The author adopted a 2 (source of WOM: friend and stranger) x2 (social
comparison: upward comparison and downward comparison) x2 (self -assessed skill:
high and low) between-subject factorial experiment to test our hypotheses. The
dependent variable of interest was purchase intention.
The whole experiment process was divided into two parts. And the sample was
from National Chiao Tung University (five classes) and Chung Hua University (four
classes). The first part of our experiment is to measure the participants’ self-assessed
skill about craft in the first week. After two weeks later, the other researchers will
appear in the class and give the same participants a formal questionnaire. After
collecting the second data, the experiment can be considered finished. By doing so,
we can make sure the participants’ skill level would not be influenced by our
manipulation of social comparison in scenario. Two weeks of time gap can obtain the
participant’s self skill level more precisely.
3.4 Manipulation of source of WOM
According to previous research, the author followed the method of closeness
relationship’s manipulation that most be used: divided the source of WOM into two
groups: close and not close. We uses “friend” as close relationship group, and
“stranger” as not close relationship group.
19
3.5 Manipulation of social comparison and Stimulus
In this study, although we interest in the effects of recommendation in such a
mass customization market, the WOM receiver still need a possible image in their
minds as a cue to participate this DIY task more possibly, that is here we have to
choose two pictures to meet our needs for both social comparison and try to get closer
to a more realistic situation. So, under this consideration, we decide to use two
handmade photo frame pictures as upward social comparison and downward social
comparison stimulus. We conduct an online pretest to pick out the two photo frame
pictures. Two criteria were considered to choose the suitable photo frame pictures: the
first, they must be perceived identical attractive; second, they must be critically
different on perceived difficulty. Following the above two rules, we select twenty
handmade photo frame pictures from “www.google.com.tw” randomly. Then, we
conduct two surveys of attractiveness and difficulty on Internet respectively. That
means a person will not judge one picture’s attractiveness and also give a difficulty
grade simultaneously.
In attractiveness survey, the author additionally analyzes the effects of gender.
From a statistic result, we found that there are four photo frame pictures that have a
significant difference between male and female (p<.05). So, we dropped these four
pictures. By doing so, we can avoid the risks of getting biases from gender difference
20
in our formal study.
After collecting the remained data, we use SPSS (paired-sample t-test) to find all
the pairs that have the same attractiveness. Then, we also use the same method
(paired-sample t-test) to analyze difficulty data. To meet above two criteria, we finally
decided a pair of photo frame pictures that have the same attractiveness (n.s. p=.759)
and were perceived significantly different difficulty (p>.000).
Eventually, two stimulus of upward and downward comparison pictures were be
confirmed in this study (see Appendix Ⅰ). Table1 summarized the information about
the two pretests.
Table 1 Results of stimulus in pretest
sample gender p-value
Attractiveness
21
3.6 Measurements
In this study, the constructs used were measured by multiple items. All items
were verified on seven-points Likert-type scale (1=strongly disagree; 7=strongly
agree).
3.6.1 Measures of Independent Variables - Source of WOM
There are two items used to measure the source of WOM. They were: “Before today, I have known this people for a few days.” and “I don’t know this person who
just talked to me.”
3.6.2 Measures of Independent Variables - Social Comparison
The author used two items to measure two levels of social comparison. They
were: “For me, it is very hard to make and design this photo frame,” and “This
word-of-mouth sender is indeed having a better artistic talent than me.”
3.6.3 Measures of Independent Variables - Self-assessed Skill
In this study, the author believed the subjective craft ability is based on
individual’s prior knowledge. To measure the skill level of each participant, the author
used a four-item scale of self-assessed ability from Yi - Chaing Huang (2007), and in
22
order to fit our study much precisely, we modified these four item slightly. These
items were more suitable for handicraft context. Finally, they were: “Compared to
other people, I think my ability about handcrafts is excellent.”，“I am confident of
choosing a beautiful handicraft”，“I think I can make a handicraft, which satisfies me”
and “As to pick out a beautiful handicraft product, I am very experienced.”
3.6.4 Measures of Dependent Variable - Purchase Intention
In previous research by Dodds, Monroe and Grewal (1991), the purchase
intention could be measured in three modified items: They were “The possibility of
paying for self-designing a DIY photo frame is high.” ,“If I were going to buy a photo
frame, I would consider paying for self-designing a DIY photo frame” ,“My
willingness to self-design a DIY photo frame is high.”
3.6.5 Realistic Check Items
In scenario’s reliability, we use two items form Shu -Han Chang (2009). They
are ”The story reflects what might happen in the real world” and “I had no difficulty
imagining myself in this situation”.
