5.5 Research Design
5.5.3 Phase Three: Final Validation Study (2 nd Study)
The objectives of the third phase, which includes one study with an independent sample, are as follows: first, the second study was designed to provide further validation for the scale developed from phase 1 and phase 2 by replicating the confirmatory factor structure. Second, the nomological validity was tested via examination of the relationships between CE and a number of potential antecedents and consequences. Third, evidence of initial validation for the proposed model of customer engagement in the online brand community was established based on the Elaboration Likelihood Model. The sampling process for the second study is explained in the following section.
5.5.3.1 Sampling Process
The second study follows a similar sampling process to the first study and employs convenience sampling by recruiting respondents via the services of Amazon Mechanical Turk (AMT). However, there are a number of rules regarding sample size that need to be considered in order to conduct Structural Equation Modelling (SEM). As discussed in Section 5.5.2.3, 150-200 is the minimum appropriate size for SEM. The process of the questionnaire design is presented in the next section.
5.5.3.2 Questionnaire Design
This section concerns with the questionnaire design. The questionnaires used for both studies are attached as appendices. Following Malhotra and Birks (2006), this section describes the different steps in order to design the questionnaire.
Information Needed
The questionnaire includes three types of information based on the current research objectives. The first type of information that was collected through the questionnaire was screening information. These questions were designed to screen respondents with respect to regular use of the online brand community to assess their eligibility to answer the questionnaire. The second
type of required information was related to demographic characteristics. The data were used to create a descriptive profile of the respondents. The final type of required information was based on the research objectives. As mentioned, the objective of the second study is to confirm the dimensionality of the scale and also re-examine the convergent and discriminant validity of the scale with respect to a new sample. Importantly, the purpose of the second study is to test nomological validity of the scale. The final type of information is collected to test the proposed customer engagement model. The model comprises potential antecedents and consequences of customer engagement in the online brand community. Data were collected regarding the constructs represented in the proposed model.
Type of Survey and Method of Administration
As mentioned in Section 5.5.2.3, a web-based survey was used to collect data. The designed questionnaire was placed on Amazon Mechanical Turk for potential respondents to complete.
Content of Individual Questions
The first section of the questionnaire comprised screening questions. Questions relating to name of the online brand community, the number of members and also the type of online brand community were asked in this section. The data were used to check the eligibility of respondents to participate in the questionnaire. The respondents were asked to write the name and number of the members and then select the type of online brand community. Different types of online brand community were explained with examples provided to respondents.
The second section of the questionnaire comprised demographic questions including gender, age, income, and educational level. The descriptive profiles of the respondents are presented according to the information collected by these questions
The third section of the questionnaire included the measurement items of the constructs in the proposed model. The items of the antecedents and consequences of customer engagement as well as the remaining items of customer engagement were included in this section. This section also included the items of the marker variable construct in order to assess common method bias. The process of selecting measurement items for each construct is presented in detail in Section 5.6.
Pre-test and Revise the Questionnaire
Prior to conducting the study of the third phase, the questionnaire was pilot tested. The purpose of carrying out a pilot test is to identify any problems for respondents when answering the questionnaire and also any problems regarding recording of the data. Furthermore, the assessment of questionnaire validity and the reliability of data can be obtained via the pilot test. And, importantly, the undertaking of preliminary analysis on the collected data by a pilot test can ensure that we achieve the research objectives. In terms of the importance of the pilot test, Bell (2005) states, “however pressed for time you are, do your best to give the questionnaire a trial run, as without a trial run, you have no way of knowing your questionnaire will succeed.” Initially, a pre-test of the questionnaire was completed by 10 PhD students competent in marketing and information systems, to help establish content validity and face validity. Their suggestions on question wording and questionnaire structure were very helpful. Their comments and suggestions were applied before conducting the pilot study.
Demographic Profile of Respondents
The report of the demographic profile for the second study is shown in Table 18. The sample comprised 45% females and 55% males, and the largest age group of the respondents was 25- 34. The results regarding the demographic profile of respondents were similar to the first study. Again as similar to the first study, the biggest segment of respondents was well educated and had obtained a postgraduate degree. In addition, the distribution of age of respondents revealed that all respondents were aged below 65 years old. The income level of respondents varied where a monthly income between $1501 and $2000 or more than $2500 were the biggest two segments. In contrast to the first study, the members of the customer-initiated communities were the larger segment of the respondents. Further details regarding the demographic profile are shown in Table 18.
Table 18 Sample Characteristics of the Second Study
Category (Items) Frequency Percentage Gender Male Female 282 225 55.6 44.4 Age Category 18 - 24 years old 25 - 34 years old 35 - 44 years old 45 - 54 years old 55 - 64 years old 65 or over 81 304 84 31 7 Zero 15.9 59.9 16.5 6.1 1.3 Zero Income =<$1001 p/m =$1001 - 1500 p/m =$1501 - 2000 p/m =$2001 - 2500 p/m =>$2500 p/m 24 26 261 43 153 4.7 5.1 51.5 8.4 30.1 Level of Education No education Primary education
Lower secondary education Intermediate secondary education Higher education Zero 9 37 48 413 Zero 1.7 11.2 9.4 81.4 Community Type-1 SNS Website 203 304 40.0 59.9 Community Type-2 Customer-initiated Company-initiated 311 196 61.3 38.6 115 | P a g e
5.5.3.3 Study Two: Quantitative Data Analysis procedures
The purpose of the second study was to provide further evidence of validity (convergent and discriminant) and reliability of the scale as well as primarily nomological validity by using a new independent sample. All the performed analyses for the first study were replicated in order to achieve the objectives of the second study. In addition, covariance-based structural equation modelling (SEM) was used to test the hypothesised relationship of customer engagement with its antecedents and consequences. The following section provides an overview of the structural equation modelling (SEM) steps.
