary the tacti-al decision (price level) in each product-market dependingn what competitive forces exist. Pricing is thus determinedy the interaction between the retailer’s strategic choices andhe market forces. Substantively, study of the interaction alsorovides a better understanding of the competitive effects ofervice quality, a relevant but sparsely researched issue inarketing.Second, we highlight the fact that the determinants of

etailer pricing decisions exist at two different levels ofbstraction. Retailer characteristics such as service qual-ty and transaction channels vary across individual retailers.

arket characteristics such as level and nature of competitionary across product-markets and are common to all retail-rs selling the particular product. Traditional linear models
annot accommodate simultaneous inclusion of the retailernd market characteristics and the interactions between them.e utilize a Hierarchical Linear Modeling (HLM) frame-ork which explicitly incorporates the main effects of retailer
Cs(S

iling 83 (3, 2007) 309–324 311

nd market level characteristics at different levels, and alsoccounts for the interactions between them (Raudenbush andryk 2002). This allows us to test not only whether, but alsohen and how in the context of various market forces, do ser-ice quality and channels of transaction influence a retailer’snline pricing decisions.

We first present the conceptual model for our analysesnd integrate the relevant literature to formulate five testableypotheses. We then describe our data collection from het-rogeneous sources reflecting information widely availableo the consumers and retailers alike and explain operational-zation of our constructs. Subsequently, we present the HLMramework which incorporates the factors at different levelsf abstraction, and follow it up with discussion of the resultsrom the analyses. We derive the managerial implicationsrom our results, and conclude with a discussion on potentialirections for future research.

Conceptual model and hypotheses

ain effects: influence of Internet retailer factors

ervice qualityVarian (2000) predicted that over time two classes of retail-

rs will emerge-low service/low price and high service/highrice. A study of the online retail market by Pan et al. (2002)nds that service quality does explain a retailer’s ability torice differentiate, albeit very little. In contrast, the theoreticalxpectations indicate service quality as an important factorn consumers’ choice of a retailer (Wolfinbarger and Gilly003). Betancourt and Gautschi (1993) suggest that retail-rs provide services in order to reduce the customer’s totalost of consumption and that these services play an impor-ant role in customers’ selection of a retailer. They present annalytical model which shows that between two retailers sell-ng the same product, a consumer would be willing to pay aigher price for the higher service quality retailer, but only asong as the consumer’s valuation of the difference in serviceuality is more than the difference in prices between the twoetailers. Based on theoretical expectations we hypothesizehat:

1. The higher the service quality of a retailer the higherhe prices charged.

hannels of transactionsEmpirical evidence from several years of data indicates

hat multichannel retailers have higher price levels than pure-lay (click only) retailers and brick-only retailers (Ancaranind Shankar 2004; Pan et al. 2003a; Ratchford et al. 2003).ue to their wide geographic presence, National Brick-and-
lick retailers can provide customers value added services
uch as the ability to order products online and pick upor in some cases return) the products in an offline store.uch services save time and provide added ease of use to

3 l of Ret

cato(alBbfi

Hp

mtpbweagaaasr

fintTtrpcip(dal

Hc

M

atortm

om

I

mttstioRu(

wesoepWbspwdawhlprtqu

Hcr

I

ciirhv(h


onsumers. These facilities, which we term as fulfillmentnd post-fulfillment facilities, lead to greater consumer trusthereby reducing the price sensitivity of consumers. On thether hand, the single distribution channel for a pure playInternet or click only) retailer, lowers the inventory costsnd can be potentially leveraged to price differentiate by costeadership. Consistent with prior studies, we expect Nationalrick-and-Click retailers to price higher because of betterrand recognition, greater trust, and provision of better ful-llment and post-fulfillment facilities. Therefore,

2a. National Brick-and-Click retailers charge higherrices than pure-play retailers.

Local Brick-and-Click retailers can provide better fulfill-ent and post-fulfillment facilities albeit in a local area and

hus have the potential to charge higher prices relative toure play retailers. Since they are expected to have lowerrand recognition than National Brick-and-Click retailers,e expect their price levels to be lower than those of retail-

rs with national presence. Local Brick-and-Click retailersim to utilize their online presence to extend their geo-raphic reach beyond that of the traditional store. This wouldttract additional consumers without incurring significantdditional costs. Based on valid expectations for both highernd lower prices, we do not specify any directional hypothe-es related to the pricing strategy of Local Brick-and-Clicketailers.

Online retailers that provide direct mail catalogs bene-t from lower costs because direct mail catalog companieseed little modification to include online ordering capabili-ies into their existing business models (Doolin et al. 2003).his addition of online ordering improves the convenience

hey provide to customer and can help the direct mail catalogetailer to differentiate and charge higher prices. Similarly,ure play retailers are adding direct mail catalogs to theirhannel portfolio because—(a) they can leverage existingnfrastructure (distribution centers, customer service and tele-hone ordering) to support their direct mail catalog channelQuick 2000), and (b) online retail managers believe thatirect mail catalogs enables them to build brand recognitionnd increase trust among consumers (Quick 2000), therebyeading to higher prices. Hence, we hypothesize that:

2b. Online retailers that provide direct mail catalogsharge higher prices than pure-play retailers.

oderating effects

The influence of retailer characteristics (service qualitynd channels of transaction) on online prices is expectedo be moderated by market characteristics such as—level
f competition, nature of competition. The main effects ofetailer characteristics explain how, within a product market,he prices vary across retailers. The moderating effects of
arket characteristics capture the differences in the influence

iufe

ailing 83 (3, 2007) 309–324

f the retailer characteristics on online prices across productarkets.

nteraction of level of competition and service qualityCohen (2000a) presents the concept of Distortion in Infor-

ation Function (DIF-ness) to measure consumer’s tendencyo use heuristics when faced with large amount of informa-ion. DIF-ness suggests that till a certain threshold consumerscan the increasing number of choices and make a choicehat maximizes their utility. Beyond this threshold, increasen alternatives can lead the consumers to use heuristics basedn brand name and service quality, to make their selection.esearch in signaling theory suggests that consumers oftense price as a proxy for quality of product and/or retailerKirmani and Rao 2000).

Considering these two aspects, we expect that when facedith small number of competitors in a product-market, retail-

rs with high service quality can easily differentiate theirervices and charge a higher price. Increase in the numberf competitors (DIF-ness) would initially force the retail-rs (especially higher service quality retailers) to lower theirrices in order to avoid losing consumers to the competitors.hen the number of competitors exceeds a certain threshold,

oth DIF-ness and signaling theory suggest that the con-umers’ use of heuristics would counteract the downwardressure on prices created by increased competition. Thus,e propose that the number of competitors in a market wouldiminish the positive influence of service quality on prices;nd the moderating influence of the number of competitorsould be nonlinear in nature. With increase in competitionigh service quality retailers would charge higher prices thanow service quality retailers. However, the difference in therices of high and low service quality retailers is expected toeduce at a diminishing rate. We expect the level of competi-ion to capture the differences in the extent to which serviceuality influences the prices charged by retailers across prod-ct markets. We hypothesize that:

3. The positive influence of service quality on pricesharged by a retailer is diminished, albeit at a decreasingate, as the number of competitors increase.

