Decision Support Systems 51 (2011) 466–479

Contents lists available at ScienceDirect

Decision Support Systems

j ourna l homepage: www.e lsev ie r.com/ locate /dss

The long tail or the short tail: The category-specific impact of eWOM onsales distribution

Jung Lee a,1, Jae-Nam Lee b,⁎, Hojung Shin b,2

a Bang College of Business, Kazakhstan Institute of Management, Economics and Strategic Research, 4 Abay Avenue, Almaty 050100, Kazakhstanb Korea University Business School, Anam-Dong Seongbuk-Gu, Seoul, 136-701, Republic of Korea

⁎ Corresponding author. Tel.: +82 2 3290 2812; fax:E-mail addresses: junglee@kimep.kz (J. Lee), isjnlee@

Hojung_shin@korea.ac.kr (H. Shin).1 Tel.: +7 727 270 4440x2059; fax: +7 727 270 4462 Tel.: +82 2 3290 2813; fax: +82 2 922 7220.

0167-9236/$ – see front matter © 2011 Elsevier B.V. Adoi:10.1016/j.dss.2011.02.011

a b s t r a c t

a r t i c l e i n f o

Article history:Received 30 September 2009Received in revised form 10 November 2010Accepted 3 February 2011Available online 17 February 2011

Keywords:eWOMLong tail theoryProduct categorizationSales distributionWilcoxon signed rank test

This paper investigates the impact of electronic word of mouth (eWOM) on sales distribution and challengesthe conventional wisdom of the long tail theory. As customers refer to eWOM to evaluate products, and eachproduct type entails a different scheme of evaluation standards, the impact of eWOM may differ by producttype. Thus, we propose a new type of product categorization based on evaluation standard objectivity andhypothesize that this categorization gradually differentiates sales distribution patterns, some of which refutethe long tail phenomenon. To validate the hypothesis, we collect data from Amazon.com, compare thedistribution of eWOM among various product types, and conduct the Wilcoxon signed rank test for statisticalsignificance. All the test results show adequate levels of significance; thus, the three hypotheses aresupported. This study sheds new light on eWOM research by developing a new approach to productcategorization and by proposing a different use of eWOM in searching for products to explain the differenteffects of eWOM on sales distribution.

+82 2 922 7220.korea.ac.kr (J.-N. Lee),

3.

ll rights reserved.

© 2011 Elsevier B.V. All rights reserved.

1. Introduction

In online shopping malls, provision of product information isundoubtedly important because customers purchase products with-out actual examination [33]. To avoid the risk of buying undesirableproducts, customers always seek for product information on the Web[2,25,56]. For online customers, electronic word of mouth (eWOM) isa primary source of product information because it delivers recent,abundant, and objective product information [7,14,34,66]. Moreover,eWOM collects and displays ratings and reviews from variouscustomers, thus assuring a customer that the reviews are true andobjective collectively [31]. It is not surprising that eWOM is anessential element of online businesses [16] and exerts a stronginfluence on consumer behavior [30,39,67]. Nowadays, customersheavily refer to eWOM [9,36,68]. According to a recent survey, 84% ofAmericans reported that online reviews affected their decision topurchase a product or service [49].

In addition to influencing an individual customer's purchasingdecision, eWOM also changes the shape of sales distribution [1]. Salesdistribution reflects the collective behavior of customers; thus, achange in the sales distribution is anticipated depending on howeWOM changes individual customers' purchasing decisions. In

business, prediction of sales distribution is important because ithelps firms properly market their goods or services to targetcustomers [53]. Along with the domination of the Pareto principleas a base rule for determining sales distribution in the last decade,there have been attempts to describe the shapes of sales distributionfor online businesses and their changes [5,17].

However, the specific impact of eWOM on sales distribution is stillinconclusive and even contradictory in the literature. For example, thelong tail theory suggests that eWOMwould thicken the tail part of thesales distribution, thus facilitating the long tail phenomenon [1].eWOM introduces unpopular but attractive products to customersand provides detailed information of the products so that it informscustomers, who would not have bought the products without eWOM,about the goods and to persuade them to make a purchase. eWOMmechanisms increase the sales of the less popular products andfacilitate the long tail phenomenon.

In contrast, eWOM may inhibit the long tail phenomenon bypromoting the sales of popular products with high ratings [56].Positive reviews persuade customers to buy the product. Customers'high ratings verify the quality of a product, and their reviews describehow the product is superior over others. Customers who look for aproduct that is proven reliable by other customers often find the oneby sorting through popular products. As a result, sales of popularproducts would rise further than those of less popular products. Inaddition, many prior studies affirm that a positive rating encourages acustomer to make a purchase [9,14]. In this context, eWOM maypromote the sales of the popular products, and thus the head part ofthe sales distribution becomes thicker, generating “the short tail.”

467J. Lee et al. / Decision Support Systems 51 (2011) 466–479

What distinguishes these two seemingly contradictory effects ofthe eWOM? Under what circumstances, does eWOM sharpen orflatten sales distribution? Ironically, the roles of eWOM in both casesare the same; it helps customers find the product that they aresearching for. If customers look for an attractive but possibly lesspopular product, eWOM helps them find the product and causes thelong tail to appear. On the contrary, if customers look for a best-sellingproduct, eWOM also helps find it. In this case, the tail of the salesdistribution would be shortened.

Predicting the patterns of sales distribution is important becausethe alternative sales distributions may lead to different strategies forfuture online businesses [21]. Under the long tail theory, Andersonemphasizes the business potential of the unpopular products bysuggesting the “selling less of more” strategy. However, if the tail canbe shortened by eWOM, the “selling less of more” strategy would notbe effective any more, and sellers would be better off focusing moreon the business potential of a small number of popular products. Forbusinesses, understanding the role of eWOM and predicting salesdistribution will make a significant contribution to their sales.

Therefore, to explain the different effects of eWOM on salesdistribution, the present study investigates customer behaviorregarding eWOM, and the characteristics of the products (i.e. producttype) they purchase. Considering product type is important because itdetermines customers' product-search patterns. For example, ifcustomers buy books, they may look for the ones that are conformingto their personal taste but are not necessarily popular products. Incontrast, if customers buy television sets, they may look for best-selling products whose quality is verified by other customers.Therefore, we expect that with different product types, onlineconsumers have different product-search patterns, and the use ofeWOM in searching for products will result subsequently in differentsales distributions.

Acknowledging the possibility of mixed effects of eWOM oncustomer behaviors, current study seeks to answer the followingquestions: (1) How do customers react to product information gatheredfrom eWOM?; (2) How does product type differentiate customerbehaviors when customers process eWOM?; and (3) How do differentcustomer behaviors result in different patterns of sales distribution? Toanswer these questions, the study proceeds as follows. First, weexplain how customers process eWOM for their online purchasedecision making. We then propose a new categorization of producttypes as a key concept, which may differentiate the impact of eWOMon customer behaviors. Next, we develop three hypotheses on thedifferent impacts of eWOM to each product type. The proposedhypotheses are validated with data gathered from Amazon.com. Wethen discuss the results of the accompanying analyses. Finally, weconclude with the business and academic implications, limitations,and future research directions.

2. eWOM and product types

2.1. eWOM processing of customers

eWOM is the collection of online feedback gathered from variouscustomers on a specific product or service. It is created by a group ofcustomers who previously purchased the product. Customers rate theproduct and voluntarily post their product reviews without explicitincentives [26]. Most of the reviewers seem to share their opinionsand experiences with good intentions [48]. Such voluntary reviewsystems help potential customers trust the contents of the reviews[31,36].

The main information in eWOM about product is quality, that is,whether this product is good or bad [20,38]. Such product informationis provided with ratings and reviews [36]. Usually, a rating is a single-digit number that quantifies the total impression on the product, anda review is the detailed description that rationalizes why the product

is rated as such. With private knowledge and personal experiences,reviewers generate qualified product information [45]. Informationstored in eWOM can be sometimes outdated, inaccurate, andirrelevant because it is generated by a variety of non-specifiedcustomers. Nevertheless, customers generally find valuable and usefulproduct information in eWOM to pre-assess the quality of a product[14].

