Available online at www.sciencedirect.

com

Expert Systems
with Applications
Expert Systems with Applications 35 (2008) 1567–1574
www.elsevier.com/locate/eswa

Online shopping recommendation mechanism and its influence

on consumer decisions and behaviors: A causal map approach
Kun Chang Lee, Soonjae Kwon *

School of Business Administration, Sungkyunkwan University, Myung Ryun 3-53, Chong No-Ku, Seoul 110-745, Republic of Korea

Abstract

Purpose of this paper: Online product recommendation mechanism (agents) are becoming increasingly available on websites to assist
consumers with reducing information overload, provide advice in finding suitable products, and facilitate online consumer decision-
making. Central of these services is consumers’ satisfaction with recommendation results. Traditional recommendation mechanism
(TRM) is based content and/or collaborative filtering approach. However, the remaining problem concerning TRM is how to analyze
the causal relationships between quantitative and qualitative factors, and investigate their impact on the central routes and peripheral
routes through which both quantitative and qualitative factors can affect customer online shopping decisions. It is well known that qual-
itative factors are hard to codify yet they have a significant effect on a customer’s decision-making process in the form of causal rela-
tionships with quantitative factors. Thus, a new online recommendation mechanism is required that incorporates qualitative factors
systematically with quantitative factors to analyze their combined influence on customers’ purchasing decision-making process. So,
our study suggest that causal maps based recommendation mechanism (CMRM).
Design/methodology/approach: ELM was applied to build hypotheses concerning how consumers’ decision satisfaction and online
shopping behavior are affected by CMRM. Specifically, the performance of the proposed CMRM is analyzed empirically by garnering
the experiment data from 250 qualified respondents who were asked to refer to the proposed CMRM before making purchasing decisions
on mobile phones.
Findings: Statistical results proved that the proposed CMRM could enhance consumers’ decision satisfaction, attitude towards the
recommended products, as well as positive purchase intentions and actual purchase.
Practical implications: CMRM can be easily implemented on the web, allowing target consumers to experience a real recommenda-
tion process. And, a wide variety of qualitative factors that seem crucial to most consumers can be pre-defined through a survey, and
incorporated into causal maps. Thus, such causal maps will improve the personalization effect on the target consumer’s purchase
intentions.
 2007 Elsevier Ltd. All rights reserved.

Keywords: Causal map; Traditional recommendation mechanism (TRM); Elaboration likelihood model (ELM)

1. Introduction 1997). The buying patterns of individual customers and

Recently, information technology has been utilized to
help companies maintain competitive advantage (Nissen
& Sengupta, 2006). Data mining techniques with recom-
mendation systems are a widely used information technol-
ogy for extracting customer’s knowledge and further
supporting marketing decisions (Balabanovic & Shoham,
*
Corresponding author. Tel: +82 16 722 8757; fax: +82 2 760 0440.
E-mail address: kwonsoonjae@naver.com (S. Kwon).
groups can be identified via analyzing customer data
(Maes, Guttman, & Moukas, 1999), but also allows a
company to develop one-to-one marketing strategies that
provide individual marketing decisions for each customer
(Lampel & Mintzberg, 1996; Murthi & Sarkar, 2003).
Recommendation systems are technologies that assist busi-
nesses to implement such strategies, and provide a type of
mass customization that is becoming increasingly popular
on the internet (Ansari, Essegaier, & Kohli, 2000; Lee &
Lee, 2005). They have emerged in e-commerce applications

0957-4174/$ - see front matter  2007 Elsevier Ltd. All rights reserved.
doi:10.1016/j.eswa.2007.08.109
1568 K.C. Lee, S. Kwon / Expert Systems with Applications 35 (2008) 1567–1574
to support product recommendation. The recommendation
systems use customer purchase history to determine prefer-
ences and identify products that a customer may wish
to purchase. Schafer, Konstan, and Riedl (2001) pre-
sented a detailed taxonomy of recommendation systems
in e-commerce, and determined how they can provide
personalization to establish customer loyalty. Generally,
recommendation systems increase the probability of cross-
selling; establish customer loyalty; and fulfill customer
needs by discovering products in which they may be
interested.
The traditional recommendation mechanism (TRM) is a
web-based system designed to help customers sort through
available products and/or services on the online shopping
malls and advise customers about what products to buy,
based on the needs expressed by the customers. Through
a variety of tasks such as defining needs, forming consider-
ation sets, making recommendations, and negotiating pur-
chases (West et al., 1999), the TRM serves to potentially
reduce the cost of thinking (Shugan, 1980), as well as the
uncertainty surrounding an online shopping purchasing
decision, and thus both reduce the difficulty of making a
choice while increasing the confidence associated with it.
