You are on page 1of 27

International Journal of Electronic Commerce

ISSN: 1086-4415 (Print) 1557-9301 (Online) Journal homepage: https://www.tandfonline.com/loi/mjec20

Modeling Consumer Purchasing Behavior in Social


Shopping Communities with Clickstream Data

Rainer Olbrich & Christian Holsing

To cite this article: Rainer Olbrich & Christian Holsing (2011) Modeling Consumer Purchasing
Behavior in Social Shopping Communities with Clickstream Data, International Journal of Electronic
Commerce, 16:2, 15-40, DOI: 10.2753/JEC1086-4415160202

To link to this article: https://doi.org/10.2753/JEC1086-4415160202

Published online: 10 Dec 2014.

Submit your article to this journal

Article views: 1728

View related articles

Citing articles: 45 View citing articles

Full Terms & Conditions of access and use can be found at


https://www.tandfonline.com/action/journalInformation?journalCode=mjec20
Modeling Consumer Purchasing Behavior in Social
Shopping Communities with Clickstream Data
Rainer Olbrich and Christian Holsing

ABSTRACT: Social shopping communities (SSCs) evolve from a linkage of social network-
ing and online shopping. Apart from direct shopping features in shopbots (e.g., search
fields), SSCs additionally offer user-generated social shopping features. These include
recommendation lists, ratings, styles (i.e., assortments arranged by users), tags, and user
profiles. Purchases can be made by following a link to a participating online shop (“click-
out”). SSCs are experiencing high growth rates in consumer popularity (e.g., Polyvore
attracts more than 6 million unique visitors per month). Thus, this business model has
received considerable venture capital in recent years. By analyzing clickstream data, we
investigate which factors, especially social shopping features, are significant for predicting
purchasing behavior within SSCs. Our logit model includes about 2.73 million visiting
sessions and shows that social shopping features exert a significant impact, both positive
and negative. Tags and high ratings have a positive impact on a click-out. In contrast,
the more lists and styles used, the less likely the user is to make a click-out. Yet, lists and
styles seem to enhance site stickiness and browsing. Moreover, the more direct shopping
features that are used, the less likely the user is to conduct a click-out. Increasing trans-
action costs and information overload could be potential reasons. We also found that
community members are more likely to make a click-out than ordinary users. This implies
that community members are more profitable.

KEY WORDS AND PHRASES: Clickstream data, online consumer purchasing behavior,
social shopping, user-generated content, virtual community.

Parallel to the increasing importance of the Internet as a shopping channel


[30, 77], the advent of Web 2.0 is rapidly moving the online landscape into a
truly consumer-driven era [8, 19, 75]. Web 2.0 provides consumers with many
methods of creating and sharing user-generated content (UGC) [23, 58, 80].
Therefore, social media such as blogs, message boards, and content-sharing
platforms are growing rapidly [8, 54]. In particular, consumers increasingly
communicate and exchange information in social networks, such as Face-
book [58, 79].
In the area of e‑commerce, this results in a linkage of online shopping and
social networking, initiating a new form of e‑commerce, that of social shop-
ping [78, 80]. In general, social shopping is about connecting consumers and
shopping together [75], which is also the focus of social shopping communi-
ties (SSCs), which can be defined as follows: An SSC is an online shopping
service that connects consumers and lets them discover, share, recommend,
rate, and purchase products [21, 38]. In contrast to traditional e‑commerce

The authors thank an anonymous collaborating company for providing the click-
stream data used in this article. Furthermore, the authors thank SAS Institute
Germany for providing SAS Enterprise Miner™ Software. Finally, the authors are
especially grateful to two anonymous IJEC reviewers for their helpful and insightful
comments.

International Journal of Electronic Commerce / Winter 2011–12, Vol. 16, No. 2, pp. 15–40.
Copyright © 2012 M.E. Sharpe, Inc. All rights reserved.
1086-4415/2012 $9.50 + 0.00.
DOI 10.2753/JEC1086-4415160202
16 Olbrich and Holsing

Figure 1. Screenshot of a Style on Polyvore

channels, such as online shops and shopbots, SSCs additionally offer user-
generated social shopping features, as well as potential interaction, so as to
initiate or simplify purchase decisions. These features include user-provided
ratings, recommendation lists, tags, and user profiles. User-generated collages,
called “styles,” are an especially popular sharing tool. A style is an assortment
of products arranged by a user, all of which normally relate to a particular
theme and can therefore enhance browsing. For example, the members of
Polyvore, an SSC focusing on fashion, have created more than 20 million styles
and continue to create more than 30,000 styles daily [65]. Thus, some brands
already use styles as an innovative marketing tool. For example, marketers
conduct a contest to increase brand awareness (e.g., Nike [66] or the apparel
brand Coach [67]). Figure 1 depicts a style of Coach’s contest themed “How
will you sparkle this season?” on Polyvore, in which consumers can create a
style with some products by Coach [67]. This style, which is modeled on the
famous actress named Blake Lively, was created by the user “krivonja” and
generated 8,730 views and 535 “likes” from other members. Overall, Coach
received 3,692 styles within one week, consisting of 103,379 likes, 13,006 com-
ments, and 204,656 page views overall [26]. Therefore, this not only led to
high visibility on the SSC itself but also increased brand and shop awareness
at decentralized places, such as social media, for example, when consumers
embedded this content on Facebook (see Figure 1, right-hand side). Further-
more, products can be searched with direct shopping features (e.g., search field
and filter mechanisms, product-detail sites offering detailed information), and
International journal of electronic commerce 17

price comparisons can be made, as well as purchases, by following a link to


a participating online shop (“click-out”).
In summary, all these user benefits lead to very high user growth rates for
SSCs (e.g., Polyvore attracts more than 6 million unique visitors per month
[46]). Thus, this business model has received considerable venture capital in
recent years [78], such as ThisNext with nearly $9 million [68]. Against this
background, it is becoming increasingly important for academics and practitio-
ners to observe what consumers do in SSCs and to analyze how to predict and
influence their behavior. Yet the academic literature on purchasing behavior in
SSCs is quite limited. Hence, our study revolves around the following research
question: What factors, especially social shopping features, are most significant for
predicting consumer purchasing behavior within social shopping communities?
Our research makes several contributions, including providing researchers
with preliminary insights into consumer use of SSCs and SSC operators, online
retailers, and manufacturers, with guidelines on the design and management
of SSCs.

Key Features of Social Shopping Communities

An SSC is a “virtual community of consumption” that revolves around


consumption-related interests [37]. SSCs were first launched about five years
ago, and the SSC business model has been evolving since then. Like most
Web  2.0 applications, SSCs cater to a specific segment of consumers [80].
Generally, most SSCs focus on women and “soft goods,” such as fashion and
lifestyle, which are hard to compare online. Because of the novelty of the
business model, we first provide a detailed overview of the key features and
describe the way revenue is generated by the SSC operator. Table 1 provides
an overview of the key features.
SSCs generally have several conventional direct shopping features, a com-
mon one being the search field, in which the user can enter key search words,
for example, “red dress.” The user can also search by means of several filter
mechanisms, such as product category, gender, brand, price, and shop. Accord-
ingly, the search results can be narrowed down considerably.
In addition to direct shopping features, a user can integrate many social
shopping features into the search process, for example, recommendation lists.
A recommendation list, created by a registered community member, contains
various products listed by a user. There are many reasons a user might wish
to create such a list; for example, the user might want to share it with friends
as a wish list for a birthday or Christmas gifts or as suggestions for wed-
ding presents. Such lists can also be used to simply collect products that are
of general interest or as recommendations for other users. The wish lists at
Amazon.com are a good example. In addition to recommendation lists, rat-
ings are another familiar and important social shopping feature. Registered
SSC users can rate products and shops. These ratings, normally on a five‑star
scale, are shown to all users.
A style is a very new type of UGC that registered users create via an editor
by means of drag and drop. Styles contain several products all of which relate
18 Olbrich and Holsing

Table 1. Description of Key Features.


