A data-driven typology of individually perceived workplace environments and employee well-being

Abstract

This study adopts a data-driven, employee-centered approach to explore how individually perceived workplace environments shape well-being. Analyzing over 1.3 million observations from 2017 to 2021, we identify nine key job demand and resource factors—including workload, emotional burden, job autonomy, organizational integrity, and support—using factor analysis. Rather than examining these dimensions in isolation, we apply Gaussian mixture models (GMM) to uncover ten distinct workplace types that reflect co-occurring patterns of demands and resources as experienced by individual employees, ranging from Grade D (least favorable) to Grade A1 (most favorable). Our findings reveal substantial variation in employee well-being—including mental health, work engagement, job satisfaction, and workplace cohesiveness—across these clusters. We further examine how changes in workplace type over time relate to changes in individual well-being, highlighting the importance of dynamic and subjective workplace conditions. By centering the employee perspective and identifying lived patterns of workplace experience, this study offers novel insights into workplace dynamics and supports more responsive, employee-oriented organizational policies and practices.

Supplementary Information

The online version contains supplementary material available at 10.1038/s41598-025-18332-z.

Introduction

The workplace plays a pivotal role in shaping employee well-being and organizational outcomes. A growing body of research highlights how the quality of the work environment influences mental health, job satisfaction, engagement, and turnover^1,2. To understand the mechanisms through which these effects occur, the Job Demands–Resources (JD-R) model has been widely adopted as a theoretical framework. It distinguishes between job demands—elements that require effort and may incur psychological or physical costs—and job resources—elements that support work goals and foster personal development^3,4. While the JD-R model has significantly advanced our understanding of how individual workplace factors relate to employee outcomes, most empirical applications examine these factors in isolation or through linear interactions. Less is known about how multiple demands and resources combine in real-world settings to form distinct types of perceived workplace environments. Furthermore, few studies examine how changes in these multidimensional environments affect employee well-being over time⁵.

Typology-based approaches offer a promising direction for integrating complex workplace characteristics into coherent categories^6–8. However, existing workplace typology studies tend to be qualitative or industry-specific, limiting their generalizability and comparability^9,10. There remains a lack of large-scale, quantitative typology studies that systematically classify perceived workplace environments across demographic groups and sectors using theoretically grounded survey data. Early models such as Karasek’s Demand–Control¹¹ and Siegrist’s Effort–Reward Imbalance¹² provided foundational typologies for understanding occupational stress by highlighting specific interactions among job characteristics. The Job Demands–Resources (JD-R) model extended these frameworks by allowing a broader range of demands and resources relevant across occupations. In addition, most workplace studies operate at the organizational or departmental level, implicitly assuming that employees share a common environment. Yet, individual workplace experiences are inherently subjective and shaped by employees’ perceptions of workload, autonomy, interpersonal relationships, and support. Even minor organizational adjustments can affect individuals differently depending on role, context, or coping style^13,14. Recent research has employed person-centered methods such as latent profile or factor mixture models to derive empirical work condition typologies and explore their effects on employee well-being. For instance, Keller et al.¹⁵ and Portoghese et al.¹⁶ demonstrate how such approaches enrich JD-R-based analyses. Building on this line of research, our study applies a data-driven clustering method to large-scale survey data to identify profiles of perceived working conditions and examine their associations with employee well-being.

Using over 1.3 million observations from a nationally mandated occupational stress survey in Japan, we extract nine key workplace dimensions via factor analysis and classify employees into ten workplace types using Gaussian mixture modeling. The typology reflects individual employees’ subjective experiences of job demands and resources, rather than organizational structures. We then examine the associations between these workplace clusters and well-being outcomes—such as mental health, job satisfaction, engagement, and workplace cohesiveness—and assess how changes in cluster membership relate to changes in employee well-being over time. In doing so, our study advances the JD-R literature by operationalizing dynamic, combined perceived workplace environments through an empirical typology framework, while offering practical insights for organizational diagnostics and policy interventions. While a number of meta-analyses have synthesized robust evidence linking job demands and resources to employee well-being, these studies primarily focus on variable-level associations^17,18. Our person-centered, clustering approach builds on this foundation by identifying naturally occurring combinations of demands and resources as experienced by individuals. Additionally, our longitudinal design allows us to examine how transitions across these profiles relate to changes in well-being over time—offering complementary insights to those provided by aggregated, cross-study syntheses.

