To test our hypotheses, the proposed structural model was estimated (IBM SPSS Amos 20.0) with trust and shared vision specified as the first-order constructs reflecting social capital, which is a second-order construct. As the control variable, R&D scale was included in the analyses.
Mediating Effects of Proactive Customer Orientation and Joint Learning
We tested for the mediating effects of customer orientation and joint learning for social capital and relationship-based radical innovation. Baron and Kenny (1986) stated that several conditions must be met to identify a partial or full mediation. First, the independent variable should affect the mediators significantly. We estimated a model using social capital as the independent variable and proactive customer orientation and joint learning as dependent variables to test this. Shown in Figure 2 (model 1), the results indicate the significant effects of social capital on proactive customer orientation (b = .47, p < .01) and joint learning capability (b
= .57, p < .01). The model fit indices include 2 = 100.153 on 41 d.f., TLI = .958, CFI = .957, and SRMR = .078. The second condition required is a significant impact of mediators on the dependent variable. In our results (model 2), both joint learning capability (b = .42, p < .01) and proactive customer orientation (b = .18, p < .05) significantly influences relationship-based innovation. The model fit indices include 2 = 22.797 on 32 d.f., TLI = 1.000, CFI = 1.000, and SRMR = .021, showing an excellent fit. The third condition is a significant direct effect of the independent variable on the dependent variable without specified mediators. In our analysis
(model 3), social capital affects relationship-based innovation significantly (b = .25, p < .01), and the model fit indices show an excellent fit with 2 = 23.283 on 13 d.f., TLI = .984, CFI
=.990, and SRMR = .062. Thus, the third condition is met.
Finally, in testing the mediation hypotheses, the significant impact of the independent variable on the dependent variable should be diminished in a partial mediation or become
non-significant in a full mediation when the mediators are added to the third model. According to our results (model 4), the coefficient of the independent variable, social capital, on the dependent variable, relationship-based innovation, decreased from .25 (p < .01) to -.10 (p > .10) with an excellent model fit, including its fit indices of 2 = 126.07 on 84 d.f., TLI = .976, CFI = .981, and SRMR = .065. As summarized in Figure 2, these results support Hypotheses 1 and 2 in the study.
--- Insert Figure 2 About Here
--- Moderation Effect of Supplier Design Responsibility and Dependence
For the moderating effects of supplier design responsibility and supplier dependence on the effects of proactive customer orientation and joint learning on relationship-based innovation, we predict that both a supplier’s high design responsibility and low dependence on the
international customer facilitate contextual settings conducive to creating higher effects of proactive customer orientation and joint learning on relationship-based innovation in Hypotheses 3a, 3b, 4a, and 4b. To test these moderating effects, we performed multi-group analyses
according to the participants’ design responsibility (i.e., yes vs. no) and by median-splitting the sample according to supplier dependence (Bentler 2005; Bollen 1989; Johnsen and Ford).
Two-group analysis was then conducted. However, the literature suggests that measurement
invariance should be assessed when multiple groups are involved in statistical analyses (Steenkamp and Baumgartner 1998). Specifically, the literature requires both configurable invariance and partial metric invariance to be supported so that a comparison of standardized path coefficients can be performed across groups, as in our study (Steenkamp and Baumgartner 1998). Therefore, we followed Steenkamp and Baumgartner’s (1998) procedure for performing measurement invariance tests.
According to the results of the measurement invariance tests, the configural invariance is supported for both two-group analyses because the combination of significantly loaded items is consistent for both groups, all factor loadings are significantly and substantially different from zero, and the factor correlations are significantly below unity across all groups for both
two-group analyses (Steenkamp and Baumgartner 1998). Subsequently, metric invariance was assessed. For both two-group analyses, all of the measurement items were metrically invariant (p
> .05) among the groups. Because partial metric invariance is a sufficient condition for a two-group comparison of standardized coefficients (Steenkamp and Baumgartner 1998), we proceed with the multiple-group analysis.
Shown in Table 4, we first estimated a two-group model based on the supplier’s design responsibility by adding and dropping an equal constraint for each hypothesized path. The chi-square difference tests show that the impact of proactive customer orientation on
relationship-based innovation is moderated by supplier deign responsibility (= 5.03, p < .05), but not by joint learning on relationship-based innovation (= .04, p > .10). The results
support Hypothesis 3a, but not Hypothesis 3b, because supplier design responsibility moderates the impact of proactive customer orientation (byes = .410, p < .01 and bno = .048, p > .05) on relationship-based innovation, but not the impact of joint learning on relationship-based innovation (byes = .362, p < .01 and bno = .442, p < .01). With the equality constraint on the
moderated path removed, the model shows a good fit (2 = 64.261 on 82 d.f., TLI = 1.000, CFI = 1.000, and SRMR = .034).
The moderating effects of supplier dependence on the relationship between proactive customer orientation and relationship-based innovation and between joint learning and
relationship-based innovation are evaluated by performing another two-group analysis based on the level of supplier dependence, which was determined by adding and dropping an equal constraint on each hypothesized path. The model estimation results show that the effects of proactive customer orientation on relationship-based innovation is moderated by supplier dependence (= 2.80, p < .10), but not by joint learning on relationship-based innovation (= .01, p > .10). Specifically, supplier dependence moderates the impact of proactive customer orientation on relationship-based innovation (blow = .281, p < .05 and bhigh = .031, p
> .05), but not by joint learning on relationship-based innovation (blow = .369, p < .01 and bhigh
= .422, p < .01), supporting Hypothesis 4a but not Hypothesis 4b. With the equality constraint on the moderated path removed, the model estimation results reveal a good model fit (2 = 107.404 on 82 d.f., TLI = .981, CFI = .986, and SRMR = .047). Table 4 presents a summary of our hypothesis-testing results.
--- Insert Table 3 About Here
---
The results of multi-group analyses suggest that proactive customer orientation has a fully moderated mediation effect (Muller, Judd, and Yzerbyt 2005) between social capital and relationship-based innovation as it only mediates the impact of social capital on
relationship-based innovation when the supplier has either high design responsibility or has low dependence on the international customer. Therefore, our Hypothesis 1 is only supported when the supplier has high design responsibility or low dependence on the international customer. In
contrast, Hypothesis 2 is fully supported since the mediating role of joint learning on innovation is not contingent on the contextual moderators.