FOCUS THEME SECTION: INTERNET MARKETING

A b s t r a c t
An Empirical Study of Online Word of Mouth The ability to exchange opinions and experi-

as a Predictor for Multi-product Category ences online is known as online word of

mouth (WOM) and has been shown in the
e-Commerce Sales literature to have the potential to impact e-
commerce sales. The purpose of this paper is
to expand previous findings by empirically
ALANAH DAVIS AND DEEPAK KHAZANCHI evaluating the impact of online WOM
attributes and other related factors (e.g.
product views, promotion, and category) on
e-commerce sales using real data from a
multi-product retail e-commerce firm.
Research has previously shown that the
introduction of online WOM on a retail e-
commerce site can positively impact product
sales. We propose and validate a conceptual
model of online WOM and its impact on
product sales and the impact of moderator
variables such as promotion, product category
and product views. It is our conclusion that
Downloaded By: [Schmelich, Volker] At: 14:34 24 March 2010

previous research on online WOM has been

INTRODUCTION advantage over traditional WOM limited as our research empirically demon-
due to its potential to reach more strates the conclusion that it is the interac-

DOI: 10.1080/10196780802044776
Message boards, chat rooms, blogs, consumers quite rapidly. tion of product category, volume and product
user feedback forums and other Previous research has shown that views, and the interaction of product views
electronic outlets for customer gen- there is a high level of consumer and product category which are statistically
erated media have become increas- acceptance and reliance on online significant in explaining changes in unit
ingly important for todays online WOM (Henning-Thurau and Walsh product sales. Pure increase in volume or
consumers to exchange opinions and 2004) and that online WOM can number of reviewer comments has no
experiences related to companies, increase product sales (Chen et al. significant effect on sales. These conclusions
products, and services with indivi- 2004, Chevalier and Mayzlin 2006, have critical implications for the practical use
duals outside their personal commu- Davis and Khazanchi 2007, Liu of online WOM in e-commerce and for
nication network of family, friends, 2006). In fact, some authors have internet marketing.
acquaintances and colleagues proposed that online WOM may Keywords: Online word of mouth, online
(Dwyer 2007). This ability to eventually take the place of tradi- consumer reviews, online recommendations,

Volume 18 (2): 130-141. [Link]

exchange opinions and experiences tional advertising (Awad et al. online marketing, online retailers
online is known as online word of 2006). Additionally, it has been
mouth (WOM), otherwise referred suggested that very little is known
to as buzz (Liu 2006) or word of about the influence of online reviews A u t h o r s
mouse (Dellarocas 2003). WOM on consumer evaluations and pur-

Copyright 2008 Electronic Markets

has been formally defined as all chase intentions of products Alanah Davis
informal communications directed (Chatterjee 2001). (alanahdavis@[Link]) is a
at other consumers about the own- PhD Candidate at the University of
The purpose of this paper is to
Nebraska at Omaha. Her current research
ership, usage, or characteristics of assess the impact of online WOM on interests include the use of collaboration
particular goods and services or their e-commerce multi-product sales and technologies both virtually and
sellers (Westbrook 1987). The dif- our ability to predict changes in sales face-to-face, and e-commerce.
ference between traditional WOM attributable to various aspects of Deepak Khazanchi
and online WOM is that the com- online WOM. While some research- (khazanchi@[Link]) is the
munication in online WOM takes ers have proposed models for the Associate Dean for academic affairs and
place in an anonymous asynchro- impact of online WOM on sales, Professor of Information Systems and
nous online environment. Online most of them have been category Quantitative Science in the College of
WOM allows consumers to share IS&T at the University of Nebraska at
specific (i.e. movies, books, music,
Omaha. Some of his current research
both negative and positive opinions video games, beer) (Chevalier and
interests include virtual project
and experiences from multiple Mayzlin 2006, Clemons et al. management, online WOM, and B2B
sources (Chatterjee 2001). 2006, Dellarocas et al. 2004, Liu assurance in extended enterprise
However, online WOM has an 2006, Zhu and Zhang 2006). An environments.
 Electronic Markets Vol. 18 No 2 131

important contribution of our research is that we use real Weatherford, Texas as well as a start rating based on
data from a multi-product retail e-commerce firm to the reviewers opinion. The review shows the date of the
develop and validate our model of online WOM and its posting as well as an image that was uploaded from the
relationship with sales. Specifically our research addresses reviewer.
the following question: What attributes of online WOM Compared with traditional advertising (e.g. TV,
can be used to predict e-commerce sales? Newspapers, etc.), WOM is apparently perceived by
The rest of the paper is organized as follows. The next consumers as being more credible than private signals
section discusses the notion of WOM and the various and is often more accessible through social networks
attributes of WOM that can be found in a priori (Banerjee 1992, Brown and Reingen 1987, Liu 2006).
literature as well as presents our initial conceptual model, Table 1 summarizes several additional noteworthy find-
followed by a description of the research design and data ings that previous WOM researchers have identified.
collection. The following section describes the two As illustrated in Table 1, a large number of extant
stages of model refinement and development followed research articles on WOM in the marketing literature
by an analysis of the collected data in relation to the final have primarily focused on studying personal relation-
proposed model. The final sections discuss the results, ships as the basis of communication about products or
conclusions, and implications of our research as well as services (Stewart et al. 1985). These ties among
suggesting directions for future research. consumers may be based on either strong or weak
personal social relationships (Brown and Reingen 1987).
In contrast, todays online consumers can exchange
BACKGROUND opinions and experiences related to companies, pro-
Downloaded By: [Schmelich, Volker] At: 14:34 24 March 2010

ducts, and services with individuals outside of their

Previous research personal communication network of family, friends,
acquaintances, and colleagues by writing their experi-
WOM relies on information that is communicated about ences and opinions in blogs, user feedback forums,
companies, products, or services among consumers. search engines, or shopping bot sites (e.g. pricescan.-
Figure 1 shows an example of online WOM in which a com). In this online environment, the personal ties are
customer has posted a review and rating based on their always weak due to the fact that the person who wrote
experience with a product. The review includes a written the review and the person relying on the review do not
opinion from a verified purchaser named Sebastian from have a personal relationship (Chatterjee 2001).