In the main findings, we identified nine key workplace dimensions—three job demands (workload, emotional burden, physical burden) and six job resources (organizational integrity, job autonomy, leadership, surrounding support, rewards, and job stability). Using Gaussian mixture modeling, we classified employees into ten distinct workplace clusters ranging from Grade D (least favorable) to Grade A1 (most favorable). We found significant and consistent differences in employee well-being outcomes across these clusters, with higher-grade clusters associated with better mental health, higher engagement, greater job satisfaction, and stronger workplace cohesiveness. Furthermore, changes in workplace cluster membership over time significantly predicted changes in well-being, underscoring the importance of dynamic workplace configurations in shaping employee experiences.

These findings contribute to the literature in several key ways. First, by applying a data-driven clustering approach, we move beyond traditional JD-R applications that examine job demands and resources in isolation. Instead, we capture how multiple workplace factors interact to form empirically grounded workplace typologies, offering a more holistic understanding of employee experiences. To the best of our knowledge, this study is the first attempt to identify the type of workplace in the JD-R framework. Second, while earlier typological models provided theory-driven profiles, our study builds on this lineage using an inductive approach grounded in large-scale, nationally mandated survey data across diverse industries, time periods, and demographic groups—enhancing the generalizability and relevance of our findings. Third, our analysis highlights the value of treating the workplace as a subjective and individualized environment. Rather than assuming homogeneity within organizations, our method reveals nuanced variation in how employees experience their working conditions, and how these experiences evolve over time. Taken together, our study extends the JD-R model by incorporating combined and dynamic workplace dimensions into a typology framework and provides practical tools for organizational diagnostics and well-being interventions.

This paper is structured as follows. Section “Literature review” introduces the workplace typologies in the existing literature and discusses the relationship between workplace change and employee well-being. Section “Methods and materials” explains the method of clustering and DiD (Difference in Difference) design to examine the impact of workplace change on employee well-being, as well as the dataset used in this study. Empirical results are discussed in Sections “Results” and Conclusion concludes.

Literature review

Workplace typology

Existing literature extensively explores workplace issues from various perspectives across different industries. Typology studies play a crucial role in understanding the diverse workplace environments, policies, and behaviors^9,10. For instance, Chan⁶ identified four types of workplace fun at the individual level, which is an important factor for employee well-being in hospital management. Hyett and Parker⁷ investigated four aspects of employee well-being at the workplace, including work satisfaction, organizational respect for the employee, employer care, and intrusion of work into private life, based on a 31-item workplace well-being questionnaire. Scharf et al.¹⁹ identified eight dimensions to predict the probability of job-related injuries in high-risk industries at the firm level. Russ-Eft⁸ proposed a typology of the elements that affect workplace learning and transfer: situational, pretraining, training design, and post-training. In addition, studies also discussed the workplace policies typology and analyzed four types of worker-friendly policies in terms of policy focus, goal, benefit, financial constraint, and target audience²⁰. However, most workplace typology research are qualitative study and usually focuses on specific industries. There is little discussion of quantitative and comparative analysis across a wider range of industries.

Theoretically, the Job Demands-Resources (JD-R) model provides a valuable framework for understanding the typology of perceived workplace environments. According to the JD-R model, workplace characteristics can be categorized into two main categories: job demands and job resources²¹. Job demands refer to aspects of the job that require sustained physical or psychological effort and are associated with costs for employees, such as high workload, time pressure, and role ambiguity. On the other hand, job resources are aspects of the job that support employees in achieving work goals, reduce job demands, and stimulate personal growth and development, such as social support, autonomy, feedback, and opportunities for skill development. In a result, the job demands and resources are combined to affect employee well-being, such as occupational stress, job satisfaction, etc.

