The VUCA environment, defined by volatility, uncertainty, complexity and ambiguity, compels firms to enhance flexibility and strengthen supply chain capabilities to sustain a competitive edge (Koç, Delibaş & Anadol 2022). Rising uncertainty has encouraged organisations to embrace digital technologies that increase information-processing capacity and partner connectivity, thereby augmenting SCVI and real-time data exchange (Troise et al. 2022; Zhou et al. 2024). Volatile, uncertain, complex and ambiguous conditions also highlight the importance of proactive SCRM: empirical evidence shows a shift from reactive to proactive risk practices that improve resilience to disruptions such as COVID-19 and geopolitical crises (El Baz & Ruel 2021). However, the realisation of SCVI and SCRM is uneven. Delphi research points to barriers such as limited data transparency and weak technology integration that keep many supply chains vulnerable to shocks (Grzybowska & Tubis 2022). This suggests that the effectiveness of firms’ responses to VUCA depends on how far they can develop and align their technology use, SCVI, SCRM and collaborative capabilities.

Based on the literature review presented in the section Literature review, the section Hypotheses development develops nine research hypotheses as follows:

The study targeted manufacturing enterprises operating in Vietnam that manage relatively complex supply chains and therefore have a stronger need for formal SCC than typical service firms.

Given resource and time constraints and the impracticality of surveying all manufacturing firms nationwide, we adopted a cluster sampling strategy. Vietnam has experienced rapid growth in industrial and export-processing zones with a diverse mix of foreign-invested and domestic manufacturers; accordingly, three representative industrial clusters were selected in Hanoi, Hung Yen and Bac Ninh.

Questionnaires were distributed to manufacturing firms via the management boards of these industrial zones.

In each enterprise, the research team approached two to three senior managers responsible for supply chain or digital technology. The questionnaire was pre-tested through semantic checking and a small pilot before the full survey. In total, 400 questionnaires were sent out, 196 responses were received, and after screening for completeness and consistency, 180 valid firm-level responses were retained for analysis. The raw response rate was 49% (196/400), and the usable rate was 45% (180/400). Data collection took place during October 2024–February 2025.

The sample is heterogeneous in terms of ownership and market participation. Private firms account for the largest share (92/180; 51.1%), and 81.1% of firms have been in operation for at least 10 years. Moreover, 145 out of 180 firms (80.6%) are involved in global supply chains, defined as having either main markets or core suppliers that are international or both domestic and international.

The sample size is adequate for PLS-SEM. Following the ‘10-times rule’, the minimum required sample equals at least ten times the largest number of arrows pointing at any latent construct. In our model, the maximum number of predictors for a construct is four, implying a minimum of 40 observations. With 180 firms, the actual sample size comfortably exceeds this threshold and is considered sufficient for complex PLS-SEM applications (Hair et al. 2019).

The questionnaire consisted of two main parts: (1) general information about the firm and the respondent, and (2) items measuring the constructs in the research model. All latent variables, business environment (BENV), digital technology adoption (TECH), supply chain visibility (SCVI), supply chain risk management (SCRM) and supply chain collaboration (SCC), were operationalised as reflective constructs.

Business environment, SCVI, SCRM and SCC were measured using a 5-point Likert scale (1 = strongly disagree; 5 = strongly agree), capturing the degree of agreement with each statement. TECH was measured using a five-level adoption scale ranging from 1 = ‘not applied’, 2 = ‘pilot testing’, 3 = ‘applied in some activities’, 4 = ‘applied in most related activities’ to 5 = ‘widely deployed with both supply-side and demand-side partners’, reflecting the extent of digital transformation in the firm’s supply chain activities.

All measurement scales were adapted from established international studies and adjusted to the Vietnamese manufacturing context to ensure content validity. Table 1 summarises the constructs, number of items, retained indicator codes and the main original sources of the scales; Table 1-A1 (see Appendix 1) provides the full wording of all items.

