Stress can be understood as a significant disruption of physiological regulatory processes caused by internal or external stimuli that necessitate an adaptive response (1). Whether stress has a positive or negative character depends on the individual’s perception and interpretation of the situation (2). Sixty percent of Germans report exposure to stress (3), with 32% stating its impact on their daily lives (4). The workplace is one of the main sources of stress (5, 6), and employee stress levels have increased in recent years (6).

However, there are differences in the perception of stress among men and women. These gender differences require individual adjustments in the context of prevention (7) as well as a better understanding of the role of gender in the context of work (8). For instance, women experience less stress than men when receiving social support (9, 10). Further, studies highlight the importance of addressing these gender-specific effects regarding stress [e.g., European Agency for Safety and Health at Work (8) and Gilbert-Ouimet (11)]. To date, studies have shown gender differences in the effects of stressors and resources on stress for small constellations of demands and resources [e.g., Padkapayeva et al. (9), Vermeulen et al. (10), Rivera (12)], as well as in samples of specific occupational groups (13–15). However, the causes of stress are manifold, and the interaction of various factors is complex. The Job Demands-Resources (JD-R) Model (16, 17) offers an explanatory approach to the development of stress in the work context.

The JD-R model provides a multidimensional framework to explain health and motivational processes. Health, motivational, and, indirectly, organizational outcomes are linked to the relationship between job demands and resources, which cluster the individual range of working conditions in an organization (18–21). Within the model, two main processes are delineated: (1) increased or inadequate job demands lead to a state of exhaustion and reduced health, and (2) a lack of or inadequate job resources reduces work-related motivation (17, 18, 22). Job demands are “physical, social, or organizational [stressors] of the job that require sustained physical or mental effort and are therefore associated with certain physiological and psychological costs” (17). Job resources represent physical, psychological, social, or organizational aspects that help achieve work goals, reduce job demands, and stimulate personal growth and development” (17).

Furthermore, an interaction between job demands and resources is described, influencing motivational and health-related follow-up processes (23). In this context, resources buffered job demands (24, 25). In addition, there are links between job demands and resources and organizational outcomes, mediated by reduced health and motivation (26, 27).

The JD-R model integrates stress and motivational theories and allows for a global application independent of specific demands, resources, or settings (16). From a cross-sectional perspective, examining the gendered effects of the multiple demand-and resource constellation prevalent in practice on the individual experience of stress remains open. To enhance comprehension of these gender differences, this research aims to examine how the stress level, as measured by subjective perceived stress, is affected by a multiple, cross-professional stressor and resource constellation in a gender-specific context. Specifically, this study aims to investigate gender differences in the effects of work stressors, including quantitative demands, qualitative demands, and working time, as well as the resources of job control, co-worker support, and quality of leadership on subjective perceived stress in a cross-professional setting.

The existing body of research has well established that job demands can increase stress and therefore act as stressors in the context of work (15, 28–30). Many different working conditions are subsumed under the multidimensional construct of job demands. These can be divided into two major categories: quantitative and qualitative demands. Quantitative demands refer to the relationship between the number and amount of job requirements and the available time to handle them (31, 32). Qualitative demands describe the quality and complexity of a work task in relation to individual abilities and skills (e.g., intellectual or emotional competencies) (32, 33). These two basic dimensions do not represent the working time. However, the working time is relevant for almost all employees in all occupations (34) and is a critical factor in mental health (35). The working time, within this study, refers to the actual amount of time spent working rather than the contracted working time.

Associations between aspects of poor mental health, such as emotional exhaustion, depression, burnout, and stress-related disorders, and high job demands have been shown for the constructs of interest, quantitative demands (35–37); qualitative demands (38) and working time (39–41). Relationships with stress experience are apparent. Thus far, an increase in quantitative demands (38, 42) and working time (43–45) has been shown as potential risk factors for experiencing stress. Based on this knowledge, it is hypothesized that:

To the author’s knowledge, a relationship with stress has not yet been demonstrated for qualitative demands as an overall construct. However, associations with a reduced mental health status, depression, and burnout have been shown in a scoping review (38). Furthermore, relationships between increasing stress and high levels of cognitive demands (46), emotional demands (47), and complexity (48) were demonstrated. Based on these findings, it is postulated that:

The buffering effect of job resources on stress has also been confirmed by several studies [e.g., Padkapayeva et al. (9), Xie et al. (15), and Huang et al. (49)]. Among the variety of job resources, some working conditions were found to be particularly important given their broad, cross-functional nature [cf. Morschhäuser et al. (34) and Rothe et al. (35)]. Job control is one of these resources. It refers to the influence that employees have over their work. This includes both the freedom to decide on the time and type of implementation of a work task (decision authority) as well as control over the use of concrete skills (skill discretion) (50). Furthermore, the resources of social support and quality of leadership are particularly important. Social support at work refers to how employees perceive the availability of co-worker support or support from direct supervisors when needed (31). It is important to note that within the scope of this study, social support from a supervisor constitutes a part of quality of leadership. However, leadership entails more than social support. According to Burr et al. (31), quality of leadership is concerned with “the next higher managers’ leadership in different contexts and domains”.

Again, associations with mental health have been well studied [e.g., Pohrt et al. (7), Brendel and Martus (51), Drössler et al. (52), and Rosen (53)]. Relationships between decreased stress levels and an adequate amount of resources were found when considering co-worker support and support from a supervisor (9, 30, 54); further aspects of employee leadership (55); as well as job control and related subscales, such as time flexibility or decision latitude (53, 56, 57). In line with these findings, the following hypotheses were formulated:

Research has highlighted gender differences in the impact of various working conditions on stress development [e.g., Padkapayeva et al. (9), (10), Lian et al. (56), and Wang et al. (58)]. These differences can be attributed to dissimilarities in exposure to job demands and resources (59, 60). Furthermore, differing appraisals between males and females, resulting from biological differences, including hormonal or genetic factors (61), as well as cognitive and affective mechanisms (62), are a cause of varying stress levels.

About a gender-specific view of the resources included, Nieuwenhuijsen et al. (36) demonstrated in their systematic review that insufficient co-worker support is positively linked to stress-related disorders in men. For women, a correlation between co-worker support and stress-related disorders was unclear. However, Padkapayeva et al. (9) found no gender-related differences in the correlation between co-worker support and job-related stress. Furthermore, within the context of quality of leadership, appreciation from supervisors has been proven to be more predictive of depressive symptoms for women (7). In contrast, social support from supervisors has only been found to reduce stress-related disorders for men. Effects of job control on stress have so far been controversial concerning gender [cf. Padkapayeva et al. (9), Steptoe and Willemsen (57), De Bruin and Taylor (63), Melin et al. (64)].

The gendered effects of stressors have been studied mainly in the context of psychological constructs but rarely specifically to stress. Quantitative demands have proven to be more critical for men compared to women regarding mental illness (12, 65, 66). An association between qualitative demands and mental illness was only observed in women (12). In contrast, separate elements of quantitative demands and qualitative demands, such as working quickly or performing complex tasks, have already been considered concerning stress in the context of gender-specificity. Quantitative demands have proven to be more relevant predictor for men compared to women regarding distress (10). Sub-aspects of both constructs were found to be more important for women predicting stress (48). Extended working time also increases the risk of depression and burnout, particularly for women (67, 68). The impact on men remains unclear, as studies have yielded conflicting results [cf. Choi et al. (67), Hu et al. (68), and Weston et al. (69)]. Even if the exact impact patterns remain unclear in some cases, gender-specific differences have been identified. Summarizing the previous findings, it is hypothesized that:

The present study was performed as secondary research using a quantitative approach based on data from the “German Study on Mental Health at Work” (S-MGA). This approach allows the hypothetical model to be tested on a representative sample of all occupational groups in Germany. The implementation was carried out according to the STROSA-2 standards (Standardized Reporting Routine for Secondary Data Analyses; see Supplementary file) (70).

The S-MGA (71), founded by the BAuA (Bundesanstalt für Arbeitsschutz und Arbeitsmedizin), is a cohort survey aimed at offering representative information on the workability and health of German employees. Employees with social insurance who were born between 1951 and 1980 were included in the survey. Sampling was carried out using a two-stage stratified random approach. Computer-assisted personal interviews with trained interviewers and paper and pencil questionnaires were used for data collection (72). The S-MGA contains information from two survey periods on workability, functional and mental health, main employment status, secondary employment, and sick leave (73). Baseline data from November 2011 to June 2012 were used for the underlying sample. This database contained 4,511 valid interviews (response rate: 35.7%) from a representative cross-sectional sample of the German working population.