Feature
category Feature Description

Direct shopping Filter Mechanisms Filter mechanisms for brand, category, gender, price, sales, shop
Search Field Field in which a visitor can enter search keywords
Social shopping List User-generated list which contains various different products
Profile User profile for publishing personal information and shopping
preferences
Rating Product User-generated average rating for products
Rating Shop User-generated average rating for shops
Style User-generated collage containing several products (assortment)
Tag User-generated tags to products and shops
Transactional Click-out Click on a link to a participating online shop (place of actual
purchase) after visiting a product-detail site

to a particular theme. Therefore, styles can be regarded as assortments selected


by users for others to browse. Each integrated product is also listed on the
side (see Figure 1, right-hand column, halfway down: “Items in this set”); by
clicking on the product, the user can view it in detail. In addition to a list of
the products, it is possible to insert pictures and colors as a background or to
write comments. To compare this feature with the offline shopping world, a
shop window displaying products with a specific theme is an appropriate par-
allel. In this context, prior studies have shown that specific functionalities and
applications in online environments can emulate consumer experiences with
bricks-and-mortar shopping [44, 45]. Note that the studies do not investigate
styles specifically, but styles generally constitute such features and can thus
contribute to emulating consumer offline shopping experiences. Both styles
and lists appear in a user’s profile and as tips for other users during their own
search processes. Thus, styles, as well as recommendation lists and ratings,
can be regarded as user-generated product recommendations.
SSCs often provide collaborative tagging as well [25]. Tags are used to an-
notate different kinds of content, such as photos and videos [57]. The tagging
of products has also become very popular because consumers benefit from
the effective sharing and organization of large amounts of information [10].
In addition to annotating different kinds of content, user-generated tags are
assigned to products, shops, and so on. Users can incorporate tags within
the search process by clicking on a tag and searching through the displayed
products.
By offering networking features such as user profiles, SSCs differentiate
themselves from other shopping services that simply offer products and
prices. Within the user’s profile, the user can publish personal information,
for example, photos or a self-description. Furthermore, a user can describe
personal preferences, for example, favorite products, brands, and shops, and
label these with a tag. In searching for these preferences, a user may find shop-
ping partners, that is, users who share the same preferences. Furthermore,
other users can leave a message in a guestbook within the profile site.
International journal of electronic commerce 19

On a product-detail site, the user obtains detailed information on a specific


product. The user can see the product price, picture, rating, tags from other
users, and information about the online shops that sell the product, for ex-
ample, shop name and rating. Finally, purchases can be made by following a
link to a participating online shop (click-out). The operator of an SSC receives
a fee from the participating online shops for each click-out (pay per click) and
actual sales (pay per sale).

Research Framework and Hypotheses


We first give a brief overview of clickstream research. We then focus on exist-
ing studies in two of the most discussed research streams: (1) search behavior
and purchase conversion and (2) the impact of UGC and electronic word of
mouth (eWOM) [8]. Accordingly, we develop our hypotheses on the factors
affecting consumer purchasing behavior in SSCs.

Clickstream Data and Consumer Behavior

Clickstream data are a powerful source of information on consumer online


behavior [9, 11]. In addition to tracking transactions such as purchases, click-
stream data also facilitate tracking earlier consumer actions, such as searching.
Existing studies have analyzed several aspects of consumer online behavior,
including information search and usage [49], consumer motives to continue
browsing [7], identification of consumer goals [47], online decision-making
processes [72], and identification of consumers with a high purchase probabil-
ity [49, 52]. In addition, these studies focus on several different types of Web
sites, for example, an automotive Web site [7] or an online shop for nutrition
products [47]. However, clickstream modelers have focused predominantly
on “search”-type products, for example, books and CDs [34, 50]. Furthermore,
the data volume is quite varied, with the number of observations ranging from
a few hundred [31] to several thousand [17].
Using clickstream data confronts researchers with a number of difficul-
ties [16], however, such as capturing the purchasing environment of consumers
and the associated data preprocessing. Accordingly, relatively few studies in
fact use such data, and therefore clickstream analysis in marketing research
is still in its infancy [8]. Hence, this study aims to make several contributions
to clickstream research, especially in the two areas discussed below.

Online Consumer Search Behavior and Purchase Conversion

In general, measuring and managing key performance metrics, such as the


number of users, view time, and conversion rates, have become crucial to
Web site managers [4, 49]. Conversion rates are not limited only to purchases
but may also entail generating leads, or, as in our study, click-outs. Thus, an
understanding of what influences these performance metrics is of considerable
20 Olbrich and Holsing

interest to researchers and Web site managers. In general, many factors can
influence behavior, including consumer involvement [60, 61], Web site design
and content [17], and both product and consumer characteristics [55]. The
following is a look at factors affecting search and purchasing behavior with
respect to SSCs and our main conversion objective of click-out. Due to data
limitations, we exclude some factors (e.g., demographic characteristics) and
focus on the following factors: involvement, visit occasion, transaction costs,
and information overload.
In the literature, consumer involvement—that is, perceived personal rel-
evance of an object [33]—is identified as an important factor affecting informa-
tion processing during searching and purchasing [47, 84]. In online settings,
involvement can be measured, for example, by the view time and number of
products in the purchasing process [20]. View time is an important performance
metric, but the research on factors affecting view time is quite limited [8, 17].
Padmanabhan et al. [59] introduced the view time visitors spent at a Web site
during a visit session. The view time seems to be significant in their modeling
approach and positively correlated with a potential purchase. Therefore, we
expect a positive correlation between the view time and a click-out. Thus,

Hypothesis 1: The longer the view time, the greater the likelihood of a click-out.

Many researchers have found that the occasion of a visit—that is, the search
motivation—is another important factor affecting purchasing behavior. For
example, Janiszewski [32] classified consumer search behavior into goal-
directed and exploratory search. Goal-directed searchers have a specific or
planned purchase in mind. Therefore, the search patterns of these consumers
focus on making a purchase decision [32]. By contrast, exploratory search
is less focused and no purchase is planned. In this context, Moe [47] used
the content of the pages viewed to make distinctions between the purchase
likelihood of consumers. She found that users can be classified by four types
of browsing behavior: directed buying, search and deliberation, hedonic
browsing, and knowledge building. With respect to our study, the cluster of
hedonic browsing is of particular interest. These users seek new stimuli, thus
enhancing impulse buying. UGC may serve as such a stimulus. Generally, we
expect a high average view time per page to be characteristic of hedonic users
who like browsing, for example, for relaxation purposes or as a moment-by-
moment activity [83]. In contrast, we hypothesize that goal-directed users have
a relatively low average view time per page because they are more frequent
shoppers, visit the SSC regularly, and thus benefit from learning effects [34].
Accordingly, they are more likely to conduct a click-out. Thus,

Hypothesis 2: The longer the average view time per page, the lower the likeli-
hood of a click-out.

Moreover, goal-directed users plan to purchase a specific product [1]. There-


fore, we expect a low usage of product-detail sites to increase the likelihood of
a click-out, because of directed search and purchase. Furthermore, we assume
International journal of electronic commerce 21

that goal-directed users intensively check the information given on a product-


detail site before conducting a click-out because of high involvement. Thus,

Hypothesis 3: The more frequently product-detail sites are visited, the lower
the likelihood of a click-out.

Hypothesis 4: The longer the average view time per product-detail site, the
greater the likelihood of a click-out.

Previous research has shown that the use of search engines and shopbots
can lower transaction costs [5, 41, 74]. These costs refer to transaction-related
activities [85] such as searching for information, negotiating, and ordering.
Transaction costs play an important role in searching and purchasing because
consumers generally choose transactions that minimize their transaction costs
[40, 69]. With regard to SSCs, direct shopping mechanisms could ease informa-
tion processing and increase search efficiency [33]. Specifically, direct buyers or
searchers use these features to access the required information quickly. Because
the search for information yields benefits, but also causes costs, consumers
will not search endlessly for information [77]. If several filtering stages are
needed to find the right product, transaction costs rise, that is, high informa-
tion costs measured in time. Therefore, high transaction costs could lead to a
termination of the purchasing process. In this way, user purchasing behavior
is rational, involving time constraints and cost-benefit trade-offs [8].
Moreover, although the Web is regarded by many researchers as useful
in searching and reducing transaction costs [1, 5, 74, 76], it is often argued
that the Web and the vast amount of information it contains can also lead to
information overload [13, 77]. Despite the high use of filtering mechanisms, it
remains possible that consumers do not find the right product or are confronted
with too much information and thus abandon the session through confusion
or frustration [13, 36].
Furthermore, the goal of consumers may be to acquire information during
their prepurchase deliberations [47, 81], with the actual purchase planned
for the next visit. This could be a reason for the high usage of direct shop-
ping features. This assumption has also been confirmed indirectly by studies
demonstrating that users who click relatively infrequently search in a very
goal-oriented manner and purchase relatively frequently [61].
In summary, we expect a high usage of direct shopping features to lower
the likelihood of a click-out. In our study we analyze the following direct
shopping features: (a) brand, (b) category, (c) search field, (d) gender, (e) price,
(f) sales, and (g) shop. Thus,

Hypothesis 5a–g: The more frequently each direct shopping feature (a–g) is
used, the lower the likelihood of a click-out.

