Female executives are particularly prone to the sleep-disturbing effect of isolated high-strain jobs: a cross-sectional study in German-speaking executives


Michael C. Gadinger, Mannheim Institute for Public Health, Social and Preventive Medicine, University of Heidelberg, Mannheim Medical Faculty, Ludolf-Krehl-Strasse 7-11, 68167 Mannheim, Germany. Tel.: +49 621 3839922; fax: +49 621 3839920; e-mail: michael.gadinger@medma.uni-heidelberg.de


This study assessed the main, curvilinear, interactive and gender-dependent effects of job demands, job control and social support in the prediction of sleep quality. Participants were 348 male and 76 female executives and managers from Germany, Austria and Switzerland. A multiple regression controlling for age, occupational hierarchy and various health behaviors was computed. On the level of the main effects of the Job–Demand–Control–Support (JDCS) model, the results indicate a sleep-promoting effect of social support. A significant three-way interaction of job demands, job control and social support was observed. This interaction confirms the buffering effect of high job control and high social support on high job demands. Further, this three-way interaction of the JDCS dimensions is moderated by gender as indicated by a significant four-way interaction. The directions of the significant interactions suggest that female executives are especially prone to react with impaired sleep quality when exposed to isolated high-strain jobs. The study seems to imply that the JDCS model is a suitable framework for the prediction of sleep quality among executives and managers. The results suggest that the JDCS model might contribute to a better understanding of the higher prevalence of poor sleep amongst female executives. Further, the results imply that high job control and high social support might help executives to maintain good sleep quality despite experiencing high job demands.


Sleep disturbances, such as insomnia, have reached epidemic proportions in developed countries. Depending on applied definitions and diagnostic criteria, prevalence rates vary between 10% and 50% (Hajak, 2001; Ohayon, 2002). The negative effects of sleep disturbances for society and individuals are tremendous. Poor sleep negatively affects physical and mental health (Doi et al., 2003; Kalimo et al., 2000; Knudsen et al., 2007; Stoller, 1994, 1997). Thus, poor sleepers tend to use medical services more frequently (Léger et al., 2002), shown as causing an annual direct economic burden of $14bn within the USA in 1995 (Walsh and Ustun, 1999).

Employee’s sleep disturbances may have substantial consequences on an organization’s economic success. Poor sleep decreases concentration, communication skills, decision making and flexible thinking (Harrison and Horne, 2000; Linton and Bryngelsson, 2000). Further, sleep deprivation may diminish job motivation leading to decreased job performance (Scott and Judge, 2006). However, sleep deprivation among executives may have especially detrimental effects as it reduces leadership qualities and competent representation of the organization (Czeisler, 2006).

One robust finding across epidemiological studies is that women show higher prevalence rates of insomnia and poor sleep (Zhang and Wing, 2006). Gender differences in prevalence rates of depression and anxiety disorders (Lindberg et al., 1997; Zhang and Wing, 2006), as well as biological (Dzaja et al., 2005) and sociological (Chen et al., 2005) variables have been successfully demonstrated as contributing to an excess in insomnia prevalence amongst women. What is unclear, however, is the effect of psychosocial work stressors on sleep problems (Knudsen et al., 2007; Linton, 2004) and their role in explaining women’s excess in prevalence rates of poor sleep quality.

One of the dominating models in the field of psychosocial work stress is the Job–Demand–Control (JDC) model (Karasek, 1979; Karasek and Theorell, 1990) and its extension, the Job–Demand–Control–Support (JDCS) model (Johnson and Hall, 1988; Johnson et al., 1989). The central tenet of the JDCS model is an increasing likelihood of mental and physical impairment with increasing job demands and decreasing job control and social support. Thus, the most adverse health outcomes can be expected in high-demand jobs with low job control and poor social support (isolated high-strain jobs). By contrast, jobs with low demands offering high job control and social support (collective low-strain jobs) are expected to cause fewer stress-related outcomes. The stress-protective effects of job control and social support are discussed as additional main effects and multiplicative interactive effects.

