3.1 Measurement Model

3.2 Structural Model

3.2.1 The Effects of PGE on POS, OP and OC

Following the measurement models introduced in the previous section, a simplified schematic of the SEM to be tested with our modeling pipeline is portrayed in Figure 1. The main aim of this model is to operationalize testable functional relationships between PGE, POS, OP, and OC.

H1.PGE will be positively related to POS.

We define a functional relationship between PGE and POS, denoted $$$ R1 $$$. Organizational support theory posits that the extent of POS exerts an influence on employees' beliefs regarding their legitimacy as organizational members (Ashforth et al. 2008). Moreover, this theory argues that a sense of legitimacy as organizational members prompts employees to form psychological and emotional bonds with the organization (Rhoades et al. 2001). Research has also demonstrated that POS is related to the perception of fairness of the organization (Glavas and Kelley 2014), and that this form of perceived favorable treatment should increase POS (Rhoades and Eisenberger 2002). Put differently, the act of treating individuals with fairness and demonstrating concern for their well-being serves as a signaling mechanism, indicating to employees the likelihood of receiving favorable treatment (Glavas and Kelley 2014). Given that the SDGs, and therefore organizational SDG engagement, entail fair and just treatment of all individuals (e.g., SDG5), with a direct emphasis on prioritizing well-being (e.g., SDG3), we expect that a similar signaling function will be present.

H2.PGE will have a positive total effect on OP.

The modeled relationship of PGE and OP in the structural model is more nuanced convoluted than it is with POS as a “target construct.” From this point onwards, we will break each working hypothesis into individual sub-hypothesis which we will describe, motivate and test sequentially via the modeling pipeline described in later sections.

H2a.PGE will be positively related to OP.

Prior research has revealed that when employees are aware that external observers positively perceive and assess their organization, it leads to an improvement in their self-esteem and a heightened sense of self-worth, in turn positively evoking one's feelings of pride (Dutton and Dukerich 1991; Oo et al. 2018). It has also been suggested that employees are likely to feel pride if they believe that external stakeholders associate the organization with a social cause (De Roeck et al. 2016). Considering that the SDGs and, consequently, organizational engagement with the SDGs, inherently encompass diverse social causes and issues (e.g., SDG1, SDG2, …), the expectation is that this engagement will positively influence employees' sense of organizational pride, given that external stakeholders might associate the organization with these overarching objectives. Therefore, we propose $$$ R2 $$$ as a testable functional relationship.

H2b.POS will mediate the relation between PGE and OP.

Theory also suggests that employees that perceive organizational support are likely to hold positive views about their organization and exhibit a favorable orientation towards it (Eisenberger and Stinglhamber 2011; Im and Chung 2018). Moreover, heightened levels of POS play a pivotal role in fostering positive psychological and emotional connections, and in exerting an influence on employees' beliefs regarding their legitimacy as organizational members (Ashforth et al. 2008; Rhoades et al. 2001). And as OP is reflective of employees' sentiments regarding their organizational membership, it can be inferred that POS contributes in part to the cultivation of OP, which we encode in the structural model via $$$ R6 $$$. Hence, the relevant indirect effect that will allow to test the role of POS as a mediator variable between PGE and OC is the product coefficient given by $$$ R1\times R6 $$$. We hypothesize that this product will indeed act to strengthen the direct effect discussed above.

H3.PGE will have a positive total effect on OC.

As previously mentioned, our focus lies primarily on examining the relationship between perceptions of organizational SDG engagement and affective OC. Similarly to H2, we will introduce several sub-hypotheses for H3.

H3a.PGE will be positively related to OC.

