*STEP 3:*

The third step was to go through the standardised residual covariance in Figure 8 after correlating errors and deleting some of the observed variables above 0.1, shown on the arrows from Behaviour to its indicators/observed variables. Different trials were conducted, and the best outcome was to eliminate the following observed variables: BEH wat 3, BEH wat 4, BEH elec 3, BEH elec 4, BEH elec 7, and BEH trans 1 in BEHAVIOUR. The resultant best fitting measurement model can be seen in Figure 9.

$$\text{t} < 12 \times \text{s} (\text{s} + 1)$$

The above equation, if confirmed indicates that the model in Figure 9 is over identified:

t= items to be identified= 24 (12 'e' + 9 factor loading + 3 latent variables)

s = number of observed variables (12)

> 24 < 78

Based on the above result, it is confirmed that the model (Figure 9) is over identified, meaning that we have more than enough observed variables (12) to identify the unobserved items (24).

**Figure 9.** Best fitting measurement model of interrelations among building occupants' environmental AKB after Step 3 (Final Measurement Model).

The list below checks all the parameters in Table 4 related to the data for Figure 9:


The above list confirms that Figure 9 is a good model fit after three steps of modifications on the model. This suggests the finding that many of the observed variables (questionnaire results) did not measure their latent variable 'Behaviour'. Therefore, occupants did not behave in an environemtally friendly manner in all of their daily actions, despite having environmental knowledge, beliefs, and attitudes.


**Table 4.** Best fitting measurement model

#### *4.4. The Structural Model*

The interrelationships among unobserved/latent variables and their observed/measured variables were analysed to reach the best fitting measurement model (Figure 9, Table 4). At this stage, where we are confident that the observed variables support their latent variables, we are now ready to create the structural model to identify and verify the interrelationships between latent/unobserved variables. For the latent variables, the direction of the arrows from Attitude and Knowledge toward Behaviour and the correlating arrows between Knowledge and Attitude were determined based on the literature review [5,10,20,60]. The observed variables remain the same as in Figure 9, as their interrelationship with their latent variables are confirmed in Table 4.

The structural model is a hypothesized/conceptualised model that defines interrelationships among the latent (unobserved) variables only, and describes how particular latent variables directly or indirectly influence other latent variables in the model [58]. The resultant structural model is shown in Figure 10.

**Figure 10.** Best fitting structural model.

In Table 5, which is the result for Figure 10, the estimate for regression weight, covariance and correlation are all below 0.40, without a significant *p*-value (the *p*-value indicates whether an observation results due to random occurrences or occurs due to a change that was made). This is because estimate values equal to or greater than 0.40 with a significant *p*-value < 0.05 indicate strong measurement, while values closer to 1 indicate a stronger measurement [42]. A smaller *p*-value < 0.05 is at the level of significance and indicates that there is stronger evidence, assuming that the null

hypothesis is true. The only estimated values with significant p-values were those for Knowledge variance at 0.952, and that for ATTITUDE variance at 1.033. Therefore, the interrelationships were not very significant among these three latent variables.


**Table 5.** Estimates/scalar estimates/maximum likelihood estimates.

\*\*\* *p*-value is significant.

The standard errors (S.E.) do not present any extremely large or small values (outliers), and as suggested by Byrne (2010) [42], the model is a reasonably good fit; on the other hand, (S.E.) should not be an extremely small value close to zero, as this indicates a poor model based on the explanation by Bentler et al. (1980) [42,55]. The critical ratios (CR) present small and/or negative values for regression weight, covariance, and correlation, and were only high for KNOWLEDGE and ATTITUDE variance; therefore, no correlations between the latent variables were totally supported. In other words, there are no strong interrelationships among AKB in this research, as the figures on the arrows between the latent variables would need to be closer to 0.1; however, each of them is supported by its observed variables and GOF indices in Table 6 (GOF measures), which is discussed in Section 5 of this paper.


**Table 6.** Summary of GOF measurement results: (i) conceptual (ii) best fitting measurement and structural models.
