Checking reflective constructs requires the assessment of individual indicators’ and latent constructs’ reliabilities, as well as the measures of convergent and discriminant validities (Hair et al., 2014a). An analysis was conducted using WarpPLS 5.0 (see Figure 4.2).
Measurement theory states how the latent variables (constructs) are measured (Hair et al., 2014). The current study employed confirmatory factor analysis. Loading and cross-loading of indicators for the sample is shown below (Hair et al., 2011) [Appendix D].
4.5.1 Reliability
Reliability, as defined by Hair et al. (2014), is the consistency of a measure - a measure is reliable when it produces consistence. In terms of reliability, Table 4.5 below shows composite reliability coefficients taking indicator loading into consideration. The generally accepted for composite reliability is 0.7 or higher (Hair et al., 1992; Nunnally & Bernstein, 1994). For the sample of online luxury consumers, all of the latent variables have composite reliabilities higher than 0.70. Cronbach’s alpha should be higher than 7.0. Hair (2012) states that Cronbach’s alpha is acceptable if it is between 0.6 and 0.7. As shown in the table below, the highlighted lower score was in SMMA-cus 0.63. As a result, the author confirmed that the composite reliability would be a more reliable measure than Cronbach’s alpha. On this basis,
Table 4.5 suggests that the reflective measurement instruments employed in this study have a satisfactory reliability.
Table 4.5 Composite Reliability and Cronbach’s Alpha
Constructs SNS
U ewe eSQ- reli
eSQ- res
eSQ- con
eSQ- EU
eSQ- prot
eSQ-
sec Att Int AP eS-
conv eS- p
off eS- p
inf Composite
reliability 0.95 0.92 0.84 0.89 0.86 0.88 0.84 0.84 0.90 0.95 0.88 0.87 0.94 0.92 Cronbach’
s alpha 0.94 0.88 0.72 0.81 0.76 0.80 0.71 0.76 0.84 0.92 0.81 0.78 0.87 0.84 Constructs eS-
site eS- fin
eS- C S eL
SM MA-
en SM MA-
int SM MA-
tre SM MA-
cus
PU PEU PBV
-soc PBV
-uti PBV
-eco Composite
reliability 0.86 1.00 0.95 0.93 0.92 0.88 0.86 0.84 0.86 0.86 0.90 0.83 0.89 Cronbach’
s alpha 0.76 1.00 0.89 0.88 0.83 0.79 0.68 0.63 0.81 0.78 0.88 0.70 0.75
Notes: SNSU = Social network site usage, eWOM = electronic word of mouth, eSQ- reli = electronic service quality- reliability, e-Sq-res
= electronic service quality- responsiveness, eSQ-con = electronic service quality-competence, eSQ- EU= electronic service quality-ease of use, eSQ - prot= electronic service quality- Portability, eSQ- sec= electronic service quality- security, Att= Attitude, Int= Intention, AP=
Actual purchase, eS- conv= electronic satisfaction convenience, eS- Poff= electronic satisfaction – product offering, eS-P - inf = electronic satisfaction – product information, eS-Site= electronic satisfaction - site, eS- fin= electronic satisfaction -financial, eS-C S= electronic satisfaction –Customer Satisfaction, eL= electronic loyalty, SMMA-en= Social marketing media activities- entertainment, SMMA-int=
Social marketing media activities-interaction, SMMA-tre= Social marketing media activities-trendies, SMMA- cus= Social marketing media activities-customisation, PU= Perceived Usefulness, PEU= Perceived Ease of use, PBV-soc= Perceived Brand value- social /emotional value, PBV-uti= Perceived Brand value- utilitarian, PBV- eco= Perceived Brand value- Economic value.
4.5.2 Convergent Validity
Convergent validity defined by Hair et al. (2014) is the extent to which a measure correlates positively with alternative measures of the same construct. In order to confirm that a model is acceptable, convergent validity loadings should be lower than 0.05 (Hair, Black, Babin, &
Anderson, 2009). The criterion used to identify a good convergent validity is an AVE of greater than 0.50 as it suggests that the latent construct can explain more than 50% of the its indicator’s variance (Henseler et al., 2009; Hair et al., 2011; Mackenzie et al., 2011; Peng and Lai, 2012; Schmiedel et al., 2014). As it shown in Appendix E, all of the loading of the sample were higher than 0.5 threshold and significant under 0.001 levels, meaning that the measurement constructs have a satisfactory convergent validity.