Since the TRM is perceived by customers to be highly cred-
ible and to have particular expertise in the decision context,
its positive impact is expected to intensify. However, the
remaining problem concerning TRM is how to analyze
the causal relationships between quantitative and qualita-
tive factors, and investigate their impact on the central
routes and peripheral routes through which both quantita-
tive and qualitative factors can affect customer online shop-
ping decisions. It is well known that qualitative factors are
hard to codify yet they have a significant effect on a cus-
tomer’s decision-making process in the form of causal rela-
tionships with quantitative factors. Thus, a new online
recommendation mechanism is required that incorporates
qualitative factors systematically with quantitative factors
to analyze their combined influence on customers’ purchas-
ing decision-making process.
To accomplish this, the first objective of this study is to
propose an alternative online recommendation mechanism
where causal relationships, possibly existing among quanti-
tative and qualitative factors that seem relevant to custom-
ers’ online shopping decisions. For this purpose, causal
mapping is introduced to analyze causal relationships
among quantitative factors and qualitative factors and help
customers revise prior decisions that were based on quanti-
tative factors alone. Quantitative factors in the case of
mobile phones include such terms as handset price, folder
type, rate of communication charge, brand, etc. In con-
trast, qualitative factors indicate customers’ tacit prefer-
ences towards body color, design, usefulness, etc., all of
which reside mentally and have causal relationships with
quantitative factors. Causal mapping is a widely accepted
methodology for analyzing causal relationships rigorously
(Eden, 1989; Nelson, Nadkarni, Narayanan, & Ghods,
2000). A causal map, therefore, looks at the pattern or
structure of assertion of causal relationships among vari-
ables and not at how or why variables or their causal rela-
tionships come to be part of the map (Ford & Hegarty,
1984). Causal maps have been shown to be relatively stable
and thus capable of providing the basis for prediction
(Bonham & Shapiro, 1976). The first objective of this study
then is to propose a causal map-driven online recommen-
dation mechanism, named CMRM, to (1) analyze causal
relationships among both quantitative factors as well as
qualitative factors, (2) help customers revise their prior
online shopping behaviors based on quantitative factors
alone, and (3) prove its validity statistically with respect
to customer decisions satisfaction and behaviors (i.e., atti-
tude towards recommended products, purchase intention,
actual purchase) in comparison with TRM.
2. Theoretical backgrounds
2.1. Recommendation mechanisms and decision making
On the Internet, recommendation mechanisms are
becoming more important in providing personalized sup-
port to a customer’s decision-making process by reducing
the difficulty associated with choosing between alternatives
(Haubl & Trifts, 2000). And, recommendation mechanisms
have been found to help consumers efficiently filter avail-
able alternatives, increase the quality of their considered
set and increase their product choice satisfaction (Haubl
& Trifts, 2000; Urban et al., 1999). Knowledge of appropri-
ate recommendation mechanisms using customer’s infor-
mation can be used to improve decision quality toward
recommendation products (Ansari et al., 2000; Tam &
Ho, 2005). Customers are typically good at selecting vari-
ables that are relevant in the decision process, but weak
at integrating and retaining large amounts of information.
Effective decision aids (i.e., recommendation systems)
should be designed to capitalize on the strengths and com-
pensate for the inherent weaknesses of their users (Hoch &
Schkade, 1996). So, customers need to examine and assim-
ilate relevant recommendation information in the decision-
making process before choice products which match with
their preference. If a product recommendation from an
information source is available to customers, they can
either decide not to consult it, consult and follow it, or con-
sult and not follow it (Senecal, Kalczynski, & Nantel,
2005). Senecal et al. (2005) investigate how different online
decision-making processes used by consumers, influence
the complexity of their online shopping behavior. So, this
paper shows CMRM effect on consumers’ decision satisfac-
tion and online shopping behavior, and increase actual
purchase.
2.2. Cognitive map
Due to incomplete and uncertain objective information,
as well as lack of sufficient knowledge, experts may find it
difficult to describe their tacit knowledge precisely. Causal
 K.C. Lee, S. Kwon / Expert Systems with Applications 35 (2008) 1567–1574
maps (Eden, Jones, & Sims, 1979; Nelson et al., 2000),
which is widely used in describe expert knowledge of spe-
cial domain, is used here as a tool for solving the problem
of recommendation systems. Also, causal maps can illus-
trate causal relationships among the factors describing a
given object and/or problem, and they can also describe
experts’ tacit knowledge about a certain object (Eden,
1989; Lee & Kwon, 2006; Montazemi & Conrath, 1986;
Nelson et al., 2000). Tacit knowledge is personal knowl-
edge embedded in individual experience and, is shared
and exchanged through direct, face-to-face contact. Tacit
knowledge can be communicated in a direct and effective
way. The proposed causal map technique has been used
to evaluate the factors that affect a given class of decisions,
and, most importantly, to enhance the overall understand-
ing within the decision maker’s environment. For this rea-
son, causal maps are used in our research to analyze and
aide decision-making by investigating causal links among
relevant domain concepts (Eden & Ackermann, 1989;
Klein & Cooper, 1982). The primary concern of causal
maps is to analyze whether the state of one element is per-
ceived to have an influence on the state of the other. In
addition, several researchers proved that causal maps are
a technically and methodologically mature technique to
solve a wide variety of unstructured decision problems.