nteraction of price level and service qualityThe price of a given product implicitly denotes the per-

eived risk and is an important determinant of consumers’nvolvement in a product (Cohen and Areni 1991). Accord-ng to Cohen (2000b), the consumers’ selection is based onisk averseness particularly when the price of a product isigh. This would lead them to prefer retailers with high ser-ice quality, who benefit from the consumers’ risk aversenessor willingness to pay higher for service quality) by chargingigher prices. The empirical evidence does indicate increase
n price dispersion with increase in average price of the prod-ct (Pratt et al. 1979). This increase in dispersion could stemrom the consumers’ willingness to pay higher prices to retail-rs who are able to reduce the perceived risk by inducing trust

l of Retailing 83 (3, 2007) 309–324 313

(lswrtwGrpa

Hco

Ic

bcifsttiGtq

steapaqtoh

Hcvp

pu

tccr

Ncfieatiped

HrN

paaimqocppo

i(similar to Pan et al. 2003b). The data for size of the retailer


Ba and Pavlou 2002). In product-markets at higher priceevels, retailers who are able to foster trust by way of betterervice quality are afforded scope for price differentiation andould charge relatively higher. Though for all products, the

etailers with high service quality will charge higher priceshan retailers with low service quality; their price premiumould increase with the price of a product (Betancourt andautschi 1993). We expect the price level of a product to

eflect the differences in the influence of service quality onrices charged by retailers across different product markets,nd propose:

4. The positive influence of service quality on pricesharged by retailers is enhanced further as the price levelf the product increases.

nteraction of nature of competition and retailerharacteristics

A differentiation strategy can be successful if—(1) theenefits provided by the differentiation are valued by theonsumer, and (2) the differentiation is unique or not eas-ly replicated by the competitors (Barney 1991). In a marketor nondifferentiable products, a retailer can offer higher con-umer benefits by providing high service quality. The abilityo charge a price premium for the higher service quality wouldhen depend on competition from retailers who provide sim-lar or higher levels of service to consumers (Betancourt andautschi 1993). We define one aspect of the nature of compe-

ition in a particular product market as the variation in serviceuality of the retailers selling that product.

As mentioned earlier, in general (for all products) the highervice quality retailers are expected to charge higher priceshan the low service quality retailers. However, the differ-nce in the prices is expected to be higher in markets withlarger variation in service quality among retailers selling aroduct (where possessing high service quality is a uniquesset) compared to markets with lower variation in serviceuality. We expect the variation in service quality to capturehe extent of difference in the influence of service qualityn prices charged by retailers across product markets. Weypothesize that:

5a. The positive influence of service quality on pricesharged by retailers is further enhanced when there is a highariation in service quality among the retailers selling theroduct.

We use number of National Brick-n-Click retailers as aroportion of total number retailers selling a particular prod-
ct to capture a second aspect of nature of competition.4
4 Our hypotheses regarding the moderating influence of market charac-eristics are drawn on the basis that online retailers use the easily availableompetitive information to formulate their pricing strategies. We do notharacterize competition in terms of the proportion of local brick and clicketailers and retailers with catalog because information on whether an online

w

rnBau

Fig. 1. Joint influence of retailer and market characteristics.

ational Brick-n-Click retailers have a higher trust withonsumers, and provide unique channel alternatives for ful-llment and post-fulfillment processes, and thus can bexpected to charge higher prices. However, this ability ofnational retailer to price higher is dependent on the compe-

ition from other National Brick-n-Click retailers. Increasen the number of National Brick-n-Click retailers selling aroduct would result in reduced ability to differentiate fromach other in terms of multichannel benefits and result iniminishing their price premiums. We hypothesize that:

5b. The premium charged by a National Brick-n-Clicketailer is diminished with an increase in the proportion ofational Brick-n-Click retailers that sell a product.

Data collection

The criteria used for identifying data to collect was: (a)roduct categories that include nondifferentiable products tovoid price variation because of product differentiation, (b)vailability of at least 20 products in each of the categories tondicate substantial penetration product in the online retail

arket, and (c) availability of a minimum of seven priceuotes for each of the selected products to ensure a base levelf competitive intensity. Using these criteria, eight productategories were selected-Books, Camcorders, DVDs, DVDlayers, PDAs, Printers, Scanners and Video Games. Table 2rovides the summary information of the data collected forperationalizing the constructs presented earlier in Fig. 1.

For each retailer we collected data for the service qual-ty ratings and the competitive intensity from Bizrate.com

as obtained from Alexa.com which provides rank of the

etailer is a local brick and click retailer or if a retailer has also has a catalog isot easily evident online. However, we would expect consumers and Nationalrick-and-Click retailers to be aware of the identity of other National Brick-nd-Click retailers through the competitive intelligence surveys that aresually undertaken in firms of this size.

314 R. Venkatesan et al. / Journal of Retailing 83 (3, 2007) 309–324

Table 2Constructs operationalized from multiple sources

Construct Data collected Source Related research using similar measures

Price Posted price for a product. Used to calculatethe price index.

Bizrate.com Pan et al. (2002), Clay et al. (2002)

Service quality Survey ratings obtained by Bizrate fromonline consumers over ten items on a10-point scale (1 = Poor, 10 = Outstanding

Bizrate.com Pan et al. (2002)

Transactional channels Dummy coded variables for characterizingthe channels through which the retaileroffers the products. The channels are firstclassified as one of the following: pure play(online only), national chain with onlinepresence, and local store(s) with onlinepresence. In addition, we also distinguishretailers that offer mail-order catalog.

Inspection of RetailerWebsites

Tang and Xing (2002)

Size Rank of retailers based on number of uniquevisitors to the online.

Alexa.com Pan et al. (2003b)

Competitive intensity Number of retailers offering an identicalproduct and with service quality ratingsavailable from Bizrate.

Bizrate.com Pan et al. (2003b)

Price level Average posted price for a product. Bizrate.com Cohen and Areni (1991), Pan et al. (2003b)Variance in service quality Coefficient of variation of the service quality

among retailers Selling a productBizrate.com

P s Inspection of RetailerWebsites

rrtalr

•

•

•

•

B

fqrv

t

w

t

Table 3Bias reduction reduces analyzable data

Productcategory

# Posted-prices # Retailers # Products

Collected Analyzed Collected Analyzed

Books 5750 2752 19 9 685Camcorder 1386 882 86 57 57DVD 9242 5738 30 15 799DVD Player 547 446 66 51 40PDA 568 479 77 57 32PSV

sbnstderof the bias reduction on the size of the final data set, yielding

roportion of NationalBrick-n-Click relations

Number of National Brick-n-Click retailerselling a product.

etailer website based on the number of unique visitors. Theetailer’s transaction channels were coded based on inspec-ion of retailer websites cross-verified with their descriptiont Bizrate.com and web21.com5 (Tang and Xing 2002). Usingocation and geographical spread of physical store(s) theetailers were classified as:

National Brick-n-Click: Presence of physical stores inmore than one state.6 (Example: Best Buy with presenceBestBuy.com)Local Brick-n-Click: physical store locations within onestate. (Example: 17th Street Photo from New York withonline presence as 17photo.com)Pure-Play: no physical stores present. (Example: buy.comand amazon.com)Catalog provider: if mail-order catalog services areoffered. (Example: Sears)

ias reduction

We collected 22,209 price quotes, for 1,880 productsrom 233 retailers. For our analyses we only used prices
uoted for items identified as “new” by retailers other thanefurb/discounters. We excluded retailers with missing ser-ice quality ratings.7 Though most retailers provide the price
5 web21.com provides a comprehensive directory listing of retailers andheir physical store locations (if any).

6 In rare cases where the website had insufficient information, the retaileras directly contact for the information.7 Bizrate requires a minimum of 30 customer surveys over last ninety days

o publish ratings information. The exclusion of retailers without service

at

ri

vbrn

rinter 1087 906 75 54 36canner 708 574 77 53 31ideo games 2921 1616 74 42 200

earch engines with product information such as—new, refur-ished, the retailers primarily offering refurbished items tendot to do so. Refurbished and used products are generallyold at lower prices since they differ from new products inerms of product quality; thereby leading to a bias in the priceispersion measure. We conducted a manual verification ofach retailer’s website to identify refurb/discounters,8 andemoved them from our analyses. Table 3 shows the impact

total of 13,393 price points suitable for analysis. To assesshe qualitative impact of the bias reduction process we exam-

atings is methodologically necessary because service rating is a criticalndependent variable in our analysis.