23
Chapter 4：Results
In this chapter, it includes the analysis of data and qualified samples’
background, manipulation check, and the reliability of the results. Through 273
effective samples, we got a reality score of average 5.04 (above 4) by using 7-points
Liket scales. Since the participants thought the scenario we designed could happen in
a real world, the following analysis and discussions were to be considered meaningful.
In this study, some statistic methodologies, such as ANOVA, Independent-Sample T
Test were being used to test our hypothesis. Statistic software SPSS that broadly used
in marketing as a tool is also being used in this study, edition is 17.00.
4.1 Background of participants
After collecting the second time questionnaires, there are total 273 samples in this study consequently qualified. All the samples are students, 62.6% are male, age
between 16 to 20 years old are 50.9%, income from NT.4,001 to NT.6000 are 30.4%
and 91.9% samples have educational level of college degree.
24
Table 2 Experience of Participants
Frequency Percentage
I am or I have ever studied design
courses.
No 240 87.9%
Yes 32 12.1%
Total 273 100%
Table 3 Experience of Participants
Frequency Percentage
I ever have related design
experiences.
No 159 58.2%
Yes 114 41.8%
Total 273 100%
Table 4 Demographics of Participants
Demographics Category Frequency Percentage
Gender Female 102 37.4%
Male 171 62.6%
Total 272 100%
Age 16-20 139 50.9%
21-25 129 47.3%
25
26-30 4 1.5%
31-35 1 0.3%
Total 273 100.0%
Income Less than 4,000 72 26.4%
4,001-6,000 83 30.4%
6,001-8,000 58 21.2%
8,001-10,000 40 14.7%
10,001-30,000 17 6.2%
30,001-50,000 2 0.7%
50,001-70,000 1 0.4%
Total 273 100%
Education degree College 251 91.9%
Graduate upward 22 8.1%
Total 273 100.0%
Occupation students 273 100%
Total 273 100%
26
4.2 Reliability
We test the reliability of each factor needed to ensure all items we used for the
same factor have internal consistency. Cronbach’s alpha is mostly used to test the
reliability, as long as the alpha value is above 0.7, it represents a good reliability of
these items in this factor. The reliability of closeness is 0.763, and comparison is
0.753.
Cronbach’s alpha value of items to assess subjects’ skill level is 0.755, and 0.875 for
reality of the scenario, purchase intention as the dependent variable in our study also
have 0.852. All Cronbach’s alpha values in this study are all above 0.7. It means this
study is reliable.
Table 5 Reliability Statistics
Construct Cronbach’s Alpha N of Items
Source of WOM .763 2
Comparison .753 2
Self-assessed Skill .755 4
Reality .875 2
Purchase Intention(DV) .852 3
27
4.3 Manipulation Check
4.3.1 Manipulation Check of Source of WOM
There are 129 and 144 samples in close relationship (friend) and not close
relationship respectively. We use Independent-Sample T Test to do the source of
WOM manipulation check. From the statistic result presented in Table 6, it shows
significantly difference between friend group and stranger group (p<.000).
Manipulation check is successful.
Table 6 Manipulation Check of Closeness
Source of
4.3.2 Manipulation Check of Comparison
There are 132 samples in upward comparison group and 141 samples in downward comparison group. Also, by Independent-Sample T Test, we again have the
result of significant difference between upward and downward comparison (p<.000).
Thus, the manipulation is also successful.
28
Table 7 Manipulation Check of Comparison
Comparison N Mean
Std.
Deviation
T
Sig.
(two-tailed)
Upward 132 3.8371 1.1892 -5.145 .000
Downward 141 4.6099 1.2860
4.3.3 Self-assessed skill
The average mean of all participants is 4.52162. We remained the number to five
dots to separate participants to two groups more precisely. By doing so, we can avoid
discarding too many effective samples. If the average mean of four skill items is
higher than 4.51262, we assigned this group as high skill people. Also, average grades
lower than 4.51262 will be considered as low skill group. High skill group mean is
5.42, and low skill group mean is 3.76.
29
4.4 Hypothesis Testing
4.4.1 Interaction between Source of WOM and Social Comparison across Different Skill Level
The author use ANOVA to examine the interaction between source of word of mouth and social comparison across different self-assessed skill. For H2a, it
suggested that for low skill customers, when an upward comparison occurred, the
closer the WOM sender, the lower the purchase intention is. H2b indicated that, for
low skill customers, when downward comparison occurred, the closer the WOM
sender, the higher the purchase intention is. As stated in H2c, we predict the purchase
intention will higher in close WOM sender than not close WOM sender, regardless of
comparison direction. In order to test the effects, a 2 (friend; stranger) x 2 (upward
comparison; downward comparison) x 2 (high skill; low skill) three way
between-subjects ANOVA was adopted to analyze the interaction.
