Social stress at work and change in women's body weight.

Social stressors at work (such as conflict or animosities) imply disrespect or a lack of appreciation and