The validity of the JDCS model has seldom been tested with respect to sleep problems. Although not conclusive (Kageyama et al., 1998; Tachibana et al., 1996), significant associations were found between sleep and demands (Akerstedt et al., 2002; Kalimo et al., 2000; Knudsen et al., 2007; Pelfrene et al., 2002; Utsugi et al., 2005), sleep and control (Kalimo et al., 2000; Pelfrene et al., 2002) and sleep and social support (Akerstedt et al., 2002; Nakata et al., 2004; Nordin et al., 2005). The interactive effects of the dimensions of the JDC(S) model on sleep problems have received even less attention. The few studies that explicitly tested interactive effects have provided inconsistent results (Kalimo et al., 2000; Nordin et al., 2005; Pelfrene et al., 2002; Utsugi et al., 2005). In summary, the existing literature has shown that the JDC(S) model seems to be significantly associated with sleep quality. What remains unclear is the potential of the JDCS model to explain the gender gap in the prevalence of poor sleep quality.

In our study, we applied the JDCS model within a sample of native German-speaking executives and managers. The aim of the study was to investigate the validity of the main and interactive effects of job demands, job control and social support in the prediction of sleep quality. Further, the gender difference in moderating the influence of job control and social support on high job demands and its influence on sleep quality were specifically explored.



Data collection for the SHAPE study (Study addressed to Highly Ambitious PErsonalities) was conducted from January 2005 to April 2006. The goal of the study was to investigate the significance of occupational characteristics on the physical and mental health of German-speaking executives. Data were collected through a self-administered, paper-based questionnaire. This questionnaire covers several health indicators, health behaviors and health relevant factors such as the work characteristics, personality traits and a stress inventory. Prior to study recruitment, the questionnaire was pretested twice with 30 subjects in a pilot study. Participants in the pretests were personally interviewed and their feedback used to refine the questionnaire.

The study population comprised a collective of middle and top executives in Germany, Austria and Switzerland. The inherent difficulty to derive a population-based representative sample of higher managers forced us to use different mechanisms to approach the target population. Thus, we used established channels of ‘trust’ to establish the current study population. For example, individuals enrolled in leadership seminars aiming at personal development were invited to take part in the study. Recruitment also occurred through presentation of the study by research team members at meetings of staff managers and senior executives within German and Austrian companies. In addition, all German military officers serving at NATO headquarters in Brussels were approached and invited to take part in the study. Inclusion criteria were reporting either: (1) an annual income above €90 000 in a private economy, (2) a salary class of at least A15 (Bundesministerium des Innern, 2007) in the public sector (which is one of the highest salary classes within the public sector and may be received for example by chief physicians) and/or (3) a high level of responsibility for their organization’s success. All participants provided informed consent.

The study population consists of 483 participants who returned questionnaires. This represents a response rate of approximately 27% among distributed surveys. Of the returned questionnaires, 87.8% or 424 participants had complete data in the dimensions of the JDCS model and sleep quality. Of this sample, 17.9% were women (n = 76). The participants worked in various branches of the German economy [most frequently in the finance sector (23.4%), transport and logistics (7.8%) and electronic industry (7.3%), as well as in the public sector and armed forces (15%)].


Job demands were measured by the ‘Overwork’ scale of the Trier Inventory on Chronic Stress (Schulz et al., 2004), which assesses quantitative overwork and shortage of time and includes eight items. Job control was measured by a three-item scale. The three items addressed the freedom to develop new ideas and solutions, the possibility of performing tasks in different ways and the scope of decision making. Social support was measured by a six-item scale. Five items assess instrumental and emotional social support provided by colleagues. These five items were adapted from the SALSA questionnaire (Rimann and Udris, 1997) and supplemented by a self-developed item asking about the general working atmosphere.

Sleep quality was measured by a five-item scale addressing global evaluation of sleep quality, problems initiating and maintaining sleep, intake of sleeping agents and occurrence of nightmares. Specifically, the English translations of the German items were: (1) ‘Generally, I sleep well’, (2) ‘I have trouble falling asleep’, (3) ‘I wake up several times per night’, (4) ‘I have nightmares’, (5) ‘I take prescribed drugs to improve my sleep’. Items 1 and 5 were included to ascertain consistency with an ongoing longitudinal study of our group (Kudielka et al., 2004a,b). Items 2 and 3 were derived from the Jenkins Sleep Problem Scale (Jenkins et al., 1988). Item 4 was included after suggestions from executives during pretesting of the questionnaire. As executives reported to frequently suffer from waking up prior than desired if allowed a full length of sleep, we included a sixth item: ‘If I do not have to get up in the morning, I feel recovered by sleeping in’. However, this last item decreased the internal consistency of the scale. Further, a factor analysis of the sleep quality items showed that the sixth item formed an independent factor and did not yield high loadings on a second factor, which encompassed the items 1–5. Therefore, the dependent variable in the present study comprised of the summary score of items 1–5 (range 5–25), with larger values indicating better sleep quality. The internal consistency was satisfactory (Cronbach’s α = 0.78). In a separate data set (Kudielka et al., 2005), a summary score of items 1–3 and 5 explained 86% of the variance of a summary score of the full Jenkins Sleep Problem Scale. The scale discriminated well from other psychosocial constructs such as self-rated health, psychosomatic complaints, depression or subjective perception of chronic stress (none of the items loading larger than 0.25 in factor analyses, data available upon request).