(Eden et al., 1979; Kwahk & Kim, 1999; Lee & Kwon,
2006; Nelson et al., 2000; Ramaprasad & Poon, 1985; War-
ren, 1995; Zhang, Wang, & King, 1994; Zhang, Chen, &
Bezdek, 1989). Therefore, causal maps can represent
experts’ beliefs and cognition about ill-structured social
relationships (Ford & Hegarty, 1984; Huff, 1990). In the
case of online shopping, it is typical for customers to con-
sider quantitative and qualitative factors simultaneously.
The problem here is how to configure causal relationships
among the factors. Since the causal relationships belong
to tacit knowledge, and using causal maps is well known
as a highly promising technique for capturing tacit knowl-
edge (Lee & Kwon, 2006; Nelson et al., 2000), causal maps
can be used effectively in designing more advanced versions
of online recommendation mechanism.
3. Research model
The proposed CMRM is based on two kinds of fac-
tors—quantitative and qualitative. In this study, the quan-
titative factors indicate product functionality-related
factors, while qualitative factors denote customers’ subjec-
tive characteristics. First, recommendations supported by
the CMRM are organized on the basis of quantitative fac-
tors alone, and then customers are allowed to incorporate
their own qualitative factors into the decision process via
causal mapping, leading to revision of the prior beliefs
about the product candidates. The final product recom-
mendation is made as follows. The CMRM is compared
with TRM with respect to customer decisions as well as
customer behavior, where the customer decision satisfac-
tion, and customer behavior includes customers’ attitudes
1569
towards the recommended product candidates, customers’
purchase intentions and actual purchase. Based on these
assertions, we develop a research model as depicted in
Fig. 1.
Before proceeding with the research model, we need to
set up a theory of how the online recommendation mecha-
nism works to affect customer decisions and behaviors by
stressing that the online recommendation mechanism
belongs to a stimuli-based decision-making environment.
The stimuli take the form of text, images, audio, anima-
tions, or video. With the stimuli, recommendations being
administered by the online recommendation mechanisms
constitute different persuasive efforts to influence customers
(Tam & Ho, 2005). To enhance the recommendation effect,
recent online recommendation mechanisms like collabora-
tive filtering, data mining, and click-stream analysis cus-
tomize their recommendations at the individual level
according to the stimuli (Ansari et al., 2000). To this end,
ELM or the elaboration likelihood model (Petty & Caciop-
po, 1981, 1986) is introduced. Strong persuasion occurs
through the central route, while weak persuasion occurs
through the peripheral route (Petty & Cacioppo, 1986).
Before being persuaded by the online recommendation
mechanisms, customers gather information through two
kinds of routes, such as the central and peripheral route.
In particular, information gathered by the central route is
known to persuade customers more effectively, according
to ELM. In other words, customers tend to elaborate on
information gathered through the central route and allow
it to influence their purchasing decisions and behaviors
more effectively. If the recommended products match with
the customers’ preferences, which are determined compos-
itely by quantitative factors as well as qualitative factors,
then the products are recognized through the customers’
central route, leading to significant decision making (Tam
& Ho, 2005). In literature, it is well known that customers
tend to give more emphasis to the qualitative factors than
the quantitative factors (Haubl & Trifts, 2000; Lee &
Benbasat, 2005). Therefore, the online recommendation
mechanisms’ persuasion effects on customers’ purchasing
decisions and behaviors will be significantly improved as
various forms of qualitative factors can be systematically
considered in the central route of the recommendation
process.
Fig. 1. Research model.
1570 K.C. Lee, S. Kwon / Expert Systems with Applications 35 (2008) 1567–1574
The CMRM’s central route is based on quantitative fac-
tors as well as qualitative factors to create strong persua-
sion. In particular, causal map is used to deal with causal
relationships among the two kinds of factors and help
revise customers’ purchasing decisions and behaviors more
systematically. According to the information processing
model (Bargh, 2002), customers’ information processing
stages include attention, elaboration, and behavior. If com-
plicated information like causal relationships among quan-
titative factors and qualitative factors are processed more
effectively by the CMRM, then a customer’s first response
will be to pay attention to the results, and then elaborate
on them, and finally try to integrate the full information
of the results and revise prior purchasing decisions and
behaviors (Hoch & Schkade, 1996). Such information pro-
cessing was empirically studied in the field of advertise-
ments (MacInnis & Jaworski, 1989).