Data were analysed using PLS-SEM with SmartPLS 4.0. Partial least squares structural equation modelling is appropriate for this study because the research model is relatively complex, the sample size is moderate, and the main aim is to explain and predict key endogenous constructs rather than to reproduce a full covariance structure (Hair et al. 2019).

The analysis followed three main steps: (1) evaluation of the measurement model, including assessments of indicator reliability, internal consistency, convergent validity and discriminant validity (section Evaluating the measurement model); (2) evaluation of the structural model, examining path coefficients, coefficients of determination (R²), effect sizes (f²) and predictive relevance (Q²) (section Evaluation of the structural model); and (3) assessment of mediating effects, testing the indirect roles of TECH, SCVI and SCRM in transmitting the influence of BENV and TECH on SCC (section Intermediary role testing).

Because all constructs were measured using a single self-reported questionnaire, we also considered common method variance (CMV). Procedural remedies were implemented by assuring respondents of anonymity and confidentiality and emphasising that there were no right or wrong answers. In addition, CMV was assessed using a full collinearity test. Following Kock (2015), the variance inflation factors (VIFs) for all latent variables should be below 3.3 to indicate that the model is free of CMV. As reported in Table 2, all VIF values ranged from 1.450 to 2.944 (for BENV1-3, TECH1-5, SCVI1-3, SCRM1-3 and SCC1-6), suggesting that CMV is unlikely to pose a serious threat to the findings.

Ethical clearance to conduct this study was obtained from the Ethics Research Committee at the School of Economics and Management, Hanoi University of Science and Technology on 16 October 2025.

Firstly, we assessed the reliability and convergent validity of the scales. As reported in Table 3, most factor loadings exceeded 0.70; only SCC1 had a slightly lower loading (0.67) but was retained because of its content relevance, which is still acceptable in exploratory settings (Hair et al. 2019). All Cronbach’s α and composite reliability (CR) values were above 0.70, and all AVEs exceeded 0.50, indicating satisfactory internal consistency and convergent validity (Fornell & Larcker 1981; Hair et al. 2019).

Secondly, discriminant validity was assessed using the Fornell–Larcker criterion and the heterotrait–monotrait (HTMT) ratio. Table 4 reports the correlation matrix and the square roots of AVE (bold values on the diagonal). In all cases, the square roots of AVE (0.82–0.87) were higher than the correlations between constructs, satisfying the Fornell–Larcker criterion (Fornell & Larcker 1981). As shown in Table 5, all HTMT values ranged from 0.274 to 0.622 and were clearly below the conservative threshold of 0.85, indicating that the constructs were empirically distinct (Henseler, Ringle & Sarstedt 2015).

Taken together, the reliability and validity tests showed that the measurement model met international standards: all constructs exhibited high internal consistency, adequate convergent validity and clear discriminant validity. BENV, TECH, SCVI, SCRM and SCC were therefore treated as distinct but related latent variables, providing a solid basis for the subsequent evaluation of the structural model in section Evaluation of the structural model.

After confirming that the measurement model achieves reliability and differentiation value, the study conducted structural model validation according to the guidance of Hair et al. (2019). The R² values of the endogenous constructs ranged from 0.051 for TECH to 0.431 for SCC, indicating low to moderate explanatory power. In addition, all Q² values were positive, suggesting acceptable predictive relevance of the model for out-of-sample prediction (Hair et al. 2019).

A screenshot of the path coefficients and the outer loadings after the initial execution of the PLS-SEM algorithm is shown in Figure 2.

The overall conformity test of the structural model shows that the SRMR (standardised root mean square residual) of the estimated model is 0.084, which is below the threshold of 0.10 and approximately the threshold of 0.08 according to strict standards, proving that the model conforms to the data at an acceptable level. In addition, no collinear problems were detected between independent structures: all VIF variance magnification coefficients in the structure model were less than 2 (the highest VIF ≈ 1.52), which is much lower than the 3.0 threshold proposed by Hair et al. (2019). This result allows the assertion that independent variables are not collinear, so the regression estimates in the model are reliable.

