Consumer's decision making process

Consumers’ Decision-Making Process and Their Online Shopping Behavior: A Clickstream Analysis

Sylvain Senecal Assistant Professor of E-Commerce and Marketing

University of Toledo 2801 W. Bancroft St., Stranahan Hall 4039, Toledo OH 43606-3390, Phone (419) 530-2422,

Fax (419) 530-2290, email: [email protected]

Pawel J. Kalczynski Assistant Professor of Information Systems

University of Toledo 2801 W. Bancroft St., Stranahan Hall 4039, Toledo OH 43606-3390, Phone (419) 530-2258,

Fax (419) 530-2290, email: [email protected]

Jacques Nantel Professor of Marketing University of Montreal

Office 4.735, 3000 Chemin de la Côte-Sainte-Catherine, Montreal (Quebec), Canada H3T 2A7, Phone (514) 340-6421, email: [email protected]

Submitted to the 6th Annual Retail Strategy and Consumer Decision Research Symposium

and to the Journal of Business Research

June 17, 2003


ABSTRACT

The objective of this study is to investigate how different online decision-making processes used by

consumers influence the complexity of their online shopping behaviors. During an online experiment,

subjects were asked to perform a shopping task on a website offering product recommendations.

Significant differences were observed between subjects’ decision-making processes and their online

shopping behavior. Subjects who did not consult a product recommendation had a significantly less

complex online shopping behavior (e.g., fewer web pages viewed) than subjects who consulted the

product recommendation. In addition, differences were also found between the online shopping behavior

of subjects who consulted but did not follow the product recommendation and subjects who consulted and

followed the product recommendation. Managerial and theoretical implications of these results are

provided.

1


1. INTRODUCTION

The objective of this paper is to investigate how different online decision-making processes used by

consumers to make a product choice influence the complexity of their online shopping behavior. When

faced with a product selection, consumers are suggested to perform an internal search (e.g., relying on

their prior knowledge of brands) and if necessary, an external search. The latter may comprise activities

such as gathering more information about brands and seeking recommendations from relevant others.

Thus, different consumers may use different decision-making strategies to make a consumption decision

(Olshavsky, 1985; Payne et al., 1993). Furthermore, consumers shopping online may modify or change

the way they search for information to take advantage of certain unique characteristics of the Internet

(Peterson and Merino, 2003). For instance, the presence of new information sources such as recommender

systems, intelligent-agent-based systems, and less easily accessible sources offline (e.g., opinions of a

large group of consumers on a specific product) may modify the way, in which consumers perform their

external information search. In this paper, we investigate the effect of different decision-making processes

on consumers’ shopping behaviors (e.g., decision time, pages visited, etc.) while performing an online

goal-directed activity, namely, the selection of a product.

When applied to the Internet, the effect of various decision-making processes on consumers’ shopping

behavior leads to interesting questions. For instance, do consumers who consult and follow an online

product recommendation have a less complex shopping behavior than consumers who do not consult or

who do consult but do not follow a recommendation? Answers to such questions have important

managerial and theoretical implications. First, they would help marketers maximize the effectiveness and

usability of their websites. For instance, if it were known that after they consult an online product

recommendation, consumers usually revisit product detail pages, hyperlinks from the product

recommendation page to these pages would facilitate consumers’ navigation and consequently, their

decision-making process. Second, Peterson and Merino (2003) and Cowles, Kiecker, and Little (2002)

argue that the Internet represents a sufficiently different retail environment where concepts such as

1

consumer information search behavior should be revisited. Thus, by investigating the effect of

consumers’ decision-making process on their online shopping behavior, this paper contributes to better

understand how consumers search for information and make their decisions online.

2. LITERATURE REVIEW

2.1. Product Recommendations and Decision-Making Processes

If a product recommendation from an information source is available to consumers, they can either decide

not to consult it, consult and follow it, or consult and not follow it. If they decide not to consult the

product recommendation, consumers would rely only on their prior knowledge or experience and on other

information about the products to make a decision. Thus, they would use an affect referral or an own-

based decision making process (Olshavsky, 1985). The former is generally favored by consumers who

already possess a strong attitude toward one option (Wright, 1975). In these cases, consumers do not base

their decisions on an exhaustive evaluation of attributes and/or alternatives, but rather on their past

experience. In fact, this is a heuristic that consists simply in accessing one’s attitude in memory in order

to make a decision. Thus, the search effort is solely internal. For the latter type of decision-making

process Payne, Bettman and Johnson (1993) suggest that consumers can use a variety of heuristics

(lexicographic, disjunctive, etc.) that may vary according to the desired decision’s accuracy and the effort

that consumers are willing to invest in the particular decision.

If consumers decide to consult and follow the product recommendation, they would use an other-based

decision-making process (Olshavsky, 1985; Rosen and Olshavsky, 1987a; Rosen and Olshavsky, 1987b).

When consumers do not have a preferred option or the capacity or the motivation to process information,

they may turn to other-based decision-making processes. Here, consumers subcontract either part or all of

their decision-making process. Solomon (1986) predicates that consumers may use surrogates to act on

their behalf for information search, evaluation of options and/or even to carry out transactions. In their

study of life insurance purchases, Formisamo, Olshavsky, and Tapp (1982) found that 71% of consumers

that followed an other-based decision-making process made their purchase choice in keeping with the

2

salesperson’s recommendation. In other-based decision-making processes, the final brand decision comes

from a recommendation source.

If consumers decide to consult the product recommendation, but to not follow it, it would represent an

owned-based or hybrid decision-making process depending on the extent of the usage of the

recommendation in their decision-making process (Olshavsky, 1985; Rosen and Olshavsky, 1987a; Rosen

and Olshavsky, 1987b). Consumers who adopt own-based decision-making processes can be influenced

by recommendations but do not rely on them exclusively to make decisions. For instance, a consumer

may ask a close friend about which attributes are important to consider for a given product (Price and

Feick, 1984), but may also gather complementary information from other information sources such as

advertising, store visit, and salespeople in order to determine the pertinent product attributes to consider.

However, if the recommendation plays a greater role in the process it would be considered a hybrid

decision-making process. In these situations, consumers use the recommendation more extensively in

their decision-making process. For instance, Rosen and Olshavsky (1987b) found evidence that

consumers use a recommended brand from a trusted information source as a benchmark to evaluate other

brands in order to find the best brand available. Based on the theory of reactance (Brehm, 1966),

Fitzsimons and Lehmann (2001) found that when consumers decide to go against a product

recommendation they experience decreased satisfaction, increased difficulty, and increased confidence

with their product choice.

