The proposed model and hypothesis testing was
conducted using PLS (Partial Least Squares) Version 3.0
[15] and AMOS 6.0 to test the measurement model
because PLS and AMOS can be regarded as complementary.
Whereas covariance-based SEM tools such as
LISREL and AMOS use a maximum likelihood function
to obtain parameter estimates, the component-based
PLS uses a least squares estimation procedure, allowing
reflective latent constructs, while placing minimal demands
on measurement scales, sample size, and distributional
assumptions [15]. PLS reports internal
composite reliability and average variance extracted
(AVE) for content validity and discriminant validity.
Based on covariance analysis, like LISREL, AMOS is
more confirmatory in nature and it provides various
overall goodness-of-fit indices to assess model fit for
convergent validity [11].
SEM is a flexible and powerful extension of the
general linear model. Like any statistical method, it
features a number of assumptions. These assumptions
should be met or at least approximated to ensure
trustworthy results. A good rule of thumb is 15 cases per
predictor in a standard ordinary least squares multiple
regression analysis [87]. Since SEM is closely related to
multiple regression in some respects, 15 cases per
measured variable in SEM is reasonable. Consequences
of using smaller samples include more convergence
failures (the software cannot reach a satisfactory
solution), improper solutions (including negative error
variance estimates for measured variables), and lowered
accuracy of parameter estimates and, in particular,
standard errors — SEM program standard errors are
computed under the assumption of large sample sizes.
Thus, our sample size of 325 is more than the minimum
number of sample size, 270 (i.e., 18 items multiplies by
15), for the AMOS estimation procedures.
Table 2 shows the internal consistency reliabilities
and correlations among constructs based on PLS
analysis. As recommended, the internal consistency
reliabilities were all higher than .7 without exception
(the minimum was .82), and the diagonal elements
(square root of the variance shared between the
constructs and their measures) were all higher than
.707 (the minimum was .77) and also higher than
correlations between target constructs and other constructs
without exception.