The structural model,
including the research hypotheses and the causal paths,
was examined using the cleansed measurement model.
Figure 1 shows the standardized LISREL path coefficients of the research model.
The normed χ 2(χ 2 to degree of freedom) was 3.156, which was not good but could be mediocre accepted [5].
The root mean squared approximation of error (RMSEA) was 0.085 lower
than 0.1, which was also mediocre fitted [21].
Root mean square residual (RMR) was 0.059
lower than 0.08, which still could be accepted [5].
Goodness of fit index (GFI) was 0.824 that was
above the recommended threshold of 0.8 [20].
The other fit indices were all satisfactory that normed
fit index (NFI) was 0.917, nonnormed fit index
(NNFI) was 0.931 and comparative fit index (CFI)
was 0.941.
These results suggest that the structural
model fitted the data adequately.
Social media interaction ties and social media commitment positively influenced normative social influence
and informational social influence and explained 33.1% and 29.8% of each variance respectively.
Normative and informational social influence were positively related to visit intention of e-commerce,
which explained 44.4% of its variance in visit intention of e-commerce. Informational social
influence and visit intention were positively associated with purchase intention of e-commerce and explained 33.6% of its variance in purchase intention. However, normative social influence did not have a positive impact on purchase intention of e-commerce (path coefficient= -0.240, t-value=-3.157).
Table 1. Results of convergent validity testing
Table 2. Results of discriminant validity testing