Based on the JD-R model, previous studies explore the determinants for occupational stress or other employee well-being indicators. For example, Jannoo et al.²² confirmed a two-factor structure, job pressure and lack of organizational support, in job stress experienced by academic staff in higher education institutions. In risk industries, higher job demand may negatively impact psychological conditions that can lead to unsafe behavior in the workplace²³. On the other hand, recent studies have started to use data-driven approaches to thoroughly examine the associations and interdependencies among various workplace factors and employee well-being. Elovainio et al.⁵ conducted a study using network analysis to examine the interconnection between work-related psychological risks and resources. The study identified clustered factors, including interactional and procedural justice and job demands, that could be targeted for efficient intervention. Mäkikangas et al.²⁴ applied latent profile analysis to distinct work engagement profiles, and investigated the job resource factors that can support remote work. Furthermore, studies also use trace data to identify the job demands of teachers and cluster different patterns based on a data-driven approach²⁵. Based on the studies mentioned above, workplace factors related to JD-R theory can be applied to a wider range of industries and topics, revealing more complex patterns of interconnectedness. Thus, grounded in the JD-R model, this study aims to provide a holistic picture of individually perceived workplace environment across multiple groups of gender, age, industry, and year.

Changing workplace environment and employee well-being

Organizational change is a continuous and dynamic process throughout the corporate lifespan, influenced by changing strategic management and corporate resources within different cultural or institutional contexts²⁶. Accordingly, the employees’ work environment also changes in response to continuous organizational change, as well as various personal and individual factors. The workplace environment factors are found to have significant relationships with employee’s well-being, especially occupational stress². The continuing organizational change is supposed to impact the employee well-being. Previous studies investigated various kinds of restructuring events, such as closure, outsourcing, merging, job cutting, etc. and concluded negative impacts on employee well-being. In particular, employees with lower income, organizational status, and skills are likely to be more negatively affected²⁷. Furthermore, multiple changes or repeated changes lead to heavier mental distress¹⁴. On the contrary, employees with an active job problem-solving style and job-change self-efficacy are found to have higher readiness for organizational change²⁸.

While the JD-R model primarily emphasizes work engagement and stress as core outcomes reflecting motivational and health impairment pathways^29–32, it has also been extended to accommodate a broader range of well-being indicators. For instance, job satisfaction is frequently conceptualized as a downstream attitudinal outcome influenced by motivational processes and job resource availability^18,33. Workplace harassment, though less commonly framed within JD-R, has been increasingly understood as a job demand that contributes to psychological strain and deteriorates well-being^34,35. Moreover, team cohesion—or workplace cohesiveness—can be considered a social resource that fosters engagement and buffers the effects of high demands³⁶. By incorporating these outcomes alongside engagement and stress, the present study adopts a comprehensive approach to well-being, consistent with evolving applications of the JD-R framework across occupational settings.

However, the existing literature does not provide direct evidence of workplace environment change at the individual level induced by organizational change. It is important to note that the perceived workplace environment faced by employees may change due to any small internal reorganization, even without significant organizational change events. Furthermore, it is important to note that workplace changes do not necessarily have an adverse effect on mental health. In fact, in some cases, such as with promotions and development opportunities, workplace changes can have a positive impact on mental health¹³. Therefore, based on the identified workplace clusters, this study aims to investigate the relationship between various patterns of changes in individual workplace environments and employee well-being.

Methods and materials

First, factor analysis is applied to reduce the dimension of workplace indicators, including 8 indicators of work burden and 23 indicators of work resource. The optimal number of factors is determined by the parallel analysis, which indicates 9 factors are optimal (see Supplementary Information Fig. S1). The factor analysis model fit well, according to the criteria RMSEA 0.035 ( < = 0.05) and Tucker Lewis Index 0.958 ( > = 0.95). We ran both oblimin, promax, and biquartimin rotations. All models fit well, but the results of oblimin rotation have the least overlapped loadings and the most suitable combination of indicators as shown in Table 1 (see correlations between factors in Supplementary Information Fig. S2).

Table 1.

