3.2 Questionnaire Generation
3.2.2 Pretest
Eight graduate students from NCTU Institute of Business and Management were invited to participate in this pretest. Most of the participants suggested that there were too many similar questions in section 1; as a result, we eliminated three questions from original list. Of course, the two religion relative items have drawn out before the pretest. Unfortunately, DSI scale demonstrated that some factor loadings of six items were not significant at 0.05 level even if the reliability was acceptable (coefficient alpha=0.6238). After discussing with professor and 5 participants, #2 and #4 of DSI scale were modified to fit for original meaning properly.
32
3.3 Obtainment of physical attribute of respondents
In general, it is difficult to obtain private physical attribute from public. One is that public unwilling to provide physical parameter because they thought that personal physical attribute will invade their privacy and probably used as illegal activities.
Another one is that analysis of physical attribute needs precision instruments and complicated processing problem. This research cooperated with Z.F. SPECTRUM TECHNOLOGIES INC., which is a company used precision instruments to acquire and analyze personal physical attributes, such as fingerprints and hair.
The report can provide a lot of physical parameters; however, we chose loop number of fingerprint to provide personal physical information due to: (1) the classified model of fingerprints are still undefined, besides, categorical data can’t be used as qualitative research, such as the shape of fingerprints (2) although hair provided more personal physical information rather than fingerprints, what parameters we should chose from hair is another problem.
3.4 Research Process
3.5 Statistic Methods
This section will introduce some statistic methods that will be used in this study.
First of all, Descriptive Statistics is a statistic technique to summary general information of variables, such as mean, variance, distribution, normality and so on. It provides researchers rough image about interested question. Graph, table, or figure is necessary to summarize and present aggregate data. Though descriptive statistics can’t provide more detail information, it’s good instrument to help researcher understand whole situation preliminary.
Factor analysis is a technique which combines lots of similar variables into each construct. Several similar variables were replaced by one factor, that is, each variable is considered as a dependent variable that is a function of some unobserved, underlying set of factors. Thus, factor analysis implies fewer factors and summarizes most of the measured information in data set.
Cluster analysis is a well-known instrument for market segmentation research;
the primary objective of cluster analysis is to classify observations into identified group by their characteristics. In general, cluster analysis usually can be divided into two major procedures: hierarchical and nonhierarchical. Though there is no absolute answer when which procedure should be choose, we use the K-means method of nonhierarchical procedure to analyze the data due to large sample (N=271).
Analysis of Variance (ANOVA) is a famous technique used to compare the means of several populations on a single measured variable. In this study we used ANOVA to examine if any variation exists in consumer innovativeness, by lifestyle
34
based typology, physical attribute based classification, and buying behavioral patterns.
Although two-way or more high-level ANOVA can be utilized, we abandon these procedures because that’s not the interest of this research.
In categorical data analysis, test of homogeneity could demonstrate whether existing significant relationship on specific variable or not among categories. Most of the time, researcher would show the contingency table and profile the difference if statistic test was significant. On the other hand, Chi-square test is the most common tool which was used to examine between two separate classifications. This study attempted to do exploratory research on this barren field; as a result, we utilized a lot of Chi-square test to identify the potential relationship.
Chapter 4 DATA ANALYSIS
The data were analyzed by software SAS 9.0. Section 4.1 demonstrated basic sample information, that is, demographic distribution and make-up. In section 4.2, the validity and reliability of questionnaire could be qualified. Section 4.3 produced the latent factors and factor scores of 30 items of lifestyle by factor analysis first, and then nonhierarchical procedure of cluster analysis been applied to form consumer typologies. We validated if existing significant relationship between lifestyle-based consumer typologies and physical-attribute-based classification in section 4.4.
Ultimately, section 4.5 examined the difference of consumer innovativeness among separate consumer typologies based on lifestyle, physical-attribute-based classification, and DC buying behavior.
4.1 Descriptive Statistics
There were totally 271 recovery questionnaires till the end of investigation.
Data consist of 267 available samples, two incomplete questionnaires, and two lost data of fingerprints of subjects. Descriptive Statistics were presented below, but the dropped sample won’t be included.