Sellers also play an important role in managing eWOM becausethey can adopt alternative methods of displaying prior reviews andopinions [22] to help customers obtain the necessary informationeasily [3]. For example, to ensure the quality of the contents, Amazon.com notifies customers about the use of real names when reviewerspost using their real names. Amazon.com also provides various searchoptions that can distinguish critical reviews from favorable reviews, ordisplay all reviews in a chronological order. This system is especiallyeffective as the rating system matures, the reviews are accumulated,and the customers become selective in exploring the reviews. Withthese systems, customers are able to find the information instantly.

With this information in eWOM, customers can make a purchasedecision. They purchase products only when the value of the productexceeds its price [41]. Accordingly, product evaluation is the mostfundamental and important step in customers' purchase decision-making process [61]. Generally, customers evaluate products in anintegrated manner using available product information, such aspictures, online advertisements, and customer reviews [54]. In onlinebusiness, eWOM is considered an effective and important source ofproduct information, which is a foundation for product evaluation.

When customers refer to eWOM to evaluate a product, customersrespond to the information in eWOM in various ways. In most cases,customers prefer products with positive reviews. Thus, the productswith higher ratings and favorable reviews generally sell better [9].However, because customer responses are not so simple for easycomprehension, researchers deepen the research and argue, forexample, that negative reviews are more influential than positivereviews [56], or extreme reviews on both ends play an important rolein customer decision making [10], and so on. Some differentiatecustomer responses by product types [24,46]. All these differentperspectives reflect various responses of customers regarding eWOM.

As stated above, most of the previous studies focus on theempirical aspect of eWOM (e.g., the correlation between rating andsales). However, if we turn our attention to customers' behavior andpurchase decision in response to eWOM, a fundamental question maybe posed on whether to follow the assessment in eWOM or not. Inother words, although customers search for products with positiveassessments, it is another issue whether or not they would believe thewords and actually purchase the products. Indeed, customers do notalways follow other customers' opinions. If customers decide tobelieve prior assessments, they would buy a product rated and rankedhighly by other customers. On the contrary, if they decide not tobelieve, they would not buy the product with highest rank. In reality,it is not always true that customers buy the product with the highestrank.

Such different responses (i.e. follow or not to follow theassessment of others) differentiate eWOM impacts and, eventually,generate different sales distributions. If customers follow others'opinions, products with higher ratings would attract more customers,whereas products with lower ratings would lose more customers. Inthis case, sales distribution would form the shape of a “thick-head andshort-tail.” In contrast, if customers do not follow others' opinions, noproduct would attract more customers. Thus, sales distribution wouldreflect the various opinions of various customers. In this case, theimpact of eWOM would not be strong, and the shape of the salesdistribution would form a “thin-head and long-tail.” Figs. 1 and 2describe these cases in brief.

Such two types of customer responses are, fundamentally,distinguished by the characteristics of the products that customers

Fig. 1. Customer's eWOM processing.

468 J. Lee et al. / Decision Support Systems 51 (2011) 466–479

intend to purchase. As customers refer to eWOM to evaluate aproduct, if products have different evaluation standards, thencustomer responses will be different as well. For example, if twocustomers have similar evaluation criteria for a certain product, it ishighly likely for them to have similar assessments on the product. Inthis case, it will be more rational for one customer to follow the othercustomer's opinion. On the contrary, if two customers possessextremely different evaluation standards, they will have dissimilarassessments and ratings. In this case, following the other customer'sopinion will not always be the best choice. In this study, we interpretthis similarity in evaluation standards as objectivity, which deter-mines the impact of eWOM and the shape of sales distribution.Objectivity is not a simple concept to define and describe, and thus wewill in detail discuss product evaluation standards and theirobjectivity in the next section.

2.2. Product attributes and its evaluation standards

To discuss the objectivity of product evaluation standards, we firstneed to explain how a customer evaluates a product. A product isevaluated by its attributes [29,62]. For product evaluation, customersexamine various attributes of the product, such as size, color, and soon [37]. Technically, product evaluation is the colligation of assess-ments of product attributes. For example, when a customer evaluatesan MP3 player, he examines each important attribute of the product,one by one such as weight, capacity, design, and warranty. He mayhave a favorable evaluation of the product in terms of its color but notin terms of its size. However, his overall evaluation of the MP3 playerwould be based on the sum of his evaluations for all the attributes(Fig. 3).

When examining a product, customers may perceive its attributesusing two types of evaluation standards: objective versus subjectivestandards. Some attributes associated with objective evaluationstandards are capacity, warranty, class, and power, whereas attributes

Fig. 2. Long tail vs. short tail.

associated with subjective evaluation standards are color, design,style, and genre.

The two standards differ on whether there is an accepted ranking-based standard for evaluation or not. For example, in the case of theMP3 player, attributes such as capacity and warranty are consideredobjective evaluation standards. With other attributes such as pricebeing the same, a 2 G MP3 player is preferred to a 1 G MP3 player,whereas a three-year warranty is preferred to a one-year warranty. Itis generally accepted that a higher capacity and a longer warranty arealways evaluated more favorably than a lower capacity and a shorterwarranty. As there are objective rankings and distinct qualitydifferences between products, we can argue that these attributesare evaluated by objective standards.

In contrast, such attributes as color and texture cannot be objectiveevaluation standards but rather subjective standards. It is becausesome customersmay prefer a yellowMP3 player to a red one, whereasothers may prefer the red one. Likewise, some may like metallicexteriors more than plastic ones, whereas others may prefer plasticexteriors. General and objective evaluation standards do not existbecause there is no distinct quality difference between red andyellow, or plastic and metal. In this case, we can argue that theseattributes are evaluated by subjective standards.

To illustrate further the use of objective and subjective standards,when a customer buys a textbook, it is commonsense that a newerversion is preferred to an old version. However, there is no consensusthat a romance book is always better than a thriller book. In this case,the version of the textbook is an attributewith an objective evaluationstandard whereas genre is a product attribute with a subjectiveevaluation standard.

For some customers, objective attributes may be perceived assubjective attributes. While most of customers prefer a large TV, somemay prefer one with a smaller screen. Some customers may evenprefer a lower-resolution camera for some unknown reason. Never-theless in this study, we do not consider such exceptional customers.Table 1 summarizes the discussions about the type of productattributes and characteristics.

Since product evaluation is the colligation of attribute evaluations,the objectivity level of product evaluation will be determined by theimportance of the objective attributes in the total product evaluation.The more important attributes, which tend to be of a higher priority

Fig. 3. Customer's product evaluation.

Table 1Product attribute types.

Product attributes with objectivestandards

Product attributes withoutobjective standards(=subjective standards)

Attribute Length, weight, class, warrantee,capacity etc.

Color, shape, style, genre, brandetc.

Description Customers have the agreedranking-based preferences foreach product

There is no agreed rankingamong products thus customersshow different preferences

Competition When there are two competingproducts, one will attract all thecustomers

When there are two competingproducts, they will share thecustomer demand

Example 2 year warrantee is preferredover 1 year warrantee. 4 G USB ispreferred over 2 G USB

Yellow shoe is not alwayspreferred over Red shoe. Thrillerbook is not always preferred overRomance book

469J. Lee et al. / Decision Support Systems 51 (2011) 466–479

and weight in evaluation, will subsequently determine the position ofthe product in the continuum.

2.3. Product category continuum

When each product is evaluated, it is evaluated through differentcombinations of the objective/subjective attributes. Each product isevaluated through their major objective or subjective attributes. Forexample, a USB memory stick is mainly evaluated through size,capacity, design, and similar attributes, whereas a camera is mainlyevaluated through brand, pixel, design, and other attributes. Thesedifferent evaluation criteria of each product reflect the relativeimportance of the attributes in the product [18]. In other words, insome product such as USB memory, objective attributes such as sizeand capacity are more importantly considered while in other productssuch as cloth, subjective attributes such as design and color are moreimportantly considered and examined.

Hence, for the current study, we characterize product types by theportion of objective attributes and formulate a continuum, the pointsof which represent the importance of the objective attributes of aproduct. Fig. 4 shows a continuum of products whose locations areindexed through the objectivity of product-evaluation standards. Onthe left end of the continuum are the products with a high portion ofobjective attributes in product evaluation, and are the products with alow portion of objective attributes. Here, “portion” can also beinterpreted as “importance” because the more important an attributeis, the higher the portion that will be assigned to it in the evaluation.

