212
5.4.1 Description of Step One
Step One (refer Figure 5.1) of the two-step preliminary analysis is referred to here as scale
evaluation, in that the data underwent a number of evaluation procedures, such as correlation
analysis, EFA and reliability analysis, before being retained for CFA.
As recommended by
Comrey (1978), pre-analysis of the suitability of the data for factor analysis is essential, in
that the data matrix must be inspected to ensure that a sufficient number of significant
correlations exist. Hair et al. (2006) points out that the data matrix can be initially tested via
measures such as the Kaiser-Meyer Olkin (KMO) measure of sampling adequacy and
Bartlett’s Test of Sphericity. KMO compares the size of the observed correlation coefficients
with the magnitude of the partial correlation coefficients and is calculated as a value between
0 and 1. A value close to 1 indicates a large number of interrelations among the variables.
Similarly, the Bartlett’s Test for Sphericity was used to test for statistical probability that the
correlation matrix had significant correlations among at least some of
the variables computed
and was indicated by a significance level less than .05 (Hair
et al.
2006).
The next stage of Step One (refer Figure 5.1) involved a closer inspection of the bivariate
correlations contained within the matrix. At this point, as indicated by Hair et al. (2006),
items not exhibiting a substantial number of correlations greater than .30 were removed as
they were not considered strong enough to be appropriate for factor analysis. Similarly, items
correlating at above .90 were also deleted so that overfitting
of the data did not occur
(Tabachnick and Fidell 1996), as is often the case when items are measuring the same thing.
Having retained only those items with correlations between .30 and .90, the data was then
considered sufficiently robust for conducting EFA.
213
Following data verification for EFA, the next stage of Step One (refer Figure 5.1) involved
conducting EFA to determine the factor structures of the data and loadings of items. In this
regard, the empirical assessment of construct validity was evaluated using contemporary
analytical guidelines recommended by Anderson and Gerbing (1988), Hair et al. (2006), and
O’Leary-Kelly and Vokurka (1998) through the examination
of factor structures,
unidimensionality and internal consistency. EFA was conducted via principal-components
factor analysis using varimax rotation, chosen on the basis that all factors were expected to be
undimensional. At this point, a similar procedure to Shi and Wright (2001) was followed
whereby factors with eigenvalues greater than 1 were identified and
items with factor
loadings less than .50 were deleted. In addition, any items exhibiting cross-loadings greater
than .04 were also removed from the analysis (O’cass and Fenech 2003). The data was then
ready for the final issue addressed in Step One of Figure 5.1, that being reliability analysis.
The final stage of Step One (refer Figure 5.1) involved conducting reliability analysis to
determine if the scale has ability to provide consistent results. Reliability tests include test-
retest method,
equivalent forms, split-halves method and internal consistency method. Of
these methods, the internal consistency method requires only one administration of the
instrument and is operationalized as the degree of inter-correlations among the items that
constitute a scale (Nunnally 1978), estimated via Cronbach’s alpha. While Sureshchandar et
al. (2001) argue that internal consistency is established if the alpha value is greater than .70.
Some advocate that the alpha value greater than .60 may be sufficient depending on the
number of the items in the scale or in the case of exploratory research (Hair
et al.
2006).
Therefore, at this stage, all scales were tested using Cronbach’s alpha
in order to determine if
they were, in fact, reliable measures of the constructs. Items meeting the alpha criteria of .70
214
(Sureshchandar
et al.
2001) were, at this point, considered reliable indicators of the constructs
and further analysis, as shown in Step Two of Figure 5.1, was initiated.
Do'stlaringiz bilan baham: 