In addition, we assume that a high usage of the home page site could also
indicate information overload. The home page could serve as an orientation in
situations of information overload, as well as the starting point for a restart of
22 Olbrich and Holsing

an exploratory search. Therefore, we expect the number of visits to the home


page to correlate negatively with the probability of a click-out. Thus,

Hypothesis 6: The more frequently the home page is visited, the lower the
likelihood of a click-out.

Impact of User-Generated Social Shopping Features on


Consumer Purchasing Behavior

Since the Internet provides many opportunities for creating and sharing UGC,
it has stimulated research in the field of eWOM communication [8]. Despite
the high value of UGC and eWOM within the purchase-decision process,
their impact has only recently become the subject of research [15, 18, 19, 24,
79, 82]. In particular, there are only a few studies on the influence of UGC on
performance measures [23]. Analysis of an SSC is especially well suited to fill-
ing this research gap. All the means of interpersonal information transmission
through social shopping features, which in fact are UGC, can be regarded as
a form of eWOM [60].
A large proportion of existing studies in this area have investigated user-
provided ratings and reviews [18, 24, 42]. Generally, the sharing of reviews
and ratings of products can increase trust, thus reducing perceived risk when
purchasing online [22]. For example, Chevalier and Mayzlin find that average
star ratings can exert a positive impact on online book sales [15]. Thus,

Hypothesis 7a: The higher the overall average product rating, the greater the
likelihood of a click-out.

Hypothesis 7b: The higher the overall average shop rating, the greater the
likelihood of a click-out.

In general, UGC such as lists and styles are being integrated steadily into
e‑commerce [8, 14]. It is applied, among other reasons, with the aim of serving
as a sales assistant [12], to increase the purchase price and promote consumer
confidence. Because UGC is faded in by the provider and the contents can be
varied, UGC can be considered as a situational factor that potentially influ-
ences consumer behavior. Fade-ins, however, particularly graphic ones, may
distract users from their originally intended searches [1]. Taking into account
the elaboration likelihood model, Park and Chung [61] referred to a periph-
eral route that distracts the user. Wang et al. [83] mentioned that the visual
system takes effect when one is not conducting a goal-directed search [83].
Non-goal-directed searchers, using the peripheral route and UGC, may be
hedonically motivated consumers [33, 63], enjoying the shopping process itself
[29]. Their search choices are often spontaneous, and the satisfaction of needs
unrelated to the product is the most important goal (e.g., they are shopping
for “play”). Therefore, UGC is important for marketing managers wishing
to attract this specific segment of consumers and to gain competitive advan-
tages. Nonetheless, novel shopping features are also unlikely to be familiar
International journal of electronic commerce 23

to users at this point and may lead to adverse reactions from users who feel
overwhelmed or undesirably influenced by the provider. Such a reaction can
occur particularly in the case of inexperienced consumers. Contrarily, there
are also experienced users who specifically want new functions, that is, they
have the trait of “novelty seeking” [13].
In summary, we assume that lists and styles lower the likelihood of a click-
out for two reasons. First and foremost, lists and styles are not designed for
direct shopping, but for enhancing browsing, particularly in the prepurchase
phase. Second, we assume that most consumers are not familiar with novel
social shopping features, so that most focus their attention on functional as-
pects of shopping motivation. Thus,

Hypothesis 8a: The more frequently lists are used, the lower the likelihood of
a click-out.

Hypothesis 8b: The more frequently styles are used, the lower the likelihood
of a click-out.

In contrast to lists and styles, tagging is a more rational form of UGC [25].
In SSCs, tags are used mainly to annotate products. Therefore, tags can be
useful in a goal-directed search, make the search process more effective, and
enable consumers to find what they are looking for [10]. Thus,

Hypothesis 9: The more frequently tags are used, the greater the likelihood of
a click-out.

At the heart of SSCs are user profiles that aim to enhance interaction be-
tween community members. Previous research shows that consumers have
several motivations to participate in communities, for example, belonging,
entertainment, and prestige [3, 22, 70, 84]. In general, the resources offered by
virtual communities can foster shopping needs satisfaction [44]. For example,
consumers can exchange opinions on company products and help one another
with specific problems, which may lead to a more personal shopping experi-
ence [22]. However, profiles are not directly relevant to shopping, especially
for nonregistered users. Thus,

Hypothesis 10: The more frequently user profiles are used, the lower the
likelihood of a click-out.

Data and Measures

The Social Shopping Community

The high-traffic SSC under consideration in this study (name undisclosed at


the request of the operator) is one of the leading worldwide SSCs and focuses
on the categories of fashion, living, and lifestyle. According to the operator, the
target audience is female. During the period of data collection, several hundred
24 Olbrich and Holsing

online shops participated and the product data base included more than 1.5
million products. The users can incorporate all common features of an SSC in
order to discover, share, and recommend products. Yet only registered users
may create UGC and interact directly with other users. As is common for most
SSCs, users do not receive any monetary incentives for creating UGC.

Clickstream Data Collection and Preprocessing

Our study uses clickstream data from the above-mentioned SSC. Because it
uses such site-centric data, the present study can be categorized as within-
site [11, 17, 48, 73], in contrast to across-site studies [31, 35, 62]. Our data span
the period from May 1, 2009, to October 31, 2009. For each session, the pages
viewed and the viewing duration of the pages are recorded. User actions are
also recorded, such as whether a user is viewing a product-detail site. Sites
that do not relate to shopping, such as the imprint or the job-offer page, are
not included. Except for the information on whether a user is logged in or
not, we do not have any data on users (e.g., demographics). Furthermore, our
data do not measure repeat visits explicitly because of data privacy concerns
of the operator. Thus, we use single visits.
Several preprocessing tasks are necessary in order to obtain a useful data set.
For this purpose, we use PHP and mySQL. First, requests from Web crawlers
and Web spiders are eliminated [16]. The most important step in preprocess-
ing the data is to link all the data from different users and transform them
into unique sessions, that is, a single visit to a Web site [81]. We aggregate
the page-level data to the session level, by using various measures that we
introduce below. Therefore, we describe the nature of user actions within each
session. Consistent with existing approaches, we remove single-page sessions
from further analysis [9, 52]. This is crucial, because sessions with only a one-
page view do not constitute a real “browsing” session. We do the same for
sessions with a view time below 1 second, because they could result merely
from user mistakes or system errors. Furthermore, we exclude sessions with
a view time greater than 45 minutes, as well as accesses by employees. After
preprocessing, the number of sessions decreases from 7,842,591 to 2,727,634.
Nevertheless, the data volume in our study is very high in comparison with
existing studies [8].

Session Measures

The aim of our study is to characterize sessions, rather than the page-to-page
decisions of a particular user. Therefore, we use page-to-page information to
generate more general session-level measures that characterize the purchasing
behavior of users. Hence, we develop four categories of session-descriptor
variables, which correspond to our hypotheses: general, direct shopping, social
shopping, and transactional. Existing studies have incorporated some of the
input variables we present, for example, view time [34, 73], so as to examine
their impact on purchase propensity. However, most of these studies consider
International journal of electronic commerce 25

only a relatively small selection of input variables and do not incorporate


detailed clickstream data. This is remarkable, because especially detailed
clickstream data variables could generate the best predictive performance
[81]. Therefore, we take detailed data into account, such as details regarding
the contents of the pages [47].
The variable category general contains typical metrics. VIEW_TIME
measures the total time in seconds spent at the site, and VIEW_TIME_AVG
measures the average time in seconds per page. LOG_IN indicates whether a
user is logged in to the site with the user’s user name and password. We use
this variable for controlling the effect of the online shopping experience [31],
which may affect consumer behavior. PROD_VIEW gives the total number
of viewed product-detail sites during the session. PROD_VIEW_TIME is the
average view time in seconds on product-detail sites. HOME reveals how
often the home page of the SSC is visited.
The variable category direct shopping reflects the way a user incorporates
direct shopping features. The search field is a common direct shopping fea-
ture. The total number of uses of the search field is counted by the variable
SEARCH_FIELD. SEARCH_CAT is the total number of pages that resulted
from filtering products via the product category filter (fashion, living, or
lifestyle). Further variables are SEARCH_BRAND, SEARCH_GENDER,
SEARCH_PRICE, SEARCH_SALES, and SEARCH_SHOP, which count the
use of the filters’ brand, gender, price, sales, and shop.
Particular attention is paid to the category social shopping. User-generated
product ratings are an important feature of SSCs. On this site, the ratings
have a five‑star scale, with 1 as the lowest value and 5 as the highest. We mea-
sure the average percentage of all products viewed by a user by the variable
RATING_PROD. This variable ranges from 0 percent (no ratings) to 100 percent
(all viewed products have a five‑star rating). We do the same for the tip-top
shops (a shop at position 1 on a product-detail site, that is, placed directly next
to the product) appearing on a product-detail site with the variable RATING_
SHOP. The total number of lists viewed by the user is counted by LIST_VIEW.
STYLE_VIEW counts the number of styles viewed. TAG_VIEW measures the
total number of pages resulting from a search action by clicking on a tag. Tags
can be found on product-detail sites, in lists and styles, and on profiles. The vari-
able PROFILE_VIEW counts the total number of pages related to a user profile,
for example, profile with photo, a friends list, or the guestbook.
Finally, the transactional category contains the variable CLICK_OUT, which
indicates whether or not the user has performed a click-out to a participating
online shop within the session. In our study, we regard this variable as an
indicator of actual purchases.