It is not easy to objectively measure or quantify the portion ofobjective/subjective attributes in each product. However, there islittle doubt that consumers characterize some products as more of an‘objective’ product and others as more ‘subjective’ in general. Thus werationalize the portion through the following references and exam-ples that have been established in prior research:

First, the purpose of purchasing can determine the portion ofobjective attributes (i.e., position of the product on the continuum). Ifa product is purchased to serve functionality and practicality like theneed to take photos, watch TV, or create storage space (e.g., utilitarian

Fig. 4. Product categ

goods), it would be evaluated through more objective standards. If aproduct is purchased to satisfy vanity, desire, and esthetic reasons likethe need to look good, feel luxurious, or enjoy ownership (e.g.,hedonic goods), it would be evaluated through more subjectivestandards. Utilitarian goods are products whose consumption iscognitively driven, instrumental and goal oriented, and purchased toaccomplish a functional and practical task [57]. Hedonic goods areproducts whose consumption is primarily characterized by anaffective and sensory experience of esthetic or sensual pleasure,fantasy, and fun [27]. This utilitarian/hedonic categorization isconsistent with our product categorization continuum, which showsthat utilitarian goods are on the left while hedonic goods are on theright.

Second, the availability of quality check before purchasing candetermine the position of the product. If a product is evaluatedthrough more objective standards, its quality will be easily evaluatedbefore purchase (i.e., search goods). If a product is evaluated throughmore subjective standards, its quality will not be easily evaluatedbefore purchase (i.e., experience goods). A search good is a product orservice with features and characteristics that are easily evaluatedbefore purchase while an experience good is a product or servicewhere product characteristics such as quality are difficult to observein advance and can only be ascertained upon consumption [43].Therefore, we can support that a search good would be on the left of acontinuum while an experience good on the right.

Lastly, the type of product differentiation can determine theposition of a product on the continuum. If products involve objectiveevaluation standards, they can be easily distinguished and ordered byquality and have objective rankings among products (i.e., verticallydifferentiated goods) [13]. However, if products involve moresubjective evaluation standards, they do not share any consensus ofrankings among themselves and, therefore, cannot be distinguishedand ordered through quality (i.e., horizontally differentiated goods)[13]. Therefore, we can argue that vertically differentiated goods willbe on the left while horizontally differentiated goods will be on theright.

These conceptualizations allow us to plot products on thecontinuum with consensus. For example, hard disk drives (HDD)whose objective attributes such as capacity and reliability areconsidered important, are located on the left side of the continuum.On the other hand, a set of cardigans, whose subjective attributes suchas color and style are considered important, is on the right side of thecontinuum. A computer can be located on the left side of the HDDbecause it possesses subjective attributes such design and sound. Auniform can be located on the right side of the continuum of clothesbecause it has less subjective attributes than ordinary clothes in termsof color and design.

We admit that our product continuum may not perfectly plot allproducts along one line that carries the customers' completeagreements because the location will vary based on the scope of theobjectives of the purchase. For example, if a customer who is used toWindows series purchases a Windows Vista, it will be located on theleft of the continuum because it is considered a better version as atypical example of vertically differentiated good [40]. However, if the

ory continuum.

470 J. Lee et al. / Decision Support Systems 51 (2011) 466–479

customer extends the scope to computer operating systems, it will beharder to have a consensus because the objectivity level of productevaluation standards will decrease. The product can then be moved tothe right side of the continuum. The scope for product categorizationis important but not easy to universally define [44].

In spite of above limitation, our product categorization can stillmake significant contributions especially for its uniqueness andcomprehensiveness. Studies on product types exist [57,59,63] andsome prior studies on eWOM use such categorization to explain thedifferent aspects of its influence [24,46]. However, extant studiesattempt to explain the various impacts of eWOM using less than threesample product types (i.e., search and experience goods in [46],household and experimental goods in [24]). Products have becomeincreasingly complicated and sophisticated and it is impossible tocategorize them into a few types only. Our categorization is based onthe objectivity continuum, which is more effective in describing thediversified effect of eWOM across products with a unified pattern.

3. eWOM impact on sales distribution

3.1. When products have objective evaluation standards

The Pareto principle (also known as the 80–20 rule) states that formany events, roughly 80% of the effects come from 20% of the causes.It is a common rule of thumb adopted in business. For example, “80%of your sales come from 20% of your clients” [6]. This rule has been themost widely accepted norm in business to explain the pattern of salesdistribution [52] and the high purchasing power of a small group ofconsumers [35]. Under the Pareto principle, for businesses to focus onthese small numbers of consumers to increase profit is effective.

In an online business environment, the so-called long tail theoryhas been proposed recently and is cited as the revolutionary violationof the Pareto rule [1]. The theory argues that low searching cost andreaching cost in an online environment substantially increase thecollective share of niche products, thereby reducing the portion ofsales possessed by those top “20%” and creating a longer tail in thedistribution of sales. Once the theory was proposed, it has soonbecome widely accepted for its new and striking idea with empiricalsupport [37].

Nevertheless, all the example products cited in the theory,including songs, books and movies, have a common feature that allof them are evaluated with high subjectivity. Intuitively, theseproducts could be one of the special cases prone to personalpreference. Customers, having a variety of tastes, cannot applyobjective standards to these products while making their purchasedecisions. When explaining the long tail theory, the authors did notexplore nor discuss the products that could be evaluated by ratherobjective standards, such as a USB drive, vacuum, and printer [1].Based on our argument above, we conjecture that the impact ofeWOM on products with objective evaluation standards is differentfrom that on products with subjective evaluation standards.

As mentioned, customers may apply similar evaluation standardsto products with objective attributes, and thus look for popular andquality-approved products. For example, when buying a vacuumcleaner, the important attributes to consider are the power, noiselevel, warranty, and so forth. Typically, stronger power, lower noise,and longer warranty are preferred by customers. As the product isevaluated upon objective standards, and customers have similarevaluation standards, customers tend to buy the popular product withcollectively positive evaluations. Knowing that other people havesimilar evaluation standards, customers consider it rational to buy aproduct with positive evaluations from other customers. As a result,customers would show a high tendency of following the eWOMevaluation, thereby making the popular product with high ratingsmore popular. In this case, unlike the cases of music and books in thelong tail theory, a popular product attracts more customers.

Subsequently, the head portion of sales distribution will thicken,and the tail part will shorten. From the reasoning, we propose thefollowing hypothesis:

H1. If products are evaluated mainly in terms of their attributes usingobjective evaluation standards, the existence of eWOMwill inhibit thelong tail phenomenon.

3.2. When products have subjective evaluation standards

As mentioned, all the products cited in the long tail theory arethose evaluated by highly subjective standards [1]. Positive evaluationwill not guarantee that other customers will purchase those productsbecause everyone has different evaluation standards. For example, thesatisfaction of other customers with a certain movie or book does notguarantee “your” satisfaction with it [35]. There are many onlineforums and discussions for those products, such as youtube.com, but itis hard to find those where the participants show the same preferencefor a certain product. There are always diverse opinions andevaluations on products, and customers are aware of, and evenenjoy, such diversity.

An important assumption pertaining to these kinds of products isthat people have different tastes and preferences. Therefore, whileenjoying diverse opinions, each customer seeks for the product thatwould best fit his/her taste. However, finding these products is noteasy because they are usually experience goods [46], which acustomer cannot examine before purchase in an online environment.In this regard, eWOM can help customers find products that would fittheir tastes the most by providing detailed information about theproducts [19]. Not only can customers check the ratings, but they canalso carefully read the opinions and reviews to knowmore about theirtrue features [15]. With eWOM, because it provides rich information,it is easier for a customer to find the product. Given the assumptionthat people have different tastes and preferences for particularproduct types, sales of the products will be more dispersed. Thisphenomenon is explained as part of the long tail theory. Thus, we havethe following hypothesis:

H2. If the products are evaluated mainly in terms of their attributesusing subjective evaluation standards, the existence of eWOM willfacilitate the long tail phenomenon.