Table 8 shows the results of three-way ANOVA, and we can find there is no
significant three-way interaction among source of word of mouth, social comparison
and self-assessed skill level (F=.237, p=.627). However, we found a main effects of
self-assessed skill (F=6.966, p<.05), so H1 is supported. To test H2a, we found no
significant effect between close and not close word of mouth sender in upward
comparison for low skill person. H2b were also not supported. To test H2c for high
30
skill people, we use a two way ANOVA and the results showed there is no significant
effect. (F=2.29, p=.133). Thus, H2c is not supported. Besides, there are two
additionally significant interactions occurred that we didn’t predict in the first place.
The two significant interactions will be further discussed in next chapter. And some
descriptive statistics were showed in Table 9.
Table 8 Three-way ANOVA of Purchase intention Source TypeⅢ Sum of Squares df Mean
Corrected Total 363.028 272
a. R Squared = .062 (Adjusted R Squared = .038)
31
Table 9 Descriptive Statistic of Purchase Intention
Self-assessed
Skill Comparison
Friend Stranger
Means Std.
Deviation N Means Std.
Deviation N
Low Up 4.66 1.075 33 4.30 1.287 36
Down 4.82 1.120 38 4.73 1.086 41 High Up 5.01 1.136 35 5.32 1.108 37 Down 4.67 1.163 23 4.50 1.083 30
32
Chapter 5：Discussion
5.1 Discussion
From the statistic results, we found there is no significant three-way
interaction among source of word of mouth, social comparison and self-assessed skill
level. However, the author uses two way ANOVA to analyze the data in upward
comparison condition (shown in Table10), and we find a significant effect between
skill and source of WOM. Figure 3 reveals that the purchase intention of low skill
people will be higher when the word of mouth information is from a friend than from
a stranger in an upward comparison condition, which suggests that low skill people
feel less threatened to a friend than to a stranger in an upward comparison. This result
is opposite to our original hypothesis, and the author thought this is a chance for
customization firms to strive. In contrast to high skill people, low skill people reveal
lower purchase intension when WOM is from a high skill friend than from a high skill
stranger. In view of this, some important implications for customization firms will be
listed in next section. On the other hand, although the purchase intention of high skill
people decreases when the word of mouth information is from a friend than from a
stranger in an upward comparison, high skill people show higher purchase intention
than low skill people in average (Figure 3).
Besides, we still have two 2X2 significant interactions as Chapter four shows
33
can be discussed as follows. The first interaction is between self-assessed skill and
comparison (F= 5.248, p=.023, Figure 4). From the Figure 4, we can see the purchase
intention of high skill people is higher in the upward comparison than in the
downward comparison. On the other hand, for low skill people, purchase intention is
higher in the downward comparison than in the upward comparison. Another
significant interaction is between self-assessed skill and source of WOM (F= 3.636,
p=.058, Figure 5). Figure 5 reveals that the purchase intention of high skill people is
higher when the WOM information is from a stranger than from a friend. For low skill
people, the purchase intention is higher when the WOM information is from a friend
than from a stranger.
From the two interactions, we can find some interesting results that are
different from our views. In the first interaction between self-assessed skill and
comparison, it reveals that in an upward comparison, high skill people have more
courage or other motives to drive them to have a higher purchase intention than in a
downward comparison. It may be due to the fact that upward comparison serves as
motivation to design a better product, yet, this speculation is to be examined in future
study.. As the low skill people, the result is consistent with our prediction.
The second interaction between self-assessed skill and source of WOM also
shows an interesting result. It shows that high skill people have more willingness to
34
try a customization product when they receive a recommendation from a stranger than
from a friend. This result is inconsistent with findings in previous literature, and
worth for further research. The fact that a stranger’s WOM information has more
significant effect than a friend’s WOM to attract high skill people to buy a
customization product may also have some important implications to firms. On the
contrary, the purchase intention of low skill people will be higher from friend’s
recommendation than from a stranger is also consistent with our prediction.
Table 10 Results of Two-way ANOVA for Upward Comparison Condition
Source TypeⅢ Sum of
Squares df Mean
Square F Sig.
