Assessment of the measurement models

In the measurement models, all the indicators are reflective because those are highly interchangeable and correlated. Before appraising the structural model, the study evaluates the quality criteria of the measurement models in a concise manner. There are two types of reliability test, one of which tests indicator reliability and the other internal consistency reliability. The outer loadings and VIF are examined for indicator reliability while internal consistency reliability is evaluated by the composite reliability, convergent validity, AVE, and discriminant validity. Moreover, discriminant validity has been tested by the cross-loadings, Fornell-Larcker criterion and HTMT (Hair, Hult, Ringle, &

Sarstedt, 2013; Hair et al., 2017; Wong, 2013).

Indicator reliability

Indicator reliability is indispensable in the measurement model. Since the study constructs are first order and reflective, the outer loadings are restrained to check reliability. Previous studies proposed that the loadings should be more than 0.70 for the manifest indicators, even though there are some exceptions. If the composite reliability (i.e., above 0.70) and the AVE (i.e., below 5) of certain constructs are established, the indicator items can be considered with below 0.70 (Bauer et al., 2014; Hair et al., 2017; Henseler et al., 2016). However, Hulland (1999); Wong (2013) substantiated that in exploratory research, reliability could be sustained by loadings above 0.40; for example, Alashban et al. (2002) reported loadings of more than 0.40. On the other hand, Capron et al. (2001) stated that, in the exploratory research, the confidence for the loading estimation is between 0.60 and 0.70. In this study, all the indicator items met the 0.70 threshold level except the six items that belong to different constructs and reach the criterion of 0.60 (See Appendix 2). Hence, the assessment shows that indicator reliability is established.

Variance inflation factor (VIF)

Multicollinearity assessment is required for the outer loadings to maintain the quality criteria. The related measurement of the collinearity is the variance inflation factor (VIF). In the regression analysis, VIF explains how much correlation exists among the predictors to maintain the quality of the loadings.

There are some restrictions; for example, the VIF value should be lower than 5.

The study evaluation shows that, in the external loadings, the VIF value of each item is below 5 (for details see Appendix 3). Similarly, each latent construct also maintains the VIF value criterion below 5 (See also Appendix 4) (Hair et al., 2017; Ringle et al., 2015).

Internal consistency reliability

The previous studies reported that composite reliability and Cronbach’s alpha can be used to test internal consistency (Bauer & Matzler, 2014; Henseler et al., 2016). However, Bagozzi and Yi (1988); Wong (2013) suggested that Cronbach’s alpha should not be used because it is more conservative to check the internal consistency in PLS-SEM. Hair, Sarstedt, Ringle, and Mena (2012) also argued against using Cronbach’s alpha. They proclaimed that composite reliability is the replacement of Cronbach’s alpha. Similarly, Henseler et al. (2016) and Sijtsma (2009) argued that Cronbach’s alpha is used to measure the sum scores instead of construct scores, which are regarded as low boundary and not indicative of actual reliability.

The Cronbach’s alpha implements that all the manifest items have equal outer loadings that are equally reliable. However, composite reliability prioritizes items with individual reliability (Hair et al., 2017). Hair et al. (2017) also stated that composite reliability is technically appropriate to measure internal consistency in PLS-SEM. Hence, this study used composite reliability to check internal consistency. Furthermore, Bagozzi and Yi (1988), Bauer and Matzler (2014), Henseler et al. (2016) and Wong (2013) proposed that the threshold level of the composite reliability should be 0.70, but that a value of 0.60 is also acceptable in exploratory research. Hair et al. (2017) also anticipated that composite reliability could also be between 0.60 and 0.70. Since all the values of the latent constructs are more than the 0.70 criterion, this study confirms that composite reliability has been established (for details, see Appendix 5).

Convergent validity

To test the convergent validity, the Average Variance Extracted (AVE) should be evaluated for each latent construct. Convergent validity is confirmed if AVE values are above 0.50 (Bauer & Matzler, 2014; Henseler et al., 2016; Wong, 2013). Since all the AVE values of the latent constructs are higher than the value of 0.50 criterion (Ringle et al., 2015), this study confirms that convergent validity has been established (See Appendix 6).

Discriminant validity

The discriminant validity is important for checking whether each indicator and latent construct is unique compared to others. It can be tested by the cross-loadings, Fornell-Larcker criterion, and Heterotrait-Monotrait Ratio (HTMT) (Hair et al., 2017; Henseler et al., 2016). The cross-loadings are the first approach to checking discriminant validity based on the construct indicators. If the value of each item that is associated with the specific construct is higher than the cross-loadings of another construct, it suggests that discriminant validity is confirmed.

In this study, the value of each construct item is greater than the items associated with the other constructs. It means that discriminant validity is confirmed by the cross-loadings (See Appendix 7).

The second necessary step is the Fornell-Larcker criterion. The logic is that each latent construct should be greater than the correlations among the latent variables (Hair et al., 2017; Wong, 2013). The evaluation shows that the value of each construct is higher than the other constructs, suggesting that discriminant validity has been confirmed (for details, see Appendix 8). Also, HTMT should be used to test discriminant validity in SmartPLS. The main fundamental theme is that the correlation between the two constructs should be less than 1.

Discriminant validity can also be rejected if the value is greater than 0.90.

Henseler, Ringle, and Sarstedt (2015) also substantiated that the value of 0.85 restricts the threshold level. This study illustrates that all the constructs confirm discriminant validity based on the 0.85 criterion (See also Appendix 9). Finally, after testing the discriminant validity by the cross-loadings, Fornell-Larcker criterion, and HTMT, this study concludes that the items and constructs are unique (Bauer & Matzler, 2014; Hair et al., 2017; Ringle et al., 2015).