*3.2. Structural Model*

The assessment of normality was done using WebPower's multivariate normality [50]. Mardia's multivariate skewness was 4.776 (z = 216.497, *p* < 0.01) and kurtosis was 60.632 (z = 10.632, *p* < 0.01), thereby indicating that the data were not multivariate normal. Thus, the researchers conducted bootstrapping with 5000 re-samples [48] to generate t- and *p*-values. Upon checking for multicollinearity, the results were 1.948 (EE), 2.379 (FC), 1.817 (PE) and 1.795 (SI). Thus, indicating that multicollinearity was not a serious issue in this study.

The values of R2 were 0.161 (Q<sup>2</sup> = 0.111) for PE, 0.331 (Q<sup>2</sup> = 0.228) for EE, and 0.377 (Q2 = 0.282) for intention to adopt solar services. Awareness was positively related to PE (β = 0.401, *p* < 0.01), EE (β = 0.575, *p* < 0.01), and FC (β = 0.606, *p* < 0.01), with the effect strongest for FC (refer Table 4). Thus, H1, H2, and H3 are supported.

**Table 4.** Structural Model.


This research proceeded to assess the effect of the four predictors on the intention to adopt solar services. PE (β = 0.238, *p* < 0.01) and FC (β = 0.281, *p* < 0.01) were significant predictions of adoption. EE and SI were insignificant. H4 and H7 were supported, whereas H5 and H6 were not supported.

Given that PLS is a prediction-oriented analytical tool, the out-of-sample prediction was assessed using PLS-Predict. The 10-fold and 10-repetition cross-validation procedure was used. Table 5 presents the results. Firstly, the assessment of Q2 for the latent variable prediction of intention to adopt; Q2 was 0.252, which was higher than 0 [51]. Thus, this research proceeded to assess the prediction of measurement items. Given that all PLS-LM values were negative, the model has high predictive power.

**Table 5.** PLS Predict.