According to social identity theory, individuals tend to develop identification within a social context, and use such grouping to distinguish themselves from others (Tajfel and Turner 1985). This process leads to the formation of a social identity. Moreover, individuals partly derive their self-concepts from social identities associated with social groups to which they belong (De Roeck et al. 2014). Given that people are naturally inclined to develop a positive social identity (Ahmad et al. 2020), it can be argued that these describe one's attributes as a member of that group (Turker 2009). In other words, these specify what one ought to think and feel, as well as how one should behave (Turker 2009). Social identity theory thus offers a logical rationale for the suggested positive correlation between organizations behaving in a responsible way and the organizational commitment exhibited by their employees, as employees commit themselves to organizations that align with and embody those attributes, and specifically that do good for the environment and wider society (Azim 2016). Considering that organizational SDG engagement fundamentally involves promoting well-being for various stakeholders, encompassing people, the planet, and broader society, it can be anticipated that it will exert a similarly positive impact on organizational commitment. Thus, it could be plausible to test for the existence of a direct effect of PGE on OC, as denoted by $$$ R3 $$$.

H3b.POS will mediate the relationship between PGE and OC, both independently and through its relationship with OP.

Research has shown that both POS and OP are positively and significantly related to OC (Glavas and Kelley 2014; Rhoades and Eisenberger 2002; Zhou et al. 2018). Organizational support theory posits that positive work experiences resulting from deliberate and voluntary actions by the organization contribute to the development of POS (Eisenberger et al. 1986). In turn, POS is expected to enhance OC through the reciprocal exchange of positive regard and care, as well as the integration of organizational membership and role status into one's social identity (Rhoades et al. 2001). Additionally, theory argues that OP leads to high levels of identification with the organization, and thus to a positive evaluation of organization status and its membership (Ashforth and Mael 1989; Tyler 1999). Consequently, employees who experience a sense of pride in their organization tend to develop psychological attachment and adhere to the values and regulations of the organization. This, in turn, fosters a heightened level of OC (Riketta 2005). Consequently, we also include $$$ R4 $$$ and $$$ R5 $$$ in the structural model.

We therefore conjecture that both POS and OP will serve as mediators for the association between employees' PGE and OC. However, the compounded nature of the modeled relations diagrammed in Figure 1 compels us to identify 3 potential concurrent channels (or indirect compound paths in SEM terminology), that can channel and conform the total indirect effect of PGE on OC. Here, our structural model implements 2 distinct pathways that go through POS starting from PGE which end at OC. The first one is the mediating relationship of POS via $$$ \mathrm{PGE}\to \mathrm{POS}\to \mathrm{OC} $$$, directly implemented via the product of path coefficients $$$ R1\times R4 $$$. Yet, there is a second indirect pathway in via $$$ \mathrm{PGE}\to \mathrm{POS}\to \mathrm{OP}\to \mathrm{OC} $$$, or equivalently, quantified via the product of path coefficients $$$ R1\times R6\times R5 $$$.

H3c.OP will mediate the relationship between PGE and OC.

Finally, as motivated, the pathway $$$ \mathrm{PGE}\to \mathrm{OP}\to \mathrm{OC} $$$ is the last one to test for in our structural model. Similarly, this pathway is implemented and quantified via $$$ R2\times R5 $$$ path coefficient product.

3.3 Sample Selection and Data Collection

A quantitative research design was developed by means of an anonymous survey questionnaire and a sampling design directed towards the employees of a Flemish regional governmental organization (a province). As a result of pandemic-related restrictions during the survey period in October 2021, online questionnaires were disseminated. A total of 1988 anonymous survey invitations were emailed to all employees, inviting them to voluntarily participate in a Dutch-language Qualtrics survey between October 13th and October 29th, 2021. From this pool, 1634 employees initiated the survey, with the first step requiring electronic informed consent in order to proceed.

Further, an additional 245 employees were excluded from further analysis as they aborted the survey after providing basic information (education, department, gender, and age), but before being informed about the study's SDG-related theme and objective. Thus, the final usable respondent pool comprised of 1389 employees of the organization, resulting in a response rate of 69.9%. This means our current modeling pipeline will lack explicit mechanisms for response sensitivity analysis, as recommended by Newman (2014) for rates below 30%. It is also worth noting that according to this author (Newman 2014), these 1387 employees can be considered partial + full respondents, which we will refer to as our “effective sample.” Furthermore, the 1389 respondents encompass employees from various departments, job levels, seniority categories, ages, and genders, enhancing the representativeness of the sample. To ensure this representation, the survey underwent extensive pre-testing rounds involving employees across all job levels, seniority categories, and demographic groups.

