CHAPTER 4: SURVEY STUDY
4.2 Survey Instruments and Data Analysis Procedures
4.2.5 Data Analysis Techniques and Tools
Statistical analysis and social network analysis were conducted to make sense of the questionnaire data. The following two sections elaborate the two methods in detail.
4.2.5.1 Statistical Analysis
Statistical procedures are basically methods of handling quantitative information so as to make that information meaningful. First, they enable researchers to organise, summarise, and describe observations. In addition, statistical procedures involve either identifying the characteristics of observed phenomenon or exploring possible correlations among two or more phenomena (Ott & Longnecker, 2008).
Given the aim of the questionnaire and the complexity of survey variables, four
statistical methods are chosen to describe the stakeholders’ perceptions with the aid of Statistical Package for the Social Sciences version 18.0 (SPSS18). Akintoye (2000) and Yang (2009) have used these methods in nonparametric analysis, and proved that these methods are able to deliver rigorous results.
Here the term “parameter” refers to a measure that describes the distribution of the population such as the mean or variance. Parametric tests are based on the assumption that we know certain characteristics of the population from which the sample is drawn (Bryman & Cramer, 2009). This research does not fulfil the parametric assumptions of normal distribution and homogeneity of variance, therefore nonparametric testing was conducted. Purposes and outcomes of different statistical analysis methods are summarised in Table 4.2.
Table 4.2 Methods of Statistical Analysis
Purpose Method Description Outcomes Identify and
The significance of the CABs that are thought to be associated without implying that one is the cause of the other. and direction of the relationship between
The correlation between CABs and respondents’
two variables that are thought to be
associated without implying that one is the cause of the other.
It is worth mentioning that this study involves several analyses of causal relationships using the correlation coefficient. This type of approach is especially useful for exploratory or other studies in settings where little is known. In order to examine the relationship between two variables, there are three prominent methods:
Pearson’s r, Spearman’s rho and Kendall’s tau. Pearson’s r can be employed only when the variables are interval and the relationships are linear. For variables at the ordinal level, such as the Likert scale significance in this research, Kendall’s tau and Spearman’s rho are available. Kendall’s tau and Spearman’s rho function the same way, except the former usually produces a slightly smaller correlation. Although Spearman’s rho is more commonly used in reports of research findings, Kendall’s tau is preferred for a more believable result when dealing with a proportion of tied ranks (Bryman & Cramer, 2009). Therefore, this research used Kendall’s tau to analyse the relation between CAB scores and respondents’ characteristics, where the latter involve a lot of tied value. On the other hand, this research adopted Spearman’s rho to analyse the causal relationship between CABs because this process is based on various CAB scores from 1 to 5 and few identical scores are involved.
Additionally, although the rho or tau could be used to describe the relationship between each of the 19 possible pairs of CAB variables, neither could provide a single measure that describes the overall relationship among all 19 variables simultaneously using a single number for comparing stakeholders’ various perceptions. Therefore, Kendall's coefficient of concordance was adopted. It is the natural extension of Spearman's rho and Kendall's tau coefficients, which evaluates the extent of agreement between two judges on the association among three or more variables (Kendall, 1955).
Finally, considering the complexity of potential value gaps among the seven key stakeholders, the Mann-Whitney test is adopted to triangulate the descriptive analysis and Kendall’s coefficient of concordance. This test is more powerful than the median test because it compares the number of times a score from one of the samples is ranked higher than a score from the other sample, rather than the number of scores which are above the median (Bryman & Cramer, 2009). This is considered useful for comparing differences on the absolute importance of CABs between two independent samples (Pallant, 2005).
4.2.5.2 Social Network Analysis
Environmental applications of Social Network Analysis (SNA) emerged in the last decade in order to understanding characteristics of social networks that increase the likelihood of collective action (Tomkins and Adger, 2004; Newman and Dale, 2004; Bodin et al., 2006). It measures and maps the relationships and flows between people, groups, organisations or other information or knowledge processing entities.
It involves actors and relations, and has been widely used in sociology, anthropology, organisational behaviour and many other domains (Liebowitz, 2005).
This research expands this knowledge to preliminary understand the prominence of seven stakeholders based on their network construct and influential supply chain partners, which lays the foundation for the comparative study across stakeholders. The concept of “degree centrality” to quantitatively analyse and visualise stakeholders’ power based on their answers to Questions 8, 9 and 10. The analysis was done by using the social network software Netminer, which takes each stakeholder as a node. If a node has many connections, it may have a large centrality score. As the length of a connection increases, the influence attenuates exponentially (attenuation factor is 0.5 in this study). It should be noted that the connection
between nodes in the network represents the strength but not the direction. This type of network is defined as a “1-mode network” in the Netminer software, and is a required input for the centrality test. The output includes a set of in-status centrality scores and out-status centrality scores, which could be mapped from the output option in Netminer. The detailed analysis procedure is introduced in the results section (CRYAM, 2009; Prell et at., 2009).