**3.1 Questionnaire, operationalization and variables of the constructs, data collection, and sample profile**

To test our research hypotheses and further explore our theoretical framework, this study utilizes survey-based empirical data from Chinese manufacturers. As part

**Figure 1.** *Theoretical framework.*

of a broader research project, a cross-sectional survey instrument was first designed. In line with the theoretical framework described in **Figure 1**, the questionnaire encompasses seven constructs, including enhancing the prevention ability of resilience, prevention and control of opportunistic behavior, manufacturing process coupling, responsible purchasing processes, and emergency-response ability; controlling dysfunctional behavior; triggering trade-off. According to related studies, we defined the operationalization of each construct and its related variables. These variables became the items or questions in the questionnaire, as shown in **Table 1**.

Based on the questionnaire, we attempted to collect related empirical data. Questionnaires were sent to 373 manufacturing companies in China. We received 231 valid responses, for a response rate of 61.93%. To characterize the profile of the companies in the sample, we investigated three characteristics—enterprise size, product type, and age of implementation of lean and audit for the promotion of supplier management robustness and resilience. According to aggregated results, the characteristics of the respondents are shown in **Table 2**.

#### **3.2 Method**

Partial least squares (PLS) analysis is a convenient method for estimating path relationship models with latent variables while including mediation effects. Because the theoretical framework of this study involves many path relationships and mediation effects, PLS is adopted as the main method to test our theoretical framework and hypotheses.

In PLS analysis, bootstrapping is used to test the statistical significance of the hypothesized relationships. The bootstrapping procedure entails generating 5000 *Enhancing the Resilience of Sustainable Supplier Management through Combination with Lean… DOI: http://dx.doi.org/10.5772/intechopen.102465*



#### **Table 1.**

*Questionnaire content, operationalization, variables of each construct, and citation source.*

subsamples of randomly selected cases with replacement. Under the analysis process, the path coefficients are generated for each randomly selected subsample, and the tvalue is calculated for every coefficient. According to the calculation results, the path coefficient and t-value are statistically significant and applied to evaluate the research hypotheses. As the analysis tool, we used SmartPLS 3.0.

In addition to PLS, we also used the Sobel test to verify the mediating effect. According to Hayes [70], a large sample may cause an error in evaluating the mediation effect when PLS is used for testing. However, the Sobel test can overcome this problem. For verification, Preacher and Leonardelli developed a free tool to perform the Sobel test. The test results verify the mediating effect. The tool is provided on the lab website of Preacher and Leonardelli.

However, to first test the theoretical framework and hypotheses with PLS, the validity and reliability of the constructs should be tested. To measure validity and reliability, factor loadings, composite reliability (CR), average variance extracted (AVE), and discriminant validity are the main indices. Regarding the requirements of the indices, the factor loadings should exceed 0.4 [71]; the CR and AVE should exceed 0.7 and 0.5 [72]. An exception is that if AVE is lower than 0.5 but higher than 0.36 and CR is above 0.7, then the situation can be accepted [73]. Discriminant validity is

*Enhancing the Resilience of Sustainable Supplier Management through Combination with Lean… DOI: http://dx.doi.org/10.5772/intechopen.102465*


**Table 2.** *Sample profiles.*

adopted to measure whether each construct can be discriminated from others' constructs. Therefore, the correlation between the constructs should be tested. If the correlation value is lower than 0.7, then every construct can be discriminated [73].

Finally, the model's goodness of fit should be measured. Regarding goodness of fit, the standardized root-mean-square residual (SRMR) is used as the main index. The SRMR was initially proposed for use in combination with CB-SEM, but it has also been extended to PLS. The SRMR is reported to be an approximate measure of model goodness of fit and has been widely adopted for this purpose. Thus, SRMR is adopted to measure the model's goodness of fit.

## **4. Test results**

#### **4.1 Construct measures**

Before testing the hypotheses, we test the validity and reliability of the constructs and discriminant validity. According to **Table 3**, all the factor loadings exceed 0.4. Therefore, the factor loadings satisfied the requirement. Regarding CR, according to **Table 3**, all the CR values exceed 0.7. Regarding AVE, the AVE values for the prevention and control of opportunistic behavior and triggering of a trade-off are higher than 0.5 and satisfy the measurement requirements. However, the values of enhancing manufacturing process coupling, responsible purchasing, emergency-response ability, and the prevention ability of resilience, as well as controlling dysfunctional behavior, are less than 0.5 but higher than 0.36. As noted by Fornell and Larcker [73], this situation may still be considered acceptable. Next, we test for discriminant validity. According to **Table 4**, the correlation values between constructs are lower than 0.7, and the test results satisfy the requirements.


#### **Table 3.**

*Construct measures assessment: composite reliability and convergent validity.*