Figure 1. Online word of mouth example

 132 Alanah Davis and Deepak Khazanchi & Empirical Study of Online Word of Mouth

Table 1. List of noteworthy examples of previous research findings from WOM studies

Article Key Findings

Amblee and Bui 2007a Expert reviews and user reviews are nearly identical in significance and importance when predicting sales;
existence of either type of review matters, however valence does not
Amblee and Bui 2007b Book sales are correlated to the number of reviews, the reputation of the author, and the reputation of
complementary goods; book ratings by readers are not a good predictor of book sales
Awad et al. 2006 Participation in and use of online WOM is complementary to participation in and use of offline WOM; the
use of online WOM is a substitute for traditional advertising
Awad and Zhang 2006 Online ratings on a website are significantly correlated with online purchases; a firms filtering strategy
effects the impact of ratings on purchases
Clemons et al. 2006 Variance of ratings plays a significant role in determining which new product sales grow the fastest in a
review of the beer industry
Hu et al. 2006 The average ratings score does not necessarily reveal a products true quality and may provide misleading
recommendations
Liu 2006 WOM activities are at their highest during the prerelease of a movie and opening week; audiences hold
high expectations before a release and become more critical during the opening week; the volume of
WOM offers explanatory power for both aggregate and weekly box office revenue more so than the
valence of WOM
Downloaded By: [Schmelich, Volker] At: 14:34 24 March 2010

Zhu and Zhang 2006 Online reviews have a significant influence on sales; one point increase in average rating is associated
with a 4% increase in sales; negative ratings have a larger impact than positive ratings; reviews are more
influential for less popular products
Chen et al. 2004 More recommendations improve sales; consumer ratings are not related to sales; the number of consumer
reviews is positively associated with sales; recommendations work better for less-popular products
Godes and Mayzlin 2004 The more conversation there is about a product the more likely someone is to be informed about it and the
greater the sales
Henning-Thurau and Walsh 2004 Consumers read online articulations to save decision making time and make better buying decisions; these
motives influence their behaviour
Anderson and Salisbury 2003 Advertising, WOM, market growth, and purchase frequency have a significant moderating influence on
adaptation rate
Chevalier and Mayzlin 2003 WOM from customers has a causal impact on consumer purchasing behaviour
Bowman and Narayandas 2001 Engaging in WOM behaviour following customer initiated contact, the median number of customers
influenced is approximately three regardless of factors; there is substantial variation in whether
customers tell others about customer initiated contact experiences and the number of people they tell
Chatterjee 2001 WOM search depends on the consumers reasons for choosing an online retailer; the influence of negative
WOM on perceived reliability and intentions is determined largely by retailer familiarity
Dellarocas 2000 The combination of controlled anonymity and cluster filtering is a powerful technique for improving the
reliability of reputation systems
Duhan et al. 1997 There are different influences on the likelihood of consumers choosing different types of recommendation
sources; choosing strong tie sources is influenced by task difficulty and prior knowledge; choosing weak
tie sources is influenced by the importance of instrumental cues and subjective prior knowledge
Ellison and Fudenberg 1995 WOM may lead people to adopt an action that is on average superior
Banerjee 1992 People will do what others are doing (herd behaviour) rather than rely on their own information
Anderson 1998 Dissatisfied customers engage in greater WOM than satisfied customers, however this difference appears
to be exaggerated
Brown and Reingen 1987 Strong ties are more influential and more likely to be used than weak ties
Westbrook 1987 Affective response is related to the favourability of consumer satisfaction in judgments, the extent of
seller directed complaint behaviour, and the extent of WOM transmission
Richins 1983 The nature of dissatisfaction, consumers attributions for blame, and perceptions of complaint situations
are related to dissatisfaction responses.

CHARACTERISTICS OF ONLINE WORD OF MOUTH awareness, attitude, expectations, perceptions, inten-

tions, and behaviour (Buttle 1998, Liu 2006). These
WOM research has shown that WOM has an influence cognitive conditions can be influenced by various
on an assortment of cognitive conditions that includes characteristics of online WOM as suggested by Liu
 Electronic Markets Vol. 18 No 2 133

(2006). For example, Liu (2006) suggests that the valence does not have an effect on product sales
WOM characteristic of volume (i.e., the total amount (Amblee and Bui 2007a, Davis and Khazanchi 2007).
of WOM) produces the cognitive consequence of However, Chen and Singh (2001) suggest that online
awareness while the WOM characteristic of valence ratings have become increasingly important because they
(i.e., whether the attitude is positive or negative) produces allow users to harvest the wisdom of the community in
the cognitive consequence of attitude. Furthermore, making decisions. Zhu and Zhang (2006) studied the
previous research has clearly demonstrated that the influence of consumer ratings on video game sales and
notions of volume and valence are two of the most showed that a higher rating by only one point was
important WOM attributes (Amblee and Bui 2007a, associated with a 4% increase in sales. Furthermore, there
Liu 2006). The following subsections will discuss is some evidence to suggest that valence is the best
volume and valence in greater detail as well as two predictor of sales among all of the WOM attributes
additional attributes of online WOM, visual cues and (Dellarocas et al. 2004).
reviewer type.