Data Analysis and Modeling Approach

Descriptive Statistics

Table 2 shows the descriptive statistics (mean, standard deviation, minimum,


maximum, and median) of the variables used. The mean view time per session
26

Table 2. Description and Descriptive Statistics of Measures.


Variable Description Mean SD Minimum Maximum Median

General          
HOME Total number of index sites (home) viewed 0.081 0.340 0.000 92.000 0.000
LOG_IN Indicator variable for log-in of a visitor 0.012 0.109 0.000 1.000 —
PROD_VIEW Total number of product-detail sites viewed 0.922 1.827 0.000 152.000 0.000
PROD_VIEW_TIME Average view time per product-detail site in sec. 17.263 58.320 0.000 1,891.000 0.000
Olbrich and Holsing

VIEW_TIME Total view time in sec. 197.168 317.878 1.000 2,699.000 72.000
VIEW_TIME_AVG Average view time per page in sec. 34.450 58.995 0.250 1,302.000 17.000
Direct Shopping            
SEARCH_BRAND Total number of usages of filter brand 0.285 2.326 0.000 252.000 0.000
SEARCH_CAT Total number of usages of filter category 1.452 6.346 0.000 451.000 0.000
SEARCH_FIELD Total number of usages of search field 1.127 2.472 0.000 401.000 0.000
SEARCH_GENDER Total number of usages of filter gender 0.685 3.789 0.000 430.000 0.000
SEARCH_PRICE Total number of usages of filter price 0.106 1.595 0.000 220.000 0.000
SEARCH_SALES Total number of usages of filter sales 0.046 0.900 0.000 234.000 0.000
SEARCH_SHOP Total number of usages of filter shop 0.114 0.838 0.000 178.000 0.000
Social Shopping            
LIST_VIEW Total number of lists viewed 0.015 0.182 0.000 112.000 0.000
PROFILE_VIEW Total number of profiles viewed 0.009 0.199 0.000 121.000 0.000
RATING_PROD Average rating for all products viewed 14.793 32.172 0.000 100.000 0.000
RATING_SHOP Average rating for all top-tip shops viewed 13.530 31.003 0.000 100.000 0.000
STYLE_VIEW Total number of styles viewed 0.010 0.156 0.000 95.000 0.000
TAG_VIEW Total number of usages of a tag 0.041 0.666 0.000 191.000 0.000
Transactional            
CLICK_OUT Indicator variable for click-out 0.386 0.487 0.000 1.000 —
International journal of electronic commerce 27

is 197.168 seconds (3.29 minutes), and the average view time per page is
34.45 seconds. Therefore, the mean number of pages visited per session is
5.72. The log-in rate is 1.2 percent (i.e., in 1.2 percent of all sessions, a user is
logged in). On average, 0.922 product-detail sites are considered. The click-
out rate is 38.6 percent. Use of social shopping features is relatively low. For
example, the mean of the variable indicating the average number of accessed
lists is just .015. These low mean values may reflect the early developmental
phase of this specific SSC and the corresponding unawareness of these new
features. In comparison, the counting variables of the more common direct
shopping features like SEARCH_FIELD (1.127) and SEARCH_CAT (1.452)
have a relatively high mean value. Overall, the mean values of all shopping
features are relatively low because there are many various features and users
do not use each at each session. This results in the value 0 for the median of
many variables.

Modeling Approach

The objective of our modeling is to determine the dependent variable


CLICK_OUT. This variable is binary coded, so we run a logistic regression,
which is conceptually simple and frequently used in marketing [6, 31].
SAS Enterprise Miner was used to estimate the logit model. The procedure
PROC_LOGISTIC uses the maximum likelihood estimation to generate relative
weights for each independent variable [71]. Our logit model, in which Λ(·) is
the inverse of the logit function, is formulated as follows:

P ( CLICK_OUT = 1) = Λ β0 + β1 ( x1 ) + … + β19 ( x19 ) . (1)



In Equation (1), β0 is the intercept. The βs (from 1 to 19) are the regression
coefficients of the predictor variables (see Table 2). Because the distributions
of the eighteen metric predictor variables are positively skewed, we use the
log transforms of all the metric predictor variables [27]. The only binary coded
variable is LOG_IN, which controls for the effect of log-in, that is, the impact
of shopping experience and community affiliation on purchasing behavior. For
accuracy assessment, we divided our sample randomly into two subsets. The
training sample contains 70 percent of the sessions and the holdout sample
the remaining 30 percent.

Results of Logit Modeling

Table 3 shows the results of the logistic regression. The model yielded a signifi-
cant likelihood ratio test (χ2(19) = 702,279.02, p < 0.0001). This implies that our
model as a whole fits significantly better than an empty model. Furthermore,
the goodness of fit is considered using the percentage of correct classifications,
that is, the hit rate and McFadden’s R2, a pseudo‑R2 statistic. The McFadden
R2 is the percentage fall in log likelihood, compared with a random effects
model with a consistent term only. In general, measures of pseudo‑R2 have to
28 Olbrich and Holsing

Table 3. Logistic Regression Results.


Standard Odds
Variable Coefficient error p-value ratio

Intercept 1.098 0.014 < 0.0001 —


General:        
HOME –3.538 0.017 < 0.0001 0.029
LOG_IN (0 vs. 1) –2.931 0.013 < 0.0001 0.003
PROD_VIEW –1.922 0.007 < 0.0001 0.146
PROD_VIEW_TIME 0.586 0.002 < 0.0001 1.796
VIEW_TIME 2.684 0.005 < 0.0001 14.641
VIEW_TIME_AVG –2.994 0.006 < 0.0001 0.050
Direct Shopping:        
SEARCH_BRAND –0.539 0.006 < 0.0001 0.583
SEARCH_CAT –1.411 0.004 < 0.0001 0.244
SEARCH_FIELD –1.840 0.004 < 0.0001 0.159
SEARCH_GENDER –0.346 0.004 < 0.0001 0.707
SEARCH_PRICE 0.191 0.009 < 0.0001 1.211
SEARCH_SALES –0.348 0.012 < 0.0001 0.706
SEARCH_SHOP –0.250 0.008 < 0.0001 0.779
Social Shopping:        
LIST_VIEW –2.042 0.026 < 0.0001 0.130
PROFILE_VIEW –5.186 0.060 < 0.0001 0.006
RATING_PROD 0.032 0.001 < 0.0001 1.032
RATING_SHOP 0.240 0.001 < 0.0001 1.271
STYLE_VIEW –1.108 0.028 < 0.0001 0.330
TAG_VIEW 0.332 0.012 < 0.0001 1.393

Number of observations 2,727,634      


McFadden pseudo-R 2 0.27*      
Percentage correct 73.4      
χ2(19) = 702,279.02, p < .0001      

* Values of pseudo-R2 between 0.2 and 0.4 indicate a good model fit [43].

