*3.4. Data Analysis*

This study employed structural equation modelling (SEM) for data analysis. SEM consists of two types, namely covariance-based SEM and partial least squares SEM (PLS-SEM) [62]. As a useful statistical tool for testing the formulated hypotheses, PLS-SEM was selected in this study to quantify the impacts of different constructs [63]. PLS-SEM has been widely employed in behavioral sciences-related research with the ability to handle non-normal data and avoid many restrictive data assumptions. For example, Liu et al. [8] adopted PLS-SEM to investigate the psychological factors influencing HESBs. Another example is that Tan [64] employed PLS-SEM to predict sustainable real estate purchasing intention with personal values and attitudes. Nomura et al. [65] also revealed the psychological driving force behind household recycling behavior. This study employs the software Smart-PLS 3 as a tool for data analysis.

The analysis process involves two steps: (1) assess the measurement model and (2) evaluate the structural model [8,66]. For the measurement modeling, composite reliability (CR), convergen<sup>t</sup> validity (CV) and discriminant validity (DV) are the common criteria to indicate the model's validity and reliability. Normally, a satisfactory value for CR varies between 0.7 to 0.9 [66]. CV can be assessed with the values of average variance extracted (AVE) and the measurement items loadings. AVE refers to a measure of the amount of variance which is captured by a latent construct in relation to the amount of variance due to measurement error [67]. The acceptable AVE value of each element should exceed 0.5 [63], while the measurement loading of a specific item should be larger than 0.4 [8,68]. DV assessment is used to confirm that each latent variable is not correlated with other latent variables [68]. The heterotrait–monotrait (HTMT) ratio is one of the common methods for DV measurement, and its index should not exceed 0.9 [8,69]. Using the bootstrap re-sampling technique, the path coe fficients (*t*-value) and levels of significance (*p* value), construct reliability and validity (CR and CV) and discriminant validity (DV) were generated in the structural model [66,70]. The analysis results are discussed in the succeeding sections.