3.3. Comparison of products with objective and subjective evaluationstandards

One of the merits of our product categorization is in itsexhaustiveness. As categorization characterizes products based onthe “importance (i.e., portion)” of their objectively measurableattributes, all products can be classified respectively by their portionswithout overlapping. Most of the categorizations in prior research(e.g., search good or experience good [46], vertically or horizontallydifferentiated goods [12], and household or experimental goods [24])often pose a limitation that the categorization criteria may neither bemutually exclusive nor exhaustive. For example, anMP3 player can beeither a search good or an experience good because it can be searchedand purchased by name and descriptions; however, it also has suchfeatures as sound and convenience, which the users cannot determinebefore purchase. In this context, our categorization is effectivebecause it allows us to categorize the products in a continuumbased on the level of evaluation objectivity. Thus, we can “compare”various products with the flexibility of using the portion of theseobjective attributes.

This benefit also leads us to draw another hypothesis. In the thirdhypothesis, we compare the influence of eWOMon two products withdifferent attributes. If the product becomes more complicated, thenumber of attributes in the product increases. Moreover, as the

471J. Lee et al. / Decision Support Systems 51 (2011) 466–479

attribute increases, it is unlikely to have similar evaluation standardsamong customers. For example, when customers purchase a laserprinter, they will checks a certain number of attributes, such asprinting speed and warranty. However, if they purchase an all-in-oneprinter, they will additionally investigate such attributes as scanningresolution, fax speed, and other attributes of a simple laser printer.The harmony of the integration of these attributes is also an importantand subjective attribute to consider. As the combinations arecomplicated and entail subjective evaluation, customers will likelyshow diverse preferences due to their different tastes. As a result, anall-in-one printer will have the higher portion of subjective evaluationstandards than the simple laser printer. From this, we can concludethat the more complicated products are, the lesser is the tendency ofcustomers to follow the eWOM, and the longer is the generated tail.Thus, we have the following hypothesis:

H3. The more product attributes that need to be evaluated, the morestrongly eWOM will exhibit the long tail phenomenon.

4. Research methodology

4.1. Sales parameterization

To investigate the patterns of sales distribution, we need aparameter for sales. Sales rankings provided by Internet shoppingsites are not adequate for the research because they are mainlydetermined by the current rather than historical sales records [50].Sales parameters for our research should reflect a product's overallpopularity, not the momentary sales rate, because the eWOM in thestudy is hypothesized to have an impact for a certain period of time.

Table 2Data description.

Objective Cases Product category # ofthe c

For H1 validation Set 1 (A) 4 G USB memory 260(B) 2 G USB memory 749

Set 2 (A) 8 G memory card 312(B) 4 G memory card 765

Set 3 (A) ~500 G external HDD 94(B) ~300 G external HDD 203

Set 4 (A) 32 G USB memory 336(B) 16 G USB memory 169

Set 5 (A) 16 G memory card 1687(B) 8 G memory card 3639

For H2 validation Set 6 (A) Recently published thriller book 108(B) Thriller book 1179

Set 7 (A) Recently published horror book 94(B) Horror book 1184

Set 8 (A) ~8 megapixel camera 456(B) ~6 megapixel camera 725

Set 9 (A) Recently published audio book 50(B) Audio book (historical fiction) 922

Set 10 (A) Recently published e-Book (SF) 110(B) e-Book (SF) 9027

For H3 validation Set 11 (C) Laser printer 783(D) All-in-one printer 716

Set 12 (C) LCD TV 1038(D) TV–DVD combo 129

Set 13 (C) Keyboard 335(D) Keyboard–mouse combo 80

Set 14 (C) Unlocked Samsung cell phone 168(D) Samsung cell phone w/ service 177

Set 15 (C) Speaker systems 250(D) Home theater systems 685

(A)Productwithweaker eWOMimpact; (B) productwith stronger eWOMimpact; (C) productwstandards.

However, in practice, obtaining longitudinal sales data is not easybecause of company policies.

Therefore, the number of reviews for each product is selected as aparameter for the general popularity level, which represents the totalsales amount. From a commonsense standpoint, the more popularproducts with high sales possess more reviews. For example, if 1% ofcustomers who purchased the product post their reviews online, it isobvious that the number of reviews will be roughly proportional tothe sales amount. In many previous studies, a positive linearitybetween the amount of referrals and the total sales has also beensupported [23].

4.2. Data collection

We collect data for two periods, 2 to 9 February 2009 and 4 to 10October 2010, from Amazon.com, the leading online shopping mall inthe world. Ideally, we should compare two sets of data collected fromtwo shopping sites (e.g., one with eWOM services and the otherwithout the service that allows sellers to sell identical products withina category) to see the impact of eWOM. However, we collected thedata sets from available online shopping malls and prepared proxycategories using various eWOM impact levels because it is difficult tofind an online shopping mall without eWOM services nowadays [26]and evenmore difficult to control the products sold in their categories.We also collected data for a short period (i.e., one week) because thedata has reflected the history of the sales. The reviews generated fromthe introduction of the product introduction were also collected andthe number of reviews is consistent with the total sales ever made.

Table 2 shows the data sets from thirty types of products thatinclude USB, memory card, books with various mediums, cameras,

products inategory

Sample size Ratio of products withat least one review

Date of collection

260 0.418 2009.2.2–2009.2.9300 0.371300 0.163300 0.14194 0.766

169 0.783300 0.375 2010.10.4–10.11169 0.592300 0.130300 0.052

108 0.556 2009.2.2–2009.2.9300 0.73194 0.457

300 0.927300 0.844300 0.79250 0.760 2010.10.4–10.11

300 0.918110 0.618300 0.764

300 0.298 2009.2.2–2009.2.9300 0.437300 0.604129 0.628300 0.59480 0.663

168 0.399 2010.10.4–10.11177 0.842250 0.532300 0.609

ithmore objective evaluation standards; and (D)productwithmore subjective evaluation

472 J. Lee et al. / Decision Support Systems 51 (2011) 466–479

printers, cell phones, and other products. The products were selectedfor multiple reasons such as popularity and availability. We choseproducts that are commonly used by various customers and generatedconsistent sales.

In data sets 1 to 10, (A)s are product categories with less eWOMimpact, and (B)s are product categories with more eWOM impact.(A) is a relatively new product category, whereas (B) covers productsthat have been sold for a relatively long time. Naturally, we assumedthat (A) is the market where the influence of eWOM is small and(B) the opposite. Moreover, we tried to control for other factors suchas price and usage by selecting (A) and (B) from the same largecategory. For example, a USB memory stick was selected for H1validation not only because it is a widely used digital device but alsobecause its capacity is being upgraded. Many people prefer sticks withhigher capacity. In 2009 February, Amazon sold 749 kinds of 2 GB USB

Fig. 5. Cumulative distribution functions of

memory sticks and 246 kinds of 4 GB. This shows that the 4 GB USBcategory is a relatively new market. Moreover, the range of pricedistributions of the two products shows that there is no significant pricedifference between the data sets. Therefore, the tendency to compare a8 GB with a 16 GB USB indicates that eWOM impact is compared whileother factors such as price and category are properly controlled.

In data sets 11 to 15, (C)s are products with more objectiveevaluation standards and (D)s are products with less objectiveevaluation standards. Generally, the overall objectivity level of theevaluation standard decreases as the number of attributes increases,as explained in H3, because it is harder to have a consensus on largernumber of attributes. Customers would have more diverse opinionson and preferences in products with more attributes. Therefore, weprepared two sets of data: one with less attributes and another withmore. For example, TV–DVD combos possess more product attributes

eWOM amount in cases 1, 2, 3, 4 and 5.

473J. Lee et al. / Decision Support Systems 51 (2011) 466–479

compared with regular TVs, thus, customers would have more diverseopinions on TV–DVD combos using less objective evaluationstandards.

We counted the number of eWOM of all products in category if thecategory possessed less than 300 products. We conducted randomsampling and collected data from 300 products if the categorypossessed more than 300 products. Seventeen among thirty productcategories have more than 300 products. The total number of productoriginally collected was 26,430 and we used 6958 data to match thenumber of reviews within one category.