Visual cues
Volume
We define visual cues as any image (a form of
According to Liu (2006), volume measures the total communication) posted by a reviewer and directed at
amount of WOM interactions. Referring back to Figure 1 other consumers when evaluating the characteristics of a
this would mean the number of comments from particular good or service. The image included with the
Downloaded By: [Schmelich, Volker] At: 14:34 24 March 2010

reviewers about a specific product or service. The reviewers rating of a product in Figure 1 illustrates this
existence of online WOM results in an increase in additional ability to verify that their dog enjoys the bean
awareness and a positive (or negative) attitude towards a bag. This notion of visual cues as an attribute of online
product resulting in a change in sales (A lvarez et al. WOM initially emerged when we asked ourselves if sales
2007). Thus, the more conversation in the form of would be impacted when reviewers comments could
online comments there is about a product, the more include images of products that they purchased online
likely someone will be informed about it, i.e., there from the e-commerce site. Our initial conclusion that
is increased awareness of the product (Godes and visual cues should be part of our online WOM model is
Mayzlin 2004). This will in consequence result in supported by research in the marketing field by Fang and
higher product sales. Many previous studies have shown Salvendy (2003) which concludes that pictures of
that the WOM volume significantly correlates with products are necessary to provide customers visual cues
consumer behaviour and market outcome (Amblee and and richer information. This belief is further supported
Bui 2007b, Anderson and Salisbury 2003, Bowman and by results from previous research of image use in an
Narayandas 2001, Liu 2006). Additionally, Mayzlin online environment. For example, a study by Zheng et
(2006) has found that rational consumers still pay al. (2002) suggests that in virtual teams exposure to
attention to anonymous on-line posts, even when it is photographs prior to team interaction increases trust.
possible for firms to pose as online consumers. For the Similarly, another study found that in newly formed
most part, researchers have concluded that online virtual teams images of team members promoted
customer reviews have a significant influence on the sale affection and social attraction (Walther et al. 2001).
of products (Awad and Zhang 2006, Zhu and Zhang Could this same idea be true for increasing trust of
2006). However, some researchers have argued that only reviewer comments and ratings (i.e., online WOM)? In
the volume of the reviews matter (Chen et al. 2004, contrast, a study of [Link] (the German Amazon e-
Duan et al. 2005). commerce shopping site) found that displaying photos
of people or employees of an e-commerce site is
generally not advisable as study participants reacted with
Valence suspicion; however this negative perception may partly
be due to the local cultural context (Riegelsberger and
Valence is the idea that WOM can be either positive or Sasse 2002). Instead of individual team members in this
negative (Buttle 1998). According to Liu (2006), example, if consumers of online products write reviews
valence measures the nature of the WOM message and and have the opportunity to upload images of their
whether it is positive or negative. In the example provided products in use, would the result be different? In this
by Figure 1, the valence of online posting is positive as vein, a recent study by Lurie and Mason (2007) has
the reviewer rated the product with five stars and has specifically related visual cue research to WOM research,
corresponding comments that are also positive. think a picture is worth a thousand words. Specifically,
Behavioural research has shown that it is unclear Lurie and Mason (2007) present the characteristics of
whether positive WOM leads to increased sales 1) visual perspective the interactivity and detailed
(Anderson 1998). Various studies have shown that extent of a product representation and 2) information
 134 Alanah Davis and Deepak Khazanchi & Empirical Study of Online Word of Mouth

context the vividness, evaluability, and information INITIAL CONCEPTUAL MODEL

framing. For example, Lands End website offers users
the ability to interact with clothes as though trying on As discussed in the previous sections, research has shown
clothes. The e-commerce site used in our study offers that WOM has an influence on an assortment of
users the ability to see products in the homes and other cognitive conditions including awareness, attitude,
locations of users who are both satisfied and dissatisfied expectations, perceptions, intentions and behaviour
with the product as another form of online WOM and (Buttle 1998, Liu 2006). These cognitive conditions
passing along to the user visual cues in terms of product can be influenced by various characteristics of online
expectations. Lurie and Mason (2007) propose that WOM (Liu 2006). For example, Liu (2006) suggests
this visual perspective leads to a customers process of that the WOM characteristic of volume produces the
decision making. cognitive consequence of awareness while the WOM
characteristic of valence produces the cognitive conse-
quence of attitude. We further propose that the
Reviewer type cognitive consequences of expectation and perception,
which have been identified in the literature (Buttle
Previous research has shown that the influence of WOM 1998, Liu 2006), can potentially influence the relation-
can be so strong that it overrides private signals and ship between online WOM attributes of visual cues and
results in individuals relying solely on the information reviewer type on product sales. It is important to note
provided by reviewers (Banerjee 1992, Ellison and here that the four of these cognitive consequences do
Fudenberg 1995). The notion of a reviewer type is not not mediate the relationship between online WOM and
Downloaded By: [Schmelich, Volker] At: 14:34 24 March 2010