Control factors. As control factors, age and occupational hierarchy were assessed. Occupational hierarchy was self-reported by subjects as lower, middle or upper management, with missing values (n = 40) coded with a third dummy variable labeled ‘missing hierarchy’.

In order to control for potential sleep relevant lifestyle factors, we measured body mass index, smoking habits (six categories ranging from ‘non-smoking’ to ‘more than 30’ per day), time spent on endurance sports activities (six categories ranging from ‘never’ to ‘more than 4 h a week’), black tea/coffee consumption (four categories ranging from ‘never’ to ‘more than six cups per day’) and ‘cola drinks’ consumption (four categories ranging from ‘never’ to ‘more than 1 L per day’).

Further, subjects were asked to categorize their average daily beer and wine/sparkling wine consumption on the basis of four categories (very rarely, 0.25–0.5 L, 0.5–1 L, >1 L). The mean (i.e. 0.375 L in category two) amount in each category was then converted into grams of pure alcohol according to standardized conversion factors (beer = 4.8 vol.%, wine = 11 vol.%) (Bühringer et al., 2000) and finally summed up into cumulative alcohol intake.

Statistical analyses

The statistical analysis comprised three steps. First, gender differences in socio-demographic, lifestyle and work-related characteristics were tested with unpaired t-tests, Mann–Whitney U-tests and chi-squared tests. Second, correlations between the JDCS dimensions and sleep quality were computed for women and men separately. Gender differences in bivariate correlations were assessed using a freeware program (Instructional Technology Group, 2007). Third, a multi-linear hierarchical regression model was used to analyze the validity of the JDCS model and the moderating effect of gender in the prediction of sleep quality. To reduce the effect of multicollinearity, we used z-transformed variables (Aiken and West, 1991). All interactive terms were formed by multiplications of the z-transformed variables (Tabachnik and Fidell, 2007). The squared main effects of the JDCS model dimensions were controlled to minimize the risk of identifying spurious moderator effects (Lubinski and Humphreys, 1990; Warr, 1990).

In model building, all independent variables were added in sequential blocks with potential predictors forced to stay in the equation. Step 1 included the control variables. In step 2, gender and the main effects of the JDCS model were added. Step 3 included the squared main effects of job demands, job control and social support. Subsequently, all possible two-way interactions among the dimensions of the JDCS model and gender were included. In step 5, all possible three-way interactions of job demands, job control, social support and gender were admitted. A last step enriched the model by a four-way interaction of gender and the JDCS constructs. Significant higher order interactions are graphically illustrated to prevent misinterpretations of their directions (Aiken and West, 1991).

spss statistical software (15th version; SPSS, Inc., Chicago, IL, USA) was used for all analyses unless otherwise stated. All tests were two tailed. A type I error probability of less than 5% was considered to constitute statistical significance.


Significant gender differences in the study and control variables were observed (Table 1). Female executives reported higher job demands, lower job control and less social support. Women also reported lower sleep quality. Further, male executives were older and over-represented in higher occupational hierarchies. When compared with their female counterparts, male managers had higher body mass index, were more likely to report high intake of coffee/black tea but consume fewer ‘cola drinks’ (Table 1).

Table 1.   Gender differences in the descriptives of the investigated variables
VariableMale (N = 348)Female (N = 76)P gender difference
  1. Gender differences investigated by unpaired t-test.

  2. Gender differences investigated by chi-squared test.

  3. §Gender differences investigated by Mann–Whitney U-test.

  4. *P < 0.05, **P < 0.01, ***P < 0.001 (two tailed).