Thus, overall the shopping behavior of consumers who consult but do not follow a product

recommendation should be more complex than those who do not consult a product recommendation since

the former base their decision on at least one more piece of information, which is the recommendation. In

addition, the shopping behavior of consumers who consult but do not follow a product recommendation

should be more extensive and complex than those who do consult and follow a product recommendation

since there is a mismatch between the recommendation and the preferred alternative, which leads to more

deliberation.

Research has shown that the type of product affects the type of information search and ultimately the

decision-making process consumers use to select a product (Bearden and Etzel, 1982; Childers and Rao,

3

1992; Formisamo et al., 1982; King and Balasubramanian, 1994; Olshavsky and Granbois, 1979). Nelson

(1970) suggests that goods can be classified as possessing either search or experience qualities. Search

qualities are those that “the consumer can determine by inspection prior to purchase” and experience

qualities are those that “are not determined prior to purchase” (Nelson, 1974 p. 730) King and

Balasubramanian (1994) found that consumers assessing a search product (e.g., a 35mm camera) are more

likely to use own-based decision-making processes than consumers assessing an experience product, and

that consumers evaluating an experience product (e.g., a film-processing service) rely more on other-

based and hybrid decision-making processes than consumers assessing a search product. Thus, by

influencing the consumers’ decision-making process, the type of product should also influence their

shopping behavior.

2.2. Online Product Recommendations

In light of research on consumers’ use of relevant others (e.g., friends) in their pre-purchase external

search efforts (Olshavsky and Granbois, 1979; Price and Feick, 1984; Rosen and Olshavsky, 1987a;

Rosen and Olshavsky, 1987b) and in consideration of the emergence of online information sources

providing personalized recommendations (Maes, 1999), Senecal and Nantel (2002) assert that online

recommendation sources can be sorted into three broad categories: 1) other consumers (e.g., relatives,

friends and acquaintances), 2) human experts (e.g., salespersons, independent experts), and 3) expert

systems and consumer decision support systems such as recommender systems and intelligent-agent-

based systems. Although the first two information sources have been researched in consumer behavior in

the past and are known to be used by and influence consumers (Ardnt, 1967; Brown and Reingen, 1987;

Duhan et al., 1997; Gilly et al., 1998; Olshavsky and Granbois, 1979; Price and Feick, 1984; Rosen and

Olshavsky, 1987a; Rosen and Olshavsky, 1987b; Still et al., 1984), recommender systems and intelligent-

agent-based systems are relatively new. These information sources can help consumers in various steps of

their decision-making process. According to Maes (1999), they can help consumers select products, select

merchants, and even automate negotiation. Peterson and Merino (2003) suggest that these latter sources,

as they become more exhaustive and efficient, will be increasingly used by consumers using the Internet

for information searches. In a study of an independent third party recommender system for pickup trucks, 4

Urban, Sultan and Qualls (1999) found that 88% of consumers using the recommender system agreed that

the recommendations provided met their needs and that 60% of consumers agreed that the system

suggested new alternatives that would not have been considered. Recommender systems have been found

to help consumers efficiently filter available alternatives, increase the quality of their considered set (i.e.,

a larger proportion of non-dominated alternatives), and increase their product choice confidence (Häubl

and Trifts, 2000; Urban et al., 1999). Finally, Cooke et al. (2002) suggest that the influence of

recommendations from these systems is moderated by contextual and product-specific information. For

instance, if a recommendation for an unfamiliar product is presented in a way that it is perceived similar

to attractive familiar products, consumers will evaluate the unfamiliar product more favorably (Cooke et

al., 2002).

2.3. Online Shopping Behavior and Clickstream Analysis

Clickstream can be defined as the path a consumer takes through one or more websites (Bucklin et al.,

2002). It can include within-site information such as the pages visited, the time spent on each page and

between-site information such as the websites visited. Thus, researchers have investigated consumer

behaviors across websites (Goldfarb, 2002; Johnson et al., 2000; Park and Fader, 2002) and within a

particular website (Bucklin and Sismeiro, 2002; Li et al., 2002; Moe, 2003; Moe and Fader, 2002; Moe

and Fader, 2001). In the latter category, some studies focused on single visits to a particular website (Li et

al., 2002; Moe, 2003), on multiple visits (Moe and Fader, 2002; Moe and Fader, 2001), or on both types

of visits (Bucklin and Sismeiro, 2002).

Within-website research has focused on clickstream or website related variables that help explain the goal

pursued by consumers who visit a website (Moe and Fader, 2002; Moe and Fader, 2001), why consumers

continue browsing on a website (Bucklin and Sismeiro, 2002), and which visitors are likely to make a

purchase (Li et al., 2002; Moe and Fader, 2002). For instance, Moe and Fader (2001) found that, based on

the clickstream of consumers who visited a bookstore and a CD store, consumer visits can be categorized

as either: 1) Directed-purchase (planned and immediate purchase), 2) Search/deliberation (planned and

future purchase), 3) Hedonic browsing (unplanned but immediate purchase), 4) Knowledge-building

(unplanned and future purchase). They found that consumers in directed-purchase visits exhibit a more 5

focused online shopping behavior by viewing less product category pages, viewing more product detail

pages within a category, spending more time on each page, repeating visits to product pages (Moe, 2003;

Moe and Fader, 2001). Thus far, no study has investigated the different consumers’ decision-making

processes within one specific type of visit, namely directed-purchase visits.

3. HYPOTHESES

As mentioned, consumers who consult but do not follow the recommendation (CNF) should have a more

complex shopping behavior than consumers who do not consult an online product recommendation (NC)

because they have more information to process. CNF should also have a more complex shopping behavior

than consumers who consult and follow a product recommendation (CF) because the recommendation

provided does not match their preferred alternative. This mismatch increases the decision difficulty

(Fitzsimons and Lehmann, 2001). Online, CNF should visit and revisit more web pages, including

product detail pages, than NC and CF when shopping online for a product. This complex navigation

pattern will lead CNF to display a navigation pattern that is less linear than NC and CF. By visiting and

revisiting more pages, CNF should follow a greater number of all the available links on a website. Thus,

their navigation pattern should also be more densely connected (i.e., compact). By visiting and revisiting

more pages, CNF should spent more time making their decision than NC and CF. Finally, CNF should

spend more time on the web pages they visit since their decision difficulty is greater than for NC and CF.