The Institute for Applied Social Sciences (infas) collected and prepared data. Direct identifiers were removed from the data before transmission to the BAuA (72). Security strategies were implemented to ensure maximum anonymity. The dataset did not contain sensitive characteristics (72). Data access to the Scientific User File provided by the FDZ-BAuA was enabled via a data exchange server. A data use agreement governed the provision and use of secondary data. Data were de facto anonymized when transferred. Data protection was ensured according to the requirements of the General Data Protection Regulation (74).

For the underlying sample, only employees in a dependent professional relationship who reported having a supervisor, clearly defined working hours, and provided information on their gender were considered. Gender differences were identified by comparing indices that are highly sensitive to sample size, so the samples of males and females were equalized. A random sample of n = 2,059 was drawn from the male dataset to avoid bias due to sample size (75, 76). Survey data from the employees subject to social insurance as of December 31, 2010, and born between 1951 and 1980, which corresponded to the selection criteria described above, were analyzed. For the gender-specific analysis, participants were grouped into males and females.

Information on gender was obtained from employment histories held by the Institute for Employment Research and verified during interviews with participants. Items from the German version of the Copenhagen Psychosocial Questionnaire (COPSOQ) (77) were used to assess quantitative demands, quality of leadership, co-worker support, and job control. All items were measured using 5-point Likert scales (e.g., always, often, sometimes, rarely, never/hardly ever). All other items were carefully selected using the top-down technique (78) by established questionnaires, scientific findings, and occupational psychology models [cf. MacKenzie et al. (79) and Weiber and Sarstedt (80)]. The variables qualitative demands and working time were each represented by global indicators using self-rated single-item measures [cf. Pattloch et al. (72)]. Both were treated as ordinal data. Working time was assessed by the number of hours worked per week and then grouped into seven categories with similar intervals.

In line with the procedure described above, subjective perceived stress was operationalized according to the cognitive-emotional level of the stress response. A total of four items from the “Scale of Positive and Negative Experience” (81), the “Psychological Well-Being Scale” (82), and the “Short-Form Health Survey” (83, 84) were used for this purpose. Items were scored on 5-to 7-point rating scales. Detailed information on the items used can be found in Supplementary file.

To control for confounding variables, factors that influence the perception of stress were also considered (63, 85). For practical purposes and to avoid the risk of misinterpretation, we controlled for the most relevant demographic categories (63). Thus, age of the participants at the time of data collection (86, 87), educational level (88, 89), income (88), children under 14 years in the household (90), caregiving (91), and hours worked in paid second jobs (92, 93) were included as confounders (see Supplementary file).

Structural equation modeling (SEM) was used as the primary analytical approach to analyze the interdependencies of multiple manifest and latent variables (80, 94). To examine disparities between genders, the underlying framework of the impact of stressors and resources on subjective perceived stress was first validated within a sample that included both males and females. Subsequently, within this framework, gender differences were assessed using a stratified sample. An a priori power analysis was conducted to determine the minimum sample size required for testing the hypothesized equation model using “A-priori Sample Size Calculator for Structural Equation Models” (95). The sample size required to achieve 80% power to identify small effects, with a significance level of α = 0.05, was n = 1,808. Hence, the sample sizes of n = 4,118 for the entire sample and n = 2,059 for the samples stratified by gender are valid to test the hypothesized equation model. Differences between these two groups were analyzed with the Mann–Whitney-U-test (96).

Data management and hypothesis testing were performed using SPSS version 29.0 and IBM SPSS AMOS 29 (Statistical Package for the Social Science, Analysis of Moment Structures). Outliers were identified as plausible and retained to avoid manipulation of results (97). Univariate non-normality was indicated by values outside the cut-offs of ±2 for skewness, ± 7 for kurtosis, and critical ratios (C.R.) > | 1.96 | (98, 99). The analyzed data contained 1,663 missing values (1.50%) (Supplement C). The missing mechanism was assumed to be missing at random (MAR) using Little’s missing completely at random (MCAR) test (100), visual inspection of missing data patterns (94), and rational reflection consideration (101, 102).