We also presented the column, “ratio of products with at least onereview,” to show that the average eWOM impact may differ amongcategories but not within the same category. For example, in data set8, 80% of the products in the category have at least one review.However, comparing sets (A) and (B) would be appropriate to showthe marginal impact of eWOM because the difference between set(A) (i.e., 0.84) and set (B) (i.e., 0.79) is not significant. The profile ofthe data sets is summarized in Table 2.

5. Analysis and results

For validation, we formulated the cumulative distribution func-tions (c.d.f.) of the number of reviews in products (A) and (B) (or Cand D) and compared the portions of the reviews possessed by thehigh-ranking products to check whether the head parts of the c.d.f. isthickened or thinned. We plotted the c.d.f. of products (A) and (B) inone figure and compared their shapes for the intuitive explanation. Allgraphs in the figures present the review shares of the upper 5–40percentile of the products to show that the two lines in each graph aredistinctively different.

To confirm the differences statistically, we conducted a Wilcoxonsigned rank test to compare the portions of the reviews shared by thehigh-ranking products [65]. The test is widely used to analyze paireddata when the assumption of a normal distribution is doubtful [11].We conducted random sampling in the larger data set to match itssample size with that of the smaller data set in the data sets where the(A) and (B) or (C) and (D) have different sample sizes (i.e., sets 1, 3, 4,

Table 3Wilcoxon signed-rank test results of cases 1, 2, 3, 4 and 5.

Mean differences Ranks N

2 G USB vs. 4 G USB Negative ranks(a) 10Positive ranks(b)

Ties(c) 15Total 26

Note: (a)4 G USBb2 G USB, (b)4 G USBN2 G USB, and (c)4 G USB=2 G USB

4 G memory card vs. 8 G memory card Negative ranks(d) 6Positive ranks(e)

Ties(f) 23Total 30

Note: (d)8 G memoryb4 G memory, (e)8 G memoryN4 G memory, and (f)8 G memory=4

300 G HDD vs. 500 G HDD Negative ranks(g) 7Positive ranks(h)

Ties(i) 2Total 9

Note: (g)500 G HDDb300 G HDD, (h)500 G HDDN300 G HDD, and (i)500 G HDD=300 G H

16 G USB vs. 32 G USB Negative ranks(a) 9Positive ranks(b)

Ties(c) 7Total 16

Note: (a)32 G USBb16 G USB, (b)32 G USBN16 G USB, and (c)32 G USB=16 G USB

8 G memory card vs. 16 G memory card Negative ranks(d) 21Positive ranks(e)

Ties(f) 8Total 30

Note: (d)16 G memoryb8 G memory, (e)16 G memoryN8 G memory, and (f)16 G memory

6, 7, 9, 10, 12, 13, 14 and 15). For example, we reduced the sample sizeof data (B) in data set 1 from 300 to 260 to match the size of data(A) through a random sampling technique. The total number ofreviews used in the study is summarized in Table 2.

5.1. H1 validation—eWOM impact on sales distribution when productshave objective evaluation standards

Fig. 5 shows the c.d.f.s of two product categories in data sets 1, 2, 3,4 and 5 with the upper 5 to 20% of the products, respectively, with theprecise portion of each case shown in the graph. Due to spacelimitation, the portions of eWOM included in the graphs weredetermined to represent approximately 80% of the total reviewamount. Table 3 presents the Wilcoxon signed rank test results of sets1, 2, 3, 4 and 5. The p-values, which are below 0.001, indicated that thedifferences shown in the graphs were all significant.

As seen in Fig. 5 and Table 3, in all five data sets, the heads of thematured markets are thicker (i.e., sharper increase of function) thanthose of the newmarkets, thus statistically and graphically supportingH1. In a cumulative distribution function, a sharper increase in theinitial stage implies higher density in the head part, which can beinterpreted as the thick head. Moreover, because the total distributionis normalized to 1, the sharper increase in the initial stage willconsequently lead to the duller increase in the tail part, whichgenerates the short tail. This shows that the matured markets have ahigher level of sales concentrations on high-ranking products.

5.2. H2 validation—eWOM impact on sales distribution when productshave subjective evaluation standards

For H2 validation, two sets from the paper-book, one from audiobook, one from e-book category and one from the digital devicecategory were selected. As explained, a book is an example of aproduct to which objective evaluation standards do not apply. Thedemand for a book varies based on individuals' preference and taste.Likewise, a camera is one of the complicated goods to which agreedevaluation standards do not apply. Since most of today's cameras

Mean rank Sum of ranks Test stat.

4 52.50 5460.00 Z=−8.559pb0.0011 105.00 105.00

50

1 31.98 1951.00 Z=−6.832pb0.0011 2.00 2.00

80G memory

2 37.49 2699.00 Z=−7.413pb0.0011 2.00 2.00

14DD

8 50.32 4931.00 Z=−8.572pb0.0011 19.00 19.00

09

8 109.50 23,871.00 Z=−12.801pb0.0010 0.00 0.00

20=8 G memory

474 J. Lee et al. / Decision Support Systems 51 (2011) 466–479

are qualified in their basic functions, such as clean-picture andprecise-focus features, potential customers are more concernedabout their design and brand, which are not easy to evaluateobjectively.

As shown in Fig. 6 and Table 4, all three data sets consistently showa sharper increase of c.d.f.s in the newly formed markets. Graphically,as proven by the Wilcoxon signed test, set 10 shows a very small yetstill significant statistical difference. The c.d.f.s in thematuredmarketsshow relatively gradual increases. These results indicate that eWOMlessens the sales concentration of high-ranking products in markets.Thus, H2 is supported.

5.3. H3 validation—comparison of products with different evaluationstandards

Products for H3 validation were selected to show the differentimpacts of eWOM with respect to the objectivity level of productevaluation standards. Product evaluation criteria become more

Fig. 6. Cumulative distribution functions of e

complicated as the number of products' attributes increases; thus,products withmore attributes were selected as those with less similarand less objective evaluation standards. For example, all-in-oneprinters have more complicated product features than simple laserprinters because the former requires more criteria for evaluation.With these rationales, we selected three product pairs, each havingone that requires relatively objective evaluation standards and onethat requires relatively less objective evaluation standards.

As seen in Fig. 7 and Table 5, all three data sets show significantlythinner heads and longer tails in products with less objectiveevaluation standards. Graphically, as proven by the Wilcoxon signedtest, set 13 shows a very small yet still significant statistical difference.Thus, H3 is supported.

6. Discussion

In the Introduction, we raised three questions about 1) customerresponse to eWOM, 2) relationship between product type and

WOM amount in cases 6, 7, 8, 9 and 10.

Table 4Wilcoxon signed-rank test results of cases 6, 7, 8, 9 and 10.

Mean differences Ranks N Mean rank Sum of ranks Test stat.

All thriller book vs. recently published thriller book Negative ranks(a) 0 0.00 0.00 Z=−7.866pb0.001Positive ranks(b) 82 41.50 3403.00

Ties(c) 26Total 108

Note: (a)recentball, (b)recentNall, and (c)recent=all

All horror book vs. recently published horror book Negative ranks(d) 0 0.00 0.00 Z=−8.008pb0.001Positive ranks(e) 85 43.00 3655.00

Ties(f) 9Total 94

Note: (d)recentball, (e)recentNall, and (f)recent=all

6 MP camera vs. 8 MP camera Negative ranks(g) 0 0.00 0.00 Z=−13.597pb0.001Positive ranks(h) 246 123.50 30,381.00

Ties(i) 54Total 300

Note: (g)8 MP camerab6 MP camera, (h)8 MP cameraN6 MP camera, and (i)8 MP camera=6 MP camera

Historical fiction audio book vs. recently published historical fiction audio book Negative ranks(a) 0 0.00 0.00 Z=−5.645pb0.001Positive ranks(b) 42 21.50 903.00

Ties(c) 8Total 50

Note: (a)recentball, (b)recentNall, and (c)recent=all

SF e-Book vs. recently published SF e-Book Negative ranks(d) 2 70.50 141.00 Z=−7.332pb0.001Positive ranks(e) 82 41.82 3429.00

Ties(f) 26Total 110

Note: (d)recentball, (e)recentNall, and (f)recent=all

475J. Lee et al. / Decision Support Systems 51 (2011) 466–479

customer response, and 3) eWOM impact on sales distribution. Theanswers to the first two questions feed into the answer for the third.The third question, which is the main research objective of this study,is embodied by three hypotheses and validated by statistical analyses.From these, we draw the following discussions.