Table 6 summarises the path coefficients and statistical significance levels for the H1–H9 hypotheses.

All β path coefficients were positive as expected and statistically significant at 5% or higher. In other words, the data support all 9 hypotheses. Specifically:

Thus, the hypothetical relationships in the model are supported by observational data. Table 6 presents the PLS-SEM estimation results for each hypothesis in detail, including path coefficient, standard error (s.e.), statistical value t, significance level p, f² index representing the effect size and R² of the related dependent variable.

The results show that the business environment in the context of VUCA (BENV) has the strongest influence on SCVI (β = 0.365, p < 0.001; f² ≈ 0.22 – medium-sized effect), implying that in the context of many fluctuations, pressure from the external environment significantly promotes information sharing and transparency in the supply chain application. This result is in line with the theoretical thesis that businesses facing a competitive and volatile environment will focus on improving transparency and visibility of the supply chain to mitigate risks. BENV also has a positive impact on the level of adoption of technology (TECH) and SCRM, although the intensity is weaker (β ≈ 0.19–0.23, p < 0.05; f² ~0.05–0.07, i.e. small effect, respectively). This suggests that environmental factors mainly indirectly affect supply chain efficiency through the promotion of intrinsic competencies such as technology, risk management and information transparency. In other words, BENV creates an incentive for businesses to develop internal capabilities, thereby improving supply chain efficiency – a thesis that is in line with the dynamic capabilities approach when businesses adapt to the environment by innovating internal capabilities.

In terms of the direct impact on SCC, intrinsic factors play a key role. Specifically, SCVI had the strongest influence on SCC (β = 0.297, p < 0.001) with an average impact size (f² ≈ 0.12). This suggests that increasing shareability and transparency in the supply chain will contribute to significantly improving the level of collaboration between parties, which is consistent with previous studies that have emphasised the role of transparent information in promoting effective collaboration and building trust among partners in the supply chain.

In addition, the level of technology adoption (TECH) also had a positive impact on the SCC (β = 0.239, p < 0.01; f² ≈ 0.10), reflecting the role of digital technologies in facilitating the connection and synchronisation of information, processes and goals among stakeholders – in line with the theoretical basis of technology presented in section Literature review and hypotheses development. Meanwhile, SCRM also had the same effect on SCC but with a weaker level (β = 0.184, p ≈ 0.05; f² ≈ 0.03) and only reached the level of statistical significance. This finding is quite surprising because, in theory, SCRM is expected to be one of the fundamental factors to promote cooperation to mitigate the impact of threats in the supply chain environment. A reasonable explanation is that the influence of SCRM on SCC is mainly promoted indirectly, through improving the level of technology adoption and SCVI. In other words, SCRM only really promotes collaboration when businesses simultaneously invest in technology and visibility to create a platform that supports risk coordination more effectively. This finding implies that businesses should not expect efficiency alone from risk management, but need to have a synchronous approach, integrating technology and information capabilities into the SCRM strategy to improve the efficiency of SCC.

Overall, the structural model results strongly support the theoretical framework: all H1–H9 are validated at a high significance level, indicating that the data are consistent with modern supply chain theory. Supply chain visibility is a key intermediary, connecting the business environment and TECH with SCC, affirming the fundamental role of information in chain coordination. Supply chain risk management’s independent contribution is modest, suggesting combining SCRM with TECH and SCVI to optimise cooperation. The findings add to the empirical evidence, increasing the reliability and generality of the model in the context that VUCA and digital transformation are reshaping cooperation in the global supply chain.

These results are consistent with prior studies that highlight the role of environmental uncertainty in driving supply chain integration and collaboration (Ivanov & Dolgui 2020; Koç et al. 2022) and support the view that visibility and digital technologies are central enablers of collaborative performance (Montecchi et al. 2021; Simatupang & Sridharan 2005).

