Understanding consumer online shopping behaviour from the perspective of transaction costs

5.4 PRELIMINARY METHOD OF DATA ANALYSIS 
This section presents the preliminary evaluation of the data via correlation analysis, EFA, 
CFA and reliability estimates.

---

209
Each variable was visually inspected for normality, skew (the degree of symmetry about the 
mean) and kurtosis (the degree of flatness or peakness of a distribution), and the presence of 
outliers. Histograms were deemed appropriate at this stage to provide the best “overall” 
picture of each variable across a small range of scores (1 to 7). In addition to visual 
inspection, each variable was analysed via test of skewness and kurtosis. Overall the data did 
not appear to be problematic, with all statistics falling within acceptable ranges. For example, 
skew and kurtosis values were between -2 and +2, indicating that the frequency distributions 
were considered normal (Pallant 2010). Similarly, the data was inspected for the presence of 
outliers and none were detected. For example, scores did not fall outside the range of 3 to 4 
standard deviations which is the recommended criteria for detecting outliers for large samples 
(Hairs
 et al.
1998). The means, standard deviations, skew and kurtosis values for each of the 
variables appear in Appendix D.
Having inspecting the data for anomalies in normality, the next step was to analyse the data 
to access the factor structures and internal consistency of the items. Two factor analysis 
techniques were identified as being appropriate for data analysis at this stage (e.g., EFA and 
CFA). Firstly, EFA is designed for the situation where the relationships between the observed 
and latent variables are not predetermined, thus warranting an exploratory approach to data 
analysis in order to discover the underlying factors. While EFA is the most conventional 
approach evident by its extensive use in marketing and consumer behaviour research (Liang 
and Huang 1998, McColl-Kennedy and Fetter Jr 1999, Chenet
 et al.
2000, Grace and O’Cass 
2001), this approach has certain limitations.

---

210
Firstly, and most importantly, EFA assigns items to factors purely on an the basis of which 
factor they load substantially, therefore, it is possible for an item to have a significant loading 
on more than one factor which, in turn, effects the identity or distinctiveness of the factor 
(Sureshchandar
 et al.
2001). Furthermore, EFA items load onto factors on a purely statistical, 
rather than theoretical, basis thereby affecting the valid identity of the factors. Secondly, as 
noted by Chandon et al. (1997), an explicit test of unidimensionality is not provided by EFA 
as each factor is defined as a weighted sum of all the available items in that dimension.
CFA, on the other hand, overcomes the abovementioned limitations in that the researcher 
specifies a model a priori, and tests the hypothesis that a relationship between the observed 
and latent variables does exist. This is extremely robust test when the researcher can postulate 
a model that draws its logic from research outputs in which reliable indicators of factors have 
previously been determined (Deeter-Schmelz
 et al.
2000, Sureshchandar
 et al.
2001). 
Furthermore, CFA offers a rigorous evaluation of dimensionality and internal consistency as 
each factor is related to only a subset of indicators (Chandon
 et al.
1997, McGee and Peterson 
2000). This being the case, and due to this study using both pre-existing measures and 
measures developed specifically for this study (e.g., items pertaining to constructs, such as 
Internet access availability and environmental uncertainty, e-service quality were adapted 
from existing measures), a two-step approach, which includes both EFA and CFA, was 
deemed appropriate. A similar procedure was adopted by Chandon et al. (1997) and Shi and 
Wright (2001) and follows two distinct steps, as shown in Figure 5.1.
The following 
discussion describes the two-step process (shown in Figure 5.1) prior to the presentation of 
results.