The following hypotheses are posited to reflect the different shopping behaviors between NC and CNF

and between CF and CNF.

H1: NC will have a less complex online shopping behavior than CNF. H1a: NC will have a less densely connected (i.e., less compact) navigation pattern than CNF. H1b: NC will have a more linear navigation pattern than CNF. H1c: NC will visit fewer pages than CNF. H1d: NC will revisit a smaller proportion of the total number of pages they visit than CNF. H1e: NC will visit fewer product detail pages than CNF. H1f: NC will need less time to make a decision than CNF. H1g: NC will spend less time per page than CNF.

H2: CF will have a less complex online shopping behavior than CNF. H2a: CF will have a less densely connected (i.e., less compact) navigation pattern than CNF. H2b: CF will have a more linear navigation pattern than CNF. H2c: CF will visit fewer pages than CNF. H2d: CF will revisit a smaller proportion of the total number of pages they visit than CNF.

6

H2e: CF will visit fewer product detail pages than CNF H2f: CF will need less time to make a decision than CNF. H2g: CF will spend less time per page than CNF.

Based on the difficulty to evaluate experience products before purchase, we suggest that consumers

shopping online for experience products will have a different shopping behavior than those shopping

online for a search product. The increased difficulty to evaluate and therefore compare experience

products will lead to more web pages being revisited and, consequently, to a less linear shopping behavior

for experience products than for search products. Thus, the following hypothesis is suggested.

H3: When shopping for experience products consumers will have a more complex online shopping behavior than when shopping for search products. H3a: Consumers shopping for a search product will have a more linear navigation pattern than those

shopping for an experience product. H3b: Consumers shopping for a search product will revisit a smaller proportion of the total number of

pages they visit than those shopping for an experience product.

4. METHODOLOGY

4.1. Sample

A convenience sample of 293 subjects was recruited by e-mail. The e-mail stated that two researchers

from a large business school were conducting a study on electronic commerce and that participants had a

chance of wining one of the products about which the experiment was designed. Potential participants did

not know in advance the types of products that were to be tested. Subjects participated in the study from

the location where they usually use the Internet. The majority of subjects were between the ages of 18 and

29 years (84%). Fifty-one percent were female, almost one third were working full time (32%); 25% of

subjects were full-time students and another 33% were part-time workers and students. On average,

subjects had been using the Internet for 4.6 years and currently used it 17 hours per week.

4.2. Procedure

7

In the first session of the experiment subjects were simply asked to complete an online questionnaire. In

the second session, subjects were asked to perform online shopping tasks on a specific website and

complete a final questionnaire. To motivate subjects to participate without mentioning the precise goal of

the experiment (i.e., the influence of recommendation on product choice), a cover story was used.

Subjects were told that a two-session experiment was being conducted to assess the commercial potential

of various products that a foreign company, which we named Maximo, was interested in introducing to

local markets via their website. Although Maximo was presented as a real European company with a

professional looking website, it was in fact a fictitious company. However, all products used in the

experiment were actual brands available online.

In addition, participants were informed that they would be asked in the second session of the experiment

to shop for products online, and that they had a one in three chance of winning one of the products

selected. This procedure was used to maximize the involvement of subjects with their online shopping

tasks. Subjects were informed that the average product value was $45. The first session questionnaire

measured their Internet usage and some demographics. At the end of the questionnaire, subjects were

asked to provide their email address and were told that they would be contacted in the following days for

the second session. Five days after the first session, subjects were sent an email with a hyperlink to the

second session website. Once on the website, they were asked to logon to the second session by entering

their email address. Following a brief introduction to the experimental website to remind them of the goal

of the study (i.e., cover story), they were then advised that within the next few minutes they would be

asked to shop on Maximo’s website.

As recommended by Nosek, Banaji and Greenwald (2002), the first online shopping task was a warm-up

task. Its goal was to familiarize subjects with the structure and functionalities of Maximo’s website (see

Figure 1 for an overview of the website and the different links between the pages). First, subjects were

shown four computer mice on the Product Category Page (CAT). In addition to a picture of each product,

the CAT page provided an overview of the four products (i.e., brand, model, and price). From that page

they could either go to Product Detail Pages (P1, P2, P3, or P4), to the Recommendation Page (RCM), or

to the Product Choice Page (CHC). If subjects elected to go to one of the Product Detail Pages, additional

product attributes and information were presented (e.g., Warranty, Size, Number of buttons, etc.). From a

Product Detail Page, they could either go back to the CAT page, go to the RCM page, or to the CHC

page. If subjects decided to click on the “Our Recommendation” button and go to the RCM page, one of

the four products was recommended. The same product was recommended to all subjects. From the RCM

page, they could either go to the CAT or CHC page. Once on the CHC page, consumers had to select their

8

final product (FIN). Before making their final selection, they could go back to any of the other pages (i.e.,

CAT, RCM, P1, P2, P3, or P4).

The warm-up task was followed by the main online shopping task. For the latter, the product type was

manipulated by using two different product classes. Based on pretest results, the search product class used

for the experiment was the calculator, and wine was used for the experience product class. Thus, subjects

were randomly assigned to a product class i.e., calculator or wine. Note that the data collection was

performed in Canada were the legal age for drinking is 18 years old. The second shopping task essentially

followed the same procedure as the warm-up shopping task. Subjects were asked to select one product out

of four within the product class and could also seek a product recommendation. After having completed

all shopping tasks, subjects were asked to complete a short final questionnaire, in which they were

prompted to guess the main objective of the experiment. They then accessed a debriefing page explaining

the actual goal of the experiment and were logged out of the second session. The debriefing page

explained the real goal of the experiment (i.e., influence of recommendations on product choices),

reassured subjects about their chance to win one of the product they selected, indicated that the collected

data would remain confidential, and that all researchers performing the study had signed a confidentiality

agreement. Finally, subjects were provided the University Ethics Committee phone number in order for

them to call if they had any questions or comments on the study.

INSERT FIGURE 1 HERE

In the above figure, CAT represents the Product Category Page, CHC – the Product Choice Page, FIN –

the Final Product Choice, RCM – the Recommendation Page, and P1 – P4 represent Product Detail Pages.

5. MEASURES

5.1. Variables

The only independent variable measured was consumers’ online decision-making process. Based on their

clickstream data, consumers were assigned to one of the following groups: 1) did not consult the product

recommendation (NC), 2) Consulted, but did not follow the product recommendation (CNF), 3)

Consulted and followed the product recommendation (CF).