Factor loadings

	Demand	Demand	Demand	Resource	Resource	Resource	Resource	Resource	Resource
	Workload	Emotional burden	Physical burden	Organizational integrity	Job autonomy	Leadership	Surrounding support	Rewards	Job stability
Proportion explained	0.1	0.1	0.05	0.2	0.17	0.14	0.12	0.09	0.03
Workload (quantity)	0.76
Workload (quality)	0.78
Work-life balance (negative)	0.33								0.25
Physical burden			0.58
Work environment			0.43
Interpersonal relationships at work		0.42
Emotional burden		0.54
Role conflict		0.81
Job control	0.21				0.38
Job match					0.82
Skill utilization					0.34
Job meaning					0.75
Opportunity for growth					0.61
Work-life balance (positive)				0.31	0.44
Support from manager						0.37	0.55
Support from coworkers							0.87
Support from families							0.48		0.2
Rewards (monetary, status)								0.7
Rewards (respect)						0.25		0.65
Role clarifies					0.25				0.27
Job stability									0.27
Manager’s leadership						0.55
Manager’s fairness						0.63
Encourage praising						0.54
Mistakes are recoverable				0.22		0.38
Trust in management				0.61
Adaptability to change				0.74
Respect for individuals				0.68
Fairness in evaluation				0.52				0.23
Approach on diversity				0.43
Career development				0.55

This table presents the factor loadings based on oblimin rotation results. Values less than 0.2 are shown in the table. The factor analysis model fit well, according to the criteria RMSEA 0.035 (≤ 0.05) and Tucker Lewis Index 0.958 (≥ 0.95). Rotation results of Promax and biquartimin refer to Appendix Table S2. Loadings with larger absolute values are bolded to highlight the primary factor associated with each indicator.

Based on Job DR model, 3 factors can be summarized to illustrate job burden, including Workload, Emotional burden, and Physical burden. Different from the original indicator groups, Workload is not only related to the qualitative and quantitative workload, but also associated with the negative impacts on work-life balance, and the degree of job control. This suggests the workload is not just about the sheer volume of tasks but also about how manageable and sustainable the work is for employees. This perspective emphasizes the importance of considering both the objective workload (quantitative and qualitative) and the subjective experience of employees when assessing and managing workload. Emotional burden consists of one original survey indicator, and the factor loadings shows that role of conflicts and interpersonal relationships at work are highly related with Emotional burden. Physical burden is linked to the individual physical conditions themselves and the external working environment. As shown in Fig. S2, workload is positively related to both emotional and physical burden, while has low correlation with workplace resources, indicating in many workplaces the allocation of resources and workload management might not be aligned.

Workplace resources are summarized into 6 factors, covering multiple aspects of task-related, leadership-related, and organizational ones. First, Organizational integrity involves a comprehensive commitment to ethical behavior and responsible conduct that goes beyond mere compliance. It shapes the organizational culture, guides decision-making, and contributes to the organization’s reputation and long-term success³⁷. Well-established organizational integrity can be reflected in day-to-day practice, which is largely determined by a corporate culture of openness, accountability, and commitment to multiple business goals and employee development opportunities³⁸. Organizational integrity, in a more specific context, is reflected in the individual behaviors of employees. This behavior is shaped by their ethical imagination, which, in turn, is influenced by the language and communication style of management. The way managers communicate values, expectations, and ethical considerations plays a crucial role in guiding and inspiring employees to exhibit behaviors aligned with the organization’s integrity framework³⁹. In our results, trust in management has a positive loading with organizational integrity, which is consistent to argument that trust in management improve organizational integrity, including organizational commitment and organizational citizenship behaviors⁴⁰. Likewise, indicators of career development, adaptability to change, respect for individuals, approach on diversity and fairness in evaluation are also important employee investments to enhance the organizational integrity⁴¹. On the other hand, higher levels of organizational integrity also lead to higher levels of workplace resilience against external shocks^42–47.

Job autonomy the extent to which employees have discretion and independence in performing their job tasks^48,49, which is positively associated with job control, job match, skill utilization, job meaning, opportunity for growth, and role clarifies. Notably, this factor is also positively associated with positive work-life balance, indicating that higher job autonomy also plays a role in enriching positive life experience.

Leadership summarizes the relationship with managers regarding the job tasks, measures the extent to which managers provide assistance, guidance, and encouragement to their subordinates as they perform their work responsibilities^49–53. Leadership has positive loadings on indicators of manager’s leadership, manager’s fairness, encourage praising, and mistakes are recoverable. Specifically, manager’s leadership indicates the opportunities to enhance employees’ capabilities, which is important for developmental leadership⁵⁴. On the other hand, manager’s fairness (sincere attitude towards employees), praising and recoverable mistakes present supportive leadership⁵⁵, an effective leadership practices to cope with potential occupational stress of employees⁵⁴.