8 The classification of “refurb/discount” retailers was also independentlyerified. Note that the presence of refurbishers can be a significant source ofias in the price information. It can certainly increase price dispersion sinceefurbishers are expected to price lower. Yet, to the best of our knowledge,o prior study has explicitly accounted for their presence in the data.

l of Retailing 83 (3, 2007) 309–324 315

ianbc

C

prraf

P

wkp

M

fsivPLSDmotdqtcet1a

to

bstcccBo

Cam

cord

erD

VD

play

erPD

APr

inte

rSc

anne

rB

ooks

DV

DV

ideo

gam

e

0.23

(0.2

1)0.

27(0

.25)

0.20

(0.2

0)0.

22(0

.18)

0.25

(0.2

4)0.

25(0

.27)

0.22

(0.2

5)0.

50(0

.96)

7.8

(1.4

0)8.

1(0

.90)

8.3

(0.6

9)8.

3(0

.74)

8.3

(0.6

7)8.

2(0

.94)

8.4

(0.7

5)8.

2(0

.71)

277

(465

)33

4(5

31)

108

(315

)13

9(3

40)

154

(379

)24

(148

)44

(106

)19

(45)

963

(671

)29

0(2

42)

354

(170

)26

2(1

54)

340

(462

)15

(11)

30(3

0)32

(11)

15(9

.8)

11(7

.6)

15(9

.2)

25(9

.3)

19(7

.9)

4(0

.9)

7(1

.4)

8(4

.5)

0.26

(0.2

2)0.

11(0

.05)

0.12

(0.0

3)0.

10(0

.02)

0.13

(0.0

6)0.

19(0

.03)

0.20

(0.1

0)0.

10(0

.05)

42

44

41

321

1012

78

72

24

4337

4642

426

1017

5751

5754

539

1542

ion;

**va

lues

are

in1,

000s

,and

repr

esen

tthe

rank

ofa

reta

iler.

Soa

low

erra

nkim

plie

sa

larg

erre

taile

r.


ned the dispersion of prices in the collected data and thenalyzed data and found no significant difference. Althougho significant difference was observed between the data, weelieve that the bias reduction provides for more reliable andredible parameter estimates in our analyses.

alculation of price index

To account for differences in price levels across differentroducts within a category, we index the price charged by aetailer for a product.9 The price index is calculated as theatio of the difference between the price charged by a retailernd the minimum price for the product to the minimum priceor the product (Eq. (1)):

INDi,j = PINDi,j − MinPricej

MinPricej

(1)

here, PINDi,j = Price index for retailer i for product mar-et j; MinPricej = Minimum price charged by any retailer forroduct market j.

easuring service quality

Bizrate.com provides consumers with fourteen measuresor service quality of a retailer. These are single-item mea-ures on a scale of 1 to 10, where 1 is very poor and 10s outstanding. Ten of these items measure the level of ser-ice quality of a retailer and include—Ease of Finding aroduct, Product Selection, Clarity of Product Information,ook and Design of the website, Shipping Options, Chargestated Clearly, Product Availability, Order Tracking, On-timeelivery, and Customer Support. We exclude four items thateasure the intentions of the shopper to revisit the website, an

verall rating of the retailer, whether the product met expec-ations, and shipping charges, since these four items do notirectly measure the service quality of a retailer. The serviceuality measure for a retailer i (Servindi) was computed ashe average of the ten items from Bizrate.com. Such a simpleomposite measure of individual items that measure differ-nt aspects of a construct has been widely used for measuringhe market orientation of an organization (Jaworski and Kohli993), and the service orientation of a retailer (Homburg et
l. 2002).
We conducted a principal component factor analysis withhe ten items and found that all the ten items loaded intone factor that explained approximately ninety percent of the

9 We do not include shipping and handling as part of the prices chargedy the retailer for three reasons: (1) for the product categories used in ourtudy, shipping and handling represents only 2 percent of the total price, andhe correlation between the prices charged by a retailer with and without theharges is greater than 0.98 for all the product categories, (2) each retailerharges differently for different shipping options and thereby making directomparison difficult, additionally the default shipping options provided inizrate.com vary from retailer to retailer, and (3) approximately 20 percentf the retailers do not report shipping and handling charges on Bizrate.com. Ta

ble

4D

escr

iptiv

est

atis

tics*

Prod

uctc

ateg

ory

cons

truc

t

Pric

ele

vel

Serv

ice

qual

itySi

ze**

Pric

ele

vel

Num

ber

ofre

taile

rsV

aria

tion

inse

rvic

equ

ality

Ret

aile

rty

peN

atio

nal

Loc

alC

lick

Tota

l

*V

alue

sin

pare

nthe

ses

repr

esen

tsta

ndar

dde

viat


for the H

vtmTeit

M

ytioiapppP

soSaeama

e1taaat

fc

puttapltalrLo

P

ei

P

wiii

ccvvce

β

w

Fig. 2. Data structure used

ariance.10 We also observed very high correlation betweenhe service quality obtained from the factor analysis and the

easure obtained as the average of the ten individual items.he substantive inferences of the study did not change withither of the service quality measure. To maintain parsimonyn our model, we thus use one measure for service quality ofhe retailer as the average of the ten items.

arket description

The descriptive statistics for the variables used in our anal-ses are provided in Table 4. From Table 4 we note that—(1)he mean of number of retailers in individual product-marketss lesser for Books, DVDs, and Videos Games as compared tother product categories. While the number of retailers offer-ng books, DVDs, and video games is lower, these retailersre on average substantially larger than the retailers in otherroduct categories, (2) the price dispersion varies from 18ercent (for printers) to 96 percent (for video games). Theserice dispersion levels are comparable to those reported byan et al. (2002).

Hierarchical linear model framework

Fig. 2 illustrates the structure of the data used in thistudy. As represented in Fig. 2, the variables used to testur hypotheses correspond to different levels of aggregation.pecifically, retailer characteristics, that is the main effects,re at the lowest level of aggregation and are measured forach retailer. The market characteristics, that is the moder-ting effects, are at a higher level of aggregation and areeasured for each product market. Such data are designated

s multilevel data (Raudenbush and Bryk 2002).The above data structure suggests that a Hierarchical Lin-

ar Modeling (Raudenbush and Bryk 2002; Steenkamp et al.999) framework should naturally accommodate our concep-ual model. In HLM, a linear regression model is specifiedt the lowest level of aggregation (Level 1). The interceptnd slope coefficients for the model at Level 1 are modeled
s a function of the variables at the next level of aggrega-ion (Level 2). We are interested in explaining variation in
10 The cronbach alpha measure of scale reliability for the ten items was alsoound to be 0.92 which is widely acknowledged as an acceptable reliabilityoefficient (Nunnaly 1978).

mVi

β

ierarchical Linear Model.

rices charged by different retailers for a particular prod-ct. The prices charged by the retailer are a function of—(a)he characteristics of each retailer such as service quality,he channels of transaction that are available to the retailer,nd the size of the retailer, (b) the characteristics of eachroduct-market such as number of competitors, average priceevels and nature of competition, and (c) the interactions ofhe retailer and the market characteristics. The retailer pricesnd characteristics that we are interested in are at the lowestevel of aggregation and hence form the basis for the Level 1egression model. The intercept and slope coefficients of theevel 1 model (the retailer level) are modeled as a functionf the Level 2 variables (market characteristics).

roposed HLM model

In the Level 1 model, the price index of a retailer is influ-nced by the various hypothesized retailer characteristics ands given by;

INDij = β0j + βij × SQi + β2j × NBCi + β3

× LBCi + β4 × Ci + β5 × Sizei + rij (2)

here SQi = service quality of retailer I; NBCi = 1 if retaileris National Brick-n-Click, 0 otherwise; LBCi = 1 if retaileris Local Brick-n-Click, 0 otherwise; Ci = 1 if online retailerprovides a mail-order catalog, 0 otherwise.