Besides, customers want to reduce the amount of cogni-
tive efforts associated with decision making, (Payne, 1982;
Shugan, 1980) because they are often unable to evaluate
all available alternatives in great depth prior to making a
decision (Beach, 1993). Recently, Lee and Benbasat
(2005) focused on customers’ negative emotions in the pro-
cess of recommendation due to the conflicts among prod-
uct attributes forcing customers to trade off some
attributes to acquire others (Luce & Bettman, 2001). This
phenomenon will occur especially when processing quanti-
tative factors and qualitative factors simultaneously.
Therefore, the CMRM relies on causal maps to help cus-
tomers compute the influence of the qualitative factors on
their prior quantitative factors-based decisions and behav-
iors in a more systematic and effective way. In summary,
depending on the arguments so far, customers will improve
their decision satisfaction (H1) by using the CMRM, and
such improved decisions will lead to an improvement in
customers’ attitudes towards the recommended products
(H2), purchase intentions (H3) and actual purchase (H4).
• H1: Compared to TRM, CMRM increases customers’
decision satisfaction.
• H2: Compared to TRM, CMRM increases customers’
positive attitude towards the recommended products.
• H3: Compared to TRM, CMRM increases customers’
intention to purchase the recommended products.
• H4: Compared to TRM, CMRM increases customers’
actual purchase the recommended products.
4. Methodology
4.1. Basics
As shown in Fig. 2, CMRM consists of two phases to
recommend products. In phase 1, CMRM computes simi-
larity processing to recommend products that satisfy cus-
tomers’ requirements about product functionality
belonging to quantitative factors. In the case of mobile
Phase 1 Phase 2
Express participants’
Display causal map
requirements for each variable
Participants input preference
Calculate the similarity
using the products’ values composed of
quantitative factors qualitative factors
Display three candidate
Calculate the satisfaction
value using participants’
recommended products qualitative factors
Display revised satisfaction
value about each
Questionnaires of
dependent variable recommended products
Questionnaires of
dependent variable
Fig. 2. Logical flow of CMRM.
phones, the product functionality includes manufacturer,
body type (folder vs sliding), price plan, body color, body
size, LCD size, waiting time, and other functional proper-
ties like MP3 and digital camera, etc. Based on customers’
preferences regarding product functionality, the CMRM
computes similarity for the candidate products, recom-
mending three candidate products.
In phase 2, the CMRM incorporates qualitative factors
to help customers revise a priori judgments about the prod-
ucts reviewed in phase 1. For this purpose, causal map is
organized to depict causal relationships among the quanti-
tative and qualitative factors.
4.2. Building of causal map
To build a causal map of CMRM, we used a four step
process suggested by Nelson et al. (2000).
Step 1: Elicitation of quantitative and qualitative factors
To determine the qualitative factors regarding
mobile phones purchase, an extensive open inter-
view was performed with 20 experts such as 14
marketing and MIS doctoral candidates, 2 mobile
phone dealers, and 4 MIS and marketing research
professors. Our interview used open interview
techniques with probes to facilitate the interview
process (Axelrod, 1976). Respondents were asked
about quantitative and qualitative factors that
seem relevant to the mobile phones purchase deci-
sion process. The interviewers did not constrain
responses to questions. Each interview lasted from
30 to 60 min. After all, seven quantitative factors
and eight qualitative factors were identified
respectively.
Step 2: Constructing CM
The causal statements are first derived from the
interviews with experts. The examples of collected
 K.C. Lee, S. Kwon / Expert Systems with Applications 35 (2008) 1567–1574
causal statements through interview with experts
are as follows; If the price of mobile phone is high,
then its practicality is low. If the perceived satisfac-
tion about mobile phone design increases, then pur-
chase intention is high. Based on the derived causal
statements among the 15 factors identified in step
1, three MIS research professors who were known
to major in causal maps drew draft causal maps
independently. The level of agreement between
the three research professors experts was measured
using Kendall’s coefficient of concordance (Seigel,
1956), which was 0.87 and it was an acceptable
level of agreement. Conceptually relevant concept
nodes were derived from the three raw causal
maps, and regrouped into common words main-
taining logical relevance that would be understood
and interpretable by other experts in the related
fields (Narayanan & Fahey, 1990). Causal rela-
tionships were also refined accordingly in the pro-
cess of concept nodes regrouping. Then all the
results like this were aggregated into a single draft
causal map.