a new concept as previous studies have empirically tested product sales, but instead explain why volume and
valence can impact online behaviour (this is why the
models of WOM which consider expert reviews versus
middle box has a dotted line). Figure 2 shows our initial
user reviews (e.g. Amblee and Bui 2007a, Kumar and
conceptual model based on previous research.
Benbasat 2006, Smith et al. 2005). In contrast, we
Figure 2 shows a + in the relationship between
define reviewer type as the nature of the individual
volume and sales because previous research has shown
participating in online WOM. Referring back to
that volume is a reliable predictor of sales (Amblee and
Figure 1, the individual relying on the review can see
Bui 2007a, Davis and Khazanchi 2007, Liu 2006). The
that the reviewer in this case was a verified purchaser.
three other dimensions of online WOM show a ? since
Online reviews originate from various sources and are
the relationship between the online WOM attributes and
generally provided by either expert reviewers or indivi-
product sales is either unknown or previous research on
duals who have had either a very negative or very positive these relationships has been inconclusive or ambiguous.
experience (Clemons et al. 2006). Researchers have Figure 2 also depicts three variables (promotion i.e.
studied the various sources that reviews come from whether or not the product is on sale, product views
including peer reviews, editorial reviews or recommen- i.e., the number of product page impressions each week,
dations, website recommendations, or online agent and product category) that have the potential of
recommendations (Kumar and Benbasat 2006, Senecal moderating the relationship between online WOM
and Nantel 2004, Smith et al. 2005). In their research, attributes and sales. Prior research on WOM has shown
Pollach (2006) asserts that reviewers may be either that the inability to deal with promotion and product
manufacturers or merchants attempting to promote views can be a liming factor in generalizing results (Davis
their own products and therefore need to convince and Khazanchi 2007). Since we were given access to real
readers that they are both experts and trustworthy. e-commerce sales data from a multi-product retail e-
Smith et al. (2005) in particular suggest that peer commerce web site we are able to overcome this
reviews (or recommendations) are the most preferred by limitation. Additionally, many proposed models have
consumers. They found that even when the peer is only been able to explain the impact of online WOM on
described as being low in rapport and expertise, peer e-commerce sales in terms of one product category (e.g.,
recommendations are still used more than editorial movies, books, music, video games, beer) (Chevalier and
recommendations (Smith et al. 2005). On the other Mayzlin 2006, Clemons et al. 2006, Dellarocas et al.
hand, Senecal and Nantel (2004) found that online 2004, Liu 2006, Zhu and Zhang 2006).
recommendation systems are the most influential source In terms of differences in the impact of online WOM
in consumers product choice process, although peer across product categories, researchers have proposed
consumers were perceived more trustworthy. Despite that consumers are likely to search online for search goods
the differences in motivation for posting reviews both such as books and toys, in fact they may prefer to shop
reviews from professional reviewers (i.e. experts) and online (Choi et al. 2006, Girard et al. 2003).
consumers (i.e. users) are nearly identical in significance Additionally, consumers like to take the time and effort
and importance when it comes to impacting product to see, smell, test, and feel experience goods like jewellery
demand (Amblee and Bui 2007a). and perfume (Choi et al. 2006, Girard et al. 2003).
 Electronic Markets Vol. 18 No 2 135

Figure 2. The impact of online word of mouth measures on potential sales: Initial conceptual model
Downloaded By: [Schmelich, Volker] At: 14:34 24 March 2010

Researchers have also found that credence goods such as et al. 2003), most of our subject firms products would
vitamins or air purifiers involve brand consciousness, risk be classified under credence products as they are all
and uncertainty; in this case consumers prefer to base significant investments in terms of price and present a
searches on WOM and/or the experience of others potential risk to the customer. This is new and is in
(Choi et al. 2006, Girard et al. 2003). In the data used contrast with previous research which has primarily
for research study, products are categorized by the focused on online WOM with regards to products that
formal attributes given by the subject organization. are generally considered search goods.
These categories include hammocks, benches,
Adirondack chairs, wind chimes, dartboards, globes,
bean bags, barstools, dog houses, pot racks, clocks, toy RESEARCH DESIGN
boxes, kitchen islands, wine racks, hand trucks and
computer desks. If we think about these in terms of Data for this research was collected from a leading multi-
product classifications such as search goods, experience product retail e-commerce company that has been in
goods and credence goods (Choi et al. 2006, Girard business since 2001 and has over 300 employees. To

Table 2. Data collected from leading e-commerce company

Data Collected Instrumentation of model variables

(Variable name) Definition (Level of Measurement)

SKU Product identification, i.e. SKU number of the product

Name Name of the product
Product Category (CatID) Identification number for the category the Random number value that ranges from 1 to the total
product is a part of number of products sold by the firm (650+)
Product Category The name of the category the product falls under Textual description
Description (CatName)
Product Views (ProdViews) The number of product page impressions that week Numerical value (Scale)
Sales (NumSold) The number of products sold that week Numerical value (Scale)
Valence (AvgRat) Average reviewer rating for the product in Numerical value (Ordinal; 1 or Low5*; *****55 or High)
the form of star ratings
Volume (NumRev) Total number of review comments for the product Numerical value (Scale)
Promotion Whether or not the product was on sale Numerical value (Ordinal; 15On sale; 05Not on sale)
Reviewer Type The type of reviewers for a product Numerical value (Ordinal; 15Verified purchaser;
(ReviewerType) 25Verified reviewer; 35Not Verified; 45Product
with more than one type of reviewer)
Visual Cues (Image) Whether or not the reviews contained reviewer Numerical value (Ordinal; 15Visual cue is present;
uploaded product images 05Visual cue is absent)
 136 Alanah Davis and Deepak Khazanchi & Empirical Study of Online Word of Mouth

Figure 3. Categories represented in data set

date the organization has served over one million different time periods. In order to minimize the effect of
customers in various categories of products. The details the holiday shopping period on online product sales, we
of the data that was collected can be found in Table 2. deliberately chose to look at sales and other related data
Online reviews are collected either from the product for January 2007 after the winter break, specifically the
Downloaded By: [Schmelich, Volker] At: 14:34 24 March 2010