Table 4-1 Demographic profile of respondents
Demographics Items Frequency Percent
Sex
Male 92 34.46 %
Female 175 65.54 %
Age
<20 yrs 8 3.00 %
21-25 yrs 21 7.87 %
26-30 yrs 54 20.22 %
36 6 yrs < Youngest child <18
yrs
55 20.68 %
Youngest child > 18 yrs,
but dependent 18 6.77 %
All children are independent 16 6.02 %
Location
Occupation
Student 10 3.76 %
Public servant 44 16.54 %
Housewife 11 4.14 %
High-Tech industry 9 3.38 %
Business 35 13.16 %
Service industry 105 39.47 %
Manufacturing 12 4.51 %
Free 12 5.64 %
Others 25 9.40 %
Monthly Income
< NT$20,000 30 11.32 %
NT$ 20,001 – NT$ 35,000 99 37.36 % NT$ 35,001 – NT$ 50,000 67 25.28 % NT$ 50,001 – NT$ 100,000 59 22.26 %
>NT$ 100,001 10 3.77 %
Regarding to Table 4-1, there were 34.46% of male and 65.54% female in this investigation. Almost half of respondents, their ages are from 26 to 40 years old (55.42%) and 37.97% are single. 77.53% of whom lived in north Taiwan, well-educated (75.28% of above junior college) and work in service industry (39.47%). Besides, they also have well monthly income (51.31% of above NT$
35,001).
4.2 Reliability and Validity
Since Goldsmith and Hofacker’s (1991) DSI scale was cited and translated into Chinese, it inevitably has to test the reliability and validity of scale. PROC CORR
38
Table 4-3 DSI scale convergent validity
Items Estimate t value
1. In general, I am among the last in my circle of friends to purchase a new digital camera.
0.7857 10.9517
2. If I heard that a new camera was available through a local store, I would be interested enough to buy it. camera, even if I hadn’t heard of it yet.
0.6170 8.0900
5. In general, I am the last in my circle of friends to know the names of the latest digital camera and relative trends.
0.7158 9.8970
6. I know more about new digital camera than other people do.
0.6114 10.0908
Cronbach coefficient alpha was acceptable (0.77), and Table 4-3 reveals that each factor loading was significant (︱t︱> 1.96), which implied that convergent validity was acceptable. Therefore, we thought that DSI scale were well-translated and applicable due to the verification of validity and reliability.
4.3 Consumer Typology
In chapter 2, the definition and development of market segmentation has been reviewed, besides, we also introduced various methods which were used to apply in this issue. First of all, all of lifestyle variables were reduced to some representative latent factors by factor analysis, then the factor scores could be used to process cluster analysis later. When both of these two steps have been completed, the analysis could go further.
4.3.1 Factor analysis
This research used 30 lifestyle items as the base of factor analysis. Principal Component Analysis was used to produce lifestyle factors; meanwhile, these factors were rotated by VARIMAX approach, which maximizes the sum of variances of required loadings of the factor matrix and tend to simplify the structures (Hair et al., 1992).
Zaltman and Burger (1975) suggested that factor’s eigenvalue should exceed one, and cumulative variance should reach 40%. After the process of PROC FASTCLUS procedure, we retained 9 factors due to Zaltman and Burger’s suggestion. Meanwhile, the cumulative variance accounts for 64% of total variance (Table 4-4)
40
Table 4-4 Eigenvalue Table of Factor Analysis
Factor Eigenvalue Proportion Cumulative
Factor1 6.648 0.2216 0.2216
Factor2 2.851 0.0950 0.3166
Factor3 1.937 0.0646 0.3812
Factor4 1.723 0.0574 0.4386
Factor 1.346 0.0449 0.4385
Factor6 1.302 0.0434 0.5269
Factor7 1.197 0.0399 0.5668
Factor8 1.156 0.0385 0.6053
Factor9 1.092 0.0364 0.6417
Since above factors have been retained to represent the latent factors with respect to lifestyle of respondents, each factor should be assigned some meaning. General speaking, the naming of factors is very subjective and vary among different researchers. Here we chose the factor loadings which value are high than .5 to label these factors.