Step 3: Validation of causal map
To prove the validity of the draft causal map
induced in step 2, member check was first applied
(Lincoln & Guba, 1985). This is done by going
back to the original three research professors
and asking their opinion about the concept nodes,
causal linkages represented on the draft causal
map. The process and purpose of this comprehen-
sive check is to test for factual and interpretive
accuracy and to provide evidence of credibility
and trustworthiness similar to internal validity in
confirmatory studies. Final causal map is depicted
in Fig. 3.
Step 4: Inference with causal map
The association or the nature of the causal linkage
between two concept nodes is analyzed using an
adjacency matrix. The adjacency matrix is an
n · n matrix, where n is the total number of con-
cept nodes in a causal map. The elements listed
’
Fig. 3. Final causal map used for recommending mobile phones.
1571
along the horizontal and vertical axes of an adja-
cency matrix will be the set of concept nodes used
in the causal map. Cells at the intersection of each
row and column contain a number indicating the
existence, direction and strength of the causal rela-
tionship between two elements (Smith & Wirth,
1992). However, we use reachability matrix for
inference with a causal map because the reachabil-
ity matrix allows the effect of one element on all
other elements within a causal map, analyzing all
the indirect relationships (Ford & Hegarty, 1984;
Smith & Wirth, 1992). By using the reachability
matrix, the CMRM can analyze the effect of cau-
sal relationships among the quantitative and qual-
itative factors on the recommendation results in a
more in-depth manner.
4.3. Experiments
A 2 · 1 factorial design with a within-subject factor was
used. The within-subject factor, indicating recommenda-
tion types such as TRM and CMRM. For experiments,
we developed a website where CMRM recommends candi-
date mobile phones for users. A competing recommenda-
tion type is TRM where customers are allowed to express
their preferences only about quantitative factors like prod-
uct functionality. Therefore, CMRM is compared with
TRM in four dimensions, including customers’ decision
satisfaction (H1), as well as attitude towards the recom-
mended products (H2), purchase intention (H3) and actual
purchase (H4).
The participants were college students who took one or
two of the five undergraduate electronic commerce courses
offered by the School of Business Administration. For the
purpose of conducting rigorous experiments, the partici-
pants were organized into two groups. The 184 pre-test
responses from the first group were used to compute simi-
larity distance among a number of mobile phone products
under consideration. The 256 responses of the second
group were used to prove research hypotheses. They partic-
ipated on a voluntary basis following the instruction that
bonus points will be given for those who completed surveys
successfully within the time limit. Of the respondents, 136
were male and 120 were female. On average, they were
24.2 years old, and they used the Internet for 21 h/week.
All the participants had previous experience purchasing
some products on the online shopping malls. In particular,
45% had previously purchased mobile phones on the online
shopping malls.
The dependent variables in the research model were mea-
sured using items from established scales. Participants
responded to a series of questions designed to measure their
affective response to the four decision satisfaction items
from Fitzsimons (2000). An established scale using semantic
differential items (Bruner, 1998; Li, Daugherty, & Biocca,
2003) was applied to measure participants’ attitudes toward
1572 K.C. Lee, S. Kwon / Expert Systems with Applications 35 (2008) 1567–1574

the recommended products as well as purchase intentions. Table 1

Actual purchase is measured by frequency of buying prod- Descriptive statistics
uct using of TRM or CMRM. The participants answered Dependent variables Reactance leave
these questions for each of the steps separately. TRM CMRM Total
The CMRM procedure consists of two phases. In phase Decision satisfaction Mean 3.85 4.78 4.32
1, the participants expressed their requirements for each SD 0.98 1.03 1.01
variable. And, CMRM displayed three candidate recom- Attitude towards the Mean 3.69 4.79 4.24
mended products with satisfaction values that are recommended products SD 1.34 1.36 1.35
researched and calculated by similarity processing using
Purchase intentions Mean 3.43 4.57 4.00
the products’ functionality (or quantitative factors). The SD 1.25 1.37 1.31
participants were asked to select one of them or none. In
Actual purchasea Mean 0.75 0.88 0.82
the final of phase 1, participants were then asked to under- SD 0.43 0.33 0.38
take questionnaires. In phase 2, the participants continued a
Actual purchasing = 1, Not purchasing = 0.
with the causal map-driven experimental session and causal
map was displayed to each participant. CMRM redis-
played the three products recommended in the previous The Pearson’s correlation coefficients indicate that the
(second) step. Given this information about the recom- four dependent variables were moderately correlated with
mended products, the participants input preference values each other. Thus, the MANOVA model was applied to test
composed of qualitative factors for each recommended the influence of the online recommendation mechanism
product. Then, the satisfaction value for each product type (TRM vs CMRM) on the four dependent variables.
model, which was calculated in the second step using the The main effect of the online recommendation mechanism
quantitative factors like production functionality, was type was significant (F(4, 252) = 134.7, p = 0.000, p <
revised by the causal map. Based on the revised satisfaction .000). Because the MANOVA results were significant, these
value attached to the recommended products, the partici- results were further analyzed using individual ANOVAs to
pants were allowed to reselect one of the three products examine the effects of the online recommendation mecha-
or none. The participants were then asked to answer the nism type on each dependent variable.
same questionnaire which was used in phase 2.
