RESEARCH METHODOLOGY
3.5 Research Techniques
3.5.2 Semi-structured Interviews
Additional non-parametric data analysis such as the Mann-Whitney U and Kruskal-Wallis tests were conducted to make judgments of the probability of observed difference between two or more groups of respondents being dependable or having happened by chance (Field, 2009). The Mann-Whitney U test is suitable here instead of the t-test, due to the expected skewed distribution in the data obtained from the Likert-scales construct measuring vulnerability and capability. For instance, when the majority of respondents select the positive anchor of ‘agree’ or ‘strongly agree’ in the construct, the distribution is expected to be negatively skewed; conversely, when the majority select the negative anchor of ‘disagree’ or ‘strongly disagree’, the distribution is expected to be positively skewed. Pallant (2013) agreed that many scales and measures used in the social sciences are not normally distributed and have scores that are skewed, either positively or negatively. It is worth noting here, however, that this does not necessarily indicate a problem with the scale, but rather reflects the underlying nature of the construct being measured. In this case, instead of violating the assumption of a normally distributed data in a parametric analysis, the used of non-parametric analysis, such as the Mann-Whitney U test, is preferred (Field, 2009; Pallant, 2013).
Correlational analysis using Spearman rho was also conducted to identify significant relationships among the vulnerability and capability factors. Correlation is a relationship measure among different factors or parties indicating the level of strength and direction of the relationship (Assaf and Al-Hejji, 2006). The Spearman rho correlation was also used to test the level of agreement or disagreement among the different groups of respondents (public organisations, consultants and contractors) on these factors. Spearman’s correlation results range between the value of 1 and −1, whereby values closer to 1 indicate a perfect positive relationship (or high degree of agreement) and -1 implies a perfect negative relationship (or disagreement) (Assaf and Al-Hejji, 2006).
3.5.2 Semi-structured Interviews
While questionnaires can provide evidence of patterns amongst large populations, qualitative interview data often produces more in-depth insights into participants’ attitudes, thoughts and actions (Kendall, 2008). Semi-structured interviews are conducted after the questionnaire
interviews with respondents also enables the researcher to obtain additional input that might not be available through the questionnaire findings. The researcher also aimed to gather the pathogenic influences (discussed in Section 2.7.1) through the interviews. Subsequently, the pathogens identified through the analysis of the interview data are compared against the pathogens laid out in Table 2.5. Semi-structured interviews were chosen to provide flexibility for the researcher to build arguments and gather as much as possible of the information required in addressing the specific issues. This also allows the researcher to meet the objective of identifying the causes and cascading effects of disruptions on public project performance.
The drawback of this approach, however, is that it might be difficult to make comparisons between the results as only a small number of interviews can take place, given time constraints. Nevertheless, this will not be an issue here, as the interview is treated as a complementary method to fill potential gaps in the questionnaire, as previously discussed. The interviews were conducted either by phone, video conference or face-to-face through the appointments agreed between the researcher and the participants. To give participants sufficient time to think about the subject matter, the interview questions (see Appendix D) were sent to them in advance.
In analysing the qualitative data, the computer software package NVivo (version 11) was used to identify relationships between existing themes and emergent new themes. These emergent themes from the interviews are useful in developing the final resilience response framework. As in previous studies (Love et al., 2010, 2011), content analysis and cognitive mapping were used to analyse and present the qualitative data. Content analysis is a technique in which the researcher interrogates data for constructs and ideas that have been decided in advance (Easterby-Smith et al., 2008). This technique is useful for the researcher to analyse key issues identified from the previous questionnaire findings. Furthermore, as different respondents may discuss different pathogenic influences based on their own situation, emergent pathogenic themes relevant to the study are also considered. Interviews were recorded and transcribed in Malay, and relevant parts were translated into English before the analysis.
Cognitive mapping was used to present the relationships between different issues gathered from the interviews. Cognitive mapping is a method of spatially presenting the data to
identify patterns that will allow the researcher to understand the relationships between the data and its significance (Easterby-Smith et al., 2008). In this study, the transcribed interviews were imported in the NVivo software. Predetermined pathogenic themes from the literature (see Section 2.7.1) were used to construct the nodes in the software. These nodes are used to represent the concept, code or themes of the data (Easterby-Smith et al., 2008). It also allows the researcher to break the pathogenic themes into sub-themes during the analysis of the interview data. Emergent pathogenic themes from the interview were also identified to reflect the respondents’ perceptions on the pathogenic influences that may not be within the predetermined pathogenic themes in the literature. The cognitive maps are then developed based on these nodes, and presented in the data analysis in Chapter 5 of the thesis.
3.5.2.1 Data Sample for Interviews
For the interviews, the non-probability method of purposive sampling was adopted, using the expert sampling approach to include particular professionals from both the public organisations and their supply chain (i.e. private organisations). These categories of professionals may have a unique, different or important perspective on the phenomenon in question, hence their presence in the sample should be ensured (Mason, 2002). The interview sample involved the participants that had responded to the earlier questionnaire. In the participant consent form distributed earlier with the questionnaire (see Appendix C), participants were given the option of whether they would be interested to take part in the subsequent interviews. This option helped the researcher to contact and assemble the relevant experts or professionals with experience working in public projects.
In terms of sample size for the interview, previous researchers (Pettit et al., 2013) conducted 10 to 40 interviews within a firm for their study on supply chain resilience. For this study, considering that the interview is supplementary to the questionnaire and is treated as qualitative data to fill any gaps in the questionnaire, 12 professionals were considered sufficient to be assessed. The interview sample comprised five professionals representing the public organisations, three engineering consultants representing the external private organisations working with public projects, and four contractors engaged by the public organisations, sufficient to represent the different groups of respondents. All respondents had
respondents were also mainly involved in the large-scale engineering projects (roadworks, bridges, dams) in which the public sector supply chain is widely involved in. The 12 respondents were sufficient to provide an overview of the current real-world scenario of public sector projects, in identifying critical pathogenic influences.
An invitation letter, participant information sheet and a consent form for signature were given to the respondents before the interview was conducted. The interviewees were given the option to withdraw from the study at any point. Names and personal information remained anonymous throughout the data collection and analysis.
3.5.3 Validity
Validity is defined as the extent to which an instrument measures what it purports to measure (Kimberlin and Winterstein, 2008). Validity and reliability estimates, as Messick (1994) suggests, “should be uniformly addressed for all assessments because they are not just measurement principles, they are social values that have meaning and force outside of measurement wherever evaluative judgments and decisions are made” (p. 13). Using mixed methods research, a panel of experts could provide the data that would allow the quantification of the consensus of those social values that are key to the audience at hand (Newman et al., 2013). Hence, in order to attain validity for this study, a panel of experts was used to pre-test the survey instrument. The validition process is iterative as the experts provide feedback, the literature is reviewed, and consensus is sought (Cronbach, 1970, 1971; Haynes et al., 1995; Newman et al., 2003). Strategies to address issues: face, content, construct, internal and external validity, are discussed below.
Face Validity
Face validity involves the evaluation of an instrument’s appearance by a group of experts to determine the extent to which the contents have actually been translated into meaningful constructs (Trochim, 2005). It is useful in establishing an instrument’s ease of use, clarity and readability (Burton and Mazerolle, 2011). Lawshe (1975) suggested a minimum of four experts for pre-test to ensure validity. In this study, five experts (three professionals from the public organisations and two researchers) were invited to participate to provide recommendations to improve the instrument. These panelists from the construction industry were deemed to be experts for the purposes of this study as each practitioner has more than 10 years of experience in the field, and both researchers possess a PhD degree. Their initial
comments on the conceptual framework are discussed Section 3.6.1 and their feedback on the survey instrument Section 3.6.2 (Table 3.6).
Content Validity
Content validity essentially addresses how representative instrument items represent the content or subject matter that the instrument seeks to measure (Newman et al., 2013). It can be estimated either qualitatively, through oral indication by a panel of experts who judge the appearance, relevance and representativeness of the survey’s elements (Netemeyer et al., 2003); quantitatively, by quantifying the degree of consensus about the survey instrument among the experts (Newman et al., 2013); or by using a combination of both methods. For this study, content validity was established using both methods in two stages. The first stage, qualitative, was the intensive review of the literature; the second stage quantified the validity of a scale, requiring the panel of experts to rate the relevance of each item.
Widely used methods to quantify content validity for multi-item scales, as in this study, are by computing Lawshe’s (1975) content validity ratio (CVR) and Waltz and Bausell’s (1983) item-level content validity index (I-CVI). The former involves requesting the panel of experts to rate items according to the degree of relevance of the items in assessing the construct that they are assigned to. A 3-point rating scale was used; 1=Irrelevant, 2=Important, but not essential, and 3=Essential. For each item, a CVR was computed, the proportion of experts that considered the items to be important or essential. The formula to calculate the ratio (Lawshe, 1975) is:
Equation 1: ne- N / 2