The research design is the framework to conduct a specific marketing research project and provides with detail on the procedures that are followed to get the information about the research objectives and problem statements. The aim of this framework is to propose a study that will test the stated hypotheses, offer possible answers to the research questions, and provide with information which is required for decision-making (Malhotra, 2004:10).
There exist two kinds of information available to marketing researchers, which are primary data and secondary data. Primary data are the data collected to express the objectives of a specific project, while secondary data are the data which were previously collected for some project other than the one at hand (Zikmund, 2003). For this study, both secondary data and primary data were used. Secondary data were used to help in the development of a conceptual model, while primary data were used in order to examine the specific problem.
Reliability means an assessment of the degree of consistency between multiple measurements of a given variable (Hair et al.,1998). They also argue that the commonly agreed-upon lower boundary for Cronbach's Alpha is 0.70, despite the fact that this value may decrease to 0.60 in cases in which exploratory research is carried out. Additional analyses are conducted to evaluate the impact of low-correlated questions on total Cronbach Alpha coefficients. This approach is employed in an effort to improve total Cronbach Alpha coefficients by identifying as well as removing questions which are unsatisfactorily correlated with the other questions that measure a variable.
The validity of the research instrument must be employed in order to confirm that the research instrument measures what it intends to measure. According to Zikmund (2003), there exist three approaches that can be utilized to deal with the evaluation of validity: content (face) validity, criterion validity and construct validity.
Content validity is the subjective agreement among professionals that a scale logically appears to evaluate correctly. When it becomes evident to experts that a measure provides proper coverage of a concept, the measure is believed to have face validity (Zikmund, 2003).
Criterion validity investigates the extent to which a measurement scale performs as expected, compared with other variables selected as meaningful criteria. These criterion variables may include demographic and psychographic characteristics, attitudinal and behavioral measures, or scores taken from other scales (Malhotra, 2004).
Construct validity is interested in the question of what a specific measuring instrument is actually measuring (Churchill, 1983). While assessing construct validity, it is critical for the researchers to have established the meaningfulness of the measure by means of convergent and discriminant validity (Zikmund, 2003).
Convergent validity is the extent to which a scale correlates positively with other measures of the same construct, whereas discriminant validity means the extent to which a particular measure does not correlate with other constructs from which it needs to vary (Malhotra, 2004). In addition to this, Zikmund (2003) states that a measure has discriminant validity when it demonstrates a low correlation with measures of different concepts.
The discriminant validity of the measuring instrument is evaluated by means of an exploratory factor analysis (EFA). An EFA is conducted to emphasize a number of variables to a manageable number that belong together and display overlapping measurement characteristics (Cooper and Schindler, 2003).
Structural Equation Modelling (SEM) method is used to evaluate hypothesized relationships in the proposed theoretical model. According to (Hair et al., 1998), SEM is a multivariate approach which examines a series of dependence relationships concurrently.
Byrne (2010) argues that there exist some aspects which differ SEM from multivariate procedures that have been used in the past. These aspects are as follows:
SEM takes a confirmatory (CFA) rather than an exploratory approach to data analysis.
SEM is capable of presenting explicit estimates of these error variance parameters, while traditional multivariate procedures cannot evaluate or correct for measurement error.
Researchers can incorporate both unobserved and observed variables with SEM approach.
There are no applied alternative methods which can be run easily for modeling multivariate relations.
Hair et al., (2010) suggest in their study a six-stage decision process which will help in conducting a research by using SEM.
Step 1: Defining individual constructs
It is essential for the researcher to review the literature thoroughly. A number of
constructs are indicated and a theoretical definition for each construct is provided through literature review. After that, the researcher should determine the scale items. He/she can either develop a scale or use the existing ones.
Step 2: Developing and specifying the measurement model
The researcher must identify each latent construct in the model, while the measured indicator items are allocated to latent constructs as well (Hair et al., 2010). In this stage, it is crucial to ask following questions:
How many indicators should be used for each of the constructs?
Should the measures be considered to be describing the construct or to be explaining
Can the validity and unidimensionality of each of the constructs be empirically
Step 3: Designing a study to generate empirical results
Having specified the basic theoretical model in terms of constructs and measured variables, the researcher must indicate issues regarding to research design and estimation. Within the scope of research design, issues such as the type of data to be analyzed (co-variances or correlations), the impact and solutions for missing data and the impact of the size of the sample need to be considered (Hair et al., 2010). Besides, it is argued that using co-variances as much as possible, as covariance
matrices provide the researcher with more flexibility. It is essential to state that SEM requires a relatively large sample size, compared to other multivariate techniques, while some of the statistical algorithms employed by SEM programs are unreliable when smaller samples are used. As for the computer programs, there are a number of available statistical computer programs (EQS, AMOS, Mpuls and CALIS) to perform SEM. For this study, AMOS has been used.
Step 4: Assessing the validity of the measurement model
As soon as the measurement model is specified, sufficient data are collected, the estimation technique that will be used is created, the researcher must make sure if the measurement model is valid. Measurement model validity is linked to two important aspects, which are goodness-of-fit (GFI) considerations and construct validity. Goodness-of-fit is used to compare theory to reality by evaluating the similarity of the estimated covariance matrix, which is the theory, to the observed covariance matrix, which is the reality. It is argued that the closer the values between the observed covariance matrix and the estimated covariance matrix, the better the model fit (Hair et al., 2010).
A main instrument to perform a measurement the differences between the observed and the estimated covariance matrixes is the Chi-square test. As Hair et al. (2010) points out, researchers benefiting from structural equation modeling will choose a relatively small Chi-square value, whereas this statistic will support a model that fits the data. The root mean square error of approximation (RMSEA), and goodness-of-fit index are used in this study to measure model fit. The GFI index is an early effort to produce a fit statistic which is not sensitive to the sample size handled, while the RMSEA is a measure that represents how well a model fits a population and not just a sample which is used for estimation (Hair et al., 2010:667).
Step 5: Specifying the structural model
Based on the literature, the structural model for the specific study is specified by assigning relationships from one construct to the other. It is done to build what dependence relationships exist among the different constructs. And then, a structural model is created that not only shows the complete set of constructs and indicators in the measurement model, but also states the structural relationships among the different constructs (Hair et al., 2010).
Step 6: Assessing structural model validity
The last step is to test the validity of the structural model. As is the case with the measurement of validity of the measurement model, the observed data is still represented by the observed sample covariance matrix. In evaluating the validity of the structural model, a new SEM estimated covariance matrix is calculated. As is the case with the measurement model, the overall fit of the structural model can also be calculated by using the Chi-square statistic. It is usually presumed that the closer the structural model goodness-of-fit to that of the measurement model, the better the structural model fit is (Hair et al., 2010).
...(download the rest of the essay above)