In the Level 1 model, the intercept term (β0j), the slopeoefficients of service quality (β1j), and the slope coeffi-ient for the indicator of National Brick-n-Click retailer (β2j)ary across product markets. The Level 2 models explain theariation across product markets in the intercept and slopeoefficients of the Level 1 model. The intercept term is mod-led as,

0j = γ00 + γ01 × LCMj + γ02 × VSQj + γ03

× PLj + γ04 × PNBCj + u0j (3)

here LCMj = Log of Number of competitors in product-arket j, PLj = Price level of product-market j, andSQ = Variance in the service quality among the retailers
j
n product-market j.The slope of service index is modeled as;

1j = γ10 + γ11 × LCMj + γ12 × VSQj

+ γ13 × PLj + u1j (4)

l of Reta

β

wr

P

oocwtdtu

mβ

mOpvotd2RmHme

A

if

tS

(med

db

H

paE

P

H

acI

P

H

oieMidH

P

M


The slope of National Brick-n-Click is modeled as;

2j = γ20 + γ21 × PNBCj + u2j (5)

here, PNBCj = Proportion of National Brick and Clicketailers in product-market j.

Substituting (3)–(5) in (2) we get,

INDij = γ00 + γ01 × LCMj + γ02 × VSQj + γ03

× PLj + γ10 × SQi + γ20 × NBCi

+ β3 × LBCi + β4 × Ci + β5 × Sizei + γ11

× [LCMi × SQi] + γ12 × [VSQj × SQi]

+ γ13 × [PLj × SQi] + γ21 × [PNBCj × NBCi]

+[u1j × SQi + u2j × NBCi] + [u0j] + rij (6)

To test the proposed nonlinear relationship between levelf competition and price premiums, we use a log conversionf ‘number of competitors’. γ00 represents the average priceharged by the retailers. rij is a homoscedastic error termith mean 0 and variance σ2

r and represents an estimate ofhe unexplained variance at the retailer level. u0j is a ran-om parameter with mean 0 and variance σ2

uj and measureshe deviation of product market j from the mean, that is thenexplained variance in prices at the product-market level.11

A Full Information Maximum Likelihood (FIML)12

ethodology is used to estimate the model parameters (γ’s,’s, and u’s) with an iterative algorithm (details on the esti-ation are provided by Raudenbush and Bryk 2002). Usingrdinary Least Squares (OLS) regression on multilevel dataresents statistical problems because—(a) the multilevel dataiolates the assumption of independence (or homogeneity)f observations required in OLS, (b) it has been shown thathe resulting estimates are biased, and (c) the estimated stan-ard errors of the effects are too small (Raudenbush and Bryk002). These smaller estimated errors in fact lead to improved-square values and also increase the significance of the esti-ated coefficients. It is important to note that by adopting theLM approach we are not only estimating a more complexodel, but also raising the requirements on robustness of the

stimates from our model.

nalysis of pricing strategies

In addition to testing the hypotheses, we are also interestedn evaluating if the interactions between retailer and marketactors provide a significant improvement in explaining price

11 One can also interpret the random effects parameter uj, as a heteroscedas-ic error term in the mixed heteroscdastic models used by Wittink (1977) andhankar and Krishnamurthi (1996).

12 We also estimated our models using a Restricted Maximum LikelihoodREML) method because when either the number of retailers per productarket or the number of total prices per product category is small, the FIML

stimates can have smaller standard errors. We did not find any substantiveifferences in the REML and FIML estimates in our data.

teWwitotHp

iling 83 (3, 2007) 309–324 317

ispersion. We accomplish this objective by estimating threeenchmark HLM Null models that are explained below.

LM Null Model I

In HLM Null Model I, the price index of a retailer i inroduct-market j (PINDij), is specified as a function of onlyn intercept term that varies across product markets. Usingq. (6) as the basis, HLM Null Model I is given by:

INDij = γ00 + u0j + rij. (7)

LM Null Model II

The HLM Null Model II assumes that the price index ofretailer is impacted only by the main effects of retailer

haracteristics. Using Eq. (6) as the basis, HLM Null ModelI is provided by:

INDij = γ00 + γ10 × SQi + γ20 × NBCi + β3 × LBCi

+ β4 × Ci + β5 × Sizei + [u1j × SQi + u2j

× NBCi] + [u0j] + rij. (8)

LM Null Model III

The HLM Null Model III assumes that the main effectsf both retailer and market characteristics influence the pricendex of a retailer. Even though it considers only the mainffects of retailer and market characteristics, the HLM Nullodel III differs from extant models in the literature because

t accommodates for the retailer and market characteristics atifferent levels of aggregation. Using Eq. (6) as the basis, theLM Null Model III is represented as;

INDij = γ00 + γ01 × LCMj + γ02 × VSQj + γ03

× PLj + γ04 × PNBCj + γ10 × SQi + γ20

× NBCi + β3 × LBCi + β4 × Ci + β5 × Sizei

+ [u1j × SQi + u2j × NBCi] + [u0j] + rij (9)

Results and discussion

odel comparison

When FIML is used for model estimation, a likelihood ratio (LR)est is appropriate for evaluating the significance of multiple fixedffects in the nested HLM models (Raudenbush and Bryk 2002).e evaluate the significance of the main effects of retailer factorsith a LR test that compares HLM Null Models I and II. The signif-

cance of the main effects of market factors is established by the LR
est between HLM Null Models II and III. Finally, the significancef the interactions between retailer and market factors is obtainedhrough a LR test between HLM Null Model III and the ProposedLM Model. We assess the explanatory power of a model by com-uting the observed variance in prices (Vp) with the variance of


Table 5aComparison of reduction in price dispersion

Camcorder(n = 882)

DVD player(n = 547)

PDA(n = 479)

Printer(n = 906)

Scanner(n = 574)

Books(n = 2752)

DVDs(n = 5738)

Video Games(n = 1616)

Likelihood valuesProposed HLM Model 503 25 430 525 92 141 2318 2935HLM Null Model 3 493 16 424 498 83 134 2270 2925HLM Null Model 2 484 13 419 481 72 97 2243 2877HLM Null Model 1 400 4 405 425 60 55 2152 2840

Likelihood ratioInteraction effectsa 19 18 12 54 18 14 96 20Main effects-market factorsb 19 6* 10** 34 22 82 54 96Main effects-retailer factorsc 167 18 28 112 24 84 182 74

R-squareProposed HLM Model 0.42 0.42 0.56 0.35 0.44 0.67 0.59 0.70HLM Null Model 3 0.34 0.35 0.48 0.21 0.37 0.62 0.54 0.62HLM Null Model 2 0.34 0.32 0.41 0.18 0.35 0.49 0.41 0.61HLM Null Model 1 0.18 0.29 0.33 0.16 0.23 0.40 0.36 0.63Simple Linear Model 0.27 0.29 0.05 0.11 0.08 0.34 0.36 0.62