JDCS dimensions
 Job demands14.2 ± 6.318.8 ± 5.8***
 Job control12.4 ± 2.911.2 ± 3.4**
 Social support26.9 ± 5.525.1 ± 8.1*
 Sleep quality21.2 ± 2.719.1 ± 4.1***
 Age46.5 ± 8.540.0 ± 8.6***
Level of hierarchy
 Upper management (%)55.3020.50***
 Middle management (%)35.7064.40
 Lower management (%)9.0015.10
Lifestyle factors
 Cigarette smoking§1.4 ± 1.11.4 ± 1.0NS (P = 0.84)
 Consumption of coffee/black tea§2.4 ± 0.8 2.1 ± 0.7**
 Consumption of ‘cola drinks’§1.2 ± 0.5 1.4 ± 0.7*
 Consumption of alcohol (grams)§30.4 ± 32.5 28.4 ± 29.8NS (P = 0.77)
 Endurance sports§2.9 ± 1.52.8 ± 1.3NS (P = 0.47)
 Body mass index25.6 ± 2.822.3 ± 3.5***

Reliability (Cronbach’s α) and intercorrelations between the study and control variables are presented separately for women and men in Table 2. In general, the highest correlations with sleep quality were observed with social support and job control. However, these two job resources yielded significantly higher correlations with sleep quality among female managers. Only amongst the women, body mass index was significantly positively associated and ‘cola drinks’ significantly negatively associated with sleep quality. By contrast, consumption of coffee/black tea was only significantly positively correlated with sleep quality among men (Table 2).

Table 2.   Reliabilty (Cronbach’s α) and intercorrelations between investigated variables
  1. *P < 0.05, **P < 0.01, ***P < 0.001 (two tailed).

1. Sleep quality (α = 0.78)          
2. Job demands (α = 0.92)−0.19***         
3. Job control (α = 0.78)0.26***−0.13*        
4. Social support (α = 0.89)0.35***−0.23***0.52***       
5. Age−0.06−0.14**0.13*0.10      
6. Alcohol consumption−0.07−0.04−***     
7. Body mass index−0.07−0.010.01−0.020.16**0.09    
8. Endurance sports−0.08−−0.06−0.14**−0.14−0.11−0.23   
9. Cigarettes−0.020.08−−0.13**  
10. Cola drinks−0.030.10−0.07−0.08−0.25***−0.11*−0.11*−0.050.15** 
11. Coffee/black tea0.11*0.080.02−0.030.14**0.100.06−0.100.14**0.09

Further, the correlations of social support with job control as well as consumption of ‘cola drinks’ with coffee/black tea were significantly highly correlated in female when compared with their male counterparts. The results of the final model predicting sleep quality are presented in Table 3. The control variables age, body mass index and consumption of coffee/black tea yielded significant beta-weights. Lower age, higher body mass index and higher intake of coffee/black tea are significantly associated with good sleep quality.

Table 3.   Results of the final regression model and adjusted R2 increases of the stepwise regression model building
PredictorSleep quality (β-weight, significance level)Adjusted R2 increase
  1. *P < 0.05, **P < 0.01, ***P < 0.001 (two tailed).

  2. High values indicate good sleep quality.

  3. Non-significant predictors are not shown.

Control variables
 Coffee/black tea0.13** 
 Body mass index0.09* 
Main effects
 Job demands−0.08 (NS)0.208***
 Job control0.00 (NS) 
 Social support0.19** 
 Gender−0.08 (NS) 
Squared main effectsNS0.046***
Two-way interactionsNS0.028***
Three-way interactions 0.010*
 Demands × control   × social support−0.26** 
 Control × social  support × gender−0.22* 
Four-way interaction
 Demands × control × social support × gender0.31**0.009**
 Adjusted R20.420 

Among the main effects of the JDCS dimension, only social support was observed to explain a significant proportion of the variance of sleep quality. A significant main effect of gender was not observed. Neither squared main effects of the JDCS dimensions nor two-way interactions were found to explain a significant proportion of the variance of sleep quality.

Three significant higher order interactions were observed. The interaction of job demands, job control and social support (Fig. 1) suggests that working in isolated high-strain jobs negatively affects sleep quality. By contrast, high job control and high social support seem to have a buffering effect on high job demands. Further, the interaction of job control, social support and gender contributes significantly to the explanation of the variance of sleep quality (Fig. 2). This three-way interaction of job control, social support and gender suggests that women – regardless of the intensity of job demands they encounter – are more severely affected by social isolation and low job control when compared with male executives.

Figure 1.

 Three-way interaction of job demands, job control and social support.

Figure 2.

 Three-way interaction of job control, social support and gender.

Additionally, the four-way interaction of gender, job demands, job control and social support was found to be a significant predictor of sleep quality (Fig. 3). As illustrated in Fig. 3, panel A, female managers are exceptionally prone to react with decreasing sleep quality when working in isolated high-strain jobs. Adjusted R2 indicates that the final model of this regression equation accounts for 42% of the variance of sleep quality.