Table 2
5. Results and discussion Repeated measure ANOVA results
Decision satisfaction DF Mean F-value p
5.1. Results squared
Within-subjects
Using SPSS windows version 12.0, Cronbach’s alpha Online recommendation 1 112.125 339.442 0.000***
was computed to assess the reliability for the four depen- mechanism type
dent variables—customers’ perceived decision satisfaction Error (online recommendation 255 0.330
(0.88), attitudes towards the recommended products mechanism type)
(0.95), purchase intentions (0.93), and actual purchase Attitude towards the DF Mean F-value p
(0.85). Furthermore, the scales for each construct were recommended products squared
tested for internal consistency, indicating that each con- Within-subjects
struct is unidimensional. Since prior research suggests that Online recommendation 1 154.111 359.263 0.000***
experience (Taylor & Todd, 1995) and gender (Gefen & mechanism type
Error (online recommendation 255 0.429 – –
Straub, 1997) influence individuals’ perceptions and use
mechanism type)
of IT, we need to test the influence of experience and gen-
der. Experience in this study includes experience purchas- Purchase intentions DF Mean F-value p
squared
ing products on the online shopping malls, as well as
purchasing mobile products on the online shopping malls. Within-subjects
Online recommendation 1 166.303 361.462 0.000***
We first tested whether demographic differences explained mechanism type
variances in the dependent variables (Suh & Lee, 2005). Error (online recommendation 255 0.460 – –
By applying multivariate regression analysis, we found no mechanism type)
significant covariate effects of gender and experience pur- Actual purchase DF Mean F-value p
chasing products online (F(12, 241) = 1.127, p = 0.456 in Squared
Wilks’ test), and gender and experience purchasing mobile Within-subjects
phones online (F(12, 241) = 0.845, p = 0.657 in Wilks’ Online recommendation 1 1.877 14.671 0.000***
test). Thus, based on the principle of parsimony, we did mechanism type
not include these variables in the subsequent analysis. Error (online recommendation 255 0.128 – –
Table 1 shows the means and standard deviations of the mechanism type)
*p < 0.10, **p < 0.05, ***p < 0.01.
dependent variables.
 K.C. Lee, S. Kwon / Expert Systems with Applications 35 (2008) 1567–1574
Further, a repeated-measure ANOVA was applied to
analyze the effects of the online recommendation mecha-
nism type and reactance level on each dependent variable.
Table 2 shows that the online recommendation mechanism
type significantly affects all the dependent variables with a
99.9% confidence level. For decision satisfaction, CMRM
(mean 4.78) is significantly better than TRM (mean 3.85),
proving H1. In terms of the recommended products, the
participants showed more positive attitudes when using
CMRM (mean 4.79) than using TRM (mean 3.69), proving
H2. Likewise, customers’ purchase intentions for the prod-
ucts recommended by CMRM (mean 4.57) were signifi-
cantly higher than that recommended by TRM (mean
3.43), proving H3. Finally, actual purchase, CMRM (mean
0.88 = 225(actual purchase)/256(total)) is significantly bet-
ter than TRM (mean 0.75–192(actual purchase)/256(total)-
total 256), proving H4. In summary, CMRM surpasses
TRM in terms of the three dependent variables with a
99.9% statistical confidence level.
5.2. Discussion
Based on the statistical results mentioned in section
above, we found the following outcomes.
First, the possible advantages of using causal maps in
online recommendation mechanism were clearly manifested
through rigorous experiments. Since there is no study utiliz-
ing causal maps that incorporates qualitative factors into
the personalized recommendation process, we suggest that
future research builds on the results of the current study.
Second, CMRM was very robust in enhancing consum-
ers’ perceived decision satisfaction and attitudes towards
the recommended products, as well as positive purchase
intentions and actual purchase. The improvement in con-
sumer satisfaction can be attributed to CMRM’s enabling
consumers to systematically analyze the possible influence
of qualitative factors on a priori quantitative factors-based
judgments. Therefore, using CMRM theoretically contrib-
utes to helping consumers elaborate on the recommended
products, proving the validity of ELM in online recom-
mendation mechanism studies once again (Tam & Ho,
2005).