Based on this sample, 3.96% of these respondents failed to fill at least one of the 17 manifest items for the PGI, 6.12% for PIGE, and 6.79% for the PGE constructs. The other latent constructs we measured exhibit a varied construct-level MVP, ranging from close to 0% for the single item constructs and 8% to 12% for the other multi-item constructs we collected from the survey. Hence, according to Newman (2014), either Full Information Maximum Likelihood (FIML) or alternative Multiple Imputation (MI) methods become necessary to tackle issues of bias and/or variance deflation.

3.4 Modeling Pipeline: Embedded PLS-SEM Estimation Pipeline Within the MICE Framework

A recent simulation study by Amusa and Hossana (2024) benchmarked several alternative techniques specifically in the context of PLS-SEM applications across different assumptions for the pattern and MVP. These authors conclude that across several missingness processes and MVP rates, regression-based imputation methods consistently outcompeted Expectation Maximization (EM) methods and other more streamlined alternative competitors across a wide range of settings. Amusa and Hossana (2024) remark that while regression-based imputation systematically outcompetes EM in terms of mean absolute error, both methods are competitive in estimator precision as MVP increases relative to available sample size. However, they limit their findings to contexts that are dealing only with reflective constructs, which aligns with our current implementation. Furthermore, these authors remark that there is an overall lack of a coherent guideline, including a full modeling pipeline to systematically address the issue.

Thus, following both Newman (2014) and Amusa and Hossana (2024), we propose the following applied modeling pipeline by integrating the framework for PLS-SEM modeling by Hair et al. (2021) within the MICE framework by van Buuren and Groothuis-Oudshoorn (2011). The following guideline presents a schema of the implemented procedure (the full modeling pipeline is documented and implemented in R 4.3.2).

4.1 Measurement Model

The detailed results for the measurement model of the 3 introduced latent scales (PGE, PIGE and PGI) are presented in Table 1. Here, the estimated reflective loadings for all 3 scales and their corresponding statistics from the implemented MICE-RF imputation pipeline for the bias-corrected PLS-SEM framework are portrayed. First, it can be observed that none of the reported $$$ fmi $$$ coefficients for any loadings in the table exceeds 0.1 (nowhere close to the 0.5 threshold detailed in previous sections). It can be noted that the observed behavior of the $$$ fmi $$$ statistics for all loadings in the model can be attributed to the fact that the within variance dominates the between variance consistently. That is, the variability that can be attributed to the bootstrap sampling through which the within variance is obtained is systematically larger than the variability that can be attributed to missing value imputation (or, the between variance).

Also, it should be noted that item 2 (eradicating hunger SDG) and item 14 (safeguarding aquatic life SDG) were excluded from further analysis in these measurement models. This decision was primarily motivated by the inherent conceptual overlap between SDG1 and SDG2 (eradicating poverty/hunger SDGs) and between SDG14 and SDG15 (safeguarding terrestrial/aquatic life SDGs), resulting in a high level of correlation between items 1 and 2, as well as items 14 and 15. We then follow the standard guidelines provided in Hair et al. (2021) to establish both composite reliability and discriminant validity for our 3 proposed scales.

Looking at the pooled loading estimators in Table 1, almost the entirety of the $$$ 3\times 15 $$$ remaining manifest indicators have a 95% confidence interval which includes 0.7, with the sole exception of PGE1. In terms of indicator reliability, this implies that we cannot rule out that our constructs are able to explain 50% of the variance of their respective manifest variables (or close to 50% for PGE1). Further, the confidence interval is decidedly north of 0.4 for absolutely all items, which Hair et al. (2021) recommend as the minimum threshold up to which content validity considerations can be valid reasons to avoid removal of any further manifest item. Lastly, the same pattern is observed for all other latent variables under consideration (i.e., OC, OP, PLS, PSS, and POS). For brevity of exposition, we defer the interested reader to Table A1 in the Appendix A, which contains the detailed statistics for all these reflective constructs. Here, the 95% confidence intervals for OC1, OP2, POS8_R do not include (but are close to) 0.7, as was the case for PGE1.