9

To test the hypotheses, several clickstream metrics were used to assess the complexity of consumers’

online shopping behavior. Some of them are based on static website complexity metrics, other use

additional attributes related to the dynamic changes recorded in the clickstream. Based on the generic

clickstream model, i.e. session id, node id, and date and time, we computed for each session: 1)

clickstream compactness, 2) clickstream stratum, 3) number of web pages visited, 4) revisited page ratio

(i.e., total number of web pages visited divided by the number of unique web pages visited), 5) number of

visits to product detail pages (see P1-P4), (5) total shopping time (6) average time per page. Since most of

these measures are self-explanatory, only the first two are explained below.

First introduced by Botafogo et al. (Botafogo et al., 1992) to measure topological characteristics of

hypertext, compactness and stratum have been adapted by McEneaney (2001) to measure user

navigational behaviors. Compactness refers to the number and distribution of hyperlinks in the hypertext.

If a hypertext is densely connected, compactness takes values close to one, while for sparsely connected

hypertexts, compactness will be close to zero. For instance, if a consumer followed all the available links

on a website during a session, he/she would have a compactness score close to one. Stratum, in turn,

refers to linearity of the hypertext. The linearity is defined by the extent, to which a graph is organized in

such a way that certain nodes must be read before others. The more linear the hypertext the closer stratum

value to one. For less linear hypertexts, stratum values are close to zero. For instance, goal-directed

consumers shopping for a specific product on a given website should have a greater stratum score than

consumers browsing between various product categories on the same website since the navigation pattern

of the former is more linear. Graph theory was used to generate compactness and stratum scores for each

subject.

5.2. Manipulation Check

Following Perdue and Summers (1986), the product type manipulation was tested during a pretest. For the

pretest, a convenience sample of 33 consumers were recruited and asked to complete an online

questionnaire. Pretest subjects were not included in the final sample. Subjects were asked to evaluate the

nature of a set of product classes (calculator, camping cooler, computer mouse, water filter system, bottle

of wine, and 35mm camera). For each product class, subjects were asked whether products could be 10

evaluated: 1) Before purchase; 2) Mostly before purchase; 3) Mostly after purchase; or 4) Only after

purchase. Results of the pretest indicated that the wine product class was mostly perceived as the

experience product (mean = 3.2, median = 3) and the calculator product class was mostly perceived as the

search product (mean = 1.4, median = 1). Furthermore, the difference between the evaluations of the two

product classes was significant (t(27) = -7.48, p < 0.001). The computer mouse was the most balanced

product category (mean = 2.1. median = 2).

6. RESULTS

Out of the 293 participants, 77 correctly guessed the goal of the experiment (i.e., the influence of

recommendations on product choices). Note that subjects who consulted a product recommendation had a

better chance of guessing the experiment’s goal since they were asked, for the purpose of a related study,

to complete a source credibility measurement scale after their shopping task. Thus, only the data from the

remaining 216 participants was used. In order to test the hypotheses, a MANOVA was performed using

consumers’ decision-making process and the product type as independent variables and their clickstream

measures as dependent variables. Contrary to the product type, the decision-making process of consumers

was not manipulated but observed. As expected, the number of subjects in each decision-making process

group was not equal. Out of the 216 participants, 85 decided not to consult the recommendation, 49

decided to consult and follow the recommendation, and 82 decided to consult but not to follow the

product recommendation. Following Keppel (1991), observations were randomly discarded in order to

have 49 participants in each group and perform a MANOVA without risking violations of the normality

and homogeneity of variance assumptions. Thus, the final sample size to test all hypotheses was 147

subjects. Table 1 provides the descriptive statistics. Results of the MANOVA suggest that there is not

interaction between the decision-making process and product type. However, a main effect of the

decision-making process on clickstream variables was observed.

INSERT TABLE 1 ABOUT HERE

11

6.1. Differences in the Clickstream of NC and CNF

H1a stipulated that consumers who do not elect to consult a product recommendation (NC) while

shopping online would have a less densely connected clickstream than those who consult, but do not

follow an online product recommendation (CNF). Although MANOVA results suggest significant

differences between the three different decision-making processes (F(2, 146) = 5.494, p < 0.01), the

contrast test comparing these two groups does not support H1a (Contrast Estimate (C.E.) = -0.036, p >

0.05). H1b suggested that NC would have a more linear navigation pattern than CNF. Results of a

contrast test support H1b (F(2, 146) = 5.975, p < 0.005; C.E.= – 0.106, p < 0.05). As illustrated in Table

1, NC had a more linear online shopping behavior (i.e., greater stratum) than CNF for both products. H1c

and H1d respectively stipulated that NC would visit fewer pages and revisit a smaller proportion of pages

than CNF. Both hypotheses are strongly supported. Further, NC visited fewer pages to perform their

shopping task (F(2,146) = 14.398, p < 0.001; C.E. = -3.097, p < 0.001) and had a smaller revisited page

ratio (F(2, 146) = 8.488, p < 0.001; C.E. = -0.146, p < 0.05) than CNF. H1e suggested that NC would

make fewer visits to product detail pages than CNF. Results support H1e (F(2, 146) = 4.422, p < 0.05;

C.E. = -1.241, p < 0.05). As shown in Table 1, more visits were made to product detail pages by CNF

than by NC for both products. H1f and H1g respectively suggested that NC would take less time to make

a product selection and would spend less time per page than CNF. H1f was supported (F(2, 146) = 3.178,

p < 0.05; C.E.= -43.404, p < 0.05). Since NC took less time than CNF to select a calculator or a bottle of

wine. H1g was not supported. No significant difference was found between NC and CNF relative to the

time they spend per page (F(2,146) = 2.471, p < 0.1; C.E. = -1.435, p > 0.05).