be interpreted with caution since they do not explain the variance in the same
way as the R2 coefficient in linear regression. Measures of pseudo‑R2 are not
as high as R2s in linear regression, such that a value of the pseudo‑R2 between
0.2 and 0.4 indicates an extremely good model fit [43]. Therefore, the value
of McFadden’s R2 in our model (0.27) demonstrates a good fit between the
estimated model and the observed empirical data. Moreover, the predictive
accuracy of our model is good, with a percentage of 73.4.
The interpretation of the coefficients is somewhat tricky. Therefore, we
interpret the results using the concept of odds ratio. The odds of an event oc-
curring is the probability that the event will occur, divided by the probability
that it will not. An odds ratio is computed by exponentiating the parameter
estimate for the predictor variable, and it can be interpreted as the multiplica-
tive change in the odds of a one-unit change in the predictor variable [2]. When
interpreting quantitative variables, it is useful to do the following: subtract 1
International journal of electronic commerce 29

from the odds ratio and multiply by 100. This yields the percentage increase or
decrease associated with a one-unit increase in the independent variable [2].
All variables are highly statistically significant at the .0001 level. Both
the view time (H1) and the average view time per page (H2) have a signifi-
cant impact. According to the magnitude of the odds ratios, the view time
(VIEW_TIME) is the most important and has a positive impact (14.641). In
contrast, the higher the average view time per page, the lower the likelihood
of a click-out. This negative impact is indicated by an odds ratio below 1 for
the variable VIEW_TIME_AVG. The odds ratio of .05 of VIEW_TIME_AVG
means that the odds of a click-out are 95 percent lower per 1‑second increase
in average view time per page. In summary, H1 and H2 are supported.
Moreover, H3 is also supported. We find that the more product-detail sites
a consumer accesses, the lower the likelihood of a click-out. This result was
expected, as a high use of product-detail sites may indicate a non-goal-directed
visit to the site, in which the consumer has not already built a consideration set
and continues browsing. The average view time per product-detail site (H4)
has a positive effect. The odds ratio is 1.796, meaning that the odds of a click-
out are 79.6 percent higher per additional second on a product-detail site.
In agreement with H6, the odds of a click-out are 97.1 percent lower per
additional visit to the home page. This implies the importance of the home
page as a tool for user orientation. Consistent with H8a and H8b, the frequent
use of lists (0.13) and styles (0.33) lowers the likelihood of a click-out. This
finding may indicate that lists and styles support site stickiness (i.e., users
stay longer on the Web site), browsing, and inspiration. This assumption is
supported by our additional finding that the average view time in sessions, in
which at least one list or style is considered, is 5.12 minutes and is thus about
56 percent higher than in sessions with no considered list or style. According
to the odds ratio of 1.393, tags have a positive effect. The more user-generated
tags are used, the greater the likelihood of conducting a click-out. Therefore,
H9 is supported.
The positive impact of the variables RATING_PROD (1.032) and RAT-
ING_SHOP  (1.271) is also in line with our assumption in H7a and H7b.
Therefore, user-provided ratings for shops and products play an important
role in consumer decision-making processes and increase the likelihood of a
click-out. Interestingly, when comparing the magnitude of the odds ratios,
a high shop rating is evidently more important than a high product rating.
This may be due to the fact that fashion and lifestyle products are up-to-date
for only a short period and therefore are not rated as frequently as shops. As
mentioned, both variables are heavily aggregated measures within a session.
Therefore, these variables should be considered carefully. However, they may
be useful in providing some initial insight into the direction of impact.
The odds ratio for user profiles (PROFILE_VIEW) has the most negative
value of all social shopping features (0.006). That is, the more user profiles
used within a session, the lower the likelihood of conducting a click-out.
Hence, H10 is confirmed.
Use of the direct shopping feature for price (H5e, SEARCH_PRICE) is the
only such feature with a positive impact on click-out. The more this feature is
used, the greater the likelihood of a click-out. This implies the importance of
30 Olbrich and Holsing

price within the purchasing process. Accordingly, our hypotheses regarding the
direct shopping features (H5(a)–(g)) are supported for six of the seven direct
shopping features. That is, the more conventional direct shopping features
are used, the lower the number of click-outs. As assumed above, increasing
transaction cost and information overload could be potential reasons. Fur-
thermore, in the case of knowledge-building, a consumers’ ongoing search
is motivated by the desire to acquire a bank of product knowledge that is
potentially useful in the future. Therefore, a consumer uses several shopping
features, but does not conduct a click-out. As expected, the use of the search
field (SEARCH_FIELD), which is often the starting point for a search, has the
most negative impact of all direct shopping features (.159).
Last but not least, we found that the binary coded control variable LOG_IN
has a significant effect on click-out. The value 0 stands for not logged-in, and 1
for logged-in. The odds ratio is 0.003 for value 0 as opposed to value 1. This
means that the probability of a click-out occurring—CLICK_OUT = 1—is lower
for the value 0 than for the value 1. In other words, logged-in users have a
higher tendency to do a click-out in comparison to non-logged-in users. One
possible explanation could be that registered community members are experi-
enced in online shopping. Hence, they visit the SSC regularly to communicate,
search, and purchase. This is in line with van den Poel and Buckinx [81], who
assumed registered clients to be among the most active ones. Our finding also
corresponds with Moe and Fader [49, 50], who demonstrated that the more
often people visit a retail site, the more likely they are to buy. Although our
data do not measure repeat visits explicitly, because of data privacy concerns
of the operator, we assume that logged-in users are repeat users.
Finally, Table 4 summarizes the results of hypothesis testing. It is evident
that only H5e, for the direct shopping feature of price, is not supported. That
is, this feature does not yield the expected negative impact on a click-out.

Discussion and Implications

Previous research has established that UGC exerts a significant impact on


economic outcomes [15, 19, 42]. However, one has to take into account that
the situational influence in which UGC occurs is moderated by consumer
involvement [60] and by both product and consumer characteristics [55]. We
enhance the research stream analyzing UGC by taking several user-generated
social shopping features into account.
We show that social shopping features have a low level of usage. Their
novelty constitutes a possible explanation. Despite the low usage, we find
that social shopping features have a significant impact on purchasing behav-
ior, both negative and positive. We find that tags exert a positive impact on a
click-out. Furthermore, and in conformity with existing studies [15, 18], we
find the same for ratings for shops and products. In contrast, lists and styles
lower the likelihood of a click-out. This finding was to be expected, as these
two features are designed for browsing and inspiration.
In line with other clickstream studies, our results show that user behavior
is consistent with rational considerations involving time constraints and
International journal of electronic commerce 31

Table 4. Results of Hypotheses Testing.


Expected
Variable Hypothesis sign Supported

General      
HOME H6 - Yes
LOG_IN (0 vs. 1) —   —
PROD_VIEW H3 - Yes
PROD_VIEW_TIME H4 + Yes
VIEW_TIME H1 + Yes
VIEW_TIME_AVG H2 - Yes
Direct shopping    
SEARCH_BRAND H5a - Yes
SEARCH_CAT H5b - Yes
SEARCH_FIELD H5c - Yes
SEARCH_GENDER H5d - Yes
SEARCH_PRICE H5e - No
SEARCH_SALES H5f - Yes
SEARCH_SHOP H5g - Yes
Social shopping    
LIST_VIEW H8a - Yes
PROFILE_VIEW H10 - Yes
RATING_PROD H7a + Yes
RATING_SHOP H7b + Yes
STYLE_VIEW H8b - Yes
TAG_VIEW H9 + Yes

cost-benefit trade-offs [8, 40]. The more direct shopping features that are
used, the lower the probability of a click-out. In addition, a consumer focus
on functional aspects and goal-directedness is also confirmed by our result
that a high average view time per page correlates negatively with a click-out.
Moreover, our data analysis yields several meaningful implications for opera-
tors and online retailers.