Second, Table 2 provides 2 sets of pooled composite reliability statistics, the $$$ rhoA $$$ and the $$$ AVE $$$ estimators (denoted $$$ \theta $$$). Given these statistics follow non-normal distributions, the table also provides the estimates for the probability for the event $$$ P\left(\hat{\theta}\le t\right) $$$ described in the previous section. Following Hair et al. (2021), these events are $$$ P\left(\hat{rhoA}\le 0.7\right) $$$ and $$$ P\left(\hat{AVE}\le 0.5\right) $$$, which would capture the probability that each individual latent construct fails to meet the standard threshold for composite reliability across re-samples/imputations. As the table portrays in every tested case, the computed probability of not meeting composite reliability standards was estimated to be effectively almost 0. For brevity of full exposition, we defer the interested reader to Tables A2 and A3 in the Appendix A.

Third, Table 3 shows a similar layout for the HTMT statistics for all cases related to our 3 constructs (i.e., PGE, PIGE, and PGI), while we defer other pairs of latent constructs (and detailed statistics) for Table A4 in the Appendix A. Similarly to the previous case, we can see that discriminant validity is certainly achieved for all cases given that the estimated probabilities of not meeting Hair et al. (2021) standards, or $$$ P\left(\hat{HTMT}\ge 0.9\right) $$$, are again virtually 0 for all tested cases. Thus, we can conclude that the tested measurement models for PGE, PIGE and PGI (as well as all other tested latent constructs) models are close to, or even beyond minimum confirmatory thresholds in some cases.

Finally, to assess potential multicollinearity issues, we examined the Variance Inflation Factors (VIF) for all latent variables in the model. As reported in Table A8 in the Appendix A, all VIF values remain well below critical thresholds, confirming that multicollinearity does not pose a concern in our analysis.

4.2 Structural Model

To assess the structural model, we start with model fit statistics portrayed in Table 4 for all endogenous latent constructs in our model. Similarly, to the patterns shown in Tables 1 and A1, here we can observe that the estimated values of the $$$ fmi $$$ statistics for the Fischer transformations of our $$$ {R}^2 $$$ coefficients fall well below 0.1 in all cases, which further confirms the stability of our estimations across imputed samples. Yet, the 95% confidence intervals for the back-transformed adjusted $$$ {R}^2 $$$ coefficients suggest that the structural model's explanatory power for POS, OP and OC can be considered “moderate” (Hair et al. 2021).

Finally, Tables 5–7 provide summary statistics for the direct, partial/total indirect and total effects relevant to each working hypotheses introduced in the previous section. As was done in previous cases, detailed statistics and additional coefficients not central for hypothesis testing are portrayed in Tables A5–A7 for each dependent variable of the structural model (viz, POS, OP, and OC). It should be noted that the effects of all additional control variables are included in these tables, but not explicitly addressed in the “Results” and “Discussion” sections. Once again, these tables corroborate the argument made based on the observed values of the $$$ fmi $$$ statistics. Furthermore, Table 5 portrays that the direct effects of PGI and PIGE on POS are not statistically significant. Yet, here we can observe that PGE does show a significantly positive “low to moderate” direct effect on POS (based on the estimated coef. and 95% confidence interval for $$$ R1 $$$), thus corroborating H1.

Moving onto Table 6, we can observe that, akin to H1, controlling for PIGE and PGI does not yield significant contributions in the structural equation for OP. Here, we can also observe the relevant direct effects, total indirect effects and total effects relevant to characterize and test H2 via H2a and H2b. On the one hand, we see that the coefficient and 95% conf. interval for $$$ R2 $$$ indicate that there is a significantly positive, but “low” effect of PGE on OP. On the other hand, we can see that these figures for the $$$ R1\times R6 $$$ composite path also show that there is a significantly “low” total indirect effect that is mediated by POS between PGE and OP. This compounds to corroborate H2, as the total effect (given by the sum of the described effects) compounds to conform a significantly positive “low to moderate” effect of PGE on OP.