For SEM, a two-stage modeling approach was used (103). First, the measurement model was assessed. Principal component analysis (PCA) and confirmatory factor analysis (CFA) were therefore carried out for the entire study sample and, regarding multi-group analysis, for each gender separately (80). Recursive models were then calculated using covariance-based single-group and multi-group SEM. The models were estimated using maximum likelihood estimation (ML) and bootstrapping with 5,000 iterations to address non-normality and ordinal data (99, 104). Model fits were assessed using Root Mean Square Error of Approximation (RMSEA), Standardized Root Mean Squared Residual (SRMR), Comparative Fit Index (CFI), and Adjusted-Goodness-of-Fit-Index (AGFI). CFI and AGFI were assumed to be ≥0.90 (80). For RMSEA and SRMR, values ≤0.08 were considered acceptable (105). Following Schreiber (106), chi-square and degrees of freedom were provided for better interpretation of the values based on them but owing to the sensitivity of chi-square to sample size, they were not included in the direct assessment for model fit.

From the provided sample of n = 7,148 with data from two survey waves, data from n = 4,118 employees were used for the current study. The entire sampling and selection process is shown in Figure 2. The final sample included 2,059 women and men each, from different occupational groups and companies covered by social security. With a median age range of 45–49 years for both males and females, the age span varied from 31 to 60 years. Approximately 72% of male and 70% of female respondents were married or in a civil partnership, and 69% of each gender lived in a household without children. The majority of respondents (male: 57.50%, female: 61.83%) had received vocational education or training, while over a third (male: 37.73%, female: 32.69%) had completed higher education (e.g., university degree). The vast majority of men worked full-time (93.88%), whereas roughly half of women were employed full-time (49.34%). Approximately 10% of both males and females reported having a second job.

Descriptive statistics on the items of the main study variables for the total sample and samples stratified by gender are presented in Table 1.

Scientific evidence highlights the importance of addressing gender differences in the perception of work-related stress [e.g., European Agency for Safety and Health at Work (8) and Gilbert-Ouimet et al. (11)]. The purpose of this study was to examine gender-specific differences in the effect of job stressors and resources on subjective perceived stress.

The overall examination of the hypothesized model, which was conducted prior to the analysis of gender differences, considering men and women simultaneously, demonstrated that high levels of quantitative demands and working time are associated with an increase in subjective perceived stress, while high co-worker support, quality of leadership, and job control are linked to a reduction in subjective perceived stress. Qualitative demands do not significantly impact subjective perceived stress. Consequently, except for hypothesis 2, these outcomes confirm hypothesis 1 and hypotheses 3–6. Hypothesized gender differences within this framework are evident for the framework as a whole, but not for each impact of stressors and resources on subjective perceived stress in detail. The effect of working time on subjective perceived stress was found to be relevant only for women, and high quantitative demands were associated with an increase in subjective perceived stress for both genders, but more so for men. No differences in either relevance or strength of effect could be demonstrated for qualitative demands, co-worker support, quality of leadership, and job control. Therefore, H7 must be rejected.

The above conclusions need to be considered in the light of some methodological issues. Due to the research design, it was not possible to control for confounding during data collection. The influence of response bias, especially social desirability, cannot be excluded given the sensitive nature of the study. However, the avoidance of interviewers and the use of questionnaires in appropriate subject areas minimize bias (73). Interviewer training was used to reduce interviewer bias and to ensure high data quality (73, 120). Though, the evaluation of working conditions was based on respondents’ self-assessments. It therefore does not provide an objective assessment of the actual working situation. Thus, reverse causality cannot be ruled out, particularly in relation to mental health and perceived stressors (121). In terms of representativeness, the sample is unbiased. Despite a response rate of around 36%, the impact of non-response bias is considered low based on a selectivity analysis (73). Selection bias is not expected due to a carefully conducted sampling process by the BAuA.

Regarding the constructs of interest, it should first be noted that the operationalization was limited by the database. Most variables could be measured using the COPSOQ as a valid and established instrument (77). However, job control failed to achieve internal consistency reliability. Moreover, well-established questionnaire scales were not available for all constructs. A careful conceptualization of subjective perceived stress, based on scientific evidence, nevertheless allows for adequate quality. Content validity can be presumed (80). However, no additional expert validation was carried out. In addition, in two cases, single items had to be used. For that reason, the measurement of qualitative demands is accompanied by a reduced information content (122). Given the lack of external criteria, criterion validity was not verified. To assess construct validity, nomological and discriminant validity were considered. As the SEM model fit confirms the underlying relationship between the constructs, nomological validity is concluded (123). HTMT values confirm discriminant validity (110). An assessment of convergent validity as a further component of construct validity was not possible as no additional measurement procedures were available. However, the AVE values do not indicate a lack of convergent validity (108). Overall, construct validity can be assumed with caution, as convergent validity cannot be definitively assessed.

