First, we test for nonlinearity by applying Ramsey’s RESET test for functional form to each partial linear model of the SEM. As quadratic effects of higher-order constructs cannot (yet) be estimated using PLS-SEM, we also apply the test to the corresponding partial linear models of the proxy SEM without higher order constructs. If the test statistic is significant, the respective partial regression might be nonlinear.
Using p = 0.05 as a threshold, there are potential nonlinear effects in the partial regression equations for the influence of Response Beliefs, Threat Beliefs, and Subjective Norm on Behavioral Intention. The same is potentially true for the effects of Distrusting Beliefs and Knowledge on Response Beliefs, and for the effects of Distrusting Beliefs and Knowledge on Threat Beliefs.
To test for nonlinearity, one can model an interaction term of the exogenous construct with itself.
For the alternate model with Response Beliefs as a quadratic influence, the effect size is smaller, as well as the t-value. This points to a linear effect of Response Beliefs on Behavioral Intention.
For the alternate model with Threat Beliefs as a quadratic influence, the effect size is smaller, as is the t-value. This points to a linear effect of Threat Beliefs on Behavioral Intention.
The Subjective Norm alternative model can only be assessed using the orthogonal and product-indicator approach, although the two-stage approach would be methodically correct. Neither of those alternate approaches shows a significant effect of a Subjective Norm quadratic term. This points to a linear effect of Subjective Norm on Behavioral Intention.
The Distrusting Beliefs - Response Beliefs alternative model can only be assessed using the orthogonal and product-indicator approach, although the two-stage approach would be methodically correct. The product-indicator approach produces a significant effect for the quadratic term, however, it is smaller and less significant than the effects of the linear term in both the original and the proxy SEM. This points to a linear effect of Distrusting Beliefs on Response Beliefs.
For the alternate model with Knowledge as a quadratic influence, the effect size is smaller, as well as the t-value. This points to a linear effect of Knowledge on Response Beliefs.
The Distrusting Beliefs - Threat Beliefs alternative model can only be assessed using the orthogonal and product-indicator approach, although the two-stage approach would be methodically correct. Both approaches produce a significant effect for the quadratic term, however, the effect sizes are significantly smaller and less significant than the effects of the linear term in both the original and the proxy SEM. This points to a linear effect of Distrusting Beliefs on Response Beliefs.
The effect of the quadratic term is miniscule and not significant. This points to a linear effect of Knowledge on Threat Beliefs.
Using the available tests, we did not find evidence for a quadratic influence of any of the exogenous variables.