First, we simplify customer responses to eWOM in a dichotomousmanner to validate its further impact on sales distribution. There havebeen numerous studies on understanding the dynamics betweeneWOM and customer behaviors, but not many of them provide anintegrative and parsimonious view. Therefore, from the literature, wedraw a concept of dichotomous response to eWOM (i.e., to follow ornot to follow others' opinions), and describe these responses in asystematic way to link conceptually its impact on sales distribution(Fig. 1). This conceptualization is further utilized in developing thehypotheses and is supported through hypothesis validation.

Second, we provide a new product category scheme thatdifferentiates customer responses from eWOM. Such customerbehavior (i.e., to follow or not to follow other customers' rating) isdetermined by the objectivity in product evaluation, and thus weformulate a continuum, the points of which represent the importance(i.e., relative weight) of the objective attributes of a product. Usingthis continuum, we can further show eWOM impact on salesdistribution (i.e., how customers' collective behavior changes theshape of the sales distribution). We also highlight the merit of ourcategorization by comparing it with the traditional categorizationtechnique. While traditional product categorizations (e.g., search/experience good, vertical/horizontal differentiation) are rather dis-crete and not exhaustive, ours is effective in explaining their positionfrom a relative perspective.

Lastly, with this product categorization and customer responses toeWOM, we draw hypotheses regarding eWOM impact on salesdistribution. If the product is evaluated by objective standards,customers tend to follow existing customers' opinions. Thus, productswith high ratings attract more customers. This customer attractionwill result to sales concentration, making a thick head and a short tailin the sales distribution. On the contrary, if the product is evaluatedmore by subjective standards, customerswill not have such consistent

behaviors, and product with high ratings will not be guaranteed toattract more customers. In this case, the sales distribution will reflectthe various preferences of many customers and, consequently, willshow a thin head and a long tail.

6.1. Research contributions and implications

The study has the following academic contributions. First, we showthat the long tail theory is not a universally applicable theory but onlysketches part of a new phenomenon brought about by eWOM in theelectronic-commerce environment. Specifically, we suggest that thelong tail impact can be observed differently across product types. Inprevious research on the long tail theory, all sample products (i.e.,music, books) are products with subjective evaluation standards. Thecommon characteristic of such products is that people have variouspreferences and do not have any objective evaluation criteria for theproducts. Having sensed such distinctive features of long tail products,we initiated our study by extending the cases to various product typesand found that the long tail phenomenon is only partly applicable andthat there are many other opposite cases resulting from eWOM. Forexample, in the case of HDD andmemory card, because the evaluationcriteria for these products are relatively objective and consistentacross customers, products with high evaluation will attract morecustomers. Thus, the sales will be concentrated more on the popularproducts, resulting in a thick head and a short tail. In short, both thelong tail and the short tail are the reflections of the customerpreferences. If customers have various preferences, the long tail willappear; if customers have consistent and unified preferences, theshort tail will appear. These tails are enhanced by the eWOM becauseit disseminates customer preferences in an effective way.

Second, we propose a new type of product categorization throughits evaluation standards. In business, product evaluation is one of themost fundamental and critical of consumer activities because itdictates customer purchase decision and subsequently affects thefirm's profits [51]. Therefore, our product categorization is not onlyeffective in differentiating eWOM impact but can also be used in otherresearch built on product evaluation activity. Furthermore, other

Fig. 7. Cumulative distribution functions of eWOM amount in cases 11, 12, 13, 14 and 15.

476 J. Lee et al. / Decision Support Systems 51 (2011) 466–479

traditional product categorization concepts, such as search orexperience goods [46], or vertically or horizontally differentiatedproducts [55], only classify products into two different categories. Adistinct advantage of our product categorization method is that itgives amore flexible and parsimonious classification of products usinga continuum of objectivity.

Lastly, we introduce a new validation technique. To test ourhypotheses, most of prior empirical research has adoptedwell-knownstatistical analyses, such as the structural equation modeling [61] forsurvey data and a variety of regressions [4,8] for sales or ranking data.However, we graphically plot the empirical cumulative distributionfunction of the sales for an intuitive comparison and conductWilcoxon signed rank tests for statistical validation. The Wilcoxontest is one of the famous nonparametric methods of analysis [60]. Webelieve that our research method is meaningful because it widens themeans of empirical verification, particularly for data that may not

meet the necessary conditions of standard statistics, such asnormality.

Based on the results of this study, we provide the followingimplications for practitioners. First, by showing the various patterns ofthe sales distributions, this study gives an idea about the appropriatesizes of shelves in online shopping malls. Conventionally, theunlimited shelves of online shopping malls have been consideredone of the advantages of the electronic-commerce environment [60].However, they are effective only for the market where consumershave subjective and diverse preferences. Thus, the long tail theory isapplicable, and low-ranking products make significant sales. For thesellers, it is still costly to manage a large number of products [28], andthus it is necessary to detect the portion of the sales generated byhighly ranked products. The more concentrated the market is (i.e.,high-ranking products cover more customers), the less incentive toextend the variety of products is expected. In addition, from the

Table 5Wilcoxon signed-rank test results of cases 11, 12, 13, 14 and 15.

Mean differences Ranks N Mean rank Sum of ranks Test stat.

Laser printer vs. all-in-one printer Negative ranks(a) 146 73.50 10,731.00 Z=−10.482pb0.001Positive ranks(b) 0 0.00 0.00

Ties(c) 154Total 300

Note: (a)all-in-oneb laser, (b)all-in-oneN laser, and (c)all-in-one=laser

LCD TV vs. TV+DVD combo Negative ranks(d) 79 40.78 3222.00 Z=−7.684pb0.001Positive ranks(e) 1 18.00 18.00

Ties(f) 49Total 129

Note: (d)combobLCD TV, (e)comboNLCD TV, and (f)combo=LCD TV

Keyboard vs. keyboard+mouse combo Negative ranks(g) 48 25.27 1213.00 Z=−4.772pb0.001Positive ranks(h) 4 41.25 165

Ties(i) 28Total 80

Note: (g)combobkeyboard, (h)comboNkeyboard, and (i)combo=keyboard

Samsung cell phone vs. Samsung cell phone w/ service Negative ranks(a) 141 71.00 10,011.00 Z=−10.302pb0.001Positive ranks(b) 0 0.00 0.00

Ties(c) 27Total 168

Note: (a)phone w/ servicebphone, (b)phone w/ serviceNphone, and (c)phone w/ service=phone

Speaker systems vs. home theater systems Negative ranks(d) 146 73.50 10,731.00 Z=−10.482pb0.001Positive ranks(e) 0 0.00 0.00

Ties(f) 104Total 250

Note: (d)home theaterbspeaker, (e)home theaterNspeaker, and (f)home theater=speaker

477J. Lee et al. / Decision Support Systems 51 (2011) 466–479

customers' point of view, the larger number of products in a marketcan cause higher searching costs [58]. Therefore, if sellers can estimatethe optimum, not maximum, number of products, it can benefit bothsellers and customers by reducing administration cost and searchingcost. In this study, we suggest that sellers should be aware of theproduct types that distinguish the optimal sizes of shelves in onlineshopping malls.

Second, this study offers strategic insights for a variety of sellersabout the management of eWOM. We show that for each producttype, the impact of eWOM is different. While most eWOM studiesfocus on the content of customer reviews, that is, favorability ofreviews [42,47] and variety of reviews [10], and so forth, not manynoticed that these reviewsmight have different impacts depending onthe type of products. For example, the findings of our study can becomparable to the observations of Elbers [17]. In fact, some of thefindings in our research show a possibility that sellers can adoptalternative product maximization strategies depending on how thetails of sales distribution would be formed. Therefore, our studysuggests that sellers should be aware of the different impacts ofeWOM for each product type, and thus cases should be differentiatedwhen managing eWOM systems.