Furthermore, surrounding support, as a complementary factor of leadership, covers the supports from not only manager, but also coworkers and family members, acknowledging the holistic nature of an individual’s life. This factor measures not only supports for job responsibilities, but a more comprehensive supportive relationships both within and outside the workplace. Rewards encompass various dimensions, including monetary, social, and psychological levels. In addition, it also has positive association with fairness in evaluation, which suggests that the perception of fairness in the evaluation process is closely linked to the effectiveness of the reward system. Job stability is originated from the similar indicator, indicating the continuity and security of employment, indicating the extent to which employees feel assured about the longevity of their positions within the organization. Job stability is also positively related to role clarifies, suggesting that when employees have a clear understanding of their roles and responsibilities, it contributes to a sense of stability in their jobs.

Based on the key factors above, we further apply Gaussian Mixture Model to classify the employees with similar workplace environment factors by using the Mclust package in R⁵⁶. This data-driven approach has been widely used in psychological field⁵⁷. After determining the workplace clusters, we applied panel data regression to compare the differences of employee well-beings across these clusters. As shown in Eq. (1), denotes six aspects of employee well beings, including mental health, high-stress risk, work engagement, job satisfaction, family-life satisfaction, harassment, and workplace cohesiveness. controls for age and gender. refers to the categorical variable of identified cluster.

1

We specify a DiD design to examine how the change of workplace clusters influence employee well-beings. As shown in Eq. (2), denotes the change of well beings compared to previous year, and denotes the change of clusters from year_t−1 to year_t. Control groups are the employees who keep their workplace unchanged, which is the reference class in the model. The treatment groups consist of all possible changes to another different clusters.

2

The sample for analysis is collected from a job stress check database, provided by Peacemind, Inc. that conducts the job stress survey and provide employee well-being support. We have 1,348,437 observations in total from 2017 to 2021. In Japan, the job stress check has been obligated for any organizations with more than 50 employees. At least, a survey with 57 questions should be conducted for all employees. The database used in this study is based on an extended version of survey with 80 questions (see Appendix Table S1), which offers more detailed workplace information than the mandatory one. Each observation in our dataset captures an individual employee’s self-reported perception of their work environment. Accordingly, the typology reflects types of workplace environments as perceived at the individual level, rather than classifications based on organizational units or departmental structures. As part of preprocessing, all continuous variables were linearly standardized to a 0–100 scale, facilitating cross-variable comparability and effect size interpretation. The panel data allows us to examine the relationship between clusters and employee well-being, especially the change of workplace types and the impacts of workplace changes on employee well-being.

Ethics approval and consent to participate

Ethical approval for the study was obtained from the Ethics Committee of PEACEMIND Inc (ref. no.: R03-001). The research was performed in accordance with relevant guidelines and regulations. The individual data regarding the stress check survey was provided with informed consent. The data used in this study does not target personal health information, and personal information is non-identifiable.

Results

Individually perceived workplace clusters

By comparing the BIC of tentative models, 10 clusters with “VVV” model tends to be the optimal one fitting better than other models (see Supplementary Information Fig. S3). Thus, we chose this model as our main results. The 10 clusters can be divided into four classes, ranging from D to A, and 10-level grades, ranging from D (the worst) to A1 (the best), according to the well-being outcomes of each cluster as shown in Table 2. The A class outperforms all other classes in terms of all well-being indicators, followed by classes B, C, and D. Grade B3 has the highest proportion of employees, with well-being indicators averaging above 50 points and a low level of high-stress risk.

Table 2.

Employee well-beings by cluster.

Except for the harassment score between cluster B2 and B3, all other pairwise T test results present significant difference in the employee well-being across clusters.

The position of the cluster for each factor is displayed in Fig. 1 (see Fig. S4 for the positions of clusters across pairwise job DR factors). Class A’s workplace resources are significantly above average. Among all the grades, A1 has the highest level of workplace resources, especially in terms of surrounding support, organizational integrity, and leadership. In addition to the high level of job resources, job demands in Grade A1 are also above average, with less workload and the lower emotional and physical burdens than most other grades. There are significant differences in job demands between Grade A2 and A3. Grade A3 employees generally have a heavier workload than Grade A2 employees, but they experience lower levels of emotional and physical burden and have sufficient job resources. Notably, surrounding supports for employees in Grade A3 are almost two times higher than those in Grade A2.