Factor1: Experience factor
There were eight significant factor loadings been included in factor1. These variables reflected the tendency to seek for novelty, new stimulation, exciting feeling, and don’t want to a boring, invariable life. Therefore factor1 represents the attempt of stimulation seeking, and enjoy the whole new experience.
Table 4-5 Factors included in Factor1
Number Item Factor loading
A23 I like the challenge of doing something I have never
done before .72
A26 I am always looking for a thrill .72
A27 I like doing things that are new and different .70 A20 I like a lot of excitement in my life .67
A14 I like trying new things .56
A2 I like outrageous people and things .55
A3 I like a lot of variety in my life .50
A30 I like my life to be pretty much the same from week
to week -.58
Factor2: Active factor
Factor2 includes four variables, which represent the feature of leading, and superiority. High score of this factor demonstrates strong attempt on being a leader, as a result, they are active and high self-esteem.
Table 4-6 Factors included in Factor2
Number Item Factor loading
A18 I like to lead others .77
A11 I have more ability than most people .75
A6 I like being in charge of a group .74
A12 I consider myself an intellectual .60
Factor3: Status factor
Factor3 included three variables, which all show the inclination to dress fashionable than others, or pursuit of latest trend, fashion event. For this reason, factor3 was named the status factor.
Table 4-7 Factors included in Factor3
Number Item Factor loading
A16 I like to dress in the latest fashions .87 A10 I dress more fashionably than most people .82 A5 I follow the latest trends and fashions .75
42
Factor4: Thriftiness factor
There are three variables included in facator4. #25 and #4 mainly describe the tendency to make things by hand, and #9 shows the attitude toward spending money.
The implicit meaning among these variables is the concept of how people spend their money, which is one part of value system. There we named factor4 the thriftiness factor.
TABLE 4-8 Factors included in Factor4
Number Item Factor loading
A25 I like to make things with my hands .84
A4 I love to make things I can use everyday .81 A9 I would rather make something than buy it .76
Factor5: Thinking factor
The factor loading of #1 was more significant higher than #7, as a result, we named factor5 mainly refer to #1. Factor5 imply the desire to explore unknown things, how the theory behind the surface, and logistic thinking.
TABLE 4-9 Factors included in Factor5
Number Item Factor loading
A1 I am often interested in theories .71
A7 I like to learn about art, culture, and history .52
Factor6: Machinery interest factor
Factor6 included two variables, one ask if respondents like to look through hardware or automotive stores (#28), another one is how much respondents interested in operation of machine (#15). As a result, the score gained in factor6 was high, which meant that respondents have high interest in machine relevance.
TABLE 4-10 Factors included in Factor6
Number Item Factor loading
A28 I like to look through hardware or automotive stores .82 A15 I am very interested in how mechanical things, such
as engines, work .75
Factor7: Self-given factor
Factor7 included #8 and #21, both of these two variables demonstrated one concept: the interest of respondent is broad or narrow. If someone interests in all kinds of things, or likes to learn everything, then the score of this factor will be low. By contrast, the score will be high if someone has narrow, limited interest, and just care about what they really concerned. Therefore, we named factor7 the self-given factor.
TABLE 4-11 Factors included in Factor7
Number Item Factor loading
A8 I am really interested only in a few things .80 A21 I must admit that my interests are somewhat narrow
and limited .72
Factor8: Family concern factor
Both of two variables within this factor involve with family. #22 asked the respondents if a woman should pay more attention to her family, then the negative sign of #19 revealed the heavy care of domestic life. Therefore factor8 represented the degree of caring family.
44
TABLE 4-12 Factors included in Factor8
Number Item Factor loading
A22 A woman’s life is fulfilled only if she can provide a
happy home for her family .65
A19 I would like to spend a year or more in a foreign
country -.56
Factor9: Critique factor
This factor only has one variable, which ask the respondents if there is too much sex on the TV. Regardless of real frequency of sex on TV, this factor reflected the critique of respondents on social subjects. If the score is high, then the degree of critique is sensitive and high. By contrast, if the score is low, it means that circumstance is acceptable, or ignored. Respondents don’t have too much critique.
