In utilizing a PLS path modeling technique, the similar two-step procedure normally conducted in structural equation modelling (SEM) was followed (Anderson and Gerbing, 1988). Through this technique, results of both confirmatory factor analysis of the model and path effect were obtained. In completing this procedure, a model validation analysis was also performed.
Results of the measurement model using a PLS algorithm (300 maximum iteration, standardized values and centroid weighting scheme) suggest that all constructs that were made up of reflective indicators are reliable with loadings all above the desired level of 0.70 (see Table 2). In SEM, a research model is said to be valid when both convergent and dicriminant validity have been achieved. Table 3 and Table 4 provide the results of these validity tests. The research model demonstrates a strong convergent validity as the latent constructs with reflective items have high composite reliability (CR) and communality. Similarly, as can be seen in Table 4, the square roots of all average variance extracted (AVE) were greater than inter-construct correlations except for the constructs of FAM and AT_BR. As the difference between the square root of AVE for FAM is less than the correlation between the two constructs merely by 0.003, the present study contends that the research model has achieved desirable discriminant validity. Using a bootstrapping technique (500 re-samples), a test on the structural model was conducted to assess the effect of each causal path, thus testing the stipulated hypotheses. As can be seen in Figure 4, except for the TRS^AT_BR path, all other causal paths are significant at 5% level of significance. Therefore, majority of the hypotheses were supported suggesting that attitude plays a mediating role for investors to decide in investing in shares of a particular company. As can be seen in Figure 4, majority of the stipulated hypotheses were supported, except for the relationship between TRS and AT_BR, while that of RISK and AT_BR turned out to be positive relationship, albeit just significant at 5% confidence level. In summary, only H2a and H3a were not supported.
Finally, the research model was analyzed in terms of its model fit. As can be seen in Table 5, the research model explains 20.0% variation in the construct of ‘intention to invest’. The Goodness-of-Fit statistic appears to be high at 0.5561 (Ringle, Wende and Will, 2009), while fit statistics for both outer model (H2) and inner model (Q2) are also high. In obtaining the cv-communality (H2) and cv-redundancy (Q2), a blindfolding procedure was run in SmartPLS. The results in Table 5 show the research model having better a measurement model (H2 = 0.770820) than the structural model (Q2 = 0.385374). As indicated by Chin, a Q2 value of greater than zero has predictive relevance, so Q2 of 0.385374 is considered far greater than this heuristic. Overall, the research model exhibits acceptable fit and high predictive relevance.
Table 2: Loadings of Reflective Indicators
|TRS 5||0.353327||INT ШУ|
|TRS 6||0.359617||MT INV1 INT INV2 INT INV3 INTJNV4 INTJNV5 Г\Т_ШУ6||0.9+5293 0.953334 0.9+1371 0.952110 0.330 552 0.909222|
Table 3: Convergent Validity
|No. of ileus||CR||AIT||Cfl nimii nihh~|
|FISK||4||09104S1||0.71907 S||0.71907 S|
Table 4: Discriminant Validity
Table 5: Model Fit Statistics
|Gustract||Structural M od el||Made! Quality|
|RISK||0.22SS73||(0.71907 S)||0.056751||0 719422||(0L163378)|
Figure 4: Results of Structural (inner) Model