This is the original estimate path model.
This is the bootstrapped path model.
This is a path model showing only the measurement model components.
Ideally, outer loadings (l) should be \(\geq\) 0.70. Loadings below 0.40 are unacceptable. AVE should be > 0.50.
Both Cronbachβs a and composite reliability rc should be \(\geq\) 0.60 and \(\leq\) 0.90. The upper threshold of acceptability is 0.95.
Discriminant validity is evaluated using the heterotrait-monotrait ratio (HTMT). The HTMT bootstrap confidence interval should not contain 1.
All indicators and constructs show sufficient convergent validity and discriminant validity. The HTMT bootstrap CIs include 1 for all the Distrusting Beliefs lower-order constructs. As they are associated with the same higher-order construct, this is permissible. All constructs except Personal Moral Norm also show internal consistency reliability, with the composite reliability for Personal Moral Norm being too high. This means that one indicator should be deleted.
COPN2 shows the highest intercorrelation with the other indicators and should therefore be deleted. All other reflective indicators can be retained as-is.
To evaluate convergent validity, a redundancy analysis is conducted. The path coefficient \(\beta\) should be \(\geq\) 0.70 and RΒ² should be \(\geq\) 0.50.
Reiterating: For sufficient convergent validity, the path coefficient should be \(\geq\) 0.70 and RΒ² should be \(\geq\) 0.50.
The variance inflation factor (VIF) should be < 5, ideally \(\leq\) 3.
Indicator weights should be significant (t \(\geq\) 1.65, bootstrapping CI does not include 0). Otherwise, the loading l should be \(\geq\) 0.50. Any positive indicator weight implies relevance.
Convergent validity is insufficient only for Perceived Self-Efficacy. There are no issues with collinearity as indicated by the VIF. All weights are significant and relevant. As further indicator deletion is unlikely to improve convergent validity for Perceived Self-Efficacy, it should be replaced by the single-indicator reflective measure.
Convergent validity cannot be evaluated.
The variance inflation factor (VIF) should be < 5, ideally \(\leq\) 3.
Indicator weights should be significant (t \(\geq\) 1.65, bootstrapping CI does not include 0). Otherwise, the loading l should be \(\geq\) 0.50. Any positive indicator weight implies relevance.
For the formative higher-order constructs, there are no issues with collinearity (VIF above 3, but below 5). The weight for Distrusting Beliefs Integrity is not significant, but the loading is sufficiently large. All other weights are significant and relevant. The formative higher-order constructs can be retained as hypothesized.
To improve internal consistency reliability of Personal Norm, COPN2 should be deleted. Because of a lack of convergent validity, the formative Perceived Self-Efficacy construct should be replaced by a single-indicator (CORB10) reflective Perceived Self-Efficacy construct.