6.2. Differences in the Clickstream of CF and CNF

12

H2a proposed that CF would have a more compact online navigation pattern than CNF. Results of a

contrast test do not support this hypothesis (C.E. = -0.038, p > 0.05). In fact, as illustrated in Table 1, the

degree of connectedness seems similar or lower for CNF than for CF. H2b suggested that CF would have

a more linear navigation pattern than CNF. Contrast test results do not support this hypothesis (C.E. =

0.043, p > 0.05). H2c and H2d respectively stipulated that CF would visit fewer pages and revisit a

smaller proportion of pages than CNF. Results of contrast tests did not support H2c (C.E. = -1.032, p >

0.05), but supported H2d (C.E. = -0.150, p < 0.05). As shown in Table 1, surprisingly CF display a

greater revisited page ratio than CNF. H2e suggested that CF would make fewer visits to product detail

pages than CNF. Results do not support H2e (F(2, 146) = 4.422, p < 0.05; C.E. = -0.010, p > 0.05). H1f

and H1g respectively proposed that CF would take less time to make a product selection and would spend

less time per page than CNF. Results of contrast tests suggest that CNF did not take more time than CF to

make a product selection (C.E. = 22.485, p > 0.05), however they did spend more time per page than CF

(C.E. = 3.967, p < 0.05). Overall, except for the time spent per page and the revisited page ratio, CF and

CNF had an identical shopping behavior.

6.3. Differences in the Clickstream of NC and CF

Based on the above results, a post hoc analysis was performed to compare the shopping behavior of CF

and NC. Scheffe test results revealed that NC had a more compact (Mean Difference (M.D.) = -0.0753,

p < 0.005) and a more linear shopping behavior (M.D. = 0.152, p < 0.005) than CF. Furthermore, NC

visited fewer pages (M.D. = -4.204, p < 0.001) and had a smaller revisited page ratio (M.D. = -0.297, p <

0.001) than CF. In addition, NC made fewer visits to detail product pages than CF (M.D. = -1.286, p <

0.05). Finally, no significant differences were found between NC and CF relative to the time they spent to

select a product (M.D. = -21.816, p > 0.05) and the time spent per page (M.D. = 2.347, p > 0.05).

6.4. The Effect of the Product Type on Consumers’ Clickstream

H3 proposed that consumers shopping online for an experience product would have a less linear

navigation pattern (H3a) and would revisit pages in a greater proportion (H3b) than consumers shopping

for a search product. H3a and H3b were not supported. Although in the hypothesized direction, there was

no significant difference between the two products and consumers’ navigation pattern linearity

(F(1,146) = 0.520, p > 0.05). In addition, no significant difference was found between the two products

and the revisited page ratio (F(1, 146) = 0.123, p > 0.05).

7. DISCUSSION

Results suggest that consumers who decided not to consult a product recommendation during their online

shopping have a less complex online shopping behavior than consumers who decided to consult the

13

product recommendation. They were found to have a more linear navigation pattern, visit fewer pages,

visit fewer product detail pages, and revisit a smaller proportion of pages they visited in order to select a

product. When consumers decided to consult an online product recommendation, the only differences

between those who followed and those who did not follow the recommendation are the time they spent

per page and their revisited page ratio. Consumers who did not follow the recommendation spent more

time per page and revisited a smaller proportion of the web pages they visited than those who followed

the product recommendation. Finally, the type of product did not significantly influence consumers’

online shopping behavior.

This study has important theoretical implications. Results surprisingly suggest that consumers who follow

an online product recommendation have a more complex online shopping behavior than those who do not

consult the recommendation and have an online shopping behavior similar to consumers who do not

follow the recommendation. In consumer research it has been traditionally assumed that consumers

follow a product recommendation in order to limit or minimize their information search either because

they lack the capacity or motivation to perform an extensive problem-solving effort (Olshavsky, 1985).

The unique characteristics of the Internet, such as information accessibility, may modify the behavior of

consumers who follow an other-based decision making process. We suggest that low information costs

associated with the Internet increase the amount of information gathered by consumers even when they

use other-based decision-making processes. For instance, in this study consumers could have visited only

two pages to receive a product recommendation, go to the final choice page and select the recommended

product. However, consumers who consulted and followed the product recommendation performed their

shopping task by visiting an average of 10 pages. Since information is more easily available online than

offline, the differences between other-based decision-making process and other decision-making process

that incorporate product recommendations (i.e., own-based or hybrid) become subtler online. Thus,

contrary to what Peterson and Merino (2003) suggest, at least in some cases consumers may search more

online than offline. Finally, in addition to static clickstream metrics (e.g., the number of pages visited), we

used compactness and stratum to assess the degree of complexity of consumers’ online shopping

behavior. These metrics, usually used to assess general navigation patterns (Botafogo et al., 1992;

14

McEneaney, 2001), have been found useful in discriminating between different online decision-making

processes. In addition to online shopping behavior, these metrics could be used to measure other

consumer-related navigation patterns. For instance, consumers seeking information on a less usable

website should exhibit lower stratum and higher compactness scores than consumers seeking information

on a more usable website. Thus, these metrics could be very useful to assess the usability of websites.

This study also has interesting managerial implications. First, as mentioned, consumers who consulted an

online product recommendation performed a much more extensive external search than consumers who

did not consult a product recommendation. Online, consumers who followed the product recommendation

seem to consult a recommendation not to minimize their search effort but to gather more information.

Thus, a website offering product recommendations should facilitate the navigation between product

recommendation pages and other product related pages, since those who consult product recommendation

are also those who visit and revisit more pages including product detail pages. Second, one difference in

the shopping behavior between consumers who followed a product recommendation and those who did

not is the time they spent per page. A potential application of this finding is for online merchants to

establish a time threshold discriminating between these two groups of consumers based on past product

recommendations and sales in order to predict which consumers would not follow the recommendation. If

consumers who consulted a recommendation exceed the calculated time per page threshold, they could

intervene in real time (e.g., live chat) to assist these consumers since they are considered the ones

perceiving more choice difficulty according to Fitzsimons and Lehman (2001).

This study has some limitations that should be kept in mind before applying the results to real market

situations. First, as with most online studies, this study used a convenience sample. Thus, due to the

possible self-selection bias it is not possible to confirm that our set of participants is a representative

sample of the population of Internet shoppers. Second, the website used in this study only contained

seven different pages and four different products. It is possible that a larger assortment of products would

yield different results. For instance, it would be interesting to see what would happen if the number of

alternatives to choose from was greater. Would the online shopping behavior of consumers who consult

and follow a product recommendation still be similar to the one exhibited by consumers who do not

15

follow the recommendation? Third, only one search and one experience product were used. Thus,

additional studies conducted with different samples, larger websites and different products would

contribute to the generalization of the present results and confirmation that they are not idiosyncratic to

this study. Finally, this study only investigated consumers’ online product choices; it did not investigate

online purchases. Thus, additional variables such as product price, product availability or delivery time

could also affect consumers’ decision-making process.