page of the website or through email (e.g. usually from a week beginning 15 January 2007 and prior to the
reply email after the customer has received the shipment holiday shopping period for Memorial Day in May
notice email). The collected reviews appear as shown in 2007, specifically the week commencing 14 May 2007.
Figure 1; including a star rating on a five point scale, Initially, the data set included about 15% of the products
review title, reviewer name and type (i.e. either verified sold from the online retailer (i.e. 546 products from 73
purchaser, verified reviewer, or not verified), reviewer categories). To allow us the ability to use the general
city and state, date of review, and reviewers overall linear model (GLM) statistical procedure effectively to
recommendation statement. Additionally, reviewers are analyze our conceptual models, the data set was further
permitted to suggest pros, cons and write various other reorganized to only include product categories where
statements in terms of best uses or bottom line over 10 products had been sold. A large number (i.e.
recommendations. Reviewers can also post images of 70%) of the categories not included in the final data set
the product in use as shown in Figure 1. were ones that had five or fewer items sold. Therefore,
The particular review system used by the subject the resulting data set included 15 product categories and
organization filters product reviews as do other review 328 unique products (refer to Figure 3).
systems (Awad and Zhang 2006). This is to say that all Table 3 provides a summary of the descriptive statistics
reviews related to product quality are posted, including relating to the first data set, (i.e. products sold the
negative ones. However, reviews are filtered to exclude during week beginning 15 January 2007).
any comments related to price, customer service issues, Table 4 provides a summary of the descriptive statistics
and other non-product quality data. relating to the data set from the second time period (i.e.
To evaluate our initial conceptual model and validate products sold during the week beginning 14 May 2007.
our final refined model, a random set of product data The data from the second time period includes the exact
was collected from the subject organization at two
Table 4. Refined model descriptive statistics (N5328)
Table 3. Initial model descriptive statistics (N5328)
Standard
Std. Range Minimum Maximum Mean deviation
Range Minimum Maximum Mean deviation
Product 102 1 103 N/A N/A
Product category 102 1 103 N/A N/A category
Product views 1326 0 1326 185.86 204.570 Product views 2554 0 2554 124.44 206.76
Promotion 1 0 1 N/A N/A Promotion 1 0 1 N/A N/A
Reviewer type 3 1 4 N/A N/A Reviewer type 3 1 4 N/A N/A
Sales 85 1 86 N/A N/A Sales 176 0 176 8.16 15.38
Valence 2 3 5 4.50 .485 Valence 4 1 5 4.44 .53
Visual cue 1 0 1 N/A N/A Visual cue 1 0 1 N/A N/A
Volume 60 1 61 3.74 5.046 Volume 110 0 110 8.54 10.48
 Electronic Markets Vol. 18 No 2 137

Table 5. GLM results for initial conceptual model (No interaction terms included)

Source Type III Sum of squares df Mean square F p-value

Intercept Hypothesis 41.513 1 41.513 1.102 .295

Error 11988.254 318.129 37.684
Reviewer type Hypothesis 11.532 3 3.844 .105 .957
Error 11153.216 304 36.688
Visual cues Hypothesis 178.088 1 178.088 4.854 .028*
Error 11153.216 304 36.688
Product category Hypothesis 2075.511 15 138.367 3.771 .000*
Error 11153.216 304 36.688
Valence Hypothesis 46.759 1 46.759 1.274 .260
Error 11153.216 304 36.688
Product views Hypothesis 878.638 1 878.638 23.949 .000*
Error 11153.216 304 36.688
Volume Hypothesis 3538.964 1 3538.964 96.461 .000*
Error 11153.216 304 36.688
Promotion Hypothesis 1.741 1 1.741 .047 .828
Error 11153.216 304 36.688
Downloaded By: [Schmelich, Volker] At: 14:34 24 March 2010

Notes: *5significant WOM attributes

same set of products and categories as from the first set (p,.05) predictors of the number of products sold.
of data; the data is just based on the sales from a different However, the online WOM attributes reviewer type,
weekly period. valence, and moderator variable promotion do not have
a significant effect on unit product sales. The lack of
significance of valence is in line with previous research
MODEL REFINEMENT AND DEVELOPMENT which has shown that it is unclear whether positive
WOM in terms of valence or average reviewer ratings
Data analysis for initial conceptual model leads to increased sales (Amblee and Bui 2007a, Amblee
and Bui 2007b, Anderson 1998).
The initial conceptual model (refer to Figure 2) was
evaluated using the data from the January time frame. As
stated previously, we utilized the GLM univariate Refined conceptual model and hypotheses
procedure and used the statistical package, SPSS, to
analyze the following initial model: Based our findings of an initial analysis of the January
data using GLM, we developed a new conceptual model.
Model: Intercept + Reviewer Type + Visual Cues (Images) + Figure 4 displays a refined, more parsimonious version of
Category + Valence (Average Rating) + Product Views + Volume the initial conceptual model after removing all indepen-
(Number of Reviews) + Promotion dent and moderator variables that were not statistically
significant in the initial analysis.
An analysis of the data to evaluate the homogeneity of Based on the refined conceptual model shown in
variances using Levenes test of equality of error variances Figure 4, we propose the following hypotheses:
shows that there is no significant difference in variance
across the groups of scores (p-value5.23). This along
with the residual plot confirms that the data is distributed
normally and that error variances are equal. An analysis of
the scatter plot of online WOM volume and product sales
shows no apparent non-linear relationship. These assess-
ments satisfy the primary data assumptions for interpret-
ing the results of the standard GLM procedure. The
results of the GLM analysis are shown in Table 5.
The statistical analysis of our initial conceptual model
shows that the main effects for online WOM attributes,
visual cues and volume, and moderator variables product Figure 4. The impact of online word of mouth measures on
category and product views are statistically significant potential sales: Refined conceptual model
 138 Alanah Davis and Deepak Khazanchi & Empirical Study of Online Word of Mouth

Table 6. GLM results for refined conceptual model

Source Type III sum of squares df Mean Square F p-value

Intercept Hypothesis 283.909 1 283.909 4.606 .035*

Error 5124.133 83.127 61.642
Visual cues Hypothesis 115.520 1 115.520 1.926 .170
Error 3947.026 65.806 59.980
Product category Hypothesis 1225.475 15 81.698 1.165 .399
Error 857.034 12.218 70.143
Volume Hypothesis 11.727 1 11.727 .216 .643
Error 15181.094 279 54.413
Product views Hypothesis 755.144 1 755.144 13.878 .000*
Error 15181.094 279 54.413
Volume * product views Hypothesis 115.444 1 115.444 2.122 .146
Error 15181.094 279 54.413
Visual cues * product views Hypothesis 838.893 1 838.893 15.417 .000*
Error 15181.094 279 54.413
Product category * volume Hypothesis 2794.326 15 186.288 3.424 .000*
Error 15181.094 279 54.413
Downloaded By: [Schmelich, Volker] At: 14:34 24 March 2010

