*STEP 1:*

In the first step, some of the observed variables with a very low factor loading were eliminated from the measurement model as follows:


This was due to the very low factor loading shown by the arrow between the latent variable and the observed variables in Figure 6. Figure 7 shows the revised measurement model, including the observed variables with higher factor loadings after Step 1, although this might not be the best measurement, and further steps as described above should be carried out.

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

**Figure 7.** Conceptual measurement model of interrelation between building occupant environmental AKB after Step 1.

The above equation, if confirmed, indicates that the model in Figure 7 is over identified: t = items to be identified= 36 (18 'e' + 15 factor loading + 3 latent variables) s=numberofobservedvariables(18)

> 36 < 171

Based on the above result, it is confirmed that the model (Figure 7) is over identified, meaning that we have more than enough observed variables (18) to identify the unobserved items (36).

Modification Indices (MI) are often used to modify models in order to achieve a better fit, but this process should be carried out carefully and with theoretical justification [58]. For the MI generated in Step 1, the threshold was set at 10, and the MI valued as equal to/greater than 11 were selected for Step 2 as shown in Table 3.