6.2. Limitations and future research

The study has limitations in the validation process, which serve asa guide in the formulation of ideas for future research. First,developing a parameter for the total sales with the number of reviewsmay generate a bias. Although a positive linear relationship betweensales and amount of referrals has been reported in a number of studies[23], it would be more persuasive to analyze the data of the actualamount of sales. Second, although we compared similar markets withhigh and low eWOM impacts, we were not able to control all otherfactors that might have an impact on the market and sales. Ideally, weshould have collected the data sets from two shoppingmalls, onewitheWOM and one without eWOM. However, it was impossible tocontrol the product types of two different shopping malls, and thus

we compromised and made the second-best choice. Third, the resultsof this study might have limited generalizability because the samplewas collected from only one shopping mall with limited productcategories. Therefore, in future research, we may consider extendingproduct categories as well as samples sizes, collecting more data frommultiple online shopping malls. For future research, in particular, theresult would bemore credible if multiple similar products are groupedinto a category. Finally, there might be other factors that we did notconsider in the study that could affect sales distribution. For example,herding behaviors of customers might have an impact on productcategories with less objective evaluation standards. Herding has beenproven to exist widely in consumer decisions of various aspects,including choosing a restaurant, which is highly differentiated inconsumer preferences [32,64].

7. Conclusion

This study shows that the impact of eWOM on sales distribution isdifferent across product types. Although the long tail theory seems todominate online business norms, we show that eWOM can changethe rule by differentiating the cases across product types. To do so, wefirst point out that customers refer to eWOM to evaluate products.We then show that because each product type entails a differentscheme of product evaluation standards, the impact of eWOM alsovaries across product types. As a result, following the advent ofeWOM, the shapes of the sales distribution for different product typesvary, thus violating the rule of the long tail. To specify the cases indetail, we also propose a new type of product categorization based onthe objectivity of product evaluation standards and hypothesize thatthis categorization will gradually differentiate the shapes of salesdistribution. For validation, we collect data from Amazon.com. Tocompare various types of products, we plot the cumulativedistribution function of eWOM that is equivalent to sales distributionand conduct a Wilcoxon signed rank test for additional statisticalsupport. All the hypotheses are supported. With the increasingimportance of eWOM in online businesses nowadays, our research

478 J. Lee et al. / Decision Support Systems 51 (2011) 466–479

shows the variety of sales distribution influenced by eWOM andsuggests a new type of product categorization that can supportproduct-specific eWOM management.

Acknowledgements

The authors sincerely thank five anonymous reviewers for theirinsightful comments and suggestions during the review process. Thiswork was supported by a 2008 research grant from the Institute forBusiness Research and Education in Korea University.

References

[1] C. Anderson, The Long Tail: Why the Future of Business is Selling Less of More,Hyperion, New York, 2006.

[2] H.K. Bhargava, D.J. Power, D. Sun, Progress in web-based decision supporttechnologies, Decision Support Systems 43 (4) (2007) 1083–1095.

[3] G.E. Bolton, E. Katok, A. Ockenfels, How effective are electronic reputationmechanisms? An experimental investigation, Management Science 50 (11)(2004) 1587–1602.

[4] E. Brynjolfsson, M.D. Smith, Frictionless commerce? A comparison of Internet andconventional retailers, Management Science 46 (4) (2000) 563–585.

[5] E. Brynjolfsson, Y.J. Hu, M.D. Smith, From niches to riches: anatomy of the long tail,MIT Sloan Management Review 47 (2006) 67–71.

[6] N. Bunkley, Joseph Juran, 103, pioneer in quality control dies, New York Times 3(2008).

[7] M.Y. Cheung, C. Luo, C.L. Sia, H.P. Chen, Credibility of electronic word-of-mouth:informational and normative determinants of online consumer recommenda-tions, International Journal of Electronic Commerce 13 (4) (2009) 9–38.

[8] J. Chevalier, A. Goolsbee, Measuring prices and price competition online: Amazonand Barnes Noble, Quantitative Marketing Economics 1 (2) (2003) 203–222.

[9] J. Chevalier, D. Mayzlin, The effect of word of mouth on sales: online book reviews,Journal of Marketing Research 43 (3) (2006) 345–354.

[10] E.K. Clemons, G. Gao, L.M. Hitt, When online reviews meet hyperdifferentiation:a study of the craft beer industry, Journal of Management Information Systems23 (2) (2006) 149–171.

[11] W.J. Conover, Practical Nonparametric Statistics, 2nd editionJohn Wiley and Sons,NY, 1980.

[12] H. Cremer, J.-F. Thisse, Location models of horizontal differentiation: a special case ofvertical differentiationmodels, Journal of Industrial Economics39 (4) (1991)383–390.

[13] H. Degryse, On the interaction between vertical and horizontal productdifferentiation: an application to banking, Journal of Industrial Economics 44 (2)(1996) 169–186.

[14] C. Dellarocas, The digitization of word of mouth: promise and challenges of onlinefeedback mechanisms, Management Science 49 (10) (2003) 1407–1424.

[15] W. Duan, B. Gu, A.B. Whinston, Do online reviews matter? An empiricalinvestigation of panel data, Decision Support Systems 45 (4) (2008) 1007–1016.

[16] P. Dwyer, Measuring the value of electronic word of mouth and its impact inconsumer communities, Journal of Interactive Marketing 21 (2) (2007) 63–77.

[17] A. Elberse, Should you invest in the long tail? Harvard Business Review 86 (7/8)(2008) 88–96.

[18] G.M. Erickson, J.K. Johansson, The role of price in multi-attribute productevaluations, Journal of Consumer Research 12 (September 1985) 195–199.

[19] J. Fong, S. Burton, Electronic word-of-mouth: a comparison of stated and revealedbehavior on electronic discussion boards, Journal of Interactive Advertising 6 (2)(2006) 61–70.

[20] C. Forman, A. Ghose, B. Wiesenfeld, Examining the relationship between reviewsand sales: the role of reviewer identity information in electronic markets,Information Systems Research 19 (3) (2008) 291–313.

[21] T. Garber, J. Goldenberg, B. Libai, E. Muller, Density to destiny: using spatialdimension of sales data for early prediction of new product success, MarketingScience 23 (3) (2004) 419–428.

[22] A. Ghose, P. Ipeirotis, Designing novel review ranking systems: predicting theusefulness and impact of reviews, Proc. of ICEC07, 2007, pp. 303–310.

[23] A. Ghose, T. Mukhopadhyay, U. Rajan, The impact of Internet referral services on asupply chain, Information Systems Research 18 (3) (2007) 300–319.

[24] L. Hankin, The effects of user reviews on online purchasing behavior acrossmultipleproduct categories, Master's Final Project Report, UC Berkeley School of Information,May 2007, http://www.ischool.berkeley.edu/files/lhankin_report.pdf.

[25] E.v. Heck, P. Vervest, Smart business networks: concepts and empirical evidence,Decision Support Systems 47 (4) (2009) 275–276.

[26] T. Hennig-Thurau, K.P. Gwinner, G. Walsh, D.D. Gremler, Electronic word-of-mouthvia consumer-opinion platforms: what motivates consumers to articulate them-selves on the Internet? Journal of Interactive Marketing 18 (1) (2004) 39–52.

[27] E.C. Hirschman, M.B. Holbrook, Hedonic consumption: emerging concepts,methods and propositions, Journal of Marketing 46 (Summer 1982) 92–101.

[28] C.F. Hofacker, E-Tail constraints and tradeoffs, Direct Marketing: An InternationalJournal 2 (3) (2008) 129–143.

[29] S.-T. Hong, R.S. Wyer Jr., Effects of country-of-origin and product-attributeinformation on product evaluation: an information processing perspective,Journal of Consumer Research 16 (1989) 175–187.

[30] N. Hu, I. Bose, Y. Gao, Liu, Manipulation in digital word-of-mouth: a reality checkfor book reviews, Decision Support Systems 50 (3) (2011) 627–635.

[31] N. Hu, L. Liu, I. Bose, J. Shen, Does sampling influence customers in online retailing ofdigital music? Information Systems and E-BusinessManagement 7 (4) (2009) 1–21.

[32] J.-H. Huang, Y.-F. Chen, Herding in online product choice, Psychology &Marketing23 (5) (2006) 413–428.

[33] A. Jøsang, R. Ismail, C. Boyd, A survey of trust and reputation systems for onlineservice provision, Decision Support Systems 43 (2) (2006) 618–644.

