CHAPTER 5: RESEARCH RESULTS
5.6 FACTOR ANALYSIS APPLIED TO THE FIVE BOUNDARY QUESTIONNAIRE CATEGORIES
CATEGORIES
The researcher was concerned with the number of emotional awareness sub-constructs that were represented and how well each item represented these domains.
First of all, the factor structure of all 70 selected Boundary Questionnaire items was examined (see Annexure B). The interrelationships among the Boundary Questionnaire variables were subsequently summarised to conceptualise the component ‘emotional awareness’ and group it into meaningful factors. Importantly, conducting a Principal Component Analysis (PCA) was not advisable as the technique assumes uncorrelated factors (Preacher, Zhang, Kim & Mels, 2013), consequently an exploratory factor analysis was conducted.
5.6.1 Exploratory factor analysis (EFA)
Since exploratory factor analysis (EFA) methods have widespread use in psychological research, the researcher conducted EFA to establish linear variates within the data and to discover degrees of difference in aircrew’s emotional awareness (Babbie, 2013). EFA was performed by analysing the covariance matrix as it produces a better-defined factor structure (Field, 2013; Preacher, Zhang, Kim & Mels, 2013).
EFA was performed using the maximum likelihood (ML) method of extraction to identify underlying emotional awareness variables. The information gained from the interdependencies between the observed variables was used to reduce the set of variables in the dataset and group the remaining traits to find the relevant latent emotional awareness factors. Results are reported in detail in the following sections.
5.6.2 KMO measure of sampling adequacy
A Kaiser Meyer-Olkin (KMO) measure of sampling adequacy was performed to determine how much variance was common variance between variables. A rotation with Kaiser normalisation was performed to aid the interpretation of the dimensions. Variables were sorted by their loading size. The diagonal elements should all be greater than 0.7 for a given pair of variables to ensure factor analysis yields distinct and reliable factors (Field, 2013; Pallant, 2006). In this study, values between 0.7 and 0.9 were used. The output magnitude loadings from the five selected Boundary Questionnaire categories ranged between 0.703 and 0.749. The dimension solution explained between 19% and 31% of the common variance, with the individual dimensions ranging between 7% and 19% (see Table 5.6).
Table 5.6 Variance accounted for by the emotional awareness constructs BQ
category
Direct oblimin rotation converged in iterations KMO Eigen- value % of variance for each factor Total variance explained Category 1 26 .749 3 1 7 18 25 Category 3 26 .729 1 2 1 7 18 6 31 Category 5 28 17 .731 1 1 1 12 9 7 28 Category 7 4 .754 2 19 19 Category 8 27 .703 2 1 14 9 23 (N = 173)
Initially, the researcher performed a promax oblique rotation, despite being aware that the sample size of 173 was barely adequate and these factors all had eigenvalues greater than 2. The data was then fitted to a factor pattern structure using oblimin rotation with Kaiser normalisation to aid the interpretation of the dimensions. Factor extraction was performed by means of Principal Axis Factoring. Where the factors correlated, sums of squared loadings could not be added to obtain a total variance. The individual dimensions only contributed between 5.6% and 19.8% variance each (Appendix A).
5.6.3 Factor analysis with direct oblimin rotation with Kaiser normalisation
As the pattern matrix is a simpler method to interpret, the researcher selected the factor loadings of the pattern matrix to decide on the factors to retain (Field, 2013). Promax rotation is used for large datasets and when the researcher expects the factors to be independent (Field, 2013). In this case, the direct oblimin rotation method was selected due to the limited sample of 173 aircrew and as the factors are bound not to be independent as the items were developed to measure one underlying construct. Oblique factor rotation was performed to discriminate between factors by effectively rotating the axes such that variables loaded maximally on only one factor (Appendix B). With direct oblimin, the degree to which factors can correlate is determined by the value of the delta constant which was kept at the recommended zero default setting (Field, 2013). The closer the communalities are to 1, the better the factors are at explaining the data (Pallant, 2006). These factors all had eigenvalues larger or close to 0.7, which is within the acceptable limit for small samples (Field, 2013) as can be seen in Appendix B.
5.6.4 Factor extraction
Item cluster extraction (ML) was performed. First, the linear components within the dataset were determined by calculating the eigenvectors of the 70 Boundary Questionnaire items. Then the eigenvalue greater than one option was selected. The eigenvalues associated with each factor represented the common variance explained by that factor. Communality before extraction is the proportion of common variance, whereas communality after extraction is the amount of variance in each of the factors identified –
which can be explained by the retained factors (Field, 2013). The researcher used the factor correlation matrix to assume some independence between the factors. The fact that correlations existed, provided evidence that the emotional awareness constructs were interrelated (Appendix B).
5.6.5 Naming of factors
The researcher analysed the content of the items after rotation that loaded highly on the same factor to identify common themes. The factor loadings were used as the basis for inputting labels to the explanatory dimensions of the new emotional awareness sub- scales. The eight extracted factors were labelled to provide a description of the underlying meaning of each emotional awareness sub-construct to enhance the interpretation of the results. The researcher named the extracted factors independently without reference to any personality model or EQ-i20 traits as emotional awareness sub-scales: involved, blend, imaginative, precise, flexible, structured thought, unstructured thought and openness (Table 5.7).
Table 5.7 Factor naming and emotional awareness trait description
BQ Category EA sub-scales BQ items retained Sub-scale description
Category 1: Sleeping, walking, dreaming
Openness 72, 92, 130 Friendly, relaxed, confident,
extraverted and open to
experience
Unstructured 1, 25, 37, 60, 82, 112, 119 Disorganised and careless Category 3:
Thoughts, feelings, moods
Involved 15 ,51 ,84, 94, 102, 106 Conscientious and focused
attention
Blend 3,62 Agreeableness and
cooperative
Imaginative 74,127,132 Creative and innovative,
resources required for
problem-solving abilities Category 7:
Neatness, exactness, precision
Exactness 43,55,76,96,108,121 Resourcefulness, complex
questioning and detail oriented for decision making
Category 8: Edges and lines
Flexible 67,142,145 Emotional stability, agreeable
and cooperative
Structured 44, 137 Conscientious, organised and
careful
As a sample size of 200 or more is required to reliably use a scree plot as a factor selection criterion (Field, 2013), an SPSS scree plot graph is not depicted (the current study had a limited sample size of 173). Once the factors had been extracted (ML), the degree to which variables loaded on the factors was determined. Factor rotation was used to discriminate between factors and the axes were rotated such that variables were loaded maximally to only one factor (Field, 2013). Factor analysis was hence performed to generate new emotional awareness scales in the aviation context from the item pool. These scales will be discussed in more detail in the next section.
5.7 DESCRIPTIVE STATISTICS FOR THE EMOTIONAL AWARENESS SUB-