**5. Results**

#### *5.1. Measurement Model Analysis*

After debugging the database and validating the instrument, the measurement model was analyzed. Considering that the model under analysis is composed of reflective items, the evaluation of the measurement model involves the respective reliability criteria: (i) item reliability, (ii) construct reliability, and (iii) internal consistency. The statistical indices used to evaluate the three criteria were: the corrected item-total correlation (CITC), the latent and cross loads, the Cronbach's alpha (CA), Composite Reliability (CR), and the Average Variance Extracted (AVE). All items had CITC indices above the recommended parameter of 0.5 (Hair et al., 2017). The respective Cronbach's alphas also had indices above 0.7, indicating the internal consistency of the constructs.

According to Henseler et al. [60], the reliability of each reflective item is measured by its latent load about its respective construct. The authors recommend that the rates be at least 0.7. Although eight of the 29 indicators in the final questionnaire had external loads below the 0.7 index (yet all above 0.6), the results of the Average Variance Extracted (AVE) show that they contribute to the measurement of their respective constructs [60]. Therefore, they were kept in the model evaluation.

The second procedure consisted of verifying the convergen<sup>t</sup> validity, estimated through the AVE. The three constructs presented indices higher than 0.50, which concludes that at least half of the variance of each construct can be attributed to the items that compose them [61,62]. Before proceeding with the discriminant validity test, the item significance test, and the internal consistency of the constructs are presented. In Table 5, latent loads of items, the composite reliability (CC), Cronbach's alpha (AC), and the average variance of the extracted constructs (AVE) are exposed.


**Table 5.** Latent loads, Cronbach's alpha, average variance extracted, and composite reliability of items and constructs.

 Note: a—Cronbach's alpha; b—Composite reliability; c—Average Variance Extracted; \* *t*-value for two-tailed test: \* 1.96 (significance level: 95%).

The discriminant validity (the extent to which items in a given construct di ffer from items in other constructs) was assessed by two criteria. First, according to the Fornell–Larcker criterion [63], the square root of the AVE indices of each of the constructs must be greater than their correlation with the others. The three investigated constructs met this condition. Table 6 shows the discriminant validity by the Fornell–Larcker criterion, the average and standard deviation of the constructs.


**Table 6.** Discriminant validity by the Fornell–Larcker criterion.

The second criterion used to assess discriminant validity was the straight-trait mono track correlation ratio [64]. The indices presented values below 0.85, as recommended by the literature [64].

#### *5.2. Structural Model Analysis*

After analysis of the measurement model, the structural model was examined. This stage of the analysis is to identify the strength, significance, the total variance explained by the endogenous constructs and predictive relevance of the model [59,64]. The multicollinearity analysis was performed by observing the Variance Inflation Factor (VIF). According to Hair et al. (2017), values between 0.2 and 5 indicate that there is no negative influence of multicollinearity between items. The values of the independent variables so as dependent varied between 1.6 and 3.9.

The predictive relevance of the model was calculated by the Stone–Geisser index (Q2). The *blindfolding* test was performed, with the omission distance of seven points, according to the recommendations of Hair et al. [58]. Cross-validation of endogenous constructs showed values above zero, EIN (0.240), and NWW (0.105). The results confirm the predictive relevance of the model. Additionally, the indicators were subjected to the Harman factor test [65]. In this procedure, all constructs are analyzed by the principal component analysis method, and, if the indicators extracted in a single factor have an explained variance percentage greater than 50%, this variance can be attributed to the method used by the measurement and not by the constructs to be measured [61,66]. In the final instrument, the explained variance percentage reached 33.7%.

The PLS algorithm analysis demonstrated that 37.6% of the variance of the endogenous latent construct in work engagemen<sup>t</sup> (EIN) can be explained by the physical factors of the environment (PFE) and the NWW. Another result observed in the analysis was that 23% of the variance of the endogenous construct can be explained by NWW physical environmental factors. Figure 2 presents the model, its structural relationships, latent loads, path coe fficients (beta), and the variance explained by endogenous constructs.

The next step in the structural model evaluation was to verify the significance of the results obtained from the standardized path coe fficients (betas) of the model. The *bootstrapping* resampling technique with 5000 samples was performed following the recommendations of Hair et al. [62]. This statistical technique aims to assess the significance of the indices found in the measurement and structural model evaluation.

At a significance level of 95% (*p* < 0.05), the validity of the standardized path coe fficients (β) was tested. Table 7 presents the hypotheses, the indices, and the decision about their validity.

At a significance level of 95%, Hypotheses 1, 2, 3 and 4 were confirmed. The results indicate the influence of PEF on both work engagemen<sup>t</sup> and NWW facets. These in turn also directly and positively influence work engagement. Hypothesis 5 was not confirmed, indicating that PEFs do not act as moderators of the relationship between NWW and work engagement.

This information provides us with evidence to support research by researchers who argue that changes in the physical work environment alone or the implementation of facets of the NWW alone are not enough to promote work engagemen<sup>t</sup> and employee performance [3,6]. This finding is one of the topics discussed in the concluding chapter of this document. Hypothesis 4, which was investigating the mediation of the facets of the NWW in the relationship between environmental physical factors

and work engagement, was confirmed. The mediation analysis is presented and discussed in the following section.


**Table 7.** Test of hypothesis relation between constructs.

a*t* value for two-tailed test: \* 1.96 (significance level: 95%); \*\* total effect.

**Figure 2.** Model results.