To test the intermediate role, the study uses a bootstrapping technique with 5000 repeating patterns in PLS-SEM. This method allows for direct estimation of the indirect effects of independent variables on dependent variables through intermediate variables, and at the same time, examines the statistical significance of these effects. In addition, the study also calculates the percentage of variance explained through intermediaries (variance accounted for – VAF) to determine the type of intermediary according to the criteria of Hair et al. (2014): if VAF > 80% is fully intermediate, 20% – 80% is partially intermediate, and if VAF < 20% (or indirect influence is not statistically significant), it is considered non-intermediate space.

The bootstrapping results show many indirect effects in the model that are statistically significant, suggesting the existence of intermediary mechanisms. Specifically, the BENV impacts SCC through three main intermediary channels: (1) through SCVI: β_indirect = 0.108, p < 0.01; (2) adoption of technology (TECH): β_indirect = 0.054, p < 0.05; (3) through SCRM: β_indirect = 0.035, p < 0.05.

All of these channels have positive and statistically significant indirect influences, suggesting that a business environment (VUCA) can drive SCC through improved transparency, enhanced technology adoption and strengthened risk management. However, the direct influence from BENV to SCC is still significant (β_direct = 0.193; p < 0.01). Therefore, this relationship is identified as partially intermediate. In addition, multi-level intermediate channels (consisting of two consecutive intermediate variables) such as BENV → TECH → SCRM → SCC (β_indirect ≈ 0.013, p > 0.1) and BENV → TECH → SCVI → SCC (β_indirect ≈ 0.016, p > 0.1) were not statistically significant, showing that the second-order indirect influence of BENV is negligible in the model.

In terms of the relationship between TECH and SCC, the results show that TECH influences SCC indirectly through two channels: SCRM: β_indirect = 0.056, p < 0.05, and SCVI: β_indirect = 0.069, p < 0.05.

These two indirect effects coexist with a significant direct effect from TECH to SCC (β_direct = 0.239; p < 0.01), indicating a partially mediated link between TECH and SCC. Between the two intermediary channels, SCVI plays a more prominent role than SCRM (β_indirect is larger and closer to β_direct), suggesting that the adoption of digital technology promotes SCC primarily through SCVI, rather than improving SCRM.

In addition, the results also show that TECH plays an intermediary role in a number of other relationships.

In particular, BENV has a strong direct effect on SCRM (β_direct = 0.188; p < 0.01) and, through TECH, partially affects SCRM (β_indirect = 0.068; p < 0.01). Accordingly, by promoting the level of technology adoption, the business environment may indirectly improve the capacity to manage risk. In contrast, the effect from BENV to SCVI is mainly direct (β_direct = 0.365; p < 0.001), while the indirect effect through TECH is very small (β_indirect = 0.052; p < 0.05), accounting for only about 12% of VAF. As a result, TECH does not play a significant intermediary role in the relationship between BENV and SCVI. Table 7 summarises the indirect effects, total effects, VAF ratios and identifies the corresponding intermediate types for the main relationships in the study model:

The results in the table show that all meaningful intermediary relationships are partially intermediate. None of the relationships reached the 80% > VAF threshold (fully mediated), and some cases of VAF <20% (such as BENV → SCVI via TECH) were identified as significantly non-mediated. For example, the indirect influence from BENV to SCC via SCVI accounts for ~50.5% of the total influence – demonstrating a clear partial mediating role. Meanwhile, the influence from BENV to SCVI via TECH only accounts for ~12.5%, indicating that TECH is hardly intermediate in this relationship.

In terms of governance, the results show that intermediary factors such as SCVI and SCRM play an important role but cannot completely replace the direct influence of input factors (such as BENV, TECH). Therefore, enterprises not only need to develop internal capabilities but also effectively manage environmental factors and underlying technologies, thereby improving the level of SCC. This conclusion is consistent with dynamic capacity theory, which emphasises that intermediate competencies are mechanisms that transform external resources into internal outcomes, but input resources still play an important fundamental role.