Implications for Operators

UGC is a fundamental part of the business model of SSC because users add value
to the operator. According to our findings, user-provided ratings for online shops
and products play a significant role in consumer decision-making processes. A
high overall rating seems to be a significant generator of click-outs. Likewise,
the use of tags exerts a positive influence on click-outs. Presumably, providing
tags supports goal-directed searching and enables consumers to find what they
are looking for. Therefore, an operator should animate users to rate products
and shops and generate tags in order to maximize the number of click-outs.
Lists and styles are novel features that are designed to enhance site sticki-
ness, stimulate browsing, and thus enhance inspiration. This conforms to our
finding that the use of lists and styles is negatively correlated with click-out,
32 Olbrich and Holsing

but the average view time in sessions, in which at least one list or style is con-
sidered, is about 56 percent higher. Despite the negative impact on click-out,
lists and styles could lead to positive long-term effects on the click-out rate and
sales. Given that a consumer may be in the prepurchase phase, a purchase is
not intended in the current session. However, lists and styles may stimulate
the examination of a product. Thus, a product could enter into a consumer’s
considered set of products and lead to a purchase in the future [47]. Therefore,
such users may be treated as potential consumers [60].
Moreover, social shopping in general could be regarded as a form of enter-
tainment [33]. Therefore, lists and styles may be important elements of Web
site design and could lead to a more emotional shopping experience. This may
be especially important for Web sites focusing on women, like SSCs. In this
context, Hansen and Jensen [28] found that women are more inclined than
men to “shop for fun.” Furthermore, social shopping features could attract the
segment of “24-hour shoppers” who love shopping online in order to explore
new ideas and goods, as identified in a recent McKinsey study [51]. Therefore,
lists and styles may enhance the business-consumer relationship and increase
loyalty rates [22], at least within these target segments.
Regarding the revenue model, the stickiness (higher view time), during ses-
sions in which at least one list or style is considered, could also lead to more
page views and thus to more advertising impressions, for example, banner
ads for specific products. Nevertheless, one further explanation of the negative
impact of lists and styles on click-out could be explained by their novelty and
information overload, especially for inexperienced users. Furthermore, goal-
oriented and time-pressed consumers may be confronted with psychological
reactance, e.g., when they feel obliged to use lists and styles. As with lists
and styles, use of profiles is also negatively correlated with click-out. Since
profiles are designed for enhancing interactivity and are not directly relevant
to shopping, an operator may have a problem monetizing this kind of UGC.
Nonetheless, profiles also enhance the site stickiness. In addition, profiles are
a core component of SSCs and, as discussed below, essential for establishing
a shopping-savvy community with a high economic value.
To avoid consumer reactance and information overload, as well as stimulate
the use of social shopping features, an operator could explain the concept of
an SSC. The explanation of social shopping features, as well as direct shopping
features, on a separate site of the SSC may be an appropriate way of achieving
this. In practice, such distinct information sites are often used. Furthermore,
an operator may improve the features, as well as content utility, for example,
by conducting usability tests, in order to improve conversion rates [4].
Moreover, our results clearly indicate that a cost-benefit perspective may be
useful for understanding and predicting purchasing behavior in SSCs. There-
fore, an operator should provide effective search tools to users; otherwise, a
user could easily abandon the session. Potential explanations could be infor-
mation overload and high transaction costs. This is shown by the result that a
high usage of direct shopping features lowers the likelihood of a click-out. In
this regard, our analysis shows that the direct shopping feature of price is the
only one that correlates positively with a click-out. Not surprisingly, product
prices seem to be an important factor within the decision-making process.
International journal of electronic commerce 33

Furthermore, we found that a frequent use of the home page exerts a strong
negative impact on click-out. This may be explained by a non-goal-directed
search, or as an indicator of information overload. Thus, providing users with
relevant information and effective search and support tools on the home page
could stimulate the user to continue searching.
Providing all relevant information regarding the purchase transaction may
also be important for product-detail sites. We found that a high average view
time per product-detail site correlates positively with a click-out. This indicates
that consumers intensively check all the information given there because this
is relevant for purchasing.
Finally, we found that logged-in users have a higher probability of making
a click-out than non-logged-in users. Logged-in users are few in number, but,
at the same time, they generally have a high economic value. It is well known
that a relatively small share of customers are responsible for a relatively large
share of profit [56]. Moreover, it is often stated that it is more important to look
after existing customers than to acquire new ones [53]. Hence, following the
strategy of community building may be an appropriate way to boost revenue.
Paying attention to community behavior and developing tools to increase
loyalty may be a prudent tactic.

Implications for Online Retailers and Manufacturers

In general, more and more consumers use infomediaries [76] as well as new
shopping channels in Web 2.0 [23, 64, 75]. Therefore, online retailers and
manufacturers must integrate this development into their marketing strategy.
SSCs could be one important component.
Our results show that the higher the overall average shop ratings on
product-detail sites, the greater the likelihood of conducting a click-out to a
participating online shop. Therefore, online retailers should encourage con-
sumers (their customers) to rate their shop in order to enhance its reputation
and trustworthiness [39]. The situation is the same for product ratings. Manu-
facturers could animate consumers to rate their products and brands.
The situation is the same for user-generated tags: the more tags used within
a session, the greater the probability of a click-out. Hence, activating consum-
ers to tag products with specific key words is another tactic that could draw
attention to specific items.
Lists and styles are also a very innovative marketing tool. Stimulating
consumers to create lists and styles with specific manufacturers’ products
could lead to a high degree of awareness. As described above, some leading
fashion brands are already experimenting with SSCs. For example, Coach’s
contest discussed earlier was very successful. Within one week, Coach received
3,692 styles, with 103,379 likes and 204,656 page views [26]. Besides the high
visibility on the SSC itself, a contest could also increase decentralized brand
and shop awareness when, for example, consumers, especially “lead users,”
embed this content in their social networks. This encourages user participation
and the creation of emotional links that enhance shop and product loyalty [22].
Furthermore, from the perspective of an operator, social shopping features
34 Olbrich and Holsing

can support one of the most important success factors in e‑commerce in the
future: decentralization [64].
Moreover, manufacturers could upload images of a new product or col-
lection before deciding to produce them, so as to obtain input from users,
especially fashion-savvy users. For this purpose, manufacturers could organize
a contest and use it as a forecasting tool, that is, base their order or product
quantities on how often their products are included in styles. In this context,
retailers and manufacturers could also license the SSC technology for use on
their own Web sites, for example, as a white label solution, with respect to
just such market-research questions.
Last but not least, the price of products is a fundamental component of
consumer decision-making processes. As our analysis shows, the direct shop-
ping filter mechanism of price correlates positively with a click-out. Thus,
online shop managers should take this into account and create appropriately
specific pricing strategies.
Overall, our findings yield preliminary insights for marketers, as well as
researchers, so that they can better assess the purchasing behavior in SSCs.
In particular, our results suggest that UGC, within the new business model of
SSC, should be taken seriously by online retailers and manufacturers in order
to boost brand or shop awareness as well as sales.

Limitations and Further Research

Our study is the first to model consumer purchasing behavior in SSCs with
detailed clickstream data, but it has limitations. First, we were unable to de-
termine whether the significance of the measures is due to the nature of result
presentation by the operator or the way the information is processed by the
user. Furthermore, we abstracted the Web site by categorizing pages, which
resulted in the loss of substantial information in terms of page content. Second,
we used aggregated measures in our models. This inevitably entails a degree
of lost information. Third, the data were obtained from an early developmental
phase of the observed SSC. Thus, consumers are often unfamiliar with the
various facilities. Observation over time would therefore be useful in order
to integrate a more representative sample into the modeling process.
A number of interesting research questions remain open. It would be useful
to investigate in greater detail the behavior of repeat users, as well as registered
users. An investigation of interaction behavior with other users and the use
of their own, self-created UGC would also be of substantial benefit. In this
manner, an operator could more accurately determine the economic value of
a particular registered user. In this respect, the use of created content by other
users and the associated revenues through, for instance, a click-out could be
allocated to a specific user. Therefore, a combination of clickstream data with
additional data sources, such as revenue, could yield additional insights.
An investigation of further categories, such as consumer electronics, could
usefully extend our results. In addition, capturing actual purchases on par-
ticipating online shops would provide a more complete picture of purchasing
behavior. Moreover, an investigation of embedded lists and styles on social
International journal of electronic commerce 35

networks like Facebook would be of great interest. For example, their impact
on conversion rates, in comparison to other online marketing activities, would
be worth considering.