Fig. 1.

Center of factors by cluster.

Class B employees have job demands and resources that are close to average levels. Specifically, employees at Grade B1 experience less workload than average and work in a corporate culture that values highest organizational integrity among class B. However, they may feel stressed about the possibility of losing their job. The primary distinctions between Grade B2 and B3 lie in the physical burdens and level of job autonomy, where employees in Grade B2 experience greater physical burdens but also enjoy higher levels of job autonomy compared to those in Grade B3.

Class C employees have below-average job resources. Although Grade C1 employees have a lighter workload and better physical conditions, they are losing trust in management due to poor organizational integrity and insufficient rewards. Other job resources, such as job autonomy, leadership, and surrounding support, are also lower. Grade C2 employees experience below average levels of emotional and physical burden. Although the workload in Grade C2 is lighter than the average, there are greater job burdens associated with low levels of job resources, particularly in terms of rewards and support. This two grades contrast with Grade A3, where the heavy workload is accompanied by abundant job resources. Regarding Grade C3, the workplace situation has worsened in terms of job demands and resources. This includes heavy workloads, significant job burdens, and a lack of organizational integrity and job autonomy.

Finally, Grade D employees face a hard demanding work environment with a workload that surpasses the average level, placing significant emotional and physical demands on them. They have limited job autonomy and receive insufficient support and recognition from management.

The ten grades of workplace clusters provide a deeper understanding of the work environment in which employees are working. Figure 2 displays the distribution of workplace clusters across groups of gender, age, year, and industry. Sector classification refers to Japan Standard Industrial Classification Rev. 13. For all groups of gender (see Fig. 2a), Grade B3 has the highest proportion, with over 25%, indicating that over one-fourth of employees have average levels of job demands and resources. The proportion of female employees in Class A is lower than that of male employees and employees who prefer not to reply. In contrast, the proportion of female employees in Class C and D is also the lowest, while employees who prefer not to reply are more likely to suffer in a worse workplace, especially the Grade D.

Fig. 2.

Workplace clusters by gender, age, sector, and year.

Regarding age groups (see Fig. 2a), the percentage of employees at Grade B3 remains high across all groups, while proportions of Classes A, C and D vary by age groups. There is a U-shaped distribution of employees in Class A, with the highest proportions in the 10s and 70s, and lower proportions in the 30s and 40s age groups. Accordingly, the 30s and 40s are also more likely to experience low grade workplaces, such as Grade D and C. More specifically, Grade A3 and C2 show opposite trends across age groups. The proportion of Grade C2 increases as age increases and reaches its highest point for those in their 60s. This suggests that employees nearing retirement age may be experiencing a lack of job resources, such as rewards and support. In contrast, the proportion of Grade A3 decreases with age. A quarter of the employees in 10s age group have sufficient job resources despite facing a heavy workload.

In relations to industry groups, Fig. 2b shows the workplace cluster distribution across 17 investigated industries. Employees in the information and communications (industry G) have the highest proportion of high grades and the lowest proportion of low grades, which is much better than in other sectors. Similarly, employees in education and learning support (industry O) are more likely to have Class A workplaces. However, they also have higher proportions in Classes C and D, and lower proportion in Class B (the average levels), suggesting an extremely polarized workplace situation in the education and learning support sector. Employees in services face more difficult working conditions than in other sectors. The worst situation is found in living-related, personal, and amusement services (industry N), with 17.3% in Grade D, 41.9% in Class C, and only 12.8% in Class A. The other industries with higher proportion in Class D include such as transport and postal services (industry H: 15.8%), compound services (industry Q: 15.1%), and other services not elsewhere classified (N.E.C.) (industry R: 12.5%).

As shown in Fig. 2c, the year-to-year change in the distribution of workplace clusters indicates a worsening workplace environment from 2017 to 2019, with increasing proportions of Class C and D and decreasing proportions of Glass A. Notably, the deterioration of the workplace environment is halted during the Covid-19 pandemic, with a significant decrease in the proportions of classes C and D in 2020, return to the level similar to 2017. However, the following year, 2021, the work environment begins to deteriorate again, with a slight increase in the proportion of classes C and D. The total percentage of employees in higher grades remain unchanged in 2021, suggesting a possible fundamental change in the workplace in the post-pandemic period. It is worth noting that significant changes in the workplace environment during the pandemic may not always result in positive outcomes⁵⁸.