16

CAT RCM

P1

CHC

P2

P3

FIN P4

Figure 1. Experimental Website Organization

Table 1. Clickstream metrics

NC CNF CF Average Calculator (i.e., Search Product) n=49 n=49 n=49 n=147

Compactness 0.445 0.467 0.543 0.482 Stratum (i.e., Linearity) 0.918 0.824 0.739 0.832 Pages 5.3 8.1 10.6 7.9 Revisited page ratio 1.15 1.26 1.55 1.31 Product Detail Pages 1.7 3.1 3.5 2.7 Time (sec). 70 89 101 86 Time (sec.) / Page 11.5 11.0 9.9 10.9

Wine (i.e., Experience Product) Compactness 0.459 0.510 0.510 0.495 Stratum 0.880 0.762 0.760 0.797 Pages 6.3 9.7 9.3 8.6 Revisited page ratio 1.18 1.36 1.37 1.31 Product Detail Pages 2.3 3.4 3.0 2.9 Time (sec). 81 149 91 107 Time (sec.) / Page 13.5 16.8 10.0 13.3

17

REFERENCES

Ardnt J. Role of Product-Related Conversations in the Diffusion of a New Product. Journal of Marketing Research 1967; 4: 291-95. Bearden W.O. Etzel M.J. Reference Group Influence on Product and Brand Purchase Decisions. Journal of Consumer Research 1982; 9 (2): 183-94. Botafogo Rodrigo A. Rivlin Ehud Shneiderman Ben. Structural analysis of hypertexts. ACM Transactions on Information Systems (TOIS) 1992; 10 (2): 142-80. Brehm J.W. A Theory of Psychological Reactance. New York, NY: Academic Press, 1966. Brown J.J. Reingen P.H. Social Ties and Word-of-Mouth Referral Behavior. Journal of Consumer Research 1987; 14 (December): 350-62. Bucklin R.E. Sismeiro C. A Model of Web Site Browsing Behavior Estimated on Clickstream Data. University of California in Los Angeles. 2002. Bucklin Randolph E. Lattin James M. Ansari And Sunil Gupta Asim Bell David Coupey Eloise Little John D. C. Mela Carl Montgomery Alan Steckel Joel. Choice and the Internet: From Clickstream to Research Stream. Marketing Letters 2002; 13 (3): 245-58. Childers T.L. Rao R. The Influence of Familial and Peer-Based Reference Groups. Journal of Consumer Research 1992; 19 (2): 198-212. Cooke A.D.J. Sujan H. Sujan M Weitz B.A. Marketing the Unfamiliar: The Role of Context and Item-Specific Information in Electronic Agent Recommendations. Journal of Marketing Research 2002; 39 (4): 488 - 98. Cowles D.L. Kiecker P. Little M.W. Using Key Informant Insights as a Foundation for E-retailing Theory Development. Journal of Business Research 2002; 55: 629-36. Duhan D.F. Johnson S.D. Wilcox J.B. Harell G.D. Influences on Consumer Use of Word-of-Mouth Recommendation Sources. Journal of the Academy of Marketing Science 1997; 25 (4): 283-95. Fitzsimons G.J. Lehmann D.R. Unsolicited Advice as an Implicit Choice Restriction: When Intelligent Agent Recommendations Yield Contrary Responses. University of Pennsylvania. 2001. Formisamo R.A. Olshavsky R.W. Tapp S. Choice Strategy in a Difficult Task Environment. Journal of Consumer Research 1982; 8 (4): 474-79. Gilly M.C. Graham J.L. Wolfinbarger M.F. Yale L.J. A Dyadic Study of Personal Information Search. Journal of the Academy of Marketing Science 1998; 26 (2): 83-100. Goldfarb A. Analyzing Website Choice Using Clickstream Data: Elsevier Science Ltd., 2002 Häubl G. Trifts V. Consumer Decision-making in Online Shopping Environments: The Effects of Interactive Decision Aids. Marketing Science 2000; 19 (1): 4-21. Johnson E.J. Moe W.W. Fader P.S. Bellman S. Lohse J. On the Depth and Dynamics of World Wide Web Shopping Behavior. University of Pennsylvania. 2000.

18

Keppel G. Design and Analysis: A Researcher's Handbook,. Upper Saddle River, NJ: Prentice Hall, 1991. King M.F. Balasubramanian S.K. The Effects of Expertise, End Goal, and Product Type on Adoption of Preference Formation Strategy. Journal of the Academy of Marketing Science 1994; 22 (2): 146-59. Li S. Montgomery A.L. Srinivasan K. Liechty J.C. Predicting Online Purchase Conversion Using Web Path Analysis. Carnegie Mellon University. 2002. Maes P. Smart Commerce: The Future of Intelligent Agents in Cyberspace. Journal of Interactive Marketing 1999; 13 (3): 66-76. McEneaney John E. Graphic and numerical methods to assess navigation in hypertext. International Journal of Human-Computer Studies 2001; 55: 761-86. Moe W.W. Buying, Searching, or Browsing: Differentiating between Online Shoppers Using In-Store Navigational Clickstream. Journal of Consumer Psychology 2003; 13 (1&2): 29-40. Moe W.W. Fader P.S. Uncovering Patterns in CyberShopping. California Management Review 2001; 43 (4): 106-17. Moe W.W. Fader P.S. Capturing Evolving Visit Behavior in Clickstream Data. University of Pennsylvania. 2002. Nelson P. Information and Consumer Behavior. Journal of Political Economy 1970; 78: 311-29. Nelson P. Advertising as Information. Journal of Political Economy 1974; 83 (July-August): 729-54. Nosek B.A. Banaji M.R. Greenwald A.G. E-Research: Ethics, Security, Design, and Control in Psychological Research on the Internet. Journal of Social Issues 2002; 58 (1): 161-76. Olshavsky R.W. Perceived Quality in Consumer Decision-making: An Integrated Theoretical Perspective. Lexington, MA: DC Heath & Co., 1985 Olshavsky R.W. Granbois D.H. Consumer Decision-making - Fact or Fiction. Journal of Consumer Research 1979; 6 (2): 93-100. Park Y. Fader P.S. Modeling Browsing Behavior at Multiple Websites. Wharton School, University of Pennsylvania. 2002. Payne J.W. Bettman J.R. Johnson E.J. The Adaptive Decision Maker. Cambridge, NY: Cambridge University Press, 1993. Perdue B.C. Summers J.O. Checking the Success of Manipulations in Marketing Experiments. Journal of Marketing Research 1986; 23 (23): 317-26. Peterson R.A. Merino M.C. Consumer Information Search Behavior and the Internet. Psychology & Marketing 2003; 20 (2): 99-121. Price L.L. Feick L.F. The Role of Recommendation sources in External Search: An Informational Perspective. Provo, UT: Association for Consumer Research, 1984