REFERENCES

1. Alba, J.; Lynch, J.; Weitz, B.; Janiszewski, C.; Lutz, R.; Sawyer, A.; and
Wood, S. Interactive home shopping: Consumer, retailer, and manufacturer
incentives to participate in electronic marketplaces. Journal of Marketing, 61,
3 (1997), 38–53.
2. Allison, P. Logistic Regression Using the SAS System: Theory and Applica-
tion. Cary, NC: SAS Institute, 1999.
3. Andrews, D.; Preece, J.; and Turoff, M. A conceptual framework for de-
mographic groups resistant to on-line community interaction. International
Journal of Electronic Commerce, 6, 3 (spring 2002), 9–24.
4. Ayanso, A., and Yoogalingam, R. Profiling retail Web site functionalities
and conversion rates: A cluster analysis. International Journal of Electronic
Commerce, 14, 1 (fall 2009), 79–113.
5. Bakos, Y.J. Reducing buyer search costs: Implications for electronic mar-
ketplaces. Management Science, 43, 12 (1996), 1676–1692.
6. Bucklin, R.E., and Gupta, S. Brand choice, purchase incidence and seg-
mentation: An integrated modeling approach. Journal of Marketing Research,
29, 2 (1992), 201–215.
7. Bucklin, R.E., and Sismeiro, C. A model of Web site browsing behavior
estimated on clickstream data. Journal of Marketing Research, 40, 3 (2003),
249–267.
8. Bucklin, R.E., and Sismeiro, C. Click here for Internet insight: Advances
in clickstream data analysis in marketing. Journal of Interactive Marketing, 23,
1 (2009), 35–48.
9. Bucklin, R.E.; Lattin, J.M.; Ansari, A.; Gupta, S.; Bell, D.; Coupey, E.;
Little, J.D.C.; Mela, C.; Montgomery, A.; and Steckel, J. Choice and the
Internet: From clickstream to research stream. Marketing Letters, 13, 3 (2002),
245–258.
10. Cattuto, C.; Loreto, V.; and Pietronero, L. Semiotic dynamics and collab-
orative tagging. In Proceedings of the National Academy of Sciences of the United
States of America (PNAS), 104, 5 (2007), pp. 1461–1464 (available at www
.pnas.org/content/104/5/1461.full.pdf+html).
11. Chatterjee, P.; Hoffmann, D.L.; and Novak, T.P. Modeling the click-
stream: Implications for Web-based advertising efforts. Marketing Science,
22, 4 (2003), 520–541.
12. Chen, Y., and Xie, J. Online consumer review: Word-of-mouth as a new
element of marketing communication mix. Management Science, 54, 3 (2008),
477–491.
13. Chen, Y.-C.; Shang, R.A.; and Kao, C.Y. The effects of information over-
load on consumers’ subjective state towards buying decision in the Internet
shopping environment. Electronic Commerce Research and Applications, 8, 1
(2009), 48–58.
36 Olbrich and Holsing

14. Cheung, M.K.; Lee, K.O.; and Rabjohn, N. The impact of electronic
word-of-mouth. Internet Research, 18, 3 (2008), 229–247.
15. Chevalier, J.A., and Mayzlin, D. The effect of word of mouth on sales:
Online book reviews. Journal of Marketing Research, 43, 3 (2006), 345–354.
16. Cooley, R.; Mobasher, B.; and Srivastava, J. Data preparation for mining
World Wide Web browsing patterns. Knowledge and Information Systems, 1, 1
(1999), 5–32.
17. Danaher, P.J.; Mullarkey, G.M.; and Essegaier, S. Factors affecting Web
site visit duration: A cross-domain analysis. Journal of Marketing Research, 43,
2 (2006), 182–194.
18. Dellarocas, C.; Zhang, X.; and Awad, N.F. Exploring the value of online
product reviews in forecasting sales. The case of motion pictures. Journal of
Interactive Marketing, 21, 4 (2007), 23–45.
19. Dhar, V., and Chang, E.A. Does chatter matter? The impact of user-
generated content on music sales. Journal of Interactive Marketing, 23, 4
(2007), 300–307.
20. Engel, J.F., and Blackwell, R.D. Consumer Behavior. New York: Holt, Rine-
hart and Winston, 1982.
21. FIDIS. Study on mobile communities. Future of Identity in the Informa-
tion Society (FIDIS) Consortium, no. 507512, 2009 (available at www.fidis
.net/fileadmin/fidis/deliverables/new_deliverables2/fidis-wp11-del11.8_
study_on_mobile_communities.final.pdf).
22. Flavian, C., and Guinaliu, M. The influence of virtual communities on
distribution strategies in the Internet. International Journal of Retail and Dis-
tribution Management, 33, 6 (2005), 405–425.
23. Ghose, A., and Ipeirotis, P. The EconoMining project at NYU: Study-
ing the economic value of user-generated content on the Internet. Journal of
Revenue and Pricing Management, 8, 2/3 (2009), 241–246.
24. Godes, D.B., and Mayzlin, D. Using online conversation to study word-
of-mouth communication. Marketing Science, 23, 4 (2004), 545–560.
25. Golder, S., and Huberman, B. Usage patterns of collaborative tagging
systems. Journal of Information Science, 32, 2 (2006), 198–208.
26. Gould-Simon, A. Designers’ love affair with Polyvore. BlackBook, June 14
(available at www.blackbookmag.com/article/designers-love-affair-with-
polyvore/19505/).
27. Greene, W.H. Econometric Analysis, 5th ed. New York: Macmillan, 2003.
28. Hansen, T., and Jensen, J.M. Shopping orientation and online clothing
purchases: The role of gender and purchase intention. European Journal of
Marketing, 43, 9/10 (2009), 1154–1170.
29. Hoffman, D.L., and Novak, T.P. Marketing in hypermedia computer-
mediated environments: Conceptual foundations. Journal of Marketing, 60, 3
(1996), 50–68.
30. Holzwarth, M.; Janiszewski, C.; and Neumann, M.M. The influence of
avatars on online consumer shopping behavior. Journal of Marketing, 70, 4
(2006), 19–36.
31. Huang, P.; Lurie, N.H.; and Mitra, S. Searching for experience on the
Web: An empirical examination of consumer behavior for search and expe-
rience goods. Journal of Marketing, 73, 2 (2009), 55–69.
International journal of electronic commerce 37

32. Janiszewski, C. The influence of display characteristics on visual explor-


atory search behavior. Journal of Consumer Research, 25, 3 (1998), 290–301.
33. Jayawardhena, C., and Wright, L.T. An empirical investigation into
e‑shopping excitement: Antecedents and effects. European Journal of Market-
ing, 43, 9/10 (2009), 1171–1187.
34. Johnson, E.J.; Bellman, S.; and Lohse, G.L. Cognitive lock-in and the
power law of practice. Journal of Marketing, 67, 2 (2003), 62–75.
35. Johnson, E.J.; Moe, W.W.; Fader, P.S.; Bellman, S.; and Lohse, G.L. On the
depth and dynamics of online search behavior. Management Science, 50, 3
(2004), 299–308.
36. Kalczynski, P.; Senecal, S.; and Nantel, J. Predicting on-line task comple-
tion with clickstream complexity measures: A graph-based approach. Inter-
national Journal of Electronic Commerce, 10, 3 (spring 2006), 121–141.
37. Kozinets, R.V. E‑tribalised marketing? The strategic implications of
virtual communities of consumption. European Management Journal, 17, 3
(1999), 252–264.
38. Laudon, K.C., and Traver, C.G. E‑Commerce: Business, Technology, Society,
5th ed. Upper Saddle River, NJ: Pearson Prentice Hall, 2009.
39. Lee, M.K.O., and Turban, E. A trust model for consumer Internet shop-
ping. International Journal of Electronic Commerce, 6, 1 (fall 2001), 75–91.
40. Liang, T.P., and Huang, J.S. An empirical study on consumer acceptance
of products in electronic markets: A transaction cost model. Decision Support
Systems, 24, 1 (1998), 29–43.
41. Liang, T.P.; Chen, H.Y.; and Turban, E. Effect of personalization on the
perceived usefulness of online customer services: A dual-core theory. In
P.Y.K. Chau, K. Lyytinen, and C.-P. Wei (eds.), Proceedings of the 11th Inter-
national Conference Electronic Commerce. New York: ACM Press, 2009, pp.
279–288.
42. Liu, Y. Word of mouth for movies: Its dynamics and impact on box office
revenue. Journal of Marketing, 70, 3 (2006), 74–89.
43. Louviere, J.J.; Hensher, D.A.; and Swait, J.D. Stated Choice Methods:
Analysis and Application. Cambridge: Cambridge University Press, 2000.
44. Macaulay, L.A.; Keeling, K.; McGoldrick, P.; Dafoulas, G.; Kalaitzakis, E.;
and Keeling, D. Co-evolving e‑tail and on-line communities: Conceptual
framework. International Journal of Electronic Commerce, 11, 4 (summer 2007),
53–77.
45. Mandel, N., and Johnson, E.J. When Web pages influence choice: Effects
of visual primes on experts and novices. Journal of Consumer Research, 29, 2
(2002), 235–245.
46. Miller, C.C. Polyvore hires former Google executive as chief. New York
Times, February 22, 2010 (available at http://bits.blogs.nytimes.com/2010/
02/22/polyvore-snags-former-google-executive-as-c-e-o/).
47. Moe, W.W. Buying, searching, or browsing: Differentiating between
online shoppers using in-store navigational clickstream. Journal of Consumer
Psychology, 13, 1/2 (2003), 29–40.
48. Moe, W.W. An empirical two-stage choice model with varying decision
rules applied to Internet clickstream data. Journal of Marketing Research, 43, 4
(2006), 680–692.
38 Olbrich and Holsing

