BUS606: Operations and Supply Chain Management

In this study, a structural equation modeling (SEM) approach, using Smart PLS statistical software, was used to test the hypotheses in the conceptual research model. Smart PLS is suitable for a small sample size and does not require normal distribution of the manifest variables. Since the current study sample size was relatively small (162), Smart PLS was found to be more appropriate and befitting of the purpose. As recommended by Anderson and Gerbing, a two-stage procedure to hypothesis testing using SEM was utilised in this study. Measurement model assessment was performed first by examining the convergent and discriminant validity of items and constructs respectively, before the testing of the hypothesised causal relationship between the research variables in the structural model.

Measurement model

To ensure convergent validity, the researcher checked if items were loaded on their respective (a priori) constructs with loadings greater than 0.600, whilst discriminant validity was checked by ensuring that there were no significant inter-research variable cross-loadings. As can be seen in Table 2, all items except for ITR4 and CC6, which approached 0.600 (0.552 and 0.575 respectively) have loadings greater than 0.600, with no cross-loadings greater than 0.850, whilst t-statistics derived from bootstrapping (300 re-samples) suggest that all loadings are significant at a probability value (pb) of 0.001. As such, this confirms that all the measurement items mostly converged well on their respective constructs and therefore are acceptable measures.

TABLE 2: Accuracy analysis statistics

ITR, Information technology resource; CC, Collaborative communication; NG, Network Governance; RL, Relation longevity; C.R., Composite Reliability; AVE, Average variance reliability; LV, Latent variable.

Note: Scores: 1 = Strongly disagree; 3 = Neutral; 5 = Strongly agree.

According to Chin, research variables should have an average variance extracted (AVE) of more than 0.500 and a composite reliability of more than 0.700 (convergent validity). The inter-construct correlations should be less than the square root of the AVE (discriminant validity). As can be seen in Table 2, all constructs exceed these criteria, with AVE and composite reliability (CR) generally equal to or greater than 0.600 and 0.800 respectively, and the square root of the AVE being at least 0.710 greater than the inter-construct correlations (Table 3). These results confirm the existence of discriminant validity of the measurements used in this study.

TABLE 3: Correlations between constructs.

ITR, Information technology resource; CC, Collaborative communication; NG, Network Governance; RL, Relation longevity

Structural model

Figure 2 and Table 4 present the current study's results of the PLS analysis. The standardised path coefficients were expected to be at least 0.200 and preferably greater than 0.300. Bootstrapping (300 re-samples) was utilised to assess the reliability of each coefficient. The results provide support for all the five hypotheses. All other path coefficients were above 0.2 and significant (pb 0.001). As indicated in Figure 2 and Table 4, the path coefficients are 0.729, 0.324, 0.611, 0.226 and 0.659 for Hypothesis 1, Hypothesis 2, Hypothesis 3, Hypothesis 4 and Hypothesis 5 respectively.

FIGURE 2: Measurement and structural model results.

TABLE 4: Results of structural equation model analysis.

ITR, Information technology resource; CC, Collaborative communication; NG, Network Governance; RL, Relation longevity

Table 4 provides the t-statistics for the hypothesised relationships. The minimum t-statistic is 13.020 and therefore exceeds the recommended threshold of 2. This further confirms the statistical significance of the posited relationships and therefore all the hypotheses are supported.

Overall, $R²$ for logistics integration (LI) and small and medium enterprise performance (SMEP) in Figure 2, indicate that the research model explains more than 51.0% of the variance in the endogenous variables. Following the formulae provided by Tenenhaus, Vinzi, Chatelin and Lauro, the global goodness-of-fit (GoF) statistic for the research model was calculated and is 0.38, which exceeds the threshold of GoF > 0.36 suggested by Wetzels, Odekerken-Schröder and Van Oppen. Thus, this study concludes that the research model has a good overall fit.