[34] H.-W. Kim, S. Gupta, A comparison of purchase decision calculus betweenpotential and repeat customers of an online store, Decision Support Systems 47 (4)(2009) 477–487.

[35] R. Koch, The 80/20 Principle: The Secret of Achieving More with Less, NicholasBrealey Publishing, London, 2001.

[36] J. Lee, J.-N. Lee, Understanding product information inference process in electronicword-of-mouth: an objectivity–subjectivity perspective, Information Manage-ment 46 (2009) 302–311.

[37] A.A. Lew, Long tail tourism: new geographies for marketing niche tourismproducts, Journal of Travel and Tourism Marketing 25 (3/4) (2009) 409–419.

[38] X. Li, L.M. Hitt, Self selection and information role of online product reviews,Information Systems Research 19 (4) (2008) 456–474.

[39] Z. Lin, D. Li, B. Janamanchi, W. Huang, Reputation distribution and consumer-to-consumer online auction market structure: an exploratory study, DecisionSupport Systems 41 (2) (2006) 435–448.

[40] A. Mas-Colell, M.D. Whinston, J.R. Green, Microeconomic Theory, OxfordUniversity Press, 1995.

[41] M.I. Melnik, J. Alm, Does a seller's e-Commerce reputation matter? Evidence fromeBay auctions, Journal of Industrial Economics 50 (3) (2002) 337–350.

[42] A. Mukherjee, W.D. Hoyer, The effect of novel attributes on product evaluation,Journal of Consumer Research 28 (3) (2001) 462–472.

[43] P. Nelson, Information and consumer behavior, Journal of Political Economy 78 (2)(1970) 311–329.

[44] J.C. Panzar, R.D. Willig, Economies of scope, The American Economic Review 71(1981) 268–272.

[45] D. Park, S. Kim, The effects of consumer knowledge on message processing ofelectronic word-of-mouth via online consumer reviews, Electronic CommerceResearch and Applications 7 (4) (2008) 399–410.

[46] C. Park, T.M. Lee, Information direction, website reputation and eWOM effect: amoderating role of product type, Journal of Business Research 62 (1) (2009)61–67.

[47] P.A. Pavlou, A. Dimoka, The nature and role of feedback text comments in onlinemarketplaces: implications for trust building, price premiums, and sellerdifferentiation, Information Systems Research 17 (4) (2006) 392–414.

[48] J.E. Phelps, R. Lewis, L. Mobilio, D. Perry, Viral marketing or electronic word-of-mouth advertising: examining consumer responses and motives to pass alongemail, Journal of Advertising Research 45 (4) (2004) 333–348.

[49] Promomagazine, Staff for PROMO Xtra, Online consumer product reviewshave Big Influence, Apr 16, 2009, http://promomagazine.com/research/online-consumer-reviews-influence-0416/.

[50] M. Rosenthal, Amazon sales rank for books: graph explains what Amazon salesranks mean, 20098 available at, http://www.fonerbooks.com/surfing.htm.

[51] S. Samiee, Customer evaluation of products in a global market, Journal ofInternational Business Studies 25 (3) (1994) 582–583.

[52] R. Sanders, The Pareto principle: its use and abuse, The Journal of ServicesMarketing 1 (2) (1987) 37–40.

[53] F.M. Scherer, Pricing, profits, and technological progress in the pharmaceuticalindustry, Journal of Economic Perspectives Summer (1993) 97–115.

[54] S. Senecal, J. Nantel, The influence of online product recommendations onconsumers' online choices, Journal of Retailing 80 (1) (2004) 159–169.

[55] A. Shaked, J. Sutton, Product differentiation and industrial structure, Journal ofIndustrial Economics 36 (1987) 131–146.

[56] S. Standifird, Reputation and ecommerce: eBay auction and the asymmetrical impactof positive and negative ratings, Journal of Management 27 (3) (2001) 279–295.

[57] M.A. Strahilevitz, G.F. Loewenstein, The effect of ownership history on thevaluation of objects, Journal of Consumer Research 25 (Dec 1998) 276–289.

[58] B.-C. Su, Characteristics of consumer search on-line: how much do we search?International Journal of Electronic Commerce 13 (1) (2008) 109–129.

[59] M. Sujan, C. Dekleva, Product categorization and inference making: someimplications for comparative advertising, Journal of Consumer Research 14(1987) 372–378.

[60] E. Turban, D. King, J. Lee, D. Viehland, Electronic Commerce 2004: A ManagerialPerspective, 3rd EditionPrentice Hall, 2004.

[61] O. Turel, Y. Yuan, C. Connelly, In justice we trust: predicting user acceptanceof e-customer services, Journal of Management Information Systems 24 (4)(2008) 125–153.

[62] V. Venkatesh, R. Agarwal, Turning visitors into customers: a usability-centricperspective on purchase behavior in electronic channels, Management Science52 (3) (2006) 367–382.

[63] M. Viswanathan, T.L. Childers, Understanding how product attributes influenceproduct categorization: development and validation of fuzzy set-based measuresof gradedness in product categories, Journal of Marketing 36 (1999) 75–94.

[64] E.A. Walden, G.J. Browne, Sequential adoption theory: a theory for understandingherding behavior in early adoption of novel technologies, Journal of theAssociation for Information Systems 10 (1) (2009) 31–62.

[65] F. Wilcoxon, Individual comparisons by ranking methods, Biometrics 1 (1945) 80–83.[66] G. Zacharia, A. Moukas, P. Maes, Collaborative reputation mechanisms for

electronic marketplaces, Decision Support Systems 29 (4) (2000) 371–388.

479J. Lee et al. / Decision Support Systems 51 (2011) 466–479

[67] J. Zhang, The roles of players and reputation: evidence from eBay online auctions,Decision Support Systems 42 (3) (2006) 1800–1818.

[68] L. Zhen, G.Q. Huang, Z. Jiang, Recommender system based on workflow, DecisionSupport Systems 48 (1) (2009) 237–245.

JungLee is anAssistant Professor in theBangCollegeofBusiness at theKazakhstan InstituteofManagement, Economics, and Strategic Research. Shewas a Researcher in theDepartment ofInformationSystemsat theNationalUniversity of Singapore. She receivedPh.D. degree inMISfromKoreaUniversityBusinessSchool,M.S.degree in InformationSystems fromtheGraduateSchool of Information of Yonsei University, and B.S. degree in Biology from Korea AdvancedInstitute of Science and Technology (KAIST). Her research interests include eWOM, trust,distrust, outsourcing, and IS economics. She has published papers in Information &Management and International Journal of Electronic Commerce, and presented severalpapers at the ICIS, AMCIS, and PACIS Conferences.

Jae-Nam Lee is an Associate Professor in the Business School of Korea University inSeoul, Korea. He was formerly on the faculty of the Department of Information Systemsat the City University of Hong Kong. He holds M.S. and Ph.D. degrees in MIS from theGraduate School of Management of the Korea Advanced Institute of Science andTechnology (KAIST) in Seoul. His research interests are IT outsourcing, knowledgemanagement, e-commerce, and IT deployment and impacts on organizationalperformance. His published research articles appear in MIS Quarterly, InformationSystems Research, Journal of MIS, Journal of the AIS, Communications of the AIS, IEEETransactions on Engineering Management, European Journal of Information Systems,Communications of the ACM, Information &Management, and others. He has presentedseveral papers at the ICIS, HICSS, ECIS, DSI and IRMA Conferences, and serves on theeditorial boards of MIS Quarterly, Information Systems Research, Electronic CommerceResearch and Applications, and Pacific Asia Journal of the Association for InformationSystems.

Hojung Shin is an Associate Professor at the Korea University Business School. Heearned his MBA and Ph.D. degrees from the Fisher College of Business at the Ohio StateUniversity. Before joining Korea University, he was a faculty at the Mendoza College ofBusiness, University of Notre Dame. His research focuses on pricing-inventory models,strategic buyer–supplier relationships, and interdisciplinary issues among operationsmanagement, information technology, and marketing. His publications appear inJournal of Operations Management, Production and OperationsManagement, EuropeanJournal of Operational Research, etc. Currently, he serves as an associate editor forDecision Sciences and Journal of the Korean Operations Research and ManagementScience Society.