156
reviews that are presented on the website,” ‘‘When I buy a
product online, the reviews presented on the website are
helpful for my decision making,” ‘‘When I buy a product
online, the reviews presented on the website make me con-
fident in purchasing the product,” and ‘‘If I don’t read the
reviews presented on the website when I buy a product
online, I worry about my decision”).
4. Research results
Two hundred and fifty undergraduate and graduate stu-
dents participated voluntarily. Their average age was 26.58
years, and 51.2% were male. The average frequency of
online purchases per month was 0.89 times. There are no
significant differences in gender (F(9,240) = 0.766,
p< 0.648), age (F(9,240) = 0.862, p < 0.560), and frequen-
cies of online purchase (F(9,240) = 0.896, p < 0.530), indi-
cating that the random assignment was successful.
Subjects were classified as either experts or novices
according to their prior knowledge dichotomized into high
and low levels. We sorted the participants based on the
number of correct responses to the 12 questions. For the
participants with the same score, we sorted them again
ordered by the subjective knowledge score. Afterwards,
we performed the median-split to divide the participants
into two groups – expert and novice. For each group, there
are significant differences for both the objective knowledge
score (M
expert
=8.68 vs. M
novice
=3.77, F(1,248) =
257.656, p < 0.001) and the subjective knowledge score
(M
expert
=5.69 vs. M
novice
=2.32, F(1,248) = 720.492,
p< 0.001). Our focus is to classify the participants as either
being the group with a relatively higher level of expertise or
the group with a relatively lower level of expertise. The
method used in this experiment is common for manipulat-
ing consumer knowledge in marketing literature[4,36]. In
this experiment, 125 subjects were experts and 125 were
novices.
Since two items to measure the perceived number of
reviews were loaded on a single factor (Cronbach’s
a= 0.952), the average of the items was used to check
whether the number of reviews was manipulated as we
intended. An ANOVA analysis indicated the presence of
the main effect of the number of reviews (M
2-review(control)
=
2.85,M
4-review
=3.89, vs. M
8-review
=4.89, F(2,247) = 198.334,
p< 0.001), indicating that the number of reviews was
manipulated as we intended.
Control variables including the perception of review
positiveness, and the general attitude toward reviews were
analyzed to see if there were significant differences among
groups. No significant difference was shown in the percep-
tion of review positiveness (F(9,240) = 0.205, p < 0.993)
and
the
general
attitude
toward
reviews
(F(9,240) = 0.036, p < 0.999). Thus, these control variables
were excluded in the following analysis.
To test hypotheses 1 (Cognitive review fit hypothesis on
experts) and 2 (Cognitive review fit hypothesis on novices),
subjects’ responses relevant to the type of review informa-
tion were examined. MANOVA was performed to check
the effects of the types of review information and the levels
of expertise on the three dependent variables: informative-
ness, usefulness, and helpfulness. The results showed that
there were significant main effects of the type of review
information (Wilk’s k = 0.384, p < 0.001) and expertise
(Wilk’s k = 0.954, p< 0.001). In addition, there was a
significant interaction effect between the type of review
information and expertise (Wilk’s k = 0.527, p < 0.001).
This interaction was significant for all the dependent
variables including informativeness (F(2,244) = 57.708,
p< 0.001), usefulness (F(2,244) = 59.682, p < 0.001), and
helpfulness (F(2,244) = 63.027, p < 0.001). Planned contrasts
showed significant differences between experts and novices.
For experts, reviews framed as being attribute-centric were
viewed as being more informative (M
attribute-centric
=5.24
vs. M
benefit-centric
=3.24, F(1,244) = 179.31, p< 0.001), useful
(M
attribute-centric
=4.96 vs. M
benefit-centric
=2.94, F(1,244)
=127.23, p < 0.001), and helpful (M
attribute-centric
=5.06
vs. M
benefit-centric
=2.90, F(1,244) = 167.12, p < 0.001) than
reviews framed as being benefit-centric. By contrast, novices
stated reviews framed as being benefit-centric were more
informative (M
benefit-centric
=5.24 vs. M
attribute-centric
=3.24, F(1,244) = 44.83, p < 0.001), useful (M
benefit-centric
=4.96 vs. M
attribute-centric
=3.90, F(1,244) =
35.03,
p< 0.001), and helpful (M
benefit-centric
=5.00 vs. M
attribute-
centric
=4.02, F(1,244) = 34.40, p < 0.001) than reviews
framed as being attribute-centric. For overall evaluation
reviews, the level of expertise has no significant effect on
the perceived informativeness (F(1,244) = 0.04, p < 0.85),
usefulness (F(1,244) = 0.62, p < 0.43), and helpfulness
(F(1,244) = 0.01, p < 0.99). The results show that experts
seek attribute information because they want to use their
prior knowledge to infer product benefits from the stated
attributes. Benefit information does not permit such infer-
ence. By contrast, novices prefer the benefits only messages
because the specification of product benefits facilitates
understanding of the given reviews. Interestingly, we found
that the difference between attribute-centric and benefit-
centric messages in terms of message preference was greater
for experts than for novices. It is consistent with a previous
study saying that experts have a clear preference structure
rather than novices [36]. Hence, hypotheses 1 (Cognitive
review fit hypothesis on experts) and 2 (Cognitive review
fit hypothesis on novices) are accepted.
Since factor analysis indicated that the two items mea-
suring purchase intention were loaded on a single factor
(eigen-value = 1.936, Cronbach’s a= 0.966), the two items
were averaged to compose a purchase intention score. The
mean and standard deviation are inTable4.
To test hypotheses 3 (Review fit hypothesis on experts’
purchase intention) and 4 (Review fit hypothesis on nov-
ices’ purchase intention), an ANOVA was performed.
The two-way interaction effect between review type and
expertise was significant as shown inTable5. The relation-
ship is shown inFig.1. For experts, purchase intention was
significantly higher in the benefit-centric condition than in
D.-H. Park, S. Kim/Electronic Commerce Research and Applications 7 (2008) 399–410
405