The practical implications of this study are as follows:
First, CMRM can be easily implemented on the web,
allowing target consumers to experience a real recommen-
dation process.
Second, a wide variety of qualitative factors that seem
crucial to most consumers can be pre-defined through a
survey, and incorporated into causal maps. Thus, such cau-
sal maps will improve the personalization effect on the tar-
get consumer’s purchase intentions.
6. Concluding remarks
This study proposed a new online recommendation
mechanism using causal maps to enhance consumers’ online
shopping decisions and behaviors by incorporating qualita-
1573
tive factors systematically into the recommendation pro-
cess. Through the empirical analysis, we found that the
proposed CMRM can enhance the four dependent vari-
ables. Moreover, the results revealed the importance of
explicitly considering the ELM concept in designing the
online recommendation mechanisms-related research
model. However, the usefulness of causal maps should not
be overemphasized when considering causal relationships
among the quantitative and qualitative factors clearly and
inducing revised recommendation results for the consum-
ers. To summarize, the CMRM surpassed TRM on all four
dependent variables.
As an important future research topic, CMRM needs to
be further tested empirically with various types of products
and services. In particular, it seems useful to compare
CMRM effects based on involvement level (high vs low).
References
Ansari, A., Essegaier, S., & Kohli, R. (2000). Internet recommendation
systems. Journal of Marketing Research, 37(August), 363–375.
Axelrod, R. (1976). Structure of decision: The cognitive maps of political
elites. Princeton, NJ: Princeton University Press.
Balabanovic, M., & Shoham, Y. (1997). Content-based, collaborative
recommendation. Communication of the ACM, 40(3), 66–72.
Bargh, J. A. (2002). Losing consciousness: Automatic influences on
consumer judgment, behavior, and motivation. Journal of Consumer
Research, 29(2), 280–285.
Beach, L. R. (1993). Broadening the definition of decision making: The
role of pre-choice screening of options. Psychological science, 4(4),
215–220.
Bonham, G. M., & Shapiro, M. J. (1976). Explanation of the unexpected:
The Syrian intervention in Jordan in 1970. In R. Axelrod (Ed.), Structure
decision (pp. 113–141). Priceton, NJ: Princeton University Press.
Bruner, G. C. (1998). Standardization and justification: Do a scales
measure up? Journal of Current Issues and Research in Advertising,
20(1), 1–18.
Eden, C. (1989). Using cognitive mapping for strategic options develop-
ment and analysis. In J. Rosenhead (Ed.), Rational analysis for a
problematic world (pp. 21–41). Chichester: Wiley.
Eden, C., & Ackermann, F. (1989). Strategic options development and
analysis (soda) – using a computer to help with the management of
strategic vision. In G. Miller (Ed.), Knowledge-based management
support systems (pp. 198–207). UK: Ellis Horwood.
Eden, C., Jones, S., & Sims, D. (1979). Thinking in organizations. London,
England: Macmillian Press Ltd..
Fitzsimons, G. J. (2000). Consumer response to stock outs. Journal of
Consumer Research, 27(2), 249–266.
Ford, J. D., & Hegarty, W. H. (1984). Decision makers’ beliefs about the
causes and effects of structure: An exploratory study. Academy of
Management Journal, 27(2), 271–291.
Gefen, D., & Straub, D. W. (1997). Gender differences in the perception
and use of e-mail: An extension to the technology acceptance model.
MIS Quarterly, 21(4), 389–400.
Haubl, G., & Trifts, V. (2000). Consumer decision making in online
shopping environments: The effects of interactive decision aids.
Marketing Science, 19(1), 4–21.
Hoch, S. J., & Schkade, D. A. (1996). A psychological approach to
decision support systems. Management Science, 42(10), 51–64.
Huff, A. S. (1990). Mapping strategic thought. Chichester, UK: John Wiley
& Sons.
Klein, J. H., & Cooper, D. F. (1982). Cognitive maps of decision-makers
in a complex game. Journal of the Operational Research Society, 33(1),
63–71.
1574 K.C. Lee, S. Kwon / Expert Systems with Applications 35 (2008) 1567–1574
Kwahk, K. Y., & Kim, Y. G. (1999). Supporting business process redesign
using cognitive maps. Decision Support Systems, 25(2), 155–178.
Lampel, J., & Mintzberg, H. (1996). Customizing customization. Sloan
Management Review, 38(1), 21–30.
Lee, Y.E., & Benbasat, I., (2005). The influence of effort, accuracy, and
negative emotions on product choice-strategies: Evaluations of rec-
ommendation agents on desktops versus handheld devices. In Pro-
ceedings of the eleventh Americas conference on information systems
(pp. 2325–2329).
