Figure 1.
Stages of Research Methodology.
Table 1.
Theoretical Constructs Description and Source.
Figure 2.
The Figure shows the items ‘used’ by Eureqa through symbolic regression setting each of the five ENG items as dependent variables (obtained using the whole data set).
Each red bar shows the number of times that particular item appeared. Clearly, the high correlation of the ENG variables among themselves is found, however in the accumulated results we observe the conspicuous role of other items as shown in Figure f. N.B. The full results for each item (n = 69) and each construct can be made available as supplementary material upon request.
Figure 3.
Data Set A – Network found as a result of the application of the model finding optimization software on each variable as a target.
This is a directed graph in which an arc connects a variable that appears as an input in a model for a target variable. There is a clear association of variables that belong to the same construct as shown by their labels but other “functional constructs” combining items from several different theory-driven consructs emerge. The size of a node is proportional to its centrality in the network measured by ‘node betweenness centrality’ A green line represents a negative relationship found and a red line a positive relationship between the two nodes (items).
Figure 4.
Data Set B – Directed network linking the variables with modularity classes computed with the same methodology as Figure 3.
The community that contains variables of the constructs of Self-brand congruency and Brand Involvement (as well as one of the Usage Intensity) is again an example of a functional construct as found by the modularity optimization algorithm. The size of a node is proportional to its centrality in the network measured by ‘node betweenness centrality’ A green line represents a negative relationship found and a red line a positive relationship between the two nodes (items).
Table 2.
Comparison of communities in Set A and Set B.
Table 3.
Comparison of communities in Set A and Whole Data Set.
Table 4.
Comparison of communities in Set B and Whole Data Set.
Figure 5.
Whole Data Set – A network generated in the same manner as presented in Figure 3 and Figure 4.
Here the majority of the Flow variables group together with all four Hedonic variables, as in Figure 3. Furthermore, most of the Brand Involvement and Self-Brand Congruency are grouped together, consistently with Figures 3 and 4. Again, the size of a node is proportional to its centrality in the network measured by ‘node betweenness centrality’ A green line represents a negative relationship found and a red line a positive relationship between the two nodes (items).
Table 5.
Adjusted Rand Index, Fleiss' Kappa and Cramer's V Results.
Table 6.
Comparison of Modularity Communities in Set A with Literature-Driven Constructs.
Table 7.
Comparison of Modularity Communities in Set B with Literature-Driven Constructs.
Table 8.
Comparison of Modularity Communities in Whole Data Set with Literature-Driven Constructs.
Table 9.
Cramer's V Results for Functional vs. Theoretical Construct Comparison.
Table 10.
Comparison of Arcs between Data Sets with safe reduction of the number of Arcs.
Figure 6.
Positive cycles found in the ‘reduced’ Whole Data Set graph.