* ** ikelihoo

tcc&MaiBr

Ncopppn(eppcvMtHg

soHtM

×

M

mT

S

phusisfd

C

bnpoiai

ref

Significant at α < 0.10, significant at α < 0.05, and all other cells for the la LR test between HLM Full Model and Null Model III.b LR test between HLM Null Models III and II.c LR test between HLM Null Models II and I.

he residuals from the corresponding model (Vr) for each productategory. The explanatory power is assessed by an R2 measure cal-ulated as [1 − (Vr/Vp)]. In addition to the HLM Null Models (I, II,

III) we also compare the Proposed Model with a Simple Linearodel incorporating only retailer characteristics (similar to Pan et

l. 2002). The variables in Simple Linear Model are service qual-ty, channels of transaction—National Brick-n-Click retailer, Localrick-n-Click retailer, existence of mail-order catalog, and size of a

etailer.13

The LR test between the Proposed HLM Model and the HLMull Model III is significant at α < 0.01 for all the eight product

ategories (Table 5a); providing strong support for the influencef the interactions between retailer and market factors on retailerrices. For six of the eight product categories we find strong sup-ort (α < 0.01) for the main effects of market factors on retailerrices. The main effects of market factors are only marginally sig-ificant for the DVD Player (α < 0.10) and PDA product categoriesα < 0.05). Finally, for all the eight product categories, the mainffects of retailer factors have a significant influence on retailerrices. We see that the Proposed HLM Model of price dispersionrovides a higher explanatory than all of the Null Models. Specifi-ally, on an average the Proposed Model explains 52 percent of theariation in retailer prices. In comparison, on average the HLM Nullodels I, II, and III approximately explain 32, 39 and 44 percent of

he variation in retailer prices, respectively. The R2 for the ProposedLM Model ranges from 21 percent (printers) to 70 percent (videoames).

Overall, the better fit between the model and the online markettructure is reflected by average improvement of 25 percent in R2

ver the empirical evidence in existing literature. The suitability of
LM structure is also evident in ability of the HLM Null Model 1
o explain greater degree of price dispersion than the Simple Linearodel.

13 The Simple Linear Model is provided by PINDij = β0 + β1 × SQi + β2

NBCi + β3 × LBCi + β4 × Ci + �5 × Size + rij.

twtatb

d ratio tests are significant at � < 0.01.

ain effects-influence of retailer characteristics

Table 5b provides the results from the estimation of the HLModel with all covariates (labeled as Proposed HLM Model inable 5a).

ervice qualityFor all the eight product categories the statistically significant

ositive coefficient for service quality indicates that retailers withigher service quality are also charging higher prices. Across prod-ct categories we find consistent evidence that retailers considerervice quality as a means for differentiating and thereby charg-ng higher prices (supporting H1). We believe that this consistentlyignificant influence of service quality is observed because ourramework accounts for retailer and market characteristics at twoifferent levels, and also incorporates their interactions.

hannels of transactionIn our analyses, a retailer is classified as pure play, National

rick-n-click, Local Brick-n-Click, and catalog. All of the chan-el classifications are used in the analyses with the exception ofure play, which is used as the base. The coefficients for eachf the retailer classifications are thus interpreted as the averagencrease/decrease in the prices relative to a pure play retailer. Over-ll, we observe that the retailer channel structure has a significantnfluence on the prices charged by retailers.

Our results support H2a. We find that National Brick-n-Clicketailers charge significantly higher prices than the pure play retail-rs in all of the eight product categories. The increase in price levelsor National Brick-n-Click retailers ranges from 6 percent (printers)o 39 percent (DVD players). If the National Brick-n-Click retailers
ere to become cost leaders and price their products in relation to
heir economies of scale, one would expect to observe lower pricesnd not the higher prices that our results indicate. We believe thathe National Brick-n-Click retailers are able to charge higher pricesecause—(1) they are able to engender trust among online shoppers

R. Venkatesan et al. / Journal of Retailing 83 (3, 2007) 309–324 319

Table 5bResults from estimation of hierarchical linear model

Camcorder(n = 882)

DVDplayer(n = 547)

PDA(n = 479)

Printer(n = 906)

Scanner(n = 574)

Books(n = 2752)

DVDs(n = 5738)

Videogames(n = 1616)

Product-market factorsLog of number of competitors 0.015 0.60** 0.13** −0.16* 0.03 0.07** −0.40** 1.65***

Variance in service quality −0.03* 0.03 0.41 −0.13 0.36* −0.46* 1.53** 0.43Price level 3.3E-5 −5.8E-3 −7.9E-6 −1.2E-4 −8E-6* −2E-3** −0.015*** −0.11Proportion-National Brick-n-Click 1.3E-05 1.70*** 0.24 0.58 3.0E-6 0.33*** 0.062 0.50

Retailer factorsService quality 0.01** 0.20** 0.02* 0.04*** 0.08*** 0.008* 0.030*** 0.15***

National Brick-n-Click 0.16*** 0.39*** 0.09*** 0.06* 0.08* 0.33*** 0.062*** 0.17*

Local Brick-n-Click 0.02 −0.03 −0.05*** −0.021 −0.02 0.25*** −0.09 0.07Catalog 0.10*** 0.12 −0.04** 0.048*** −0.02 0.05*** −0.02* −0.08Size −33.7*** −34.1*** −5.7* −3.1 −2.3 −5.8** −2.4*** −1.3**

Interaction effectsPrice level × service quality 8.2E-6 4.2E-5 −1E-4 −9.2E-4 −2.0E-5 2E-3* 1.7E-3*** 0.012Log of number of

competitors × service quality6.7E-3 −0.07*** −0.03* −0.025** −0.02*** −0.03*** −0.041*** −0.22***

Variance in servicequality × service quality

0.04*** 0.27*** 0.29*** 1.81** 0.22** 0.07 0.43*** 1.41**

Proportion of National −1.03** −4.71*** −0.48*** −0.35 0.15 −0.13*** −0.59** −0.26*

*

gpatisaaarS

i(tcIltBbatsra

tu

rsi

5lDaranspBlr(acr

M

mcir

Brick-n-Click × NationalBrick-n-Click

Significant at α < 0.10, **significant at α < 0.05, ***significant at α < 0.01.

iven their national presence and brand recognition, and (2) appro-riate use of technology enables these retailers to provide shoppersdditional convenience in terms of being able to switch channels ofransaction from pre-ordering to post-fulfillment, for example order-ng online and taking delivery offline at a nearby store. Arguably,uch conveniences are part of a higher level of service quality thatretailer offers and the national retailers are clearly able to chargepremium for these additional services.14 These findings provide

dditional empirical support to previous findings that multichanneletailers charge higher prices than pure-play retailers (Ancarani andhankar 2004).

The Local Brick-n-Click retailers are observed to be—(a) charg-ng significantly lower prices (than pure play retailers) in the PDA5 percent lower) product category, (b) significantly higher prices inhe books product category (25 percent higher), and (c) no signifi-ant difference in their price levels for the remaining 6 categories.n summary, we find that for a majority of product categories ana-yzed, Local Brick-n-Click retailers are not charging a premium forhe facilities that their multichannel operation provides. The Localrick-n-Click retailers seem to be using their Internet operations foretter geographic reach, as there are minimal marginal costs in oper-ting an Internet retail store. In this regard, it should also be notedhat for customers outside of the geographic reach of the physicaltore, the Local Brick-and-Click retailer is equivalent to a pure-playetailer because none of the advantages of multichannel transactions
re available.
Finally, retailers providing mail-order catalogs in addition toheir website are found to be charging higher prices for three prod-ct categories (camcorder - 10 percent higher, printers and books -

14 Note that in addition to the price levels all National Brick-and-Clicketailers also charge the sales taxes in almost all states. Consequently, con-umers are de-facto paying a premium significantly higher than that indicatedn our results.