Figure 3.

 Four-way interaction of job demands, job control, social support and gender.


Scope of the study

The present study aimed at elucidating the main, interactive and gender-dependent effects of job demands, job control and social support in the prediction of sleep quality among managers and executives. We sought to clarify whether the three-way interaction of the dimensions of the JDCS model predict sleep quality beyond the additive main effects. In addition, we aimed to evaluate whether gender moderates the composite sleep effect of job demands, job control and social support.

Principal findings

On the level of the main effects, only social support significantly predicted evaluations of sleep quality in the final regression model. The finding of a significant sleep-improving main effect of social support is consistent with results of most previous studies (Akerstedt et al., 2002; Nakata et al., 2004; Nordin et al., 2005). By contrast, the main effects of job demands and job control were no longer associated with sleep quality when interactive effects of the JDCS dimensions were added to the regression model.

Even though female managers reported significantly lower sleep quality, the main effect of gender was not significant in the fully adjusted model. Gender lost its significance with the inclusion of interactive effects in the regression model. Thus, it seems as if it is not gender per se but rather the (gender-dependent) interactions of the JDCS dimensions that account for female manager’s reduced sleep quality.

We observed three higher order interactions. The substantial and significant interaction of job demands, job control and social support (Fig. 1) reflects the buffer hypothesis of the JDCS model: isolated high-strain jobs are associated with the most adverse effects on sleep quality. By contrast, the combination of high job control and high social support has the strongest buffering effect on the sleep-disturbing outcome of high job demands.

The four-way interaction of gender, job demands, job control and social support examined the question of whether these observations hold equally for both gender. Fig. 3 pinpoints to where in the JDCS model gender differences may matter. As illustrated in panel A, female managers experience greater adverse effects on sleep quality when working in isolated high-strain jobs when compared with their male counterparts. An observation that holds equally for both genders is that the provision of either job control (panel B) or social support (panel C) has a substantial buffering effect on the negative outcomes of increasing job demands. The size of this ‘partial buffering effect’ is comparable with the size of the composite buffering effect of high job control and social support (panel D) among women. Amongst men, the composite effect of high job control and high social support seems to be slightly greater in size than any of the partial buffering effects. The result that women react with disturbed sleep when exposed to job environments both with low job control as well as low social support was further supported by a significant three-way interaction of social support, job control and gender (Fig. 2). This three-way interaction of job control, social support and gender suggests that women – regardless of the intensity of job demands they encounter – are severely affected by social isolation and having only little possibilities of making own decisions. By contrast, male managers seem to be nearly unaffected by the interplay of low job control and low social support. Male managers only seem to suffer from this combination if high job demands are added (Fig. 3).

The findings of interactions consistent with the JDCS theory challenge the position by de Jonge and Kompier (1997) on the difficulty to interpret interactive effects of the JDCS dimensions. Furthermore, our results question the suggestion by Taris (2006) to cease research on interactive effects within the JDCS framework due to the absence of empirical evidence. By contrast, we propose intensifying research on interactive effects, as they seem to contribute to an understanding of women’s increased probability of poor sleep quality.

Possible explanations

Which possible explanations for the findings suggesting that female managers are particularly prone to the sleep-disturbing effect of isolated high-strain jobs are offered by the existing literature? More balanced effort–reward ratios may help male managers to compensate the adverse effects of isolated high-strain jobs (Siegrist, 1996): female managers generally receive lower salaries (Greenglass, 2002) while expected to perform at higher standards than their male colleagues (Davidson and Burke, 2000). Additionally, women may experience barriers in promotion to senior management positions (Daily et al., 1999; Fielden and Cooper, 2002; Lee and James, 2007). Our data support the idea of higher effort–reward imbalance as male executives reported lower job demands but assigned themselves more frequently to high-salary classes and upper management positions when compared with female executives. This explanation receives further support by findings of Ostry et al. (2003) revealing that the interaction of dimensions of the JDC model and the Effort–Reward Imbalance (ERI) model leads to a better prediction of self-reported health status. Utsugi et al. (2005) observed that the JDC model has a higher predictive value among women, whereas the ERI model had a higher predictive value among men. Applying Ostry et al.'s (2003) and Utsugi et al.'s (2005) results to our empirical findings, it is reasonable to assume that male managers (who are more affected than women by gratifications such as job promotion and high salary) have a better chance of compensating the detrimental effects of isolated high-strain jobs.

