Chapter 4: Research Results
4.3 Cluster Analysis
國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
4.3 Cluster Analysis
Cluster analysis was performed using a two-stage cluster-analysis technique to derive content gratification, social-relation gratification and social gratification.
Groups and clusters were formed using Ward’s method, followed by a K-means procedure to decide the adequate attributes of clusters. Finally, we will discuss the results of each cluster.
4.3.1 Cluster Method
We adapted the two-stage clustering approach to conduct the cluster analysis by using SPSS 21.0 software for our model. First, we used hierarchical clustering and Ward’s clustering method to decide the best number of clusters. The hierarchical clustering connects objects to form clusters based on their distance, and the Ward method is a hierarchical agglomerative technique that is used to identify the initial cluster solution. There are the means of each factor, and the best number of clustering is two. The Clustering results are shown in Table 11.
Table 11. Clustering results by hierarchical clustering method
Cluster 1 2 3 4 5
Agglomeration coefficient
8229.80 7199.15 6768.25 6471.41 6213.42
Change Rate 14.3% 6.36% 4.58% 4.15%
The cluster seeds formed by the Ward solution were submitted to a K-means procedure, which is an iterative partitioning technique that is used to group the data until the optimal solution is found. We used the K-means method to partition the input data set into two clusters according to the result of hierarchical clustering. We also conducted a t-test to identify the degree of resonance by different types of resonance dimensions using mean scores in three resonance-behavior dimensions. We separately labeled each cluster based on the results of resonance. Cluster A was labeled “high status” in resonance because the individuals in this cluster were more likely to engage in resonance behavior such as “like,” “reply,” and “share.” On the other hand, cluster B was labeled “low status” due to low scores on resonance. The results of each mean are shown is Table 12. As you can see, the amount in cluster A is more than in cluster B.
‧
Table 12. t-Test of Resonance in Dimensions
Cluster Factors
The t-test results demonstrate that users in social networks have different levels of content gratification, social-relation gratification and self-presentation gratification.
As can be seen in Tables 13–15, low-resonance clustering, in contrast to
high-resonance clustering, has high degrees of homophily and trust (t = -11.5, A < B).
On the other hand, high-resonance clustering is characterized by high degrees of content gratification, self-presentation gratification, and purchase intention.
Table 13. t-Test of Two Clustered Groups on Content Gratification
Cluster
Factors
Content Gratification
Utilitarian Value Hedonic Value
Cluster A 3.58 4.02
Cluster B 2.69 3.95
T values 17 9.3
Difference of Mean A > B A > B
Table 14. t-Test of Two Clustered Groups on Social-Relation Gratification
Cluster
Factors
Social-Relation Gratification
Social Tie Homophily Trust Normative Influence
‧
Table 15. t-Test of Two Clustered Groups on Self-Presentation Gratification and Purchase Intention
Cluster
Factors
Self-Presentation Gratification Purchase Intention Self-Presentation Purchase Intention
Cluster A 3.99 4.1
Cluster B 3.62 3.42
T Values 13.01 16.006
Difference of Mean A > B A > B
After deciding the cluster of resonance, we separately applied the structural equation model to test the relationship in high resonance clustering and low resonance clustering to decide the dominated factors. All details are shown in Table 16. In high resonance cluster, all factors in gratifications are significant expect for homophily.
Resonance is significantly related to purchase intention. Besides, the most significant variables to cause resonance are self-presentation, utilitarian value in content
gratification and social tie in social relation gratification. On the other hand in low resonance cluster, all factors in gratifications are significant expect for hedonic value in content gratification, homophily in social relation gratification and trust in social relation gratification. The results of clustering results by structural equation model are same with the structural equation model statistical results of research model.