L

ephitr

percent higher). However, for two product categories they chargeower prices than pure play retailers (PDA - 4 percent lower, andVDs - 2 percent lower). These results indicate that providing andditional channel of transaction (such as a catalog) does influenceetailer pricing strategies, but the effect of providing a catalog inddition to a website varies by product category. Therefore, we doot find support for H2b. In summary, our results provide conclu-ive evidence that National Brick-n-Click retailers charge higherrices than pure-play retailers, while the same is not true for Localrick-n-Click retailers and online retailers with mail-order cata-

ogs. Our results imply that while previous research has classifiedetailers in the online markets as either pure-play or brick-and-clickPan et al. 2002; Tang and Xing 2002), there is finer distinctionmong brick-and-click retailers, with only the national brick-and-lick retailers consistently charging higher prices than pure-playetailers.

oderating influence of market characteristics

Given that the interaction between retailer characteristics andarket characteristics are significant for a majority of the product

ategories, we do not discuss the individual main effects of thenfluence of market characteristics on the pricing strategies of theetailers.

evel of competition and service qualityIn seven of the eight product categories (all product categories

xcept camcorders) the coefficient of the interaction between com-
etitive intensity and service quality is negative and significant,ence supporting H3. The significance of coefficient indicates that anncrease in the level of competition in a product market diminisheshe positive impact that service quality has on the price charged by aetailer. However, the negative influence of level of competition has

3 l of Ret

di

P

bhqnhtDfisroffpvfl

Nq

bucwpwIip

Nn

erpnoprvcp

T

oahigat

cptttsbq

�

gvpracitFP

c(nt1lclq

I

oafccu(Asaaanull hypothesis that the pooling is appropriate (for α = 0.01). Wefind that the coefficients of pure-play only retailers are similar tothe brick-n-click retailers, which implies that the factors outlinedin our study impact the pricing decisions of retailers irrespective oftheir channel structure.


iminishing returns, indicating a weakening strength of the decreasen the price charged with increasing number of competitors.

rice level and service qualityWe hypothesized that the consumers’ selection of a retailer is

ased on risk averseness particularly when the price of a product isigh. This would lead consumers to prefer retailers with high serviceuality, who would in turn benefit from the consumers’ risk averse-ess (or willingness to pay higher for service quality) by chargingigher prices. The results indicate a positive and significant rela-ionship between price level and service quality for only Books andVDs. Therefore, for six of the eight product categories we do notnd support for H4. Similar to Pan et al. (2003b) we do not find atrong support for the moderating influence of price levels on theelationship between service quality and prices. The weak influencef price levels on the premium charged by retailers could be thator products with higher price levels, consumers can search onlineor product information with ease. Once the consumers decide on aroduct to purchase, they make a choice of the retailer based on thearious characteristics of the retailers offering the product. There-ore, we are unlikely to observe a moderating influence of priceevels in our analyses.

ature of competition: variance in service quality and serviceuality

We find significant support for a positive interaction effectetween variance in service quality of retailers selling the prod-ct and the service quality of a retailer (H5a) for seven productategories (all product categories except books). The retailersho provide a high level of service quality further increase therice levels with increase in scope for service differentiation,hich is typically marked by presence of low quality retailers.

n such product-markets with high variation in service qual-ty it is easier for high service quality retailers to justify theirremiums.

ature of competition: proportion of national retailers andational retailer

For six of the eight product categories (camcorders, DVD play-rs, PDA, Books, DVDs, and Video Games), national retailerseduce their prices as the proportion of national retailers in theroduct-market increases (H5b). Our results show that while theational retailers charge higher prices than others, they do sonly as long as competition from other national retailers in theroduct-market is low. An increase in the number of nationaletailers in the product-market diminishes the scope for ser-ice differentiation for national retailers in terms of multichannelharacteristics, thereby exerting a downward pressure on theirrices.

otal effect of service quality

An important objective of our study is to evaluate the effectf service quality of a retailer on prices charged. We propose thatfter accounting for market characteristics service quality would
ave a positive influence on the prices charged. In order to makenferences regarding the influence of service quality we investi-ate the total impact of service quality that includes both the mainnd the moderating effects. Evaluation of the total effects is par-icularly important because the moderating influences of market p
ailing 83 (3, 2007) 309–324

haracteristics are found to have opposing forces - level of com-etition is observed to negatively impact the relationship, whereashe nature of competition is found to positively impact the rela-ionship between retailer’s service quality and prices charged. Theotal effects of service quality measures the impact of a change inervice quality of a retailer on the prices charged, and is obtainedy taking the first derivative of Eq. (6) with respect to serviceuality;15

PIND

ServQi

= γ10 + γ11 × [Log(#of competitors)i] + γ12

× [VServindj] + [u1j] (11)

For each product category, we plotted the values of the derivativeiven by (11) for all the observations, that is for all the observedalues of level of competition and the nature of competition. Thelots indicate that the sign of the derivative is positive across theange of the data used in our study. We can safely conclude that forll product categories, and over a wide range of values of level ofompetition and nature of competition, service quality has a positivenfluence of the prices charged. The change in the derivative, overhe various values of level and nature of competition are shown inig. 3 for two representative product categories—PDAs and DVDlayers.

Fig. 3 shows that for both DVD Players and PDAs, the pricesharged by a retailer increases with the variation in service qualitynature of competition) and decreases at a diminishing rate with theumber of retailers (level of competition. Specifically, we observehat as the number of retailers offering a product increases beyond1 for DVD Players, and 20 for PDAs, the diminishing impact of theevel of competition on prices charged is minimized. Overall, thehange in prices charged by the retailer over the range of values forevel and nature of competition is positive for an increase in serviceuality.

ndependent analyses of pure-play and brick-and-click retailers

It can be argued that brick-and-click retailers have a larger setf constraints, such as price consistency between the offline storesnd Internet stores that would impact their pricing strategies. Theseactors are not accounted for in our analyses and can potentiallyonfound the observed results. To eliminate any possibility of suchonfounding effects, we separately estimated the model coefficientssing only the pure-play retailers and only brick-and-click retailersboth national and local). The results of our analyses are provided inppendix A. Table A1 shows that the results from the separate analy-

es are similar to the results obtained by pooling both pure-play onlynd brick-and-click retailers. We conducted a pooling test (Kumarnd Leone 1988) to examine whether data for pure-play retailersnd brick-n-click retailers can be combined. We could not reject the

15 For Books and DVD players we find a significant positive influence ofrice levels, which is also included in equation 10 for calculating total effects.