Impacts on employee well-being

We constructed panel data to compare the difference in the relationships between clusters and employee well-being, as shown in Table 3 (see regression results based on an unbalanced full sample in Supplementary Information Table S3). Grade B3 is set as the reference group. The results present significant differences in employee well-being across clusters, indicating that the 10 clusters are effective in distinguishing employee well-being regarding mental health, high stress, work engagement, job satisfaction, family-life satisfaction, harassment, and workplace cohesiveness. Given that the outcome variables are standardized on a 0–100 scale, the coefficient estimates can be interpreted as meaningful effect sizes. For example, employees in Grade A1 report 5.732 points higher work engagement—equivalent to 11.22% relative to the B3 group average—and 4.138 points higher mental health (7.9%) compared to those in Grade B3 (see Table 3). These differences indicate substantial variations in employee well-being across workplace types. Figure S5 shows the position of clusters across well-being outcomes. The levels of employee well-being across clusters remain stable over the years (see Supplementary Information Fig. S6).

Table 3.

Panel data regression results.

	Mental health	High stress	Work engagement	Job satisfaction	Family-life satisfaction	Harassment	Workplace cohesiveness
Cluster A1	4.138***	− 0.026***	5.732***	5.561***	3.095***	1.016***	6.069***
	(0.080)	(0.004)	(0.081)	(0.092)	(0.093)	(0.108)	(0.100)
Cluster A2	2.331***	− 0.012***	2.060***	2.338***	0.819***	0.866***	2.317***
	(0.067)	(0.003)	(0.068)	(0.077)	(0.078)	(0.090)	(0.084)
Cluster A3	1.656***	− 0.017***	2.506***	2.683***	1.152***	0.998***	1.949***
	(0.048)	(0.002)	(0.048)	(0.055)	(0.055)	(0.064)	(0.060)
Cluster B1	1.317***	− 0.033***	1.615***	1.124***	− 0.193***	− 0.145*	0.813***
	(0.059)	(0.003)	(0.059)	(0.067)	(0.068)	(0.079)	(0.073)
Cluster B2	0.561***	− 0.021***	0.742***	0.808***	− 0.299***	0.343***	0.348***
	(0.047)	(0.002)	(0.048)	(0.054)	(0.055)	(0.064)	(0.059)
Cluster B3 (Reference cluster)
Cluster C1	− 1.249***	0.021***	− 4.078***	− 4.225***	− 0.309***	− 0.815***	− 2.813***
	(0.068)	(0.003)	(0.069)	(0.078)	(0.079)	(0.092)	(0.085)
Cluster C2	− 1.369***	0.012***	− 1.635***	− 2.076***	− 0.780***	− 1.026***	− 1.733***
	(0.039)	(0.002)	(0.040)	(0.045)	(0.046)	(0.053)	(0.049)
Cluster C3	− 4.459***	0.152***	− 3.277***	− 4.388***	− 0.417***	− 1.110***	− 2.397***
	(0.046)	(0.002)	(0.046)	(0.053)	(0.053)	(0.061)	(0.057)
Cluster D	− 5.647***	0.232***	− 5.280***	− 6.799***	− 0.472***	− 4.664***	− 6.185***
	(0.050)	(0.002)	(0.051)	(0.057)	(0.058)	(0.067)	(0.063)
Age fixed	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Gender fixed	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Observations	277,320	277,320	277,320	277,320	277,320	277,320	277,320
R2	0.117	0.064	0.118	0.125	0.01	0.028	0.079

The cluster B3 with the most proportion is set as reference cluster. The significant level is denoted as follows: *p < 0.1; **p < 0.05; ***p < 0.01.

The heatmaps in Fig. 3 summarizes the DiD results of the effects of changing workplace clusters on each of the employee well-being indicators, including mental health, work engagement, job satisfaction, family life satisfaction, harassment, and workplace cohesiveness. There are 90 patterns of workplace changes and the coefficients in each cell are the effects on well-being compared to those that keep their workplace cluster unchanged. Overall, the greater the change in workplace grades, the stronger the impact on well-being, which is evident in mental health, work engagement, job satisfaction, and workplace cohesiveness. The largest positive longitudinal change is found in work engagement, which increased by 17.22 points when moving from Grade D to A1, while the largest decline was a 14.28-point drop in workplace cohesiveness from A1 to D.