19

Rosen D.L. Olshavsky R.C. The Dual Role Of Informational Social Influence: Implications for Marketing Management. Journal of Business Research 1987a; 15: 123-44. Rosen D.L. Olshavsky R.C. A Protocol Analysis of Brand Choice Strategies Involving Recommendation. Journal of Consumer Research 1987b; 14 (3): 440-44. Senecal S. Nantel J. Online Influence of Relevant Others: A Framework. RBC Financial Group Chair of E-Commerce, HEC Montreal, University of Montreal. 2002. Solomon M.R. The Missing Link: Surrogate Consumers in the Marketing Chain. Journal of Marketing Research 1986; 50 (4): 208-18. Still R.R. Barnes J.H. Kooyman M.E. Word-of-Mouth Communication in Low-Risk Product Decisions. International Journal of Advertising 1984; 3 (4): 335-46. Urban G. Sultan F. Qualls W. Design and Evaluation of a Trust Based Advisor on the Internet. e-Commerce Research Forum, MIT. 1999. Wright P. Consumer Choice Strategies: Simplifying Vs. Optimizing. Journal of Marketing Research 1975; 12 (1): 60-67.

20


8. TECHNICAL APPENDIX FOR REVIEWERS

8.1. Hypertext Representation

We adopt the graph theory to conveniently represent and manipulate hypertext-organized websites.

Traditionally, a graph G is defined as a set of vertices V and set of edges E, which link pairs of vertices.

In other words G=(V,E). Let us define hypertext H as a labeled directed connected graph, with a set of

nodes N and a set of links L such that for any two nodes there exists a directed path that connects these

nodes.

We observe that based on the above assumptions, the minimum number of links for ||N|| nodes in

hypertext H is equal to ||N||-1. The real hypertext, though, is a directed connected multiple graph

(multi-digraph), which means there may be more than one link in the same direction between any two

nodes. Therefore we make an assumption that in hypertext H only one directed link (ni,nj) between ni and

nj such that ni∈N, nj∈N is permitted. In this way the hypertext becomes a simple digraph with the

maximum number of unique hyperlinks equal to ||N||2 (self-loops permitted). In practice, the number of

unique hyperlinks will be somewhere between ||N||-1 and ||N||2.

One can build an adjacency matrix to represent the hypertext graph as defined above. The adjacency

matrix of hypertext H is a matrix with rows and columns labeled with hypertext nodes, havung a 1 or 0 in

position ni, nj according to whether ni and nj are adjacent or not (Chartrand, 1985 p. 218). For a hypertext

with no self-loops the matrix will have 0s on the diagonal.

21

Table 2 below presents an adjacency matrix for the Website used in our online shopping experiment.

Table 2. Adjacency matrix for the experimental website

To From CAT CHC FIN RCM P1 P2 P3 P4 CAT 0 1 0 1 1 1 1 1 CHC 1 0 1 1 1 1 1 1 FIN 0 0 0 0 0 0 0 0 RCM 1 1 0 0 0 0 0 0 P1 1 1 0 1 0 1 1 1 P2 1 1 0 1 1 0 1 1 P3 1 1 0 1 1 1 0 1 P4 1 1 0 1 1 1 1 0 For a given adjacency matrix, a distance matrix may be built. The distance matrix has a structure similar

to the adjacency matrix, from which it is constructed. Each cell in a distance matrix contains the

minimum number of link traversals (clicks) required to reach from node ni to nj (Floyd, 1962). For each

node nj that cannot be reached from another node ni this distance is assumed to be infinite or ∝. Infinite

values, however, are difficult to handle and Botafogo et al. (1992) proposed to substitute them with ||N||,

hence transforming the distance matrix into a converted distance matrix (CDM).

Table 3. Converted distance matrix (CDM) for the experimental website

To From CAT CHC FIN RCM P1 P2 P3 P4 CAT 0 1 2 1 1 1 1 1 CHC 1 0 1 1 1 1 1 1 FIN 8 8 0 8 8 8 8 8 RCM 1 1 2 0 2 2 2 2 P1 1 1 2 1 0 1 1 1 P2 1 1 2 1 1 0 1 1 P3 1 1 2 1 1 1 0 1 P4 1 1 2 1 1 1 1 0 The converted distance matrix may be further transformed into several hypertext complexity metrics.

8.2. Hypertext complexity metrics

In order to compute compactness and stratum values for a given hypertext, its CDM must undergo several

transformations. Namely, converted in-distance (CIN), converted out-distance (COD), relative in-

centrality (RIC), relative out-centrality (ROC), status, contra-status and prestige values must be computed

for each node.

22

Converted in-distance and converted out-distance are centrality metrics used to identify important

destinations with many paths leading to them and important departure points respectively. Let Cij denote

the element of the converted distance matrix that stores the distance from node ni to nj. The converted in-

distance and converted out distance values (Botafogo et al., 1992) for each node are given by

, . (1) ∑=i

ijj CCID ∑=j

iji CCOD

The above metrics are scale-sensitive, and the authors proposed converting them into relative in-centrality

and relative out-centrality respectively. Formally (Botafogo et al., 1992),

RICi=CD/CIDi and ROCi=CD/CODi, where ∑∑=i j

ijCCD . (2)

Based on the above metrics, the compactness of hypertext H is given by

( )

( )2i j

ij2

1NN

C1NN

Cp−

−−

=∑∑

. (3)

The concepts of status and contrastatus have been adopted by Botafogo et al. from an earlier work by

Harary (1959). The metrics were originally applied to measure hierarchies in social theory and they

translate well onto the hierarchical organization of hypertext.