R. Venkatesan et al. / Journal of Retailing 83 (3, 2007) 309–324 321

ervice q

T

cbhn2adppkassbatdi(

flifrootims

M

fismif

TC

Fig. 3. Total effects of s

Implications, limitations and future research

heoretical implications

The results from this study contribute significantly to theurrent understanding of the sources of price differentiationy online retailers. Our study extends previous findings onow retailer characteristics such as service quality and chan-els of transaction (Pan et al. 2002; Ancarani and Shankar004) impact the prices. We also show how market char-cteristics such as number of competitors influence priceispersion in online markets. In particular, our proposed com-rehensive model explains greater amount of variance inrices and provides empirical evidence that retailer and mar-et characteristics interact to determine the prices charged byretailer. Our key findings include: (1) a positive influence ofervice quality on prices across several product categories; (2)imilar to prior findings (Pan et al. 2002), an increase in num-er of competitors induces a downward pressure in prices forll retailers, albeit at a decreasing rate; (3) for the first time in
he literature, we show how the increase in scope for serviceifferentiation is leveraged by high service quality retailers toncrease their prices particularly in face of high competition;4) interestingly we find that low service quality retailers in
iaae

able 6omparison of premiums over segments

uality on online prices.

act seek higher prices in markets characterized by either aarge number of competitors and low variation in service qual-ty or in markets with a large variance in service quality andew competitors, and (5) National Brick-and-Click retailerseduce their prices in face of increasing competition in form ofther National Brick-and-Click retailers. The findings fromur study pertain to retailer characteristics, market charac-eristics and their interactive effects, thereby highlighting themportance of incorporating the hierarchical structure of the

arket in a unified empirical model when studying pricingtrategies of online retailers.

anagerial implications

We expect that in a wide variety of product-markets ourndings will serve as benchmarks for retailer’s choices ofervice quality and channels of transaction. Such bench-arks are especially useful for established retailers who are

n the midst of adding new products to their portfolio, andor new entrants jointly deciding market entry, service qual-
ty and channel related investments and prices. To providen illustration of such a benchmark analysis, we conductedpost-hoc analysis with comparison of price levels across
ight product-market segments.


Table A1Results from estimation of Hierarchical Linear Model

Camcorder DVD player PDA Printer Scanner Books DVDs Video games

Pure-playProduct-market factors

Log (number of competitors) 0.012 0.43*** 0.18*** 0.14*** 0.07*** −0.10 −0.24* 1.03***

Variance in service quality −0.001 −0.05*** −0.57** −0.29 0.49 0.35* −1.21** −0.41Price level 4.5E-6 −5.4E-3 −1.8E-6 −3.2E-4 −4.1E-6 −6E-6* −0.007** −8.0E-2**

Retailer factorsService quality 0.07** 0.018** 0.015* 0.19*** 0.14** 0.006*** 0.09* 0.23*

Size −31.4** −35.1*** −2.6* −3.8 −3.2 −4.2 −2.2*** −1.9***

Interaction effectsPrice level × service quality 4.8E-6 5.5E-5 −2.0E-4 −1.0E-5 −7.2E-6 −1E-4 2E-4** −0.008Log (number ofcompetitors) × service quality

−5.1E-3 −0.04*** −0.03** −0.06*** −0.04*** −0.01*** −0.015* −0.11**

Variance in servicequality × service quality

0.05* 0.34*** 0.31*** 0.04* 0.13** 0.01 0.09* 1.11**

Brick-n-Click onlyProduct-market factors

Log (number of competitors) 0.14 0.96*** 0.04** 0.02** −0.14 0.18 0.02*** 0.06Variance in service quality −0.009 0.02 0.29 −0.59 0.24 0.02 1.9 0.62Price level 2.2E-5* −4.5E-4 −2.8E-6 −3.1E-4 −5E-6 −5.7E-3 −0.005 −9.2E-3**

Retailer factorsService Quality 0.05 0.31*** 0.07* 0.01*** 0.08** 0.01** 0.08*** 0.10**

Size −37.0 −36.1*** −6.7*** −2.7 −2.1 −6.1*** −1.2*** −1.1

Interaction effectsPrice level × service quality 7.1E-6 −3.7E-5 −9E-5 −3.7E-5 2.2E-5 1.3E-3 8.5E-3 −3.6E-3Log (number of competitors)× service quality

−7E-3 −0.10*** −0.028 −0.009* −0.01*** −0.06* −0.058** −0.28**

Variance in service quality −0.03* 0.21** 0.28*** 0.03 0.01 0.009 0.32** 1.96**

*

ceeuoT

p(o

1

2

cv

1

2

tqSg

× service quality

Significant at � < 0.10, **significant at � < 0.05, ***significant at � < 0.01.

We performed a median split of service quality, level ofompetition and variance in service quality to obtain two lev-ls (High and Low) for each of the three factors resulting inight segments. We then compared the mean price premi-ms charged by retailers across these segments. The resultsf this comparison (based on a MANOVA) are provided inable 6.

From Table 6 we obtain the following insights on how theremium of a retailer with a certain level of service qualityhigh, low) varies across market conditions (level and naturef competition).

. Low service quality retailer charges lowest premium inmarkets with low competition and little scope for servicedifferentiation (cell 4). Any change in intensity of compe-tition and/or availability for service differentiation (cells1, 2, and 3), yield increased premiums for the retailer.It needs to be emphasized that though the premiumsincrease, they are not different from each other, that islevel of premium is not significantly different across cells1, 2, and 3.

. In contrast, a high service quality retailer charges high-est premium in markets characterized by low competitionand low scope for differentiation (cell 8). Additionally,the high service quality retailer is able to seek similar

idta

high premiums in highly competitive markets with a highpotential for service differentiation (cell 5).

We obtain the following insights on how the premiumsharged between the high and low service quality retailersaries within same market conditions,

. High service quality retailer charges significantly higherthan the low service quality retailer when—(a) levelof competition and scope for differentiation are high(cell 5 vs. cell 1), and (b) both level of competitionand scope for service differentiation are low (cell 8 vs.cell4).

. Most importantly, under no market condition is the lowservice quality retailer able to charge significantly higherpremium than the higher service quality retailers.

The above analyses, clearly identifies the market condi-ions that are suitable for retailers with a given level of serviceuality and transaction channels for charging higher prices.uch an analysis maps out the decision space for retailers,iving them quantifiable measures of tradeoffs that are crit-
cal for making informed entry in a product market, pricingecisions and investment levels for strategic positioning inerms of service quality. Finally, we note that these insightsre based on post-hoc analyses of the results from our concep-

l of Reta

tb

L

sdcdttnnmiaiictcftitpcasoft

G

A

B

B

B

B

B

B

C

C

C

C

C

C

C

C

D

eH

J

K

K

L

M

NP

P

P


ual model and serve as guidelines for detailed examinationy future research.

imitations and future research

Current price dispersion research, which includes thistudy, is limited by not having access to supply side costata. Future research needs to address revenue measures thatonsider the actual transacted prices. It is possible for theispersion of transacted prices are made is different fromhe dispersion in retailer prices. Further, we believe that fac-ors such as product life cycle and timing of market entryeed to be analyzed (Pan et al. 2002, 2003b), implying theecessity of longitudinal data. Note that pricing is one of theany strategic decision criteria that a retailer has to manage

n order to optimize profits. For instance, Internet retailersre also interested in building a market share, and in ensur-ng consumer retention. Future studies should compare thenfluence of service quality across these different decisionriteria and investigate the process through which they leado optimal profits for retailers. Our results indicate channelhoice affects the pricing strategy of the online retailer. Inact we find that retailers providing multiple channels ofransaction also charge higher prices. Future research shouldnvestigate whether providing multiple channels of transac-ion also increases the revenues of the retailers, and whyroviding multiple channels of transaction affects consumerhoice and willingness to pay. Given that price dispersions a metric intrinsically signals the confluence of consumerearch, retailer pricing strategies and informational efficiencyf markets, we hope that our integrated approach will be use-ul to other researchers examining the interactions betweenhese competing forces in electronic markets.

Acknowledgements

The authors thank V. Kumar, Jim Marsden, and Alokupta for their helpful comments.

Appendix A. Separate Analysis of Pure-Play andBrick-And-Click Retailers

(See Table A1).

References

ncarani, Fabio and Venkatesh Shankar (2004). “Price levels and pricedispersion within and across multiple retailer types: Further evidence
and extension,” Journal of Academy of Marketing Science, 32 (2) 176–187.
a, Sulin and Paul A. Pavlou (2002). “Evidence of the effect of trust buildingtechnology in electronic markets: Price premiums and buyer behavior,”MIS Quarterly, 26 (3) 269–289.