49. Moe, W.W., and Fader, P.S. Capturing evolving visit behavior in click-
stream data. Journal of Interactive Marketing, 18, 1 (2004), 5–19.
50. Moe, W.W., and Fader, P.S. Dynamic conversion behavior at e‑commerce
sites. Management Science, 50, 3 (2004), 326–335.
51. Monroe, S.; Sinclair, D.; and Wachinger, T.A. Understanding online
shoppers in Europe. McKinsey Quarterly, May 2009 (available at www
.mckinseyquarterly.com/Understanding_online_shoppers_in_Europe_
2353/).
52. Montgomery, A.L.; Li, S.; Srinivasan, K.; and Liechty, J.C. Modeling on-
line browsing and path analysis using clickstream data. Marketing Science,
23, 4 (2004), 579–595.
53. Mozer, M.C.; Wolniewicz, R.; Grimes, D.D.; Johnson, E.; and Kaushan-
sky, H. Predicting subscriber dissatisfaction and improving retention in the
wireless telecommunications industry. IEEE Transactions on Neural Networks,
11, 3 (2000), 690–696.
54. Nielsen Media Research. Online engagement deepens as social media
and video sites reshape the Internet, Nielsen reports. April 22, 2009 (avail-
able at www.nielsen.com/us/en/insights/press-room/2009/online_
engagement.html).
55. Nikolaeva, R., and Sriram, S. The moderating role of consumer and
product characteristics on the value of customized on-line recommenda-
tions. International Journal of Electronic Commerce, 11, 2 (winter 2006–7),
101–123.
56. Niraj, R.; Gupta, M.; and Narasimhan, C. Customer profitability in a
supply chain. Journal of Marketing, 65, 3 (2001), 1–16.
57. Nov, O., and Ye, C. Why do people tag? Motivations for photo tagging.
Communications of the ACM, 53, 7 (2010), 128–131.
58. O’Reilly, T., and Battelle, J. Web squared: Web 2.0 five years on. Web 2.0
Summit, O’Reilly Media, Sebastopol, CA, 2009 (available at http://assets
.en.oreilly.com/1/event/28/web2009_websquared-whitepaper.pdf).
59. Padmanabhan, B.; Zheng, Z.; and Kimbrough, S.O. Personalization
from incomplete data: What you don’t know can hurt. In F. Provost and R.
Srikant (eds.), Proceedings of ACM SIGKDD International Conference on Knowl-
edge Discovery and Data Mining. New York: ACM Press, 2001, pp. 154–163.
60. Park, D.; Lee, J.; and Han, I. The effect of on-line consumer reviews on
consumer purchasing intention: The moderating role of involvement. Inter-
national Journal of Electronic Commerce, 11, 4 (summer 2007), 125–148.
61. Park, J., and Chung, H. Consumers’ travel website transferring behav-
iour: Analysis using clickstream data-time, frequency, and spending. Service
Industries Journal, 29, 10 (2009), 1451–1463.
62. Park, Y.H., and Fader, P.S. Modeling browsing behavior at multiple Web
sites. Marketing Science, 23, 2 (2004), 280–303.
63. Parsons, A.G. Non-functional motives for online shoppers: Why we
click. Journal of Consumer Marketing, 19, 5 (2002), 380–392.
64. Peters, K.; Albers, S.; Asselmann, D.; and Schäfers, B. E-commerce re-
visited: The impact of an uncoupled consumer buying process on retailing.
Marketing—Journal of Research and Management, 5, 2 (2009), 85–104.
International journal of electronic commerce 39

65. Polyvore. Overview (available at www.polyvore.com/cgi/about/).


66. Polyvore. Contests: Sports are delicious (available at www.polyvore
.com/sports_are_delicious/contest.show?id=87455/).
67. Polyvore. Blake Lively (available at www.polyvore.com/blake_lively/
set?id=14492336/).
68. Rao, L. Social shopping site ThisNext raises $1.2 million. TechCrunch,
January 28, 2010 (available at http://techcrunch.com/2010/01/28/social-
shopping-site-thisnext-raises-1-2-million/).
69. Rindfleisch, A., and Heide, J.B. Transaction cost analysis: Past, present,
and future applications. Journal of Marketing, 61, 4 (1997), 30–54.
70. Rothaermel, F.T., and Sugiyama, S. Virtual Internet communities and
commercial success: Individual and community-level theory grounded
in the atypical case of TimeZone.com. Journal of Management, 27, 3 (2001),
297–312.
71. SAS Institute. SAS/STAT 9.1 User’s Guide. Cary, NC: SAS Institute, 2004.
72. Senecal, S.; Kalczynski, P.J.; and Nantel, J. Consumers’ decision-making
process and their online shopping behavior: A clickstream analysis. Journal
of Business Research, 58, 11 (2005), 1599–1608.
73. Sismeiro, C., and Bucklin, R.E. Modeling purchase behavior at an
e‑commerce website: A conditional probability approach. Journal of Market-
ing Research, 41, 3 (2004), 306–323.
74. Smith, M.D., and Brynjolfsson, E. Consumer decision-making at an
Internet shopbot: Brand still matters. Journal of Industrial Economics, 49, 4
(2001), 541–558.
75. Stephen, A.T., and Toubia, O. Deriving value from social commerce net-
works. Journal of Marketing Research, 42, 2 (2010), 215–228.
76. Su, B.C. Consumer e‑tailer strategies at on-line shopping comparison
sites. International Journal of Electronic Commerce, 11, 3 (spring 2007), 135–159.
77. Su, B.C. Characteristics of consumer search on-line: How much do we
search? International Journal of Electronic Commerce, 13, 1 (fall 2008), 109–129.
78. Tedeschi, B. Like shopping? Social networking? Try social shopping.
New York Times, September 11, 2006 (available at www.nytimes.com/
2006/09/11/technology/11ecom.html?fta=y/).
79. Trusov, M.; Bucklin, R.; and Pauwels, K. Effects of word-of-mouth ver-
sus traditional marketing: Findings from an Internet social networking site.
Journal of Marketing, 73, 5 (2009), 90–102.
80. Turban, E.; King, D.; and Lang, J. Introduction to Electronic Commerce, 3d
ed. Upper Saddle River, NJ: Pearson Education, 2011.
81. Van den Poel, D., and Buckinx, W. Predicting online-purchasing behav-
ior. European Journal of Operational Research, 166, 2 (2005), 557–575.
82. Villanueva, J.; Yoo, S.; and Hanssens, D.M. The impact of marketing-
induced versus word-of-mouth customer acquisition on customer equity
growth. Journal of Marketing Research, 45, 1 (2008), 48–59.
83. Wang, K.; Wang, E.T.G.; and Farn, C.K. Influence of Web advertising
strategies, consumer goal-directedness, and consumer involvement on Web
advertising effectiveness. International Journal of Electronic Commerce, 13, 4
(summer 2009), 67–95.
40 Olbrich and Holsing

84. Wang, Y., and Fesenmaier, D.R. Assessing motivation of contribution in


online communities: An empirical investigation of an online travel commu-
nity. Electronic Markets, 13, 1 (2003), 33–45.
85. Williamson, O.E. The Economic Institutions of Capitalism: Firms, Markets,
Relational Contracting. New York: Free Press, 1985.

Rainer Olbrich (rainer.olbrich@fernuni-hagen.de) is a professor of marketing at


the University of Hagen, Germany. He received his Ph.D. in business administration
from the University of Münster in 1992. He has held the Chair of Marketing at the
University of Hagen since 1997. He has worked as a consultant for various companies
and organizations. His research interests include consumer goods marketing, retail
marketing, search engine marketing, and social commerce. He has published a number
of books, e.g., Marketing and Marketing-Controlling, in these areas, and his papers have
appeared in various academic journals.

Christian Holsing (christian.holsing@fernuni-hagen.de), is a Ph.D. candidate in


the Department of Marketing at the University of Hagen, Germany. He studied business
administration at the Universities of Vienna and Bielefeld, where he achieved the degree
Dipl.-Kfm. His business experience includes a two-year apprenticeship in wholesale
and foreign trade at Germany’s largest grocery retailer, the founding of an e‑commerce
start-up, various internships (e.g., at Telefónica), and marketing consulting experience
in the retail industry at Information Resources Inc. (IRI). His research interests include
online consumer behavior, social commerce, online business models, electronic word
of mouth, interactive marketing, retail marketing, and Web analytics.

You might also like