Visual cues * product Hypothesis 903.887 13 69.530 1.278 .226

category Error 15181.094 279 54.413

Notes: *5significant WOM attributes

Hypothesis 1: Products with higher volume (i.e. number of online Model: Intercept + Visual Cues + Product Category + Volume +
customer review comments) have a higher number of online Product Views + Visual Cues * Product Category + Volume * Visual
purchases (sales). Cues + Volume * Product Views.
Hypothesis 2: Products with a higher number of reviewer
submitted visual cues (i.e. images) have a higher number of We specifically include interaction terms in this analysis
online purchases. to test specifically our hypotheses relating to the
Hypothesis 3: Products with a higher number of product views moderating effects of product views and product
have a higher number of online purchases. category on changes in product sales. Table 6 displays
Hypothesis 3.1: Product views moderates the relationship the results from the GLM analysis.
between volume and sales. That is, the relationship between
The analysis of data once again demonstrates equality
volume and sales will change depending on the number of
product views. of error variances. Specifically, an analysis of the data to
Hypothesis 3.2: Product views moderates the relationship evaluate the homogeneity of variances using Levenes
between visual cues and sales. That is, the relationship between test of equality of error variances shows that there is no
visual cues and sales will change depending on the number of significant difference in variance across the groups of
product views. scores. The following section presents the statistical
Hypothesis 4: Category will have an effect on the number of analysis and results of the refined conceptual model in
online purchases. detail in relation to our hypotheses.
Hypothesis 4.1: Category moderates the relationship between
volume and sales. That is, the relationship between volume and
sales will change depending on category of product. DISCUSSION OF RESULTS
Hypothesis 4.2: Category moderates the relationship between
visual cues and sales. That is, the relationship between The following discussion evaluates our online WOM
visual cues and sales will change depending on category of model overall and the related hypothesis.
product.
Hypothesis 1: Products with higher volume (i.e. number of online
customer review comments) have a higher number of online
Data analysis for refined model purchases (sales).

To test our refined model (refer to Figure 4) and the As shown in Table 6, we fail to reject the null hypothesis.
related hypotheses, we used the May data, mentioned Thus, the main effect for the online WOM attribute,
earlier, and implemented it for the GLM procedure in volume, is not significant by itself in explaining variances
SPSS as follows: in product sales. This conclusion contradicts extant
 Electronic Markets Vol. 18 No 2 139

research results regarding the positive impact of online are similar to previous research in that the more popular
WOM volume on sales Researchers have previously a specific product, the more likely sales will increase
shown that volume measured in terms of the number of (Chevalier and Mayzlin 2003, Zhu and Zhang 2006).
online customer reviews does have a significant influence However, it is important to note that our research shows
on the online sale of products (Awad and Zhang 2006, that this relationship is moderated by the presence or
Zhu and Zhang 2006). In fact, a study from Duan et al. absence of reviewer uploaded images (visual cues)
(2005) suggests that only the volume matters in relation available for a product. Figure 5 illustrates the post hoc
to sales interest, not whether the reviews are positive or analysis of visual cues as they relate to product sales.
negative. One explanation for the difference in results is Clearly, many of the product categories with images
that previous researchers have primarily focused on a seem to have higher sales.
single product or product type such as movies, books,
music, video games, beer (Chevalier and Mayzlin 2006, Hypothesis 4: Category will have an effect on the number of
Clemons et al. 2006, Dellarocas et al. 2004, Liu 2006, online purchases.
Zhu and Zhang 2006). Based on our findings it can be Hypothesis 4.1: Category moderates the relationship between
concluded that the interaction of product category and volume and sales. That is, the relationship between volume and
volume is more important for multi-product ecommerce sales will change depending on category of product.
sites. Though increased volume of online WOM can Hypothesis 4.2: Category moderates the relationship between
increase the buzz about a product (there is a high visual cues and sales. That is, the relationship between visual
cues and sales will change depending on category of product.
correlation), increased sales are not necessarily guaran-
teed for all products with consumer comments.
Downloaded By: [Schmelich, Volker] At: 14:34 24 March 2010

Our overall hypothesis about the main effect, product

Hypothesis 2: Products with a higher number of reviewer category, is not supported by the data. This implies that
submitted visual cues (i.e. images) have a higher number of the type of product (category) by itself is not an
online purchases. important factor in predicting e-commerce product
sales. However, we do show that the interaction of
Table 6 shows that we fail to reject the null hypothesis, products category and volume (i.e. number of com-
i.e., the main effect for visual cues is not statistically ments about a product) has a significant effect on
significant in explaining variances in product sales. Thus, product sales (i.e. Hypothesis 4.1 is supported) and the
product pages that include reviewer submitted images product category does not moderate the relationship
are not by themselves enough to effect product sales. between visual cues and product sales (i.e. Hypothesis
4.2 is not supported).
Hypothesis 3: Products with a higher number of product views The previous results imply that product categories
have a higher number of online purchases. with higher online WOM represented by volume of
Hypothesis 3.1: Product views moderates the relationship between comments will have a higher number of purchases than
volume and sales. That is, the relationship between volume and those categories that have lower volume of comments. This
sales will change depending on the number of product views. is new to WOM research in that the proposed models have
Hypothesis 3.2: Product views moderates the relationship
between visual cues and sales. That is, the relationship between
visual cues and sales will change depending on the number of
product views.