P

iling 83 (3, 2007) 309–324 323

arney, Jay (1991). “Firm resources and sustained competitive advantage,”Journal of Management, 17 99–120.

aye, Michael R. and John Morgan (2001). “Information gatekeepers on theInternet and the competitiveness of homogeneous product markets,” TheAmerican Economic Review, 91 (3) 454–474.

aye, Michael R. and John Morgan (2003). “Red Queen Pricing Effects inE-Retail Markets,” http://ssrn.com.

aylis, Kathy and Jeffrey M. Perloff (2002). “Price dispersion on the Inter-net: Good firms and bad firms,” Review of Industry Organization, 21305–324.

etancourt, Roger and David Gautschi (1993). “Demand complementarities,household production, and retail assortments,” Marketing Science, 9 (2)146–161.

arlton, Dennis W. and Judith A. Chevalier (2001). “Free Riding and SalesStrategies for the Internet,” NBER Working Paper # 8067.

hen, Pei-yu and Lorin M. Hitt (2003). “Understanding Price Dispersionin Internet-Enabled Markets,” Working Paper, The Wharton School ofBusiness, University of Pennsylvania, Philadelphia, PA 19104.

hintagunta, Pradeep (2002). “Investigating category pricing behavior at aretail chain,” Journal of Marketing Research, 39 (2) 141–154.

lay, Karen, Ramayya Krishnan and E. Wolff (2001). “Prices and pricedispersion on the web: Evidence from the online book industry,” Journalof Industrial Economics, 49 (4) 521–539.

lay S Karen, Ramayya Krishnan, E. Wolff and D. Fernandes (2002). “Retailstrategies on the web: Price and non-price competition in the online bookindustry,” Journal of Industrial Economics, 50 (3) 351–367.

ohen, Joel B. and Charles S. Areni (1991). “Affect and consumer behav-ior,” in Handbook of Consumer Behavior, T. S. Robertson and H. H.Kassarjian eds. Englewood Cliffs, NJ: Prentice Hall., 188–240.

ohen, Marcel (2000a). “The impact of brand selection on price competition- A double-edged sword,” Applied Economics, 32 (5) 601–609.

ohen, Marcel (2000b). “Consumer involvement - driving up the cost,”Consumer Policy Review, 10 (4) 122–125.

oolin, Bill, Laurie McLeod, Bob McQueen and Mark Watton (2003).“Internet strategies for established retailers: Four New Zealand case stud-ies,” Journal of Information Technology Cases and Applications, 4 (5)3–20.

Marketer (Aug., 2004). “Online Selling and CRM,” eMarketer.omburg, Christian, Wayne D. Hoyer and Martin Fassnacht (2002). “Service

orientation of a retailer’s business strategy: Dimensions, antecedents, andperformance outcomes,” Journal of Marketing, 66 (4) 86–101.

aworski, Bernard J. and Ajay K. Kohli (1993). “Market orientation:Antecedents and consequences,” Journal of Marketing, 57 (3) 53–70.

irmani, Amna and Akshay R. Rao (2000). “No pain, no gain: A criti-cal review of the literature on signaling unobservable product quality,”Journal of Marketing, 64 (2) 66–79.

umar, V. and Robert P. Leone (May, 1988). “Measuring the effect of retailstore promotions on brand and store substitution,” Journal of MarketingResearch, 25 178–185.

al, Rajiv and Mikolas Sarvary (1999). “When and how is the Internet likelyto decrease price competition,” Marketing Science, 18 (4) 485–503.

ullaney, Timothy J. (Feb. 17, 2003). “The Web is Finally Catching Profits,”in BusinessWeek.

unnaly, J. (1978). Psychometric Theory, New York: McGraw-Hill.akes, Ariel (2003). “A reconsideration of hedonic price indexes with

an application to PCs,” American Economic Review, 93 (5) 1578–1596.

an, Xing, Brian T. Ratchford and Venkatesh Shankar (2002). “Can price dis-persion in online markets be explained by differences in E-Tailer servicequality,” Journal of Academy of Marketing Science, 30 (4) 433–445.

an, Xing, Brian T. Ratchford and Venkatesh Shankar (2003). “The evolu-tion of price dispersion in Internet retail markets,” Advances in Applied
Microeconomics: Organizing the New Industrial Economy, 12 85–105.
an, Xing, Brian T. Ratchford and Venkatesh Shankar (July, 2003b). “WhyAren’t the Prices of the Same Item the Same at Me.com and You.com?:Drivers of Price Dispersion Among E-Tailers,” Working Paper, Univer-sity of Maryland, College Park, MD 20742.

http://ssrn.com/

3 l of Ret

P

P

P

Q

R

R

S

S

S

S

T

V

W

Wmeasuring, and predicting eTail quality,” Journal of Retailing, 79 (3)


an, Xing, Brian T. Ratchford and Venkatesh Shankar (2004). “Price dis-persion on the Internet: A review and directions for future research,”Journal of Interactive Marketing, 18 (4) 116–135.

orter, Michael (2001). “Strategy and the Internet,” Harvard BusinessReview, 79 63–78.

ratt, John, David Wise and Richard Zeckhauser (May, 1979). “Price differ-ences in almost competitive markets,” Quarterly Journal of Economics,93 189–211.

uick, Rebecca (Nov. 30, 2000). “New Page in E-Retailing: Catalogs—Forget Gee-Whiz Technology; Web Stores Say Snail Mail is Cool Toolof the Future,” Wall Street Journal, B1.

atchford, Brian T., Xing Pan and Venkatesh Shankar (2003). “On the effi-ciency of Internet markets for consumer goods,” Journal of Public Policyand Marketing, 22 (1) 4–16.

audenbush, Stephen and Anthony S. Bryk (2002). Hierarchical LinearModels: Applications and Data Analysis Methods, Thousand Oaks: CA:Sage Publications, Inc.

hankar, Venkatesh and Ruth Bolton (2004). “An empirical analysis of
determinants of retailer pricing strategies,” Marketing Science, 23 (1)28–49.
hankar, Venkatesh and Lakshman Krishnamurthi (1996). “Relating pricesensitivity to retailer promotional variables and pricing policy: An empir-ical analysis,” Journal of Retailing, 72 (3) 249–272.

Z

ailing 83 (3, 2007) 309–324

mith, Michael (2001). “The Law of One Price? The Impact of IT-EnabledMarkets on Consumer Search and Retailer Pricing,” Working Paper,Carnegie Mellon University.

teenkamp, Jan-Benedict E.M., Frenkel ter Hofstede and Michel Wedel(April, 1999). “A cross-national investigation into the individual andnational cultural antecedents of consumer innovativeness,” Journal ofMarketing, 63 55–69.

ang, Fang-Fang and Xialon Xing (2001). “Will the growth of multi-channelretailing diminish the pricing efficiency of the web?,” Journal of Retail-ing, 77 319–333.

arian, Hal R. (2000). “Market structure in the network age,” in Under-standing the Digital Economy, Erik. Brynjolfsson and Brian Kahin eds.Cambridge, MA: MIT Press.

ittink, Dick R. (May, 1977). “Exploring territorial differences in the rela-tionship between marketing variables,” Journal of Marketing Research,25 145–155.

olfinbarger, Mary and May C. Gilly (2003). “e-TailQ: Dimensionalizing,

183–198.eithaml, Valarie, A. Parasuraman and Arvind Malhotra (2002). “Service

quality delivery through web sites,” Journal of Academy of MarketingScience, 30 (4) 275–362.

do market characteristics impact the relationship … of retailing 83 (3, 2007) 309–324 do market...

Documents