Corrected Model 21.086a 3 7.029 5.261 .002
Intercept 3269.108 1 3269.108 2446.788 .000
Self-Assessed Skill 16.558 1 16.558 12.393 .001
Source of WOM .026 1 .026 .020 .889
Self-Assessed Skill *
Source of WOM 3.903 1 3.903 2.921 .090
Error 183.043 137 1.336
Total 3486.778 141
Corrected Total 204.129 140
a. R Squared = .103 (Adjusted R Squared = .084)
35
Figure 3 Interaction between Source of WOM and Self-Assessed Skill in Upward Comparison
36
Figure 4 Interaction between Self-Assessed Skill and Comparison
Figure5 Interaction between Self-Assessed Skill and Source of WOM
37
5.2 Implication
Although our original prediction was not supported, we can still see interesting
results fortunately. For example, for low skill customers, we think it is hard to push
them to buy a self-design consumption experience in prediction, because of lack of
intrinsic motivation and threats from close friend’s upward comparison. However, we
got totally different conclusion from this prediction. Low skill people will have higher
purchase intention when the word of mouth information is from a friend than from a
stranger in an upward comparison. This means the word of mouth effects on low skill
person is more effective when it is from a high skill friend than from a high skill
stranger. This may due to that low skill feel less threatened in an upward comparison
with a friend than with a stranger. On the other hand, for high skill person, although
they will feel threatened from a friend in upward comparison, their purchase intention
is still averagely higher than low skill people. The author thought the inner motivation
of high skill people drives themselves to have higher purchase intention when facing
an upward comparison.
From above discussions, the author think high skill level is important in the
customization market. We believe more experiences can help increasing skill, as long
as enough experiences accumulated, low skill person may possibly become high skill
person and have higher purchase intention to a customization product. By doing some
38
activities to help accumulate customer’s experience and lift their skill level is
important for firms. We suggest customization firms to do some promotions, such as
low-price promotion, to attract low-skill customers. In addition, firms can provide
some training program and instructions to low-skill customers when they design and
make a customization product. Help customers to have high skill must be one of the
important things in mass customization market.
5.3 Limitation and Future Research
We can draw some implications from results in this study, but there are still some
limitations. First, as we mentioned earlier in chapter three, in order to avoid
participants to be affected by our experiment scenario (manipulation of upward
comparison or downward comparison), we measured their self-assessed skill points
before the formal questionnaire. So, two times qualified samples are much fewer than
we imagined. Some people may only have filled out only one time questionnaire, such
kind of data became useless. One cell of our study is only 23 samples, so the result
may be unstable.
Second, there are much more male sample than female (male=171vs.
female=102). This causes problems for equal preference of stimulus picture in male
and female people. There is a significant difference between gender (t=2.065, p<.04).
However, we also find another significant difference between easy and hard photos
39
(t=-4.484, p<.000). We thought this is because the avoidless noise from scenario they
just involved, or the critical curiosity to peek at other people’s picture. Nevertheless,
we discovered an interaction between two picture and gender (F=5.838, p<.016), from
this two-way ANOVA result, we see the main effect of gender has disappeared
(F=2.18, p=.141), but the main effects of two photos’ appealing still remains (F=22.77,
P<.000). The problem of photo mentioned above is not showing in our pretest.
Another possibility is the sample number of pretest is not enough. It should be taken
into account in future research.
The third is the difficulty of designing an appropriate scenario. We tried to design
a more interactive scenario for participant to show a recommended behavior, such as
animation, but this may cause more distractions, for example, the actor’s gender and
or their looks.
The fourth limitation is that people have different levels of involvement to this
product. When people received a recommendation of a product, the level of
involvement in themselves may affect the intention to buy or consume this product. In
future research, involvement can be measured or be controlled to ensure each
participant was affected only by manipulation.
Thus we know the result is more consist with SEM model, but we think there
still exists other possible outcomes about the effects of word of mouth in mass
40
customization market. Other different and more suitable scenario or manipulations
can be reconsidered in the future. The effect of word of mouth to such a mass
customization area is interesting.
41
Reference
Argo, J. J., K. White, et al. (2006). Social Comparison Theory and Deception in the Interpersonal Exchange of Consumption Information. Journal of Consumer Research, 33(1), 99-108.
Arnaud De Bruyn, G. L. L. (2008). A multi-stage model of word-of-mouth influence through viral marketing. International Journal of Research in Marketing, 25(3), 151-163.
Bansal and Voyer (2000). World-of-mouth processes within a services purchase decision context. Journal of Service Research, 3(2), 166-177.
Berry, L. L. (1980). Services marketing is different. Business, 30(May/June), 24-29.
Brown, J. J. and P. H. Reingen (1987). Social Ties and Word-of-Mouth Referral Behavior. Journal of Consumer Research, 14(3), 350-362.
Chang, S. H. (2009). The impact of Customer Participation on Attribution of Service Outcomes: The Moderating Role of Self-Efficacy."
Darren, W. D. and C. P. Moreau (2007). Thinking Inside the Box: Why Consumers Enjoy Constrained Creative Experiences. Journal of Marketing Research, 44(3), 357-369.
Deng, X., J. W. Hutchinson. (2009). Aesthetic self-design: Just do it yourself. Working
Deng, X., J. W. Hutchinson. (2009). Aesthetic self-design: Just do it yourself. Working