Lee, K. C., & Kwon, S. J. (2006). The use of cognitive maps and case-
based reasoning for B2B negotiation. Journal of Management Infor-
mation Systems, 22(4), 337–376.
Lee, H. J., & Lee, J. K. (2005). An effective customization procedure with
configurable standard models. Decision Support Systems, 41(1),
262–278.
Li, H., Daugherty, T., & Biocca, F. (2003). The role of virtual experience
in consumer learning. Journal of Consumer Psychology, 13(4), 395–408.
Lincoln, L. S., & Guba, E. G. (1985). Naturalistic inquiry. Thousand Oaks,
CA: Sage Publications.
Luce, M., & Bettman, J. R. (2001). In J. W. Payne (Ed.), Emotional
decisions: Tradeoff difficulty and coping in consumer choice. Chicago:
University of Chicago Press.
MacInnis, D. J., & Jaworski, B. J. (1989). Information processing from
advertisements: Toward an integrative framework. Journal of Market-
ing, 53(4), 1–23.
Maes, P., Guttman, R. H., & Moukas, A. G. (1999). Agents that buy and
sell: Transforming commerce as We know it. Communications of the
ACM(March), 81–91.
Montazemi, A. R., & Conrath, D. W. (1986). The use of cognitive
mapping for information requirements analysis. MIS Quarterly, 10(1),
45–56.
Murthi, B. P. S., & Sarkar, S. (2003). The role of the management sciences
in research on personalization. Management Science, 49(10),
1344–1362.
Narayanan, V. K., & Fahey, L. (1990). Evolution of revealed causal maps
during decline: A case study of admiral. In A. S. Huff (Ed.), Mapping
strategic thought (pp. 107–131). Chichester, UK: John Wiley & Sons.
Nelson, K. M., Nadkarni, S., Narayanan, V. K., & Ghods, M. (2000).
Understanding software operations support expertise: A revealed
causal mapping approach. MIS Quarterly, 24(3), 475–507.
Nissen, E. M., & Sengupta, K. (2006). Incorporating software agents into
supply chain: Experimental investigation with a procurement task.
MIS Quarterly, 30(1), 145–166.
Payne, J. W. (1982). Contingent decision behavior. Psychological Bulletin,
92, 382–402.
Petty, R. E., & Cacioppo, J. T. (1981). Attitudes and persuasion: Classic
and contemporary approaches. Dubuque, IA: W.C. Brown.
Petty, R. E., & Cacioppo, J. T. (1986). The elaboration likelihood model
of persuasion. Advanced Experiment Social Psychology, 19, 123–
205.
Ramaprasad, A., & Poon, E. A. (1985). Computerized interactive
technique for mapping influence diagrams (MIND). Strategic Man-
agement Journal, 6(4), 377–392.
Schafer, J. B., Konstan, J. A., & Riedl, J. (2001). E-commerce
recommendation applications. Journal of Data Mining and Knowledge
Discovery, 5(1–2), 115–152.
Seigel, S. (1956). Nonparametric statistics for the behavioral sciences. New
York: McGraw-Hill.
Senecal, S., Kalczynski, P. J., & Nantel, J. (2005). Consumers’ decision-
making process and their online shopping behavior: A click stream
analysis. Journal of Business Research, 58, 1599–1608.
Shugan, S. M. (1980). The cost of thinking. Journal of Consumer Research,
7, 99–111.
Smith, K. L., & Wirth, A. (1992). Measuring differences between cognitive
maps. Journal of the Operational Research Society, 43(12), 1135–
1150.
Suh, K. S., & Lee, Y. E. (2005). The effects of virtual reality on consumer
learning: An empirical investigation. MIS Quarterly, 29(4), 673–697.
Tam, K. Y., & Ho, S. Y. (2005). Web personalization as a persuasion
strategy: An elaboration likelihood model perspective. Information
Systems Research, 16(3), 271–291.
Taylor, S., & Todd, P. (1995). Assessing IT usage: The role of prior
experience. MIS Quarterly, 19(4), 561–570.
Warren, K. (1995). Exploring competitive futures using cognitive map-
ping. Long Range Planning, 28(5), 10–21.
West, P. M., Ariely, D., Bellman, S., Bradlow, E., Huber, J., Johnson, E.,
et al. (1999). Agents to the rescue? Marketing Letters, 10(3), 258–
301.
Zhang, W. R., Chen, S. S., & Bezdek, J. C. (1989). Pool2: A generic system
for cognitive map development and decision analysis. IEEE Transac-
tions on Systems, Man and Cybernetics, 19(1), 31–39.
Zhang, W. R., Wang, W., & King, R. S. (1994). A-Pool: An agent-
oriented open system shell for distributed decision process modeling.
Journal of Organizational Computing, 4(2), 127–154.