Managing the pathways to inclusive sustainability through accounting system information: Role of circular economy Practices and green growth

2.1. Theoretical Underpinning

The theoretical frameworks of stakeholder theory (Freeman, 1984) and the resource-based view (Barney, 1991) are employed in this study for the proposed model investigation. These two theories offer significant insights into the internal and external drivers of organisational transformation and sustainable practices. The resource-based view provides valuable insight into enhancing an organisation’s performance through its existing resources and internal expertise (Corbett and Claridge, 2002). The resource-based view centres on the organisation as the primary unit of analysis, positing that organisations possessing rare, valuable, and unique assets can achieve a sustainable competitive advantage by executing innovative, value-generating strategies that are challenging for rivals to replicate (Barney, 1991).

This study illustrates how the resource-based view emphasises AIA’s role in improving efficiency through process optimisation, minimising inefficiencies, and augmenting operational capability. By integrating artificial intelligence with accounting information system, PSOs can enhance resource efficiency and refine decision-making processes. This study emphasises how PSOs can leverage AIA potential to foster innovation, hence advancing societal and environmental objectives. By utilising these internal resources, PSOs can adopt CEE and IGG strategies to attain SDG 16. According to Cornelissen (2023), stakeholder theory is articulated as a strategy instrument applicable across all sectors, notwithstanding its origins in the business domain (Moloney and McGrath, 2020). Stakeholder theory addresses ethical considerations, stakeholder interests, and organisational management to promote ethical, effective, and pragmatic approaches for managing uncertainties (Freeman and Dmytriyev, 2017). This theory offers researchers a framework to perform comprehensive and insightful investigations concerning stakeholders (Tallberg et al., 2022). Stakeholder theory motivates authors globally to do further research to comprehend valuable insights for organisations (Freeman, 1984). Furthermore, its theory aids strategic management in disseminating strategies by including other views and domains, including environmental considerations (Donaldson and Preston, 1995). The alignment of sustainable management and stakeholder theory is evidenced by their shared long-term perspective and strategic planning characteristics. As stated by Dumay et al. (2010), PSOs should assume a crucial role in promoting sustainable development. PSOs are obligated to meet the expectations linked to fulfilling the requisite requirements of organisational effectiveness and efficiency in public service delivery. Besides, PSOs should include sustainability strategies that mitigate adverse environmental and social effects while fostering a more sustainable society. The use of digital technology is presumed to enhance the incentive of various stakeholders to engage, foster reciprocal contributions, share perceived decision-making power, and augment the potential public value of co-creation (Lember et al., 2019). In a nutshell, the resource-based view suggests that PSOs should optimize their internal resources and capabilities, while stakeholder theory emphasizes the importance of engaging stakeholders and using these resources accountably to advance SDG 16.

2.2. Conceptual Framework

4.1. Measurement Scale Development

The questionnaire was first composed in English and subsequently translated into Vietnamese by bilingual specialists to guarantee linguistic precision. To evaluate content validity, three academic experts in sustainable development examined the questionnaire with a 7-point Likert-type scale. A pilot test involving 30 individuals was conducted following initial comments. A small-scale pilot study was conducted with three principal objectives: to evaluate the validity of the measuring scales; to enhance the clarity and quality of the survey questions; and to modify and finalize questionnaire. Multiple modifications were implemented to improve clarity, guarantee accurate replies, and reduce potential biases prior to the deployment of the final survey. The reliability scores for all constructs above 0.70, deemed acceptable. The research variables were indicated in Table 1.

Inclusive green growth. This investigation utilised five items that were derived from those suggested by Bouma and Berkhout (2015); Kourula et al. (2017); Chapman and Shigetomi (2018) to evaluate IGG.

Circular economy practices. This investigation utilised five items that were derived from those suggested by Kuzma et al. (2021); Hazen et al. (2021); Pizzi et al. (2022) to evaluate CEE.

Artificial intelligence-powered accounting information system. This investigation utilised five items that were derived from those suggested by Abu Afifa et al. (2025).

Sustainable Development Goal 16. This investigation utilised five items that were derived from those suggested by Greenland et al. (2023).

4.2. Target Population and Data Collection

The research model and hypotheses were evaluated utilising data from PSOs in Vietnam. Based on the perspective of Huynh et al. (2025), Vietnam has been acknowledged as an expeditiously industrialising nation in Southeast Asia. The nation is experiencing structural transformation propelled by government-led development, digitisation efforts, and a heightened focus on innovation-driven growth (Mai et al., 2024). According to Thai et al. (2025), the anticipated investment demand in Vietnam is forecast to reach approximately $30 billion by 2030. The establishment of a comprehensive infrastructure system is essential for the nation to attain sustainable development objectives and improve national competitiveness in a coordinated and contemporary fashion, meeting the demands of socio-economic advancement (Thai et al., 2025). Vietnam is implementing a significant governmental and provincial reorganisation to enhance administrative efficiency, governance, and economic competitiveness.

Effective 1 March 2025, Vietnam’s Government organizational structure was streamlined from 18 to 14 ministries and from four to three ministerial-level agencies, with the aim of reducing institutional overlap and enhancing inter-agency coordination. Significant mergers encompass the consolidation of the Ministry of Finance with the Ministry of Planning and Investment, alongside the establishment of new entities such as the Ministry of Ethnic and Religious Affairs. The government merged 63 provinces and cities into around 34 larger administrative units at the provincial level, facilitating enhanced resource distribution and local governance. The elimination of district-level administration effective 1 July 2025 has established a two-tier governance structure – provincial and commune – thereby lowering bureaucracy and expenses. Successful implementation of these reforms necessitates robust coordination, openness, and stakeholder engagement to assure stability, uphold investor confidence, and actualize Vietnam’s goal for a modern, responsive government. Examining Vietnam is essential for comprehending its distinctive developmental trajectory and for extracting insights relevant to other developing economies undergoing analogous transformations. Recently, Vietnam has enacted numerous policies to manage, efficiently use, and sustainably employ natural resources, fuels, and raw materials, promoting the creation of environmentally friendly products that are renewable, reusable, and recyclable (Quang et al., 2025).

According to Barbera et al. (2024), global academic experts acknowledge accounting as a tool of authority and regulation inside organisations and society. Accounting systems have become increasingly prevalent in the public sector, frequently offering the tools and techniques to convert public sector reform concepts into routine practices and behaviours (Miller and Power, 2013). In addition, accountants significantly contributed to the measurement, disclosure, and assurance of all organisational information, particularly with corporate social responsibility (Huang and Watson, 2015). Collectively, accountants can serve as a crucial element in fostering sustainable growth within the public sector by ensuring transparency, accountability, and the effective allocation of public resources. As a result, accountants were chosen as the participants in this study. Before administering the survey, participants were apprised of the study’s objectives and their entitlement to decline participation or withdraw from the study at any moment. Targeted questionnaires were distributed to accountants of PSOs to gather empirical data using a combination of convenience and snowball sampling techniques. As noted by Hair et al. (2018), large samples may cause statistical significance to become overly sensitive, resulting in a type I error. A large sample size could render any link statistically significant, regardless of its actual validity (Hair et al., 2018; Kline, 2023). In this regard, Boreham et al. (2020) asserted that the integrity of a sample was more contingent upon the meticulous selection of respondents than on the sample size itself. Hair et al. (2018) argued that the sample-to-variable ratio indicated a requisite minimum observation-to-variable ratio of 5:1. In the meantime, ratio of 15:1 or 20:1 was recommended as more favourable (Hair et al., 2018). In this research, the ratio of 20:1 was employed. Data were gathered from September 2024 to April 2025. The questionnaire was distributed to 770 individuals. A total of 723 responses were gathered, of which 712 were complete and appropriate for the study. The final sample size was determined by removing 11 responses due to incompleteness or inadequate information. Table 2 provided a detailed summary of the demographic data gathered during the survey.

4.3. Data Analysis Technique

This research utilised IBM AMOS 28, SPSS 30.0, Smart-PLS 4.1.0.9, and fsQCA 4.1 for the assessment of the suggested model. A descriptive analysis was conducted utilising SPSS software to assess the fundamental attributes of the data. Dash and Paul (2021) asserted that covariance-based structural equation modelling (CB-SEM) using AMOS-produced superior model fit indices relative to partial least squares structural equation modeling (PLS-SEM), which has been evolving its fit indices. Therefore, CB-SEM was utilised to evaluate the model fit indices in this research. The measurement and structural models were analysed using the PLS-SEM technique to validate the associations between independent and dependent variables. PLS-SEM was selected for its appropriateness in assessing causal links among latent constructs and for quantifying the degree and importance of these interactions using path coefficients (Oroni et al., 2025). Despite its efficiency as a robust analytical tool, PLS-SEM’s reliance on mean-centred symmetric model estimate may mask nuanced effects within empirical data, particularly in asymmetrical or nonlinear connections (Sukhov et al., 2023). To mitigate these constraints and enhance comprehension of the data, the study additionally utilised fsQCA, an asymmetric methodology. The final method employed in this work was fsQCA utilising version 4.1 software by Ragin (2023), which provided an asymmetric analytical approach. This methodology adeptly addressed theoretical complexity while offering both theoretical and practical insights into the dependent variable (Pappas and Woodside, 2021). The fsQCA was employed to investigate random data configurations, as it did not depend on pinpointing a singular “optimal” model. According to Chanda et al. (2025), fsQCA facilitated the simultaneous evaluation of many causal configurations, allowing for the analysis of nonlinear and intricate interactions that the SEM method may neglect.

5.1. Common Method Bias

To mitigate this common method bias (CMB), two approaches were implemented: statistical measures and procedural measures. Requirements for procedural measures included the elucidation of questionnaire items and the assurance of respondents’ confidentiality. To mitigate social desirability bias, respondents were guaranteed anonymity and confidentiality. By deliberately rearranging the presentation order and omitting construct titles from the survey, the questionnaire was intended to obfuscate the distinct association between constructs. The variance of the first factor in Harman’s single-factor analysis was 38.441%, which was below the recommended threshold of 50% in the context of statistical methods. Additionally, the variance inflation factor (VIF) was implemented to assess comprehensive collinearity in this investigation, requiring a value of less than 3.3 (Kock 2015). The collinearity analysis of this study revealed that the VIF fell within the range of 1.684–2.315, which was substantially lower than the 3.3 threshold. Consequently, these findings indicated that the results of this investigation were not influenced by CMB. As a result, this approach enhanced confidence in the quality of the data and the reliability of the subsequent analysis.

5.2. Model Fit

CFA was conducted using the maximum likelihood technique, and the support of AMOS 28.0. A thorough assessment of the model’s fitness, considering several parameters, was conducted through CFA. The CFA outcomes were assessed based on five key model fit indices namely the Chi-square/df ratio, goodness-of-fit index (GFI), comparative fit index (CFI), Tucker–Lewis index (TLI), and root mean square error of approximation (RMSEA). All indicators for model fitness of the proposed model presented in Figure 2 adhered to the specific criteria, rendering the model acceptable across all aspects. Consequently, the Chi-square/df was 1.410 (criterion: ≤ 3), RMSEA was 0.024 (criterion: < 0.080); GFI was 0.969 (criterion: > 0.9), CFI was 0.991 (criterion: > 0.9), and TLI was 0.989 (criterion: > 0.9) (Kline, 2023).

Figure 2

(Source: Extracted from statistical software and authors’ own study)

5.3. Measurement Model Assessment

This investigation followed the procedures established by Hair et al. (2024) for assessing the internal consistency, reliability, and convergent validity of the measurement model. This study utilised Cronbach’s alpha, composite reliability (rho_c), and composite reliability (rho_a) for evaluation. Cronbach’s alpha and composite reliability (rho_c) must exceed the criterion of 0.7 (Hair et al., 2024). Table 3 revealed that Cronbach’s alpha values varied from 0.848 to 0.891, while composite reliability (rho_c) values ranged from 0.892 to 0.920. Composite reliability (rho_a) must exceed 0.7 (Dijkstra and Henseler, 2015). The composite reliability values (rho_a) varied from 0.849 to 0.892, as demonstrated in Table 3. Convergent validity was evaluated by the average variance extracted (AVE) and factor loadings. Table 3 indicated that all model constructs had an AVE ranging from 0.622 to 0.696, surpassing the threshold of 0.5. Consequently, these outcomes confirmed convergent validity (Hair et al., 2024). Furthermore, the factor loadings displayed in Table 3 varied from 0.768 to 0.852, surpassing the 0.708 threshold (Hair et al., 2024). Thus, the findings validated the establishment of convergent validity and construct reliability.

Discriminant validity was evaluated using the heterotrait–monotrait ratio (HTMT) and the Fornell–Larcker criterion. To meet the Fornell–Larcker criterion for discriminant validity, the square root of AVE for each construct must exceed its correlations with all other constructs in the model (Fornell and Larcker, 1981). The results in Table 4 confirmed the existence of discriminant validity for all constructs in the proposed model. The discriminant validity of the measurement model was evaluated using the HTMT. In accordance with Henseler et al. (2015), the threshold for HTMT values must not exceed 0.9. According to the HTMT ratio, Table 4 demonstrated that the measurement model exhibited discriminant validity.

5.4. Structural Model Assessment

Table 5 and Figure 3 illustrated the coefficient of determination (R ²), cross-validated redundancy (Q ²), and effect size (f ²). Figure 2 displayed the outcomes of PLS-SEM, resulting in an R ² value of 0.278 for CEE. In other words, over 20% of the variance in CEE could be accounted for by AIA. The value of R ² was 0.453 for IGG. As such, over 40% of the variance in IGG could be accounted for by the AIA and CEE. The value of R ² was 0.309 for SDG_16. As such, over 30% of the variance in SDG_16 could be accounted for by the AIA, CEE, and IGG. Cross-validated redundancy, or Q ², quantified the predictive importance of exogenous constructs, with values exceeding 0 signifying that the model possessed predictive relevance. The values of Q ² were 0.172 for CEE; 0.304 for IGG; and 0.189 for SDG_16, signifying that the model possessed predictive relevance (Hair et al., 2024). According to Table 5, the direct effects of AIA on CEE (β = 0.528; t-value = 19.406; p-value = 0.000) was significant with a large effect (f ² = 0.386) and on IGG (β = 0.390; t-value = 11.796; p-value = 0.000) was significant with a medium effect (f ² = 0.201). Conversely, the direct effect of AIA on SDG_16 (β = 0.080; t-value = 2.199; p-value = 0.028) was significant with a small effect (f ² = 0.006). Although the effect magnitude was modest, it nonetheless signified that AIA exerted a significant direct impact on SDG_16. The direct effect of CEE on IGG (β = 0.380; t-value = 12.314; p-value = 0.000) was significant with a medium effect (f ² = 0.191). The direct effect of CEE on SDG_16 (β = 0.077; t-value = 2.166; p-value = 0.030) was significant with a small effect (f ² = 0.005). Although the effect magnitude was modest, it nonetheless signified that CEE exerted a significant direct impact on SDG_16. In addition, IGG was confirmed to affect SDG_16 (β = 0.454; t-value = 11.634; p-value = 0.000) in a significant and positive manner with a medium effect (f ² = 0.163). Hence, H1–H6 were supported. Hence, H1–H6 were supported.

Figure 3

(Source: Extracted from statistical software and authors’ own study)

The significance of AIA’s indirect effects on IGG and SDG_16 through CEE were assessed. Given that the direct effect of AIA on IGG was also supported and that the indirect effect was significant (β = 0.201; t-value = 10.231; p-value = 0.000), it disclosed that CEE partially mediated the relationship between AIA and IGG. Given that the direct effect of AIA on SDG_16 was also supported and that the indirect effect was significant (β = 0.041; t-value = 2.141; p-value = 0.032), it disclosed that CEE partially mediated the relationship between AIA and SDG_16. Hence, H7 and H8 were supported.

5.5. Fuzzy-set Qualitative Comparative Analysis

Calibration. The preliminary stage in fsQCA analysis was calibration (Pappas and Woodside, 2021). The initial phase involved data calibration, wherein raw data (survey data) was converted into scores representing set memberships (Ali et al., 2025). The initial 7-point Likert scale was recalibrated to a range of values between 0 and 1. A score of 0 signified complete non-membership, a score of 1 denoted full membership, and a score of 0.5 showed the crossover point. In accordance with Pappas and Woodside’s (2021) recommendation, this study calculated 95%, 50%, and 5% of the metrics as three anchors and adjusted the original data into full membership, crossover, and full non-membership, respectively. However, Tho and Trang (2015) proposed setting the full membership threshold at a rating of 6, the full non-membership threshold at a rating of 3, and the crossover point at 4.5. This study established the definitive non-membership criterion at 3, as opposed to the threshold of 2 utilised by Ordanini et al. (2014). This was because Vietnamese employees demonstrated a propensity to choose the right side (strongly agree) of the scale when answering survey questions (Tho and Trang, 2015).

Concerning the impacts of AIA and CEE on IGG, for a causal configuration or a singular condition to be deemed sufficient, consistency must surpass 0.80 and coverage must exceed 0.20 (Pappas and Woodside, 2021). The findings from fsQCA displayed in Table 6 highlighted that the three fsQCA solutions (complex, parsimonious, and standard) produced identical configurations. The scores for consistency and coverage of AIA and CEE in Table 6 surpassed the recommended thresholds (Consistency > 0.80, Coverage > 0.20). The findings indicated that none of these antecedent requirements above the 0.9 consistency criterion for need, signifying that they did not represent or resemble necessary conditions for IGG achievement. In light of these findings, we conducted a further analysis of the interplay between factor combinations and IGG achievement utilising fsQCA 4.0 software.

Researchers using fsQCA were advised to present two analyses: one with a relatively permissive consistency threshold and another with a more restrictive consistency threshold. This study conducted an additional analysis in accordance with this proposal, employing a consistency threshold of 0.90. A comprehensive examination of the configurations underscored the fact that none of the conditions could be considered adequate for the attainment of the IGG. However, it was possible to combine these conditions. This was an INUS condition, which was a condition that was insufficient yet necessary for the outcome, despite being superfluous in itself (Mackie, 1965). The configuration in Table 7, characterised by a significant integration of AIA and CEE as fundamental elements, exhibited a consistency of 0.912807 and a raw coverage of 0.474271. This combination underscored a robust pathway towards elevated IGG attainment, signifying the collective significance of these elements.

Concerning the impacts of AIA, CEE, and IGG on SDG_16, the findings from fsQCA displayed in Table 8 highlighted that the three fsQCA solutions (complex, parsimonious, and standard) produced identical configurations. The scores for consistency and coverage of AIA, CEE, and IGG surpassed the recommended thresholds (consistency > 0.80, coverage > 0.20). The findings indicated that none of these antecedent requirements above the 0.9 consistency criterion for need, signifying that they did not represent or resemble necessary conditions for SDG_16 achievement. In light of these findings, we conducted a further analysis of the interplay between factor combinations and SDG_16 achievement utilising fsQCA 4.0 software.

Table 9 illustrated several solutions identified that contributed to significant attainment of SDG_16, classified into concise solutions.

Configuration 1, characterised by a significant integration of AIA and CEE as fundamental elements, exhibited a consistency of 0.915166 and a raw coverage of 0.423493. This combination underscored a robust pathway towards elevated SDG_16 attainment, signifying the collective significance of these elements.

Configuration 2, characterised by a significant integration of AIA and IGG as fundamental elements, exhibited a consistency of 0.87086 and a raw coverage of 0.522778. This combination underscored a weak pathway towards elevated SDG_16 attainment.

Configuration 3, characterised by a significant integration of CEE and IGG as fundamental elements, exhibited a consistency of 0.902587 and a raw coverage of 0.485375. This combination underscored a robust pathway towards elevated SDG_16 attainment, signifying the collective significance of these elements.

6.1. Discussion and Implication

6.2. Limitation and Agenda for Future Research

The research was inevitably impeded by certain limitations. The execution of this investigation in Vietnam was the primary constraint. Consequently, future research may focus on a variety of national and cultural contexts to facilitate comparisons with the findings of the current study. Second, the participants in this study were selected through a combination of non-random convenience and snowball sampling techniques. Thus, the generalizability of our findings may be improved through additional research that utilizes systematic sampling techniques to obtain a large sample size. Third, the data used for analysis were self-reported, which further restrict the generalizability of the research findings. The responses to the accountant survey were extremely beneficial; however, the findings may have been influenced by respondents’ subjective perceptions and potential common method bias. In addition, because the study adopted a cross-sectional design, it could not fully capture changes in the relationships among the variables over time. Consequently, future research should employ longitudinal designs to examine the stability and development of these relationships across different time periods. Finally, future research should integrate additional mediating or moderating variables to provide a more comprehensive explanation of the proposed relationships.