To put it simply, the status of a node ni in hypertext H is defined recurrently as the number of nodes

adjacent to ni (immediate neighbors) plus twice the number of nodes adjacent to the nodes adjacent to ni,

excluding those counted previously, plus three times the number of nodes adjacent to them, again

excluding those counted previously, and so on. The contrastatus is similar to status except that what

makes it bigger is not the number of immediate and non-immediate neighbors of ni but the number of

nodes other than ni, to which ni is an immediate and non-immediate neighbor. Therefore, for a given

converted distance matrix C, the status and contrastatus of node ni are given by

( )∑ <=i

ijiji ,0C,NCifS , ( )∑ <=j

ijijj ,0C,NCifCS . (4)

23

Furthermore the prestige for each node ni in hypertext H is given by Pi=Si-CSi. Because Pi may be

negative for some nodes the absolute value of prestige may be computed for each node. Then, the

absolute prestige of hypertext H is given by

∑=i

iPAP . (5)

The absolute prestige metric is good for describing hierarchically-organized structures. However, as

Botafogo et al. note in (1992), cycles in hypertext may affect the usefulness of the absolute prestige

metric. Hence, they propose to use the linear absolute prestige (LAP) given by

−=odd is N if

4NN

even is N if4NLAP 3

3

, (6)

to compute the stratum. As a result, the stratum of hypertext H is given by

LAPAPSt = . (7)

In sum, compactness focuses on the density of links in hypertext, while stratum – on the hierarchical

structures in the digraph. Table 4 below presents the converted distance matrix extended with the metrics

necessary to compute compactness and stratum values.

Table 4. Calculating compactness and stratum for the experimental website

To From

CAT CHC FIN RCM P1 P2 P3 P4 COD ROC S P |P|

CAT 0 1 2 1 1 1 1 1 8 14.4 8 2 2 CHC 1 0 1 1 1 1 1 1 7 16.4 7 1 1 FIN 8 8 0 8 8 8 8 8 56 2.1 0 -13 13 RCM 1 1 2 0 2 2 2 2 12 9.6 12 6 6 P1 1 1 2 1 0 1 1 1 8 14.4 8 1 1 P2 1 1 2 1 1 0 1 1 8 14.4 8 1 1 P3 1 1 2 1 1 1 0 1 8 14.4 8 1 1 P4 1 1 2 1 1 1 1 0 8 14.4 8 1 1 CID 14 14 13 14 15 15 15 15 115 AP= 26 RIC 8.2 8.2 8.8 8.2 7.7 7.7 7.7 7.7 CS 6 6 13 6 7 7 7 7 For the website used in our study, the compactness and stratum values are 0.85 and 0.2 respectively.

24

8.3. Clickstream data

Clickstream may be defined as a sequence of nodes visited by a consumer in a shopping session. More

formally, for a given set of hypertext nodes { }N21 n,...,n,nN = clickstream is the mapping of N onto the

set of natural numbers, resulting in a sequence Skji ,....nn,n,nS = consisting with as many elements as

there were node visits (McEneaney, 2001). Note that clickstream may contain more than ||N|| elements as

a single node may be visited more than once in a single session.

Depending on the information needs of the online store management, technical infrastructure and stuff

they employ to handle the store’s website, clickstream data may record the finest details of online

shopping behavior such as every move of the mouse pointer or scroll-up/down action or just time-

stamped facts of moving from one node to another in a certain session. The detail-level or granularity of

clickstream data determines the information potential of the clickstream database. The finest is granularity

(detailed data) the more information can be acquired by processing clickstream data.

For the purpose of our experiment we used a simple clickstream database model that consists of one table

with three attributes: (1) unique identifier of a session (2) date and time of user action (click), and (3)

URL of the node visited.

8.4. Analyzing clickstream data

As clickstream databases follow the relational design principles, most processing may be done with a

relational database data manipulation language, namely the Structured Query Language or SQL.

However, some queries, especially those which capture the dynamic nature of clickstream may be very

complex. Moreover, some metrics are too complex for SQL and clickstream data must be processed by an

external (non-relational) application. This was the case with some of the online shopping behavior

complexity metrics we applied in this study.

8.5. Complexity Metrics

Below are more detailed descriptions of just two of the above metrics proposed by McEneaney in (2001):

clickstream compactness and clickstream stratum. The original names of the metrics “path compactness”

and “path stratum,” slightly diverge from a general graph theory notion (as noted by their author), hence

25

the change of “path” to “clickstream.” Clickstream compactness and clickstream stratum are pseudo-

dynamic versions of compactness and stratum once defined for static hypertext. We begin with the

concept of modified distance matrix for representing clickstream data.

8.5.1 Clickstream distance matrix

Let be the subset of unique hypertext nodes in a given clickstream sequence NN ⊂ Scba ,....nn,n,nS =

where { }N21 n,...,n,nN =in ∈ . The clickstream matrix is a (M NN × ) matrix, whose elements Mij

contain integer numbers representing the number of transitions from node ni to nj (McEneaney, 2001).

Next, all values greater than 1 are substituted with 1 for all elements of matrix M. More formally,

. Finally, a distance matrix is built for M according to the rules defined by

Botafogo et al. in (1992). As a result a clickstream distance matrix

)Mmin(1,i ij∀∀ jM ij ←

( )NNC × emerges. Note that infinite

values in C are substituted with ˆ N rather than ||N||.

McEneaney (2001) points out that although the original clickstream record may be lost due to the above

transformations, each node visited and each link traversed is represented in the matrix, thus reflecting the

general structure of user navigation.

8.5.2 Clickstream compactness Despite certain limitations of representing clickstream in a way similar to representing a static website

discussed in (McEneaney, 2001), the metrics may be applied to assess and compare complexity of various

sequences in a given website. The clickstream compactness is given by

( )

( )2i j

ij2

1NN

C1NN

pC−

−−

=∑∑

. (8)

In our study we assume the closer the value is to one the more complex is the online shopping

behavior. In turn, the closer the value is to zero, the less complex is the behavior.

pC

8.5.3 Clickstream stratum The clickstream stratum is given by

26

PAL

PAtS = , (9)

where the elements of the ratio are computed according to formulas (4) (5) and (6) with C used instead

of and

ijˆ

ijC N instead ||N||. We assume that the closer is the value to zero the more complex is online

shopping behavior. The closer the value is to one (strictly linear clickstream) the less complex the

behavior.

tS

References

Botafogo Rodrigo A. Rivlin Ehud Shneiderman Ben. Structural analysis of hypertexts. ACM Transactions on Information Systems (TOIS) 1992; 10 (2): 142-80. Chartrand Gary. Introductory graph theory (Unabridged and corr. ed.). New York: Dover, 1985. Floyd R. W. Algorithm 97: shortest path. Communications of the ACM 1962; 5: 345. Harary F. Status and contrastatus. Sociometry 1959; 22: 23-43. McEneaney John E. Graphic and numerical methods to assess navigation in hypertext. International Journal of Human-Computer Studies 2001; 55: 761-86.

27

Consumer's decision making process

Marketing

Transcript of Consumer's decision making process