The results in Table 6 show that the overall hypothesis is

supported and that the main effect, product views, is
statistically significant in explaining the increase in
product sales. Furthermore, we can see that the
hypothesis about the effect of the interaction between
product views and volume on product sales is not
rejected. However, the effect of the interaction between
product views and visual cues on product sales is
statistically significant and we reject the null.
These results suggest that, in general, product pages
with a higher number of customer views will produce a
higher number of purchases. This conclusion seems to
make intuitive sense in that more the number of
customers viewing a product page the more popular
that product may be and therefore the more likely sales
of that product will increase. Our results in this regard Figure 5. The estimated marginal means of numbers of units sold
 140 Alanah Davis and Deepak Khazanchi & Empirical Study of Online Word of Mouth

all been category specific (i.e. movies, books, music, or satisfaction with purchase. Another important area of
video games, beer) in the impact of online WOM on researcher that is still unanswered is why valence is
sales (Chevalier and Mayzlin 2006, Clemons et al. 2006, sometimes a predictor of sales (Chen and Singh 2001,
Dellarocas et al. 2004, Liu 2006, Zhu and Zhang 2006). Dellarocas et al. 2004, Zhu and Zhang 2006) and
sometimes not (as in this study). Finally, it might be
interesting to further explore the impact of online WOM
CONCLUSION, IMPLICATIONS AND FUTURE on unit sales for products grouped by classification
RESEARCH schemes such as search goods versus credence goods.

The overarching aim of this study is to explore the

influence of online WOM attributes on e-commerce References
sales using actual data from a multi-product retail e- lvarez, L. S., Martn, A. M. D. and Casielles, R. V. (2007)
A
commerce website. We propose and validate a con- Relationship Marketing and Information and
ceptual model to explain the impact of online WOM Communication Technologies: Analysis of Retail Travel
attributes on product sales. Our research contributes to Agencies, Journal of Travel Research 45(4): 453.
the field in five key ways: (1) the use of real e- Amblee, N. and Bui, T. (2007a) Freeware Downloads: An
commerce multi-product sales data to study the impact Empirical Investigation into the Impact of Expert and User
of online WOM on change in product sales (as opposed Reviews on Demand for Digital Goods, Americas
to prior research studies which have mostly used Conference on Information Systems, Keystone, Colorado.
perception measures), (2) the accounting of moderator Amblee, N. and Bui, T. (2007b) The Impact of Electronic-
Downloaded By: [Schmelich, Volker] At: 14:34 24 March 2010

variables such as promotion, product views, and product Word-of-Mouth on Digital Microproducts: An Empirical
category in our models, (3) the statistically conclusive Investigation of Amazon Shorts, 15th European Conference
finding that main effects due to online WOM attributes on Information Systems, St Gallen, Switzerland.
volume, valence, visual cues, and reviewer type do not by Anderson, E. W. (1998) Customer Satisfaction and Word of
themselves explain changes in product sales, and (4) the
Mouth, Journal of Service Research 1(5): 517.
statistically conclusive finding that moderator variables
Anderson, E. W. and Salisbury, L. C. (2003) The Formation
(product category and promotion) do not by themselves
of Market-level Expectations and its Covariates, Journal of
explain changes in product sales, but, product views by
Consumer Research 30(1): 11524.
themselves can explain some of the change in product
Awad, N. F., Dellarocas, C. and Zhang, X. (2006) The Digital
sales. Furthermore, products that are visited by custo-
Divide of Word of Mouth, 1st Midwest United States
mers and have visual cues in the form of reviewer images
Association for Information Systems Conference (MWAIS
are more likely to result in unit sales, and (5) the
01), Grand Rapids, Michigan.
statistically conclusive finding that specific products that
Awad, N. F. and Zhang, J. (2006) A Framework for
also have higher online WOM in the form of volume of
Evaluating Organizational Involvement in Online Ratings
comments can impact their sales.
Communities, 1st Midwest United States Association for
One of the most important implications for practice is
the idea that e-commerce companies need to have a stra- Information Systems Conference (MWAIS01), Grand
tegy of combining efforts to increase product views and the Rapids, Michigan.
volume of comments on specific product pages. This Banerjee, A. V. (1992) A Simple Model of Herd Behavior,
research also points out the value in allowing reviewers Quarterly Journal of Economics 107(3): 797817.
to upload product images, probably boosting the cognitive Bowman, D. and Narayandas, D. (2001) Managing Customer-
consequence of expectations about a product. initiated Contacts with Manufacturers: The Impact on Share
Our research confirms and provides a better under- of Category Requirements and Word-of-mouth Behavior,
standing of previous conclusions about the attributes of Journal of Marketing Research 38(3): 28197.
online WOM and its potential impact on product sales Brown, J. J. and Reingen, P. H. (1987) Social Ties and Word-
and also puts into play the notion that there are other of-mouth Referral Behavior, Journal of Consumer Research
factors (i.e. product views and categories) that influence 14(3): 350.
multi-product e-commerce sales and work in conjunc- Buttle, F. A. (1998) Word of Mouth: Understanding and
tion with online WOM attributes such as visual cues and Managing Referral Marketing, Journal of Strategic
volume. Marketing 6(3): 24154.
Further research needs to be conducted to verify our Chatterjee, P. (2001) Online Reviews: Do Consumers use
overall conclusion that online WOM by itself will not Them?, in: M. C. Gilly and J. Myers-Levy (eds) Association
increase sales but that such a result can only be obtained for Consumer Research 2001 Proceedings, Provo, UT:
by a conjoint increase in product views and an increase in Association for Consumer Research, pp. 12934.
online WOM volume for specific product categories. Chen, M. and Singh, J. P. (2001) Computing and Using
Another extension to this research may be to better Reputations for Internet Ratings, proceedings of the 7th
understand how online WOM effects customer loyalty ACM conference on Electronic Commerce, Tampa, Florida.
 Electronic Markets Vol. 18 No 2 141