Fig. 3.

Impacts of workplace shifts on employee well-being.

The impact of work on family life satisfaction is weaker compared to other well-being indicators and shows inconsistent results when compared to other job-related indicators. For instance, the change from Grade D to B1 has negative impacts, contrary to expectations. The workplace environment itself can hardly predict the level of satisfaction in personal life. Improving the workplace, such as changing to a higher grade of cluster, has limited impact on harassment issues. The change from grade D to grade A1 does not significantly impact the level of harassment. The most effective workplace changes are observed when transitioning from the lowest grades (D, C3, and C2) to grades A2 and A3. It is generally beneficial to move from grade D to a higher-grade workplace to alleviate harassment, except for grades A1 and B1. Employees who move to grade D are likely to experience harassment.

We also conducted several robustness analyses based on a set of cut-off levels (0.25, 0.5, and 0.75) of cluster uncertainty and the results do not change the conclusions about the impact of cluster change on employee well-being (see uncertainty distribution of clustering results in Supplementary Information Fig. S7).

Conclusion

This study provides a comprehensive examination of individually perceived workplace typology and its impact on employee well-being across various dimensions. By applying a data-driven approach to classify workplace environments into distinct clusters, we have identified key factors that contribute to the quality of the workplace experience, including job demands and resources. Nine workplace factors are summarized based on a large-scale employee workplace and well-being database of over 1,300,000 observations from 2017 to 2021. The factors include three job demands: workload, emotional burden, and physical burden, and six job resources: organizational integrity, job autonomy, leadership, surrounding supports, rewards, and job stability. The workplace is further classified into ten grades from D to A1 in four classes from D to A, which can significantly distinguish job-related employee well-being. By comparing workplace clusters across gender, age, industry, and year, this study provides a holistic picture of what types of workplace environments exist in specific groups. In addition, the dynamically changing workplace clusters are found to significantly affect job-related employee well-being, including mental health, work engagement, job satisfaction and workplace cohesiveness. However, the issue of workplace harassment needs to be explored more deeply beyond the workplace environment.

Overall, our study underscores the value of adopting a person-centered, data-driven clustering approach to identify distinct workplace types and better understand their implications for employee well-being. In contrast to meta-analytic findings that typically examine isolated variables across studies, our approach captures how individual employees subjectively experience different constellations of job demands and resources. This enables a more holistic and grounded view of the workplace that accounts for intra-organizational variation and individual heterogeneity. By focusing on employee-centered workplace profiling, our findings support the design of more tailored and responsive organizational strategies—ones that address the specific needs of different employee groups across gender, age, and industry, and adapt to changes triggered by external shocks. This typology-based diagnostic tool thus empowers organizations to pursue workplace improvements that are not only evidence-based but also aligned with the lived experiences of employees.

Several limitations should be considered when interpreting our findings. First, although our clustering approach is grounded in the JD-R model, the empirically derived workplace types may be context-specific, reflecting Japanese cultural and organizational norms. As such, generalizability to other countries or labor markets may be limited, and future studies should validate the typology in diverse cross-cultural settings. Second, while our longitudinal design captures workplace type changes, we did not examine external or organizational-level factors (e.g., industry shocks, leadership change) that may drive these changes. Third, personal coping strategies and psychological resilience, which may moderate the impact of workplace change, were not measured. Lastly, while we observed positive well-being changes during the COVID-19 pandemic, caution is needed in interpreting these results without further post-pandemic follow-up. These limitations point to important directions for future research, including cross-national comparative studies, integration of multilevel (organizational and individual) predictors, and the refinement of JD-R-based typologies to enhance both theoretical precision and applied value.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

All authors read and approved the final manuscript. J.X. conducted the analysis and prepared the primary manuscript. X.D.-P. participated in drafting the manuscript and critically revised the manuscript. S.M. supervised the study.

Data availability

The stress check survey data used in this study were provided by PEACEMIND Inc. under a licensing agreement and are not publicly available due to data use restrictions. However, the data may be made available by the corresponding author upon reasonable request and with permission from PEACEMIND Inc.

Declarations

Competing interests

The authors declare no competing interests.