An alternative avenue of explanation is provided by Ely (1994). He stresses the sociocultural context within which women work, combining social identity theory (Tajfel and Turner, 1985) with theory of token status (Kanter, 1977; Spangler et al., 1978). According to social identity theory, members of social groups (female executives) may perceive clear and abiding status differences in comparison with other relevant groups (male executives). This perception leads to comparison with in-group members (other female executives) aiming to maintain a positive self-image. The downside of such self-enhancing in-group comparison is their threat to in-group solidarity and within-group social support (William and Giles, 1978). A second threat arises according to Kanter’s theory of token status from female managers’ experience of being largely under-represented relative to male managers. Token status is accompanied by a salient risk of social isolation. The combining element of these two theories is a relative excess of the risk of social isolation for female executives. By contrast, male executives can often rely on influential informal networks providing assistance and important information to facilitate work performance (Fielden and Cooper, 2002) and to reduce work stress.

Strengths and weaknesses

Several caveats of the present study require consideration. In this study, sleep quality was measured with a five-item scale that in its present form has not been previously validated. However, four items were taken either from the Jenkins Sleep Problem Scale (Jenkins et al., 1988) or included in previously published studies from our group (Kudielka et al., 2004a, 2005). Only one item that assessed the frequency of nightmares was new. Excluding the new item, our scale was observed to explain 86% of the variance of the full Jenkins Sleep Problem Scale in an independent data set (Kudielka et al., 2005), suggesting convergent validity. Discriminant validity is indicated by factorial analyses showing that our full sleep quality scale is distinct from other health-related constructs, such as depression or psychosomatic complaints. The cross-sectional nature of this study permits causal inferences. From a statistical perspective, the observed associations could arise from other unknown confounding variables or differ in direction. For example, it is theoretically conceivable that lower sleep quality increases the threat of social isolation or reduces the manageability of job demands. However, several relevant independent variables are either fixed (gender) or change over a much larger time frame than the typical recall period of sleep quality (age and occupational hierarchy). Moreover, a systematic review on longitudinal studies assessing the JDC(S) model (de Lange et al., 2003) suggests that the results from longitudinal and cross-sectional studies are similar, supporting the notion that the direction of effect is from the dimensions of the JDCS to employees health. A questionnaire-based study such as the present one is prone to criticism that the observed associations are inflated by the common variance (e.g. female managers who report low social support in the context of high-control, high-demand jobs would tend to report more psychosomatic complaints than their male counterparts). In line with Spector’s (2006) criticism of the common variance argument, it is difficult to explain why the common variance should affect a particular subgroup of the study population and thus give rise to a false-positive observation of relevant higher order interactions. The subgroup issue leads to a further limitation of the present study. Despite being one of the largest investigations into sleep quality amongst executives, the number of female executives in the present study was limited. While our sample represents the female-to-male ratio found in managerial positions in Germany (Hoppenstedt, 2007), a larger proportion of women would have increased the power to detect effects that remained non-significant in the present study.

Are the observed effects relevant? For example, the four-way interaction accounts for 0.9% of the variance of psychosomatic complaints. This amount of additionally explained variance beyond main effects is within the R2 increases of 1–3% typically found in studies with relevant interactive effects (Champoux and Peters, 1987). Due to difficulties in the verification of moderator effects, Evans (1985) suggests that R2 increases of 1% should be regarded as relevant. Moreover, we took several statistical precautions to prevent increases in inflated R2 values, e.g. by controlling for squared main effects, all lower order interactions and by deriving interactive terms from multiplications of z-standardized values. This provides further protection against spurious moderator effects (Jones and Fletcher, 1996; Lubinski and Humphreys, 1990), multicollinearity (Schaubroeck and Fink, 1998) and liberal statistical modeling (Tabachnik and Fidell, 2007).


In conclusion, the present study implies that high job control and high social support might help executives to mitigate the adverse effects of a demanding job on sleep quality. Further, interactive effects of the JDCS model might help to explain the gender gap in sleep quality differences. Interventions targeted at enhancing social support and job control might complement non-pharmacological strategies such as giving information about sleep hygiene (Atlantis et al., 2006), offering cognitive–behavioral therapy (Harvey et al., 2002) and developing corporate sleep policies limiting scheduled work time for management personnel (Czeisler, 2006). However, longitudinal and intervention studies aiming at elucidating the causal effects of increasing job resources on women’s and men’s sleep are needed.


The SHAPE study was supported by a grant of the Stifterverband für die Deutsche Wissenschaft der Medizinischen Hochschule Hannover, project number: H 410 7103 9999 12337.

Conflict of interest