Table 16. SEM Statistical Results of High Resonance Clustering
Path Beta T Statistics R-squared Support
H1: Utilitarian Value ->Resonance 0.36*** 4.02***
0.67
Yes
H2: Hedonic Value ->Resonance 0.11* 1.78* Yes
H3: Social Tie ->Resonance 0.28*** 3.83*** Yes
H4: Homophily ->Resonance 0.07 0.02 No
H5: Trust Value ->Resonance 0.08* 1.56* Yes
H6: Normative Influence ->Resonance 0.21** 2.02** Yes H7: Information Influence ->Resonance 0.16* 2.00* Yes H8: Self-presentation ->Resonance 0.42*** 4.51*** Yes H9: Resonance ->Purchase Intention 0.67*** 8.74*** 0.46 Yes Standard errors in parentheses. *p < 0.05; **p < 0.01; ***p < 0.001
‧
國立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
Table 17. SEM Statistical Results of Low Resonance Clustering
Path Beta T Statistics R-squared Support
H1: Utilitarian Value ->Resonance 0.11* 1.18**
0.48
Yes
H2: Hedonic Value ->Resonance 0.03 0.01 No
H3: Social Tie ->Resonance 0.34*** 3.77*** Yes
H4: Homophily ->Resonance 0.00 0.00 No
H5: Trust Value ->Resonance 0.00 0.00 No
H6: Normative Influence ->Resonance 0.17** 2.27** Yes H7: Information Influence ->Resonance 0.11* 1.13* Yes
H8: Self-presentation ->Resonance 0.28** 3.19** Yes
H9: Resonance ->Purchase Intention 0.46*** 7.22*** 0.39 Yes Standard errors in parentheses. *p < 0.05; **p < 0.01; ***p < 0.001
4.3.3 Analysis Matrix of Demographic Information in Clustering
To examine the meaningfulness of the cluster solution, we used demographic information to see whether differences exited between the two clusters on gender and platform experiences. We used cross-table testing by SPSS 21.0 to verify whether there is a relation between the row variable and the column variable. Cross-table is a two-way table consisting of columns and rows to analyze categorical data. The cells of the table would report the frequency counts and percentages for the number of respondents in each cell.4.3.3.1 Comparison and Differentiation of Clusters Based on Gender
After conducting the cross-table testing on different types of clusters, we found that there were significant differences among the clusters. Comparisons between two clusters revealed that the percentages of males and females are different. As for the high status of resonance clustering, we found that the percentage of females was larger than that of males. On the other hand, we found that the percentages of males were a little higher than females in low-status clusters of resonance. The results are shown in Table 18.‧
Table 18. Describing Resonance Clustering of Demographic in Gender Information
Clustering Male Female Total Number
Cluster A 76 (46.1%) 131 (57.7%) 207 (52.8%)
Cluster B 89 (53.9%) 96 (42.3%) 185 (47.2%)
Total Number 165 (100%) 227 (100%) 392 (100%)
4.3.3.2 Comparison and Differentiation of Clusters Based on Platform Experience
We also tested the platform experience on clustering by cross table to examine the whether there are different types of pattern in each clustering. Surprisingly, there exits an interesting phenomenon. Compared with low-resonance clustering,
respondents showed whether they have much richer experience or less experience in using social network platforms in high status of resonance clustering. As for low status cluster, respondents presented an average status from 1 year to 5 years. All results are shown in Table 19.
Table 19. Describing Resonance Clustering of Demographics in Platform Experience Information
4.3.3.3 Comparison and Differentiation of Clusters Based on Revenue
We also used the cross-table analysis to test the effect of demographic revenue information on clustering. Compared with a high degree of resonance, we found people who have high revenue are not likely to engage in resonance behavior. On the other hand, people who have low revenue are likely to engage in resonance behavior on social-network platforms. All results are shown in Table 20.‧
國立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
Table 20. Describing the Demographics of Resonance Clustering Based on Revenue Clustering Less than $10,000 $10,001– $20,000 Over $20,001 Total
Cluster A 72 (55.4%) 91 (58.7%) 44 (41.1%) 207 (52.8%) Cluster B 58 (44.6%) 64 (41.3%) 63 (58.9%) 185 (47.2%)
Total 130 (100%) 155 (100%) 107 (100%) 392 (100%)
All chi square test of demographic variables on resonance is shown in Table 21.
All demographic variables in resonance clustering are significant.
Table 21. The effect of demographic variables on resonance by Chi Square test
Variables Pearson Chi-Square DF Asymp. Sig. (Two-sided)
Gender 12.04 2 0.005
Platform experience 16.25 8 0.000
Revenue 14.82 8 0.000
Standard errors in parentheses. *p < 0.05; **p < 0.01; ***p < 0.001