Chen, P. S., Wu, S. and Yoon, J. (2004) The Impact of Online Godes, D. and Mayzlin, D. (2004) Using Online
Recommendations and Consumer Feedback on Sales, 25th Conversations to Study Word-of-mouth Communication,
International Conference on Information Systems, Marketing Science 23(4): 54560.
Washington, DC. Henning-Thurau, T. and Walsh, G. (2004) Electronic Word
Chevalier, J. and Mayzlin, D. (2003) The Effect of Word of of Mouth: Motives for and Consequences of Reading
Mouth on Sales: Online Book Reviews, Yale School of Customer Articulations on the Internet, International
Management, Working Paper Series: Economics & Journal of Electronic Commerce 8(2): 5174.
Marketing ES#28 & MK#15, available online at: Hu, N., Pavlou, P. A. and Zhang, J. (2006) Can Online
[Link] Reviews Reveal a Products True Quality? Empirical Findings
Chevalier, J. A. and Mayzlin, D. (2006) The Effect of Word of and Analytical Modeling of Online Word of Mouth
Mouth on Sales: Online Book Reviews, Journal of Communication, proceedings of the 7th ACM conference
Marketing Research 43(9): 9. on Electronic Commerce, Ann Arbor, Michigan.
Choi, S. Y., Choi, B. and Lee, H. (2006) Categorizing Kumar, N. and Benbasat, I. (2006) The Influence of
Commercial Products for Customer Oriented Online Recommendations and Consumer Reviews on Evaluations
Retailing, Computers and Industrial Engineering 51(1): of Websites, Information Systems Research 17(4): 42539.
90101. Liu, Y. (2006) Word of Mouth for Movies: Its Dynamics and
Clemons, E. K., Gao, G. G. and Hitt, L. M. (2006) When Impact on Box Office Revenue, Journal of Marketing 70:
Online Reviews Meet Hyperdifferentiation: A Study of the 7489.
Craft Beer Industry, Journal of Management Information Lurie, N. H. and Mason, C. H. (2007) Visual Representation:
Systems 23(2): 14971. Implications for Decision Making, Journal of Marketing
Downloaded By: [Schmelich, Volker] At: 14:34 24 March 2010

Davis, A. and Khazanchi, D. (2007) The Influence of Online 71(1): 16077.

Word of Mouth on Product Sales in Retail e-Commerce: An Mayzlin, D. (2006) Promotional Chat on the Internet,
Empirical Investigation, 13th Annual Americas Conference Marketing Science 25(2): 15563.
on Information Systems (AMCIS13), Keystone, Colorado. Pollach, I. (2006) Electronic Word of Mouth: A Genre
Dellarocas, C. (2000) Mechanisms for Coping with Unfair Analysis of Product Reviews on Consumer Opinion Web
Ratings and Discriminatory Behavior in Online Reputation Sites, 39th Hawaii International Conference on System
Reporting Systems, 21st International Conference on Sciences, Hawaii.
Information Systems (ICIS21), Brisbane, Queensland, Richins, M. L. (1983) Negative Word of Mouth by
Australia. Dissatisfied Consumers: A Pilot Study, Journal of
Dellarocas, C. (2003) The Digitization of Word of Mouth: Marketing 47(1): 6878.
Promise and Challenges of Online Feedback Mechanisms, Riegelsberger, J. and Sasse, M. A. (2002) Face it Photos
Management Science 49(10): 140724. dont make a Web Site Trustworthy, CHI Short Talks,
Dellarocas, C., Awad, N. F. and Zhang, X. (2004) Exploring Minneapolis, Minnesota.
the Value of Online Reviews to Organizations: Implications Senecal, S. and Nantel, J. (2004) The Influence of Online
for Revenue Forecasting and Planning, 25th International Product Recommendations on Consumers Online
Conference on Information Systems (ICIS25). Choices, Journal of Retailing 80(2): 15969.
Duan, W., Gu, B. and Whinston, A. B. (2005) Do Online Smith, D., Menon, S. and Sivakumar, K. (2005) Online Peer and
Reviews Matter? An Empirical Investigation of Panel Data, Editorial Recommendations, Trust, and Choice in Virtual
working paper, Austin, TX: University of Texas. Markets, Journal of Interactive Marketing 19(3): 1537.
Duhan, D. F., Johnson, S. D., Wilcox, J. B. and Harrell, G. D. Stewart, D. W., Hickson, G. B., Ratneshwar, S., Pechmann, C.
(1997) Influences on Consumer Use of Word of Mouth and Altemeier, W. (1985) Information Search and
Recommendation Sources, Journal of the Academy of Decision Strategies among Health Care Consumers,
Marketing Science 25: 283. Advances in Consumer Research 12(1): 25257.
Dwyer, P. (2007) Measuring the Value of Electronic Word of Walther, J. B., Slovacek, C. L. and Tidwell, L. C. (2001) Is a
Mouth and its Impact in Consumer Communities, Journal Picture Worth a Thousand Words? Photographic Images in
of Interactive Marketing 21(2): 6377. Long-term and Short-term Computer-mediated
Ellison, G. and Fudenberg, D. (1995) Word of Mouth Communication, Communication Research 28: 10534.
Communication on Social Learning, Quarterly Journal of Westbrook, R. A. (1987) Product/consumption-based
Economics 110(1): 61243. Affective Responses and Postpurchase Processes, Journal of
Fang, X. and Salvendy, G. (2003) Customer-centered Rules Marketing Research 24(3): 25870.
for Design of e-Commerce Web Sites, Communications of Zheng, J., Veinott, E., Bos, N., Olson, J. S. and Olson, G. M.
the ACM 46(12): 33236. (2002) Trust without Touch: Jumpstarting Long-distance
Girard, T., Korgaonkar, P. and Silverblatt, R. (2003) Trust with Initial Social Activities, CHI Letters 4(1): 1416.
Relationship of the Type of Product, Shopping Zhu, F. and Zhang, X. (2006) The Influence of Online
Orientations, and Demographics with Preference for Consumer Reviews on the Demand for Experience Goods:
Shopping on the Internet, Journal of Business and The Case of Video Games, 27th International Conference
Psychology 18(1): 10120. on Information Systems (ICIS27), Milwaukee, WI.