4.5.3 Discriminant Validity
Discriminant validity examines the extent to which two indicators under the same construct are correlated (Hair et al., 2010; Hair et al., 2014a). The reflective indicators require two types of validities; namely convergent and discriminant validity (Hair et al., 2011). Construct validity allows the researcher to ensure that the set of indicators indeed measure the latent construct they intend to measure (Henseler et al., 2009). Hair et al. (2010) state that validity explains the latent variable is represented by its indicators.
It can be checked by looking at the diagonal of each indicator in relation to the latent construct. This can be achieved through the Average Variance Extracted by the latent construct (AVE). The criterion used to identify a good convergent validity is an AVE of greater than 0.50, as it suggests that the latent construct can explain more than 50% of the its indicator’s variance (Henseler et al., 2009; Hair et al., 2011; Mackenzie et al., 2011; Peng &
Lai, 2012; Schmiedel et al., 2014). This method is known as Fronell and Larcker (1981) criterion and it was found to be widely used in research (Ringle, Sarstedt & Straub, 2012). In this case, the square root of AVE of the latent construct should be higher than other constructs along the diagonal (Hulland, 1999; Ketkar et al., 2012; Peng & Lai, 2012). Second, the indicator’s loading with its latent constructs should be higher than the remaining cross loadings (loading with other latent variables) (Hair et al., 2011; Hair et al., 2014a; Schmiedel et al., 2014). In Appendix E, the square root of AVEs are given in diagonal. The principle is that the square root of the AVE for each latent variable should be higher than any of the correlations of that respective latent variable.In the sample, all of the square roots of AVEs are higher than the correlations of that respective latent variable. This indicates that all the questions in the survey were understood and answered correctly as intended.
4.5.4 Collinearity
Collinearity, as defined by Hair et al. (2014), is when two indicators are highly correlated.
Besides the Validity and Reliability tests, Kock and Lynn (2012) suggest to conduct a full collinearity test. According to Hair et al. (2014a), collinearity occurs when two or multiple indicators (multicollinearity) are highly correlated (redundancy among constructs). In PLS- SEM, Kock and Lynn (2012) recommend using the full variance inflation factor (VIF) for each predictor construct to assess the full collinearity. The authors also argued that a full collinearity test can also be used to assess the common method bias. In Table 4.6 below, variance inflation factor VIF scores of each dependent variable were mentioned for the online luxury consumers’ sample. The R-squared coefficients of the consumers’ outcome are between. The full collinearity check was completed and this test is based on the variance inflation factors for each of latent variables. The normal scores of VIF are 5 or lower. From a flexible perspective, VIF should be lower than 10 (Hair et al., 2009) in the online luxury consumers sample; the highest VIF (4.882) was recorded for the eSQ-con variable and the second highest was (4.162) for the e-SQ-res variable.
Table 4.6 Full Collinearity VIF
Constructs SNS U
eW OM
eSQ- reli
eSQ- res
eSQ- con
eSQ- EU
eSQ- prot
eSQ-
sec Att Int AP eS-
conv eS- p
off eS- p
inf Full
collinearity VIF for construct
1.30 1.61 3.15 4.16 4.88 2.88 3.31 3.70 3.59 3.49 2.00 2.27 2.74 2.28 eS-
site eS- fin
eS-
C S eL SMM
A- en SM MA-
int SM MA-
tre
SMM
A- cus PU PEU PBV -soc
PBV -uti
PBV -eco 3.58 3.13 3.84 2.38 2.62 3.96 2.25 2.72 1.38 1.80 3.34 2.02 2.69 Notes: SNSU = Social network site usage, eWOM = electronic word of mouth, eSQ- reli = electronic service quality- reliability, e-Sq-res electronic service quality- responsiveness, eSQ-con = electronic service quality-competence, eSQ- EU= electronic service quality-ease of use, eSQ-prot= electronic service quality- Portability, eSQ- sec= electronic service quality- security, Att= Attitude, Int= Intention, AP=
Actual purchase, eS-conv= electronic satisfaction convenience, eS- Poff= electronic satisfaction – product offering, eS-P - inf = electronic satisfaction – product information, eS-Site= electronic satisfaction - site, eS- fin= electronic satisfaction -financial, eS-C S= electronic satisfaction –Customer Satisfaction, eL= electronic loyalty, SMMA-en= Social marketing media activities- entertainment, SMMA-int=
Social marketing media activities-interaction, SMMA-tre= Social marketing media activities-trendies, SMMA- cus= Social marketing media activities-customization, PU= Perceived Usefulness, PEU= Perceived Ease of use, PBV-soc= Perceived Brand value- social /emotional value, PBV-uti= Perceived Brand value- utilitarian, PBV- eco= Perceived Brand value- Economic value.