As a step further, the confirmation of discriminant validity is crucial for assessing the scientific data's authenticity. Ketchen [104] suggested that the discriminant validity required the cross-correlations between latent constructs (LTCs) to be less than their reflective (self) correlations. In the present case, cross-correlation values of all constructs were less than their reflective correlation values (Table 3). Following Hair et al. [105], the discriminant validity is satisfied based on this criterion. Moreover, an advanced discriminant validity test by Henseler et al. [102] is used for further verification. This is known as the heterotraitmonotrait ratio (HMR) of correlations. It calculated the pairwise cross-correlations between the constructs (Table 4). At 90% confidence interval, all the cross-correlations are found within the range of confidence interval, demonstrating that the discriminant validity is established. HMR is the most recent test and it has been reported to perform better than the Fornell-Larcker [102] criterion. Since the discriminant validity is proved valid, the path analysis can be carried out.

**Table 3.** Discriminant validity results based on Fornell and Larcker [106] criterion.


**Table 4.** Discriminant validity testing based on the Heterotrait-Monotrait Ratio.


Notes: CI: confidence interval. The brackets [] contain the confidence intervals at 90%.

#### **4. Main Results**

The path modeling-based results are shown in Table 5 and Figure 3. The structural model was evaluated after the measurement model were proven to be reliable and efficient. As a primary condition, the R-square was generated for each of the constructs. R-square measures the variations captured by each of the non-exogenously discovered constructs to

communicate the structural model's predictive capacity. As a rule of thumb, a magnitude no less than 0.25 has been proposed to be an average score, whereas a magnitude below 0.13 is insufficient to pass this criterion in the behavioral sciences. In contrast, the badness of outcome is exhibited by any score less than or equal to 0.03 [48]. In the present case, the R-square value is 0.807, which is well above 0.25, satisfying the path model's first criterion (Table 5).


**Table 5.** Path modeling analysis and post-estimation criteria results.

Notes: PC: path coefficient. \* *p* < 0.05, \*\* *p* < 0.05, \*\*\* *p* < 0.01, VIF: variance inflation factor.

**Figure 3.** Path modeling-based estimation of coefficients. Notes: \* *p* < 0.10, \*\* *p* < 0.05, \*\*\* *p* < 0.01. Solid line denotes significant path, while dashed line denotes insignificant one. Source: Authors' elaboration.

Next, Stone–Geisser's Q-square criterion was used explore the LTCs' predictive relevance [107,108]. The non-negative range score reflects the LTCs' predictive relevance [102]. Further, the predictive relevance's relative impact is given by the degree of this criterion. A Q-square > 0.35 indicates that the exogenous constructs imparted adequate prediction for their respective endogenous constructs [97]. The magnitude of the measured Q-square (0.365) proved the accuracy and precision of the structural model. The path coefficients analysis is taken as a further prerequisite. In the structural model, an approximate path coefficient score above 0.1 indicates a significant contribution of a respective variable to the dependent variable [103]. After that, f-square is obtained, determining the effect size to characterize a construct's contributing ability. Based on the f-square score, exogenous constructs define the identified differences in endogenously defined LTCs [109].

The path modeling does not require the prior existence of a normal distribution principle. Alternatively, this principle is followed by the bootstrap-based parameter estimation method to evaluate the importance of external loading and ICFs' path coefficients. The bootstrapping method scrutinizes nearly 4 × <sup>10</sup><sup>4</sup> samples extracted from the initial sample using the "with replacement" alternative for estimating every bootstrapped sample. This bootstrapping procedure involves generating a probability distribution for manipulating the variances and standardized residuals. To assess the validity of path coefficients, the null hypothesis of *H*<sup>1</sup> = *H*<sup>2</sup> = *H*<sup>3</sup> = *H*<sup>4</sup> = *H*<sup>5</sup> = *H*<sup>6</sup> = *H*<sup>7</sup> = *H*<sup>8</sup> = *H*<sup>9</sup> = 0 was assessed against the alternative of *H*<sup>1</sup> = *H*<sup>2</sup> = *H*<sup>3</sup> = *H*<sup>4</sup> = *H*<sup>5</sup> = *H*<sup>6</sup> = *H*<sup>7</sup> = *H*<sup>8</sup> = *H*<sup>9</sup> = 0. For decision-making, the probabilities equal to or less than the statistical magnitude of 0.05 are considered significant at a 5 percent level [102]. To this end, the only null hypothesis retained was *H*<sup>6</sup> = 0, while the remaining were successfully rejected (Table 5). In other words, all the ICFs contributed to the WAPP of individuals, except for the moral values.

The path coefficients-based relative significance of the ICFs of individuals' WAPP is depicted in Figure 4. The ICF of peer groups' beliefs is highest ranked, followed by a lack of trust in political will, mythical attitude towards pandemic, and so on. The moral values are the lowest-ranked ICF. This ranking of significance is based on the strength of the path coefficients. For illustration, the magnitudes of path coefficients are provided as peer groups' beliefs = 0.710, lack of trust in political will = 0.652, mythical attitude towards pandemic = 0.581, pandemic knowledge = 0.509, self-efficacy = 0.472, risk-averse behavior = 0.421, perceived risk = 0.399, and ease of pandemic prevention adoption = 0.105. However, the coefficient of moral values remained insignificant and lowest (0.015). And thus, moral values imparted a neutral contribution to the individuals' WAPP.

In summary, a lack of trust in the political will and a mythical attitude towards the pandemic are the dominant inhibitors of individuals' WAPP. Meanwhile, the other ICFs are revealed as the driving forces of individuals' WAPP, except moral values which highlighted a neutral role in determining the individuals' WAPP. Peer groups' beliefs and pandemic knowledge are discovered as the main driving forces of individuals' WAPP (Figure 5).

**Figure 4.** Ranking the significance of intention-based critical factors (ICFs) affecting individuals' willingness to adopt pandemic prevention (WAPP). Source: Authors' elaboration.

**Figure 5.** Path coefficients-based classification of factors into driving forces, inhibitors, and neutral factors. Source: Authors' elaboration.