TABLE 4-13 Factors included in Factor9
Number Item Factor loading
A17 There is too much sex on television today .77
4.3.2 Cluster analysis
Since the factor scores of each respondent had been computed by PROC FACTOR procedure, cluster analysis can go further to the next step of data analyzing.
Because of the data amount were above 200, hence we utilized K-means method of nonhierarchical procedure to process these information.
General speaking, the primary query of cluster analysis is how many clusters should be chose. However, there is still no consensus among researchers. One of popular rules is CCC criterion, which showed in Table 4-14. Since the CCC value of four clusters design was best, therefore we decided to segment respondents into four clusters with factor centroid.
Table 4-14 Clustering statistics compare Table
Number of Cluster Pseudo F R-squared CCC
3 clusters 17.86 0.16053 -6.369
4 clusters 18.65 0.21837 -5.845
5 clusters 17.82 0.26755 -7.052
6 clusters 17.79 0.31029 -7.242
After the process of cluster analysis, table 4-15 presented factor means of each cluster, and then we could name each cluster by the centroids of factor scores.
TABLE 4-15 Factor Means of Clusters
Factor1
-0.02053 -0.77332 -0.48635 -0.71144
Cluster2 (Maker)
0.21993 -0.16094 -0.37363 0.98759
Cluster3 (Achiever)
0.00015 0.72124 0.60536 -0.05315
Cluster4 (Peace amateur)
-0.18086 -0.11935 0.01040 -0.38098
Factor5
0.54432 -0.80786 0.01682 -0.08044 0.41883
0.22878 0.32075 0.49552 -0.14804 0.13566
-0.15075 -0.03392 -0.46692 -0.30613 0.42594
-0.37917 0.24390 0.02775 0.48922 -0.80556
46
Cluster1: Hedonismer
In cluster1, these respondents emphasize on factors such as “Machinery interest,”
“Active,” and “Thriftiness”. However, all of above factor means are negative. As a result, respondents who belong to this cluster don’t thought themselves are smart, capable people, and they have few interest in machinery. Besides, they are willing to spend money if they need something rather than making by themselves.
Cluster2: Maker
In cluster2, the factor means of “Thriftiness,” “Self-given,” and “Status” are significant higher than others. Therefore, respondents of this cluster revealed heavy attempt of saving; meanwhile, they don’t inclined to chase fashion, and only concerned about what they really interested.
Cluster3: Achiever
The respondents of cluster3 produced high factor mean of “Active,” “Status,” and negative “Self-given”. In contrast to cluster1, the respondents of cluster3 believe that they are smart, intellectual, and superior to other people. Rather than narrow interest, they also inclined to have widespread interest. In addition to positive, active characteristics, they also like to dress fashionable, seeking for vogue.
Cluster4: Peace amateur
These respondents care about “Family concern”, and have little “Critique” on societal issue. However, they are willing to pay rather than respondents who belong to cluster2. In essence, this cluster tends to be conservative, and adaptable.
4.4 Proposition test
4.4.1 Description of cluster
Since the respondents have been separated into four segments based on lifestyle factors respectively, we first profile these four clusters on demographic characteristics by a series of contingency tables. Table 4-15 display that demographic variables were significant except for age, FLC, education, and occupation.
In cluster1, almost four out of five were female in opposition to 60% of others.
In terms of education, we find that even if it is not significant in Chi-square test, the percentage of education above college was 64% in cluster3, rather than 48%, 47%, 37% in other clusters.
Moreover, the personal monthly income was quite different among clusters.
There were almost 73% of respondents who were belong to cluster3, their average monthly income was above NT$ 35,000; however, in cluster2, the percentage of monthly income above NT$ 35,000 was less than 40%. In detail, the highest percentage of average monthly income (52.31%) in cluster2 fall into the interval
“NT$ 25,000- NT$ 35,000” Besides, the percentage of monthly income below NT$
20,000 was especially high (16.44%) in cluster4 rather than other clusters.