5.5.3.4 Structural Equation Modelling and Hypotheses Testing
The Structural Equation Model (SEM) is:
“A collection of statistical techniques that allow a set of relationships between one or more independent variables (IVs) and one or more dependent variables (DVs) to be examined” (Tabachnick and Fidell, 2014: 655).
SEM includes a group of statistical models that explain the relationship among the constructs (IVs and DVs). This group uses causal analysis, path analysis and confirmatory factor analysis (CFA), etc.
In the first step of the SEM technique, the researcher proposes a set of relationships between independent and dependent variables drawn upon theories and prior research. In these sets of relationships, independent variables predict dependent variables. In addition, a variable that is dependent in one relationship is independent for another relationship. In fact, SEM expresses these ‘dependence relationships’ among the sets of relationships in which a variable plays the role of both dependent and independent variable in two different relationships. Then, a series of structural equations is defined from the proposed relationships for each dependent variable (Hair et al., 2010). Therefore, the current research adopted SEM to analyse the model and test the hypotheses. There are different steps to follow in order to analyse the SEM. These steps have been proposed by different scholars such as Kline (2011) and Hair et al. (2010) and Tabachnick and Fidell (2014). The current research follows the six stages for analysing the structural equation model proposed by Hair et al. (2010). These six basic steps are listed below and these processes are shown in Figure 15:
1. Defining individual constructs
2. Developing the overall measurement model 3. Designing a study to produce empirical results 4. Assessing the measurement model validity 5. Specifying the structural model
6. Assessing structural model validity
Defining the Individual Constructs
What items are to be used as measured variables?
Develop and Specify the Measurement Model
Make measured variables with constructs Draw a path diagram for the measurement model
Designing a Study to Produce Empirical Results
Assess the adequacy of the sample size Select the estimation method and missing data
Assessing Measurement Model Validity
Assess line GOF and construct validity of measurement model
Measurement Model Fit?
Refine measures and design a new study
Proceed to test structural model with stages 5 and 6
Specify Structural Model
Convert measurement model to structural model
Assess Structural Model Validity
Assess the GOF and significance, direction, and size of structural parameter estimates
Measurement Model Fit?
Refine model and test with new data
Draw substantive conclusions and recommendations Stage one Stage two Stage three Stage four Stage five Stage six
Figure 15 Different Stages of SEM (adapted from Hair et al., 2010)
Defining Individual Constructs
It is necessary to have an appropriate measurement theory in order to obtain a reasonable result from SEM (Hair et al., 2010). It is crucial to explain how hypotheses are developed from the theories and how the model is proposed. However, this stage evolves operationalisation of constructs, developing scales and pre-testing the measures. The current research has explained how the constructs are conceptualised and the related items adapted from the prior studies.
Developing the Overall Measurement Model
When the scales and measures for constructs have been developed, the next stage is to specify the measurement model. It is preferable to show the measurement model by a diagram. In the diagram, the measures (observed variables) are assigned to their construct (latent variable). This stage is also known as model specification. As explained in Section 5.7, the current research uses Amos 22.0 for SEM analysis and this software specified the measurement model.
Designing a Study to Produce Empirical Results
The next stage after specifying the basic model is to consider issues regarding research design and model estimation. In terms of research design, there are three main issues: sample size, the type of data and missing data, which were explained in the previous section. Similarly, there are three issues regarding model estimations: the model structure, used estimation technique and the computer program for SEM analysis.
Hair et al. (2010) state that “among the most important steps in setting up a SEM analysis is determining and communicating the theoretical model structure to the program”. All the approaches perform the same function and, as mentioned, the current research uses Amos 22.0 for both measurement model and the structure of model, which are completely graphical.
Assessing the Measurement Model Validity
In stage four of the SEM analysis, after specifying a measurement model, collecting the required data and selecting the estimation technique, the next stage is to test the model’s validity. This stage includes two validity tests: (1) construct validity and (2) goodness-of-fit of the proposed model. Both were described in Subsection 5.5.2.5.
Specifying the Structural Model
In stage two, the indicators were assigned to the related constructs, which is called specifying the measurement model. However, in this stage, the relationships between constructs (latent variables) that are “based on the proposed theoretical model” are assigned to each other (Hair et al., 2010). Thus, this stage is called the structural model specification. In this stage, single- headed and also directional arrows are added to represent the proposed hypothesis in the current research model. The current research has suggested the hypotheses in Chapter Four and they show the relationships between constructs. These dependence relationships are specified in this stage. After this stage, we will have a confirmatory factor analysis (CFA) model in which the relationships between constructs are specified based on the proposed hypotheses. It is necessary to have both the measurement model and the structural model together in order to estimate the SEM.
Assessing the Structural Model Validity
In the final stage, the structural model that was explained in the previous section is tested as well as the hypotheses. It is crucial to test the first measurement model and then test the structural model. Therefore, two requirements are needed before this stage: first, testing the validity of the measurement model and second to test if the model has an acceptable model fit.
In order to test the validity of the structural model, the estimated parameters are needed to make sure that the structural model is validated. The process of testing validity is the same as in stage 4. The only difference is related to covariance matrix (Hair et al., 2010). In the measurement model, it is assumed that all constructs are correlated to each other, while in the structural model, which is based on the hypotheses, there is no correlation between some of the constructs. The criteria mentioned in Table 15 are used for establishing the validity of the structural model and therefore “at least one absolute index and one incremental index” as well as the chi-square goodness-of-fit (GOF) test (Hair et al., 2010) are required.