Further to declare is that both independent and dependent variables were collected within the same measurement context. To test for common method bias, Harman’s-single-factor-test was conducted using PCA (124). The extraction of one factor explains 21.69% of the total variance. As this value is well below the threshold of 50%, no common method bias can be proven (125).

Finally, the statistical analysis needs to be discussed in terms of internal validity. The conditions for imputation were generally favorable and the missing rate was low (1.50%). Thus, no significant bias in imputed values is expected despite the violation of the normal distribution assumption (138). In the presence of ordinal variables, ML estimation tends to produce inflationary χ²-values, underestimated factor loadings; correlations; and standard errors, and biased error variances (126, 127). However, the effects are likely to be small due to bootstrapping and the large sample size. In terms of testing for group differences, it should be noted that the measurement models were valid for both groups. Equalization of the sample sizes in gender-stratified groups avoided biased chi-square difference tests.

Strengths and weaknesses beyond the biases discussed above are outlined below. With the S-MGA, this study is based on a very well-documented dataset with transregional representation, which allows for comprehensive conclusions across occupations and sectors. As the S-MGA focuses on the relationship between work and mental health, its data were well suited to this study. The outstanding sample size enables precise and robust estimates, even within stratified samples. Moreover, it is essential to highlight the comprehensive approach. Through the multi-faceted analysis of interconnections using SEM, a comprehensive view of the working conditions that arise collectively in practice can be obtained. Including working conditions with cross-functional implications further ensures a high level of generic quality while focusing on issues of particular relevance.

Finally, the inclusion of quality of leadership and subjective perceived stress should be emphasized at this point. Previous research has mostly focused on social support provided by supervisors. However, the impact of supervisor behavior on employee health extends beyond social support. These determinants are considered in this study. Considering subjective perceived stress, it is further possible to identify approaches for counteracting negative demands at an early stage. In contrast to the consideration of long-term consequences of stress, such as burnout, this results in implications that not only address stress but also prevent a variety of stress-related consequences.

Alongside these strengths, there are some limitations to be aware of. First of all, the period of the data collected must be stated. As data were retrieved 2011–2012, the socio-economical context might have changed. Nevertheless, regarding job stressors and resources, studies indicate that gender roles have hardly changed as women (care work, mental load) and men also showed traditional role models in relation to Covid-19 when both worked at home (128, 129). Even though a comprehensive combination of working conditions was included, they still only represent a limited section of workplace reality. Thus, this study provides excellent initial guidance for implementations, but further investigations within the individual professional contexts are recommended. In addition, the cross-sectional design does not allow conclusions to be drawn about causal effects. Longitudinal studies are therefore needed (130, 131). Finally, some limitations regarding the transferability of the results need to be considered.

Due to the representative sample, taking into account the limitations outlined above, the findings can be applied to the German labor market almost without restriction. However, other countries differ from German conditions in terms of cultural values and norms, role models, labor law, or even the design of working conditions. Differing gender-specific patterns have been identified in the correlation between work characteristics and mental health across various countries (117). Therefore, the transferability of the results to other countries might be limited.

The restricted age range of the underlying sample, which is between 31 and 60 years, further limits the transferability. The sample does not include young workers, such as apprentices, or workers over the age of 60. However, these target groups are also represented in the labor market. It cannot be excluded that there are other interrelationships for employees beyond the age range considered here. Particularly in view of the increasing importance of apprentices in the next few years (132) and an increase in the retirement age, it is advisable to address these groups in further research.

It should also be noted that the data was collected several years ago. Since then, the generational distribution of employees has changed. As a result, the given stressors and resources are confronted with different role models, ways of thinking, and behaviors. Additionally, working conditions have evolved due to the COVID-19 pandemic and technological advances (133–135). This development impacts the considered stressors and resources. For instance, the requirements for a high quality of leadership in times of remote working and the increasing use of artificial intelligence are different from those of a few years ago.

A final point to consider is the combination of stressors and resources. In terms of effect strength relating to gender differences, the results appear to rely on the specific combination of working conditions analyzed. Therefore, this study offers valuable guidance for initial interventions for a wide range of workplaces by considering key conditions of high generic quality. However, varying contexts are found to impact gender differences in stress differently (136). Thus, for a precise comprehension of gender differences in a certain workplace, it may be necessary to conduct a specific analysis that takes into account additional relevant stressors and resources. If there are specific findings available for the target group of interest, these should be prioritized.