The reliabilities of the different measures included in the model ranged from 0.84
to 0.95 thus indicating good internal consistency.
Since reliability does not ensure validity (Anderson and Gerbing 1982, Hair et al
1998) hence more reliable measures like construct/composite reliability provide the
researcher with greater confidence that the individual indicators (i.e., items loading on the
latent construct) are all consistent in their measurement. Construct/composite reliability
can be expressed in mathematical form as the sum of the standardized loading square
divided by the summation of the sum total of standardized loading square and indicator
measurement error (Hair et al 1998).
4.2.6.3 Convergent and discriminant validity
Convergent and discriminant validity are both considered subtypes of construct
validity. Convergent validity is defined as the degree to which concepts that should be
related theoretically in fact are interrelated (Campbell and Fiske 1959). Discriminant
validity is the degree to which concepts that should not be related theoretically are,
in fact, not interrelated in reality (Campbell and Fiske 1959).
In order to show that theoretically similar constructs are highly intercorrelated, in
other words similar constructs converge, studies have indicated that their average
variance extracted (AVE) estimates should be higher than 0.50 (Fomel and Larcker 1981,
Hair et al 1998). AVE greater than 0.50 is an indicator of convergent validity. Table 4.6
shows the AVE for all the five constructs used in this study. Average variance extracted
can be calculated as the summation of the standardized loading square divided by the sum