Table 4-16 Cluster Profile based on Demographic Characteristics Cluster1
48
Occupation 13.6598 (0.1350) Public servant 18.18 15.38 20.00 13.51
Business 13.64 3.08 20.00 13.51
Service industry 43.18 43.08 37.33 36.49
Others 25.00 38.46 22.67 36.49
Monthly income 25.3230
(0.0026**)
<NT$ 20,000 6.82 9.23 9.33 16.44
20,001-35,000 43.18 52.31 17.33 38.36
35,001-50,000 25.00 20.00 34.67 23.29
>NT$ 50,000 25.00 18.46 38.67 21.92
4.4.2 Relationship of physical-attribute classification with segment identity
The next step of data analysis was to examine if physical attribute classification was significant in differentiating the various clusters identified. First of all, we classified 267 subjects into 4 segments by cumulative percentage of25%, 50%, and 75% of total finger ridge count (TFRC). These segments were named TFRC1, TFRC2, TFRC3, and TFRC4, respectively. Then four clusters based on lifestyle factors were examined with four physical attribute based segments. Table 4-16 demonstrated the Pearson Chi-square test.
50
TABLE 4-17 Relationship of Physical Attribute (TFRC) with clusters Number
(Col Pct) Cluster1 Cluster2 Cluster3 Cluster4 Total (Col Pct)
Note: N=259, DF=9, χ2 =10.6664, p-value=0.2993
The insignificant χ2 showed that physical attribute might not be able to discriminate the various clusters adequately. However, from Table 4-16 we also find an interesting arrangement, that is, the distributive weigh of TFRC of each cluster was a little different and specific. For example, 35.56% of respondents in cluster1 belong to TFRC1, which meant their TFRC were less than 25% of total. Besides, 32.31% of respondents in cluster2 belong to TFRC4, which meant the TFRC these subjects possessed were above 75%. Others were highlight in Table 4-16.
4.5 Consumer innovativeness on lifestyle based, physical attribute based, and DC buying behavior
As we mentioned, previous studies mainly focused on definition and measurement of innovativeness; however, we sought to examine the likely relationship between consumer innovativeness and latent variables, such as lifestyle, physical attribute, and buying behavior in this exploratory research.
To being with, the clusters based on lifestyle factors were analyzed with consumer innovativeness by one-way ANOVA, and then followed by physical attribute classification and DC buying behavior. In addition, further post hoc analysis was utilized to describe the difference among segments, or levels in detail.
4.5.1 Consumer innovativeness on lifestyle based typology
In Table 4-17, F value is 2.71 and p-value is below .05, which meant that consumer innovativeness was significant different among four clusters. Next step, Scheffe’s test was used to examine the difference of any two pairs (Table 4-18).
Table 4-18 Consumer Innovativeness on Lifestyle based clusters
Source DF Sum of
Squares
Mean
Square F Value Pr > F
Cluster 3 4.4436 1.4812
2.71 0.0458*
Error 254 138.9315 0.5470
Corrected Total 257 143.3751
Note: N=258, *:p<.05 **:p<.01 ***:p<.001
Table 4-18 demonstrated the number and mean of each clusters, the outcome showed that consumer innovativeness of cluster3 was highest, whereas the score of cluster1 was lowest. Besides, the grouping column meant that there is no significant different score between cluster3 and cluster2, cluster2 and cluster4, cluster3 and cluster4, cluster2 and cluster1, cluster4 and cluster1. Nevertheless, the consumer innovativeness of cluter3 was significant higher than the score of cluster1.
52
TABLE 4-19 Scheffe multiple comparisons (Lifestyle typology)
Cluster N Mean Grouping
Cluster3 74 2.7995
Cluster2 65 2.5821
Cluster4 74 2.5755
Cluster1 45 2.4185
Furthermore, the question if consumer innovativeness was varied due to distinct physical attribute was another issue we interested. The most of studies which investigated the innovativeness were related to psychology, or organizational behavior;
on the other hand, biotechnology or genetics concerned with disease, race, and psychology. Hence, this research attempted to explore the relationship of consumer innovativeness on physical attribute classification, which was segmented by TFRC.
Table 4-19 demonstrated that F value is 3.07 and p-value is below the significant level of .05, as a result, consumer innovativeness was significant different among physical attribute segments.
4.5.2 Consumer innovativeness on physical attribute based classification
Table 4-20 Consumer Innovativeness on physical attribute based segments
Table 4-20 Consumer Innovativeness on physical attribute based segments