Finally, Table 7 confirms that PIGE and PGI do not compound a total effect on OC either. Further, we will focus on characterization and testing of H3 via H3a, H3b, and H3c as done in the previous case. We start with $$$ R3 $$$, which indicates that we should reject H3a given that there is no significant direct effect that we observe from PGE to OC. Instead we can clearly see that all the estimated coefficients for partial indirect effects (e.g., $$$ R1\times R6\times R5 $$$, $$$ R1\times R4 $$$, and $$$ R6\times R5 $$$) are positive and statistically significant. Together, they do compound to what can be called a “low to moderate” total indirect effect of PGE on OC (since the 95% conf. interval does not exceed 0.25). However, an interesting observation from the partial direct effects is that both POS and OP (as antecedents of OC in the model) play the role of full mediators for the total effect PGE. In other words, even if there is a “low” effect of PGE on OC, it can be fully understood and quantified by understanding the direct and indirect links of PGE both with POS and OP.

5.1 Theoretical Implications

This study explores the potential effects of organizational engagement with the SDGs on employee attitudes (RQ1) and examines employees' perceptions of this engagement (RQ2). Grounded in stakeholder theory, organizational support theory, and social identity theory, we outline pathways through which Perceived Organizational SDG Engagement (PGE) may influence attitudes like Perceived Organizational Support (POS), Organizational Pride (OP), and Organizational Commitment (OC). Our empirical findings provide strong evidence supporting these attitudinal outcomes.

The results confirm H1, showing a significant positive direct effect of PGE on POS. Additionally, PGE has a significant total effect on OP, both directly and through POS, supporting H2, H2a, and H2b. This highlights that employees' perceptions of organizational SDG engagement enhance their perceived organizational support and pride. However, the direct effects of the controls PGI and PIGE on POS and OP are insignificant, suggesting that individual preferences may not influence these attitudes, contrary to prior research (van Dick et al. 2019). Regarding H3, no direct effect of PGE on OC is found, leading to the rejection of H3a. However, PGE has a significant total effect on OC, fully mediated by POS and OP, supporting H3, H3b, and H3c. This underscores the indirect pathways through which organizational SDG engagement shapes employee commitment.

These findings are particularly significant given the limited research on organizational SDG engagement's impact on stakeholders. Existing studies have primarily focused on external stakeholders (Barta et al. 2023; Foroudi et al. 2022; Stauropoulou et al. 2022), with little exploration of employee outcomes. To our knowledge, this study is the first to examine how organizational SDG engagement affects employees, marking a crucial step toward a deeper understanding of its organizational impact. Our research enriches the theoretical landscape and sets the stage for future studies on employee perceptions and outcomes in this context.

A second key theoretical contribution lies in conceptualizing and operationalizing employee SDG perception scales. Given the lack of validated employee-level measures, our study introduces three 17-item scales: Perceived SDG Importance (PGI), Perceived Importance of Organizational SDG Engagement (PIGE), and Perceived Organizational SDG Engagement (PGE). These scales address a critical research gap by capturing employees' nuanced perceptions of organizational SDG engagement, providing a foundational framework for future studies on SDG perceptions.

This research also offers a notable theoretical contribution by extending the current literature on employee outcomes within the realm of OS concepts, such as CS and CSR. Importantly, our study is contextualized within public organizations, providing insights into the attitudinal outcomes of public sector employees, a demographic that has historically received limited attention in prior research on OS (Gelderman et al. 2017; Lozano and Barreiro-Gen 2023). By testing whether some of the private sector effects known from OS research also apply to the public sector, we validate these effects in a new context, thereby enhancing the robustness and scope of current research findings. This approach incidentally responds to calls from authors who strongly recommend that empirical SDG research be extended to diverse contexts (Mahajan et al. 2024).

Finally, we have succeeded in implementing a fully integrated, customizable modeling pipeline for dealing with the issue of missing values within PLS-SEM research in a principled manner. As part of this effort, we have provided a step-by-step documentation integrating the seminR framework within the MICE framework using RF to robustly specify conditional imputation models. As remarked by Amusa and Hossana (2024), the lack of similar implementations makes our effort valuable for practitioners not only within our particular research domain, but also in similar application contexts across different disciplines.

5.2 Limitations and Perspectives for Future Research

Our study has limitations that suggest avenues for future research. First, the use of self-administered surveys at a single point in time may introduce Common Method Bias (CMB), potentially affecting response accuracy. Although we mitigated this risk by ensuring respondent anonymity (Podsakoff et al. 2003), self-reported data remain a limitation. Additionally, relying on a single data collection method restricts the depth of analysis. Future research should incorporate multiple data collection methods and advanced statistical techniques to enhance validity.

Second, focusing exclusively on employees within a single organization limits the generalizability of our findings. Potential cultural biases highlight the need for broader studies across diverse organizational settings and stakeholder groups. Comparative analyzes across different contexts, along with examining factors like organizational size, industry, and location, would provide deeper insights into SDG engagement's impact on employees. Furthermore, while we believe our scales apply to various stakeholders; this could not be empirically tested. Nonetheless, our study lays a strong foundation for future research to validate these scales across different environments, ensuring their robustness and applicability in diverse organizational and cultural settings.

Third, our study's timing during the COVID-19 era presents a limitation. Research shows the pandemic negatively affected employee mental health (Hamouche 2023), potentially influencing behaviors and attitudes unrelated to organizational SDG engagement. This may have contributed to the “low to moderate” results observed. Future studies should conduct comparative analyzes across different time frames to distinguish the lasting effects of the pandemic from the true impact of SDG engagement on employee attitudes and behaviors.

Fourth, the deliberate exclusion of the term “SDG” in the proposed scales offers advantages but also introduces limitations. Without explicit reference to the SDGs, we cannot directly assess the instrument's impact. Instead, our insights focus on the underlying themes of the SDGs. Future research should develop scales that explicitly mention “SDGs” to improve methodological rigor and allow for comparative analysis. This approach would help identify potential differences in outcomes and clarify the impact of explicitly referencing the SDG framework in research. Additionally, this paper does not provide any commentary on the SDGs themselves. We do not evaluate their effectiveness, address their limitations and criticisms, or discuss the feasibility of their realization, all of which have been widely debated (Arora-Jonsson 2023; Ivic et al. 2021; Leal Filho et al. 2023). Future studies on stakeholder effects of organizational SDG engagement should carefully consider these critiques and the broader context of the SDGs, as they can influence how stakeholders engage with and respond to the goals.

Finally, to the best of our knowledge, our modeling pipeline adheres to the state of the art in reducing biases due to missing values within PLS-SEM. Although residual biases may persist, our approach offers a robust foundation for fellow researchers. While further refinement can be the subject of future research, we hope our approach facilitates others in the broader academic community to navigate these challenges effectively.

5.3 Practical Implications

This study highlights the practical importance of organizations aligning with the SDGs, showcasing the dual impact of such engagement. Beyond contributing to global betterment, SDG engagement positively influences employee attitudes. Understanding these attitudes allows organizations to tailor SDG strategies effectively, fostering a positive workplace environment. It is crucial that employees perceive SDG efforts as genuine rather than tokenistic for them to be impactful. This underscores the importance of communication, with SDG reports addressing both external and internal stakeholders to foster engagement and understanding across all levels.

In addition, the introduction of individual-level SDG scales in the study provides a practical instrument for assessing and enhancing employee attitudes, enabling organizations to execute more targeted and impactful sustainability initiatives. These scales not only facilitate the measurement of employee perceptions and attitudinal dynamics, but also serve as a valuable tool for evaluating employees' comprehension of the undertaken SDG efforts and shedding light on whether they deem such initiatives of paramount importance. This dual functionality equips organizations with actionable insights to refine communication strategies and ensure that employees grasp the profound significance of SDG initiatives.

Overall, the study emphasizes the broader benefits that extend beyond societal contributions, emphasizing the value of SDG engagement for organizational success and employee well-being. Nevertheless, the authors emphasize the following. It is crucial to acknowledge that the SDGs should not be solely instrumentalized to influence employee attitudes. Sincere and genuine commitment to societal betterment must remain at the forefront of such organizational endeavors and is, as this study highlights, in fact the foundation of positive employee attitudes.