Methodology

decision of others (Rogers & Cartano, 1962) and also can affect members in the social community because of special techniques, knowledge, personalities and other uniqueness (Kotler, 2001).

Because of the Internet, the online opinion leaders begin to appear and exert their influence towards others through the Internet.

(3) Virtual communities of consumption

Rheingold (1993) defined virtual communities as online social aggregations which emerge when enough Internet users discuss long enough, with sufficient human feelings, to form online relationships.

According to previous researches, the most well-known typology of virtual communities was illustrated by Hagel and Armstrong (1997) who classified virtual communities into four types, including virtual communities of interest, of relationship, of fantasy and of consumption.

This study only focuses on the virtual communities of consumption, which refer to virtual communities focusing on facilitating consumption, serving to some kinds of commercial purposes, and encouraging participants to communicate and interact with others so as to make transactions (Hagel & Armstrong, 1997).

(4) Purchase intention

Purchase intention refers to the likelihood that a consumer will purchase a specific product (Ajzen & Fishbein, 1975; Dodds, Monroe, &

Grewal, 1991; Schiffman & Kanuk, 2000). It can be used to predict the future purchase behavior of the individuals (Ajzen & Fishbein, 1975).

method.

(1) Methods for data collection

 Literature review method

Through the literature review, the current situation is analyzed and the research questions are identified. For Part 1, based on the original models, the model for this thesis is built up and the hypotheses are explained. Then, the questions in the questionnaire are designed.

For Part 2, based on the existing identification approaches, the approach is selected for this study and the indicators of opinion leaders are decided.

 Small group discussion

For Part 1, through the small group discussion, the readability problems and confusing words of the Chinese version of the questionnaire are purified.

 Questionnaire

For Part 1, this thesis chooses to send out questionnaire through Internet for the empirical study.

 Snowball sampling

For Part 2, after selecting the specific virtual community of consumption randomly, the snowball sampling is used for collecting one-week data for the continue analysis. This method will avoid the bias problem by considering all the members in the network in a specific time period.

(2) Data analysis methods For Part 1:

After obtaining the original data through the formal questionnaire, a variety of statistical methods to analyze the data and to validate the model are used. The relationships between the variables, the proposed research assumptions and theoretical models will be empirical testified.

Part 1 uses Structural Equation Modeling (SEM) for data analysis, through the statistical analysis software, called SPSS 23 and AMOS 21.0.

these pieces of software are commonly used in the social sciences. The

specific data analyses can be divided into three steps:

 Descriptive statistics analysis

It is to describe the overall information of the respondents to the questionnaire. The questionnaire collects data of the distribution of the sex, age, education level, incomes and other information of these respondents. With the information, a better understanding of these respondents will be obtained.

 Validity analysis and reliability analysis

In order to evaluate the measurement of the items in the questionnaire, the validity and reliability analysis are used. For validity analysis, the KMO & Bartlett’s Test of Sphericity of factor analysis and the factor loadings of the data are measured. On the other hand, for reliability analysis, the Cronbach’s α is used.

 Structural Equation Modeling (SEM)

The SEM is used to testify the model and to analyze the relationships among the variables. This study adapts the two-way approach for structural equation modeling (SEM) proposed by Anderson and Gerbing (1988). The first step is to test the measurement model, namely to test the latent constructs and their respective observed variables. Then the second step is to test the relationship of each variable, namely to test the hypotheses. The goodness-of-fit measures used in this research include ^2/df, GFI, RMSEA, AGFI, NFI, CFI, IFI.

For Part 2:

After the literature review on the related approaches for identifying opinion leaders, the Social Network Analysis is selected. The software used for data analysis includes UCINET 6.0, Netdraw, and SPSS 23. To be more specific:

 Social Network Analysis (SNA)

SNA is used to study the structure of the virtual community and to identify the opinion leaders. Also, the influences from these opinion leaders towards the eWOM communication are analyzed.

The research process of SNA is generally composed of eight steps, including:

1) Defining the research questions and the research focuses;

2) Determining the network boundary and relationship dimensions;

3) Selecting or designing the research tools;

4) Collecting the original data by using snowball sampling;

5) Converting the data to meet the format for the SNA software;

6) Using the software to run the data analyses for studying the structure of the virtual community;

7) Using the software to run the data analyses for gathering data related to the indicators for identifying opinion leaders;

8) Making comparison and identifying the opinion leaders;

9) Calculating and comparing the data of both having and deleting opinion leaders;

10) Giving out discussions and conclusions

As for the step 7) for this thesis, the analyses include: analysis of tie strength, analysis of the small-world phenomenon, analyses of centrality, identification of opinion and the analysis of influences of opinion leaders on eWOM dissemination.

 Frequency analysis

In particular, the frequency analysis is used during the process of analyzing the In-degree centrality and Out-degree centrality.

1.4. Thesis outline and technical routes