Chapter 4. RESEARCH DESIGN AND METHODOLOGY
4.6. Data analysis techniques
The data gathered was analysed qualitatively and quantitatively (Subsection 4.6.1 and 4.6.2). Qualitative techniques were applied to make sense of meanings. Content analysis was used to analyse the semi-structured interviews whilst descriptive and inferential statistics were used for the analysis of the questionnaire survey. Statistical Packages for Social Sciences (SPSS) software was used to analyse the data quantitatively.
4.6.1. Qualitative data analysis via content analysis
One of the challenges in qualitative research is data analysis. Literature describes a number of tools and techniques (Miles and Huberman, 1994) that must be selected based on the objectives of the research. Since the research at this stage was more exploratory than confirmatory in nature, “content analysis” for analysing the interview transcripts was chosen. Fourth-nine interviews were conducted and there were over a hundred pages of interview transcripts to analyse. The collected data was coded and analysed using content analysis, based on the guidelines provided by Gillham (2000), and Strauss and Corbin (1998). Data from the interviews were analysed immediately after each interview to identify constant and regular themes. The inductive process was used as Yin (1994) suggested, looking for consistent themes that emerge from the data, and was supplemented by the deductive process to ensure the data was not misconstrued or misinterpreted. This overall iterative approach was used successfully within an interpretive methodological paradigm to identify clusters and emergent themes or categories whilst maintaining the richness of the data (Huberman and Miles, 2002). Manual coding was used in this research instead of computerised coding. Manual coding entail reading text and extracting user-specified information deemed relevant to its content and / or context (Carley, 1990). However, as Morris (1994) claims, manual coding in content analysis is more reliable, but time consuming. The following are the main reasons for coding data manually, in this research study.
Number of interviews conducted was fairly low (49)
The interviewees were asked a different number of question (refer to Appendix B). The different groups of participants used different words on the same subject e.g. information management for word knowledge management)
The findings from the review of literature were also taken into account when analysing the content of the transcribed interview data. This allowed synthesising the literature to identify and divergence of theory vs. practices, if any.
4.6.2. Quantitative data analysis via statistical techniques of analysis
The quantitative aspects of the responses to the questionnaire were analysed using SPSS version 16, in the University’s mainframe computer. SPSS was chosen for this research as its software was the easiest to learn and use, and it had a data editor that resembled. SPSS providing familiarity to the researcher. It could also perform most of the general statistical analysis required, which was well suited and adequate for this particular research. Another important reason why SPSS was chosen was because it could easily create and customise graphs that could be pasted into other documents such as Word, Excel or Powerpoint. The data collected from the survey was analysed using non- parametric statistical or ‘ranking’ tests. These differ from parametric tests in that, the assumption made, or conclusion drawn, are regardless of the shape of the population, whereas parametric tests assume that the scores are drawn from a normally distributed population (Siegal, 1956). All usable response data was analysed using these tools. Several types of statistical analysis were undertaken (Table 4.23). A descriptive analysis was performed in order to describe the data in a meaningful way, for example, the number of employees in an organisation, or the number of years of experience a participant had in the construction industry. Descriptive statistics such as mean, percentages and frequencies were used in the study.
An inferential statistical analysis (e.g. Spearman correlation) was carried out to check whether the scores could be inferred to the general population (all contractors in Malaysia). In this study, the Spearman correlation test was used as follows:
The Spearman’s rank correlation coefficient was used to determine the relationship between two quantitative variables measured in an ordinal scale. For example, the relationship between knowledge sharing approaches and the size of the organisation.
As a result of the non-normal nature of the data distribution, a Spearman’s Rho correlation was used instead of the Pearson correlation, as Salkind (2004) stated, “when the data is ordinal, the suitable test for correlation is Spearman’s rank coefficient”.
Correlation coefficients indicate the strength of the association between the variable under investigation. The sign (+ or -) indicates the direction of the relationship. The value can range from -1 to +1, with +1 indicating a perfect positive relationship, 0 indicating no relationship, and -1 indicating a perfect negative or reverse relationship (Hair et al, 2006). The interpretation for the value of a correlation coefficient can be found by referring to the work undertaken by Salkind (2004) (Table 4.22). Table 4.23 summarises the data analysing method used for this study.
Table 4.22 : Interpreting a correlation coefficient
Size of correlation Coefficient general interpretation
.8 to 1.0 Very strong relationship .6 to 0.8 Strong relationship .4 to 0.6 Moderate relationship .2 to 0.4 Weak relationship .0 to 0.2 Weak or no relationship
Source: Salkind (2004)
Table 4.23 : Summary of the data analysis methods used
Data collection techniques Analysing method Analysing techniques Software Semi-structured
interview Coding Content analysis Manually
Questionnaire survey Descriptive statistics Comparison of mean Calculation of frequency Cross tabulation SPSS 16 Inferential statistics (non parametric test) Normality Spearman’s coefficient of correlation Kolmogorov-Smirnov test SPSS 16 SPSS 16
Normality
In order to test for the normal distribution of response data, a Kolmogorov-Smirnov test for all dependent and independent variables was conducted. In this study, all of the items were confirmed not to be normally distributed, therefore, a non-parametric test was used. Although normality of variables is not always required for analysis, the solution is usually improved if the variables are all normally distributed (Pallant, 2001). Since the variables indicated a significant result (sig. value ≤ 0.05) and, ordinal data was used in this study, non-parametric techniques were considered more suitable for the analysis. Table 4.24 shows the result of the normality test on the variables. The significant p-value is less than 0.05 indicating the data is not normally distributed.
Table 4.24 : Tests of normality
Items Kolmogorov-Smirnova Shapiro-Wilk
Statistic df Sig. Statistic df Sig.
Formal approaches .060 384 .002 .987 384 .002 Informal approaches .081 384 .000 .988 384 .003 Challenges in setting up .066 384 .000 .987 384 .002 Challenges in implementing .049 384 .027 .991 384 .022 Readiness to set up .067 384 .000 .993 384 .091 Readiness to implement .086 384 .000 .989 384 .006 Important of knowledge sharing .083 384 .000 .975 384 .000 Contribution of knowledge
sharing .074 384 .000 .977 384 .000
Organisational Structure .113 384 .000 .961 384 .000 Organisational Culture .085 384 .000 .974 384 .000 Human resource practices .121 384 .000 .962 384 .000
a. Lilliefors Significance Correction
The elaboration and findings of the data collected from the questionnaire will be discussed further in Chapter 5-9.