Page 112 - 33-2

Service Innovation in the IT Service Industry: Social Influence and Relationship Exchange Perspectives

of the study constructs. The results of the measurement model evaluation display a very
2
good fit with the data: χ (157) = 284.72, p ≈ .00, RMSEA = .052, NNFI = .97, CFI = .98,
AGFI = .87. The factor loading of each item on its representative construct exceeds .5 (p
< .001), which demonstrates adequate validity and reliability. Because the present study
uses multi-item scales to measure each construct, we examine the internal consistency
of these measures through Composite Reliability (CR) and Average Variance Extracted
(AVE) (Bagozzi and Yi, 1988). Both well exceed the respective standard benchmarks of

.60 and .50, which suggests good internal consistency. We also use three different methods
to evaluate discriminant validity among all the constructs. First, as shown in Table 2, the
square roots of the AVE for each latent variable can be seen on the diagonal. These values
should be greater than the bivariate correlations between the latent variable and all other
latent variables; that is, the diagonal values (square roots of the AVE for each construct)
are greater than non-diagonal elements in that same row or column. Second, also shown
in Table 2, the 95% confidence interval of the correlation between any two latent variables

2
did not include one (Anderson and Gerbing, 1988). Finally, for each pair of factors, χ -
value for a measurement model constraining their correlation to one is compared with a
2
baseline measurement model without this constraint. We perform a χ -difference test for 15
pairs of variables (see Appendix Table A2), and each test results in a significant difference.
The above tests demonstrate that all of the construct measures in the measurement model
achieved discriminant validity.

4.2 Results

The structural model evaluation provides a satisfactory fit with the observed data
2
(χ [173] = 337.23, p ≈ .00; RMSEA = .057, NNFI = .96, CFI = .97, AGFI = .86). Table
3 displays all the indexes of model fit. Figure 2 indicates the path coefficients of the
structural model.
H1 and H3 predict that EML and TMX positively influence innovation intention
through value congruence. We find both EML and TMX have a significant relationship
with value congruence (γ = .39, p < .001; γ = .37, p < .001, respectively). Additionally,

we also find that value congruence is positively related to innovation intention (β = .29,
p < .001). Our results thus support H1 and H3. Similarly, H2 and H4 propose that EML
and TMX positively influence innovation intention through felt obligation. We find both
EML and TMX have a positive relationship with felt obligation (γ = .52, p < .001; γ = .14,
p < .1, respectively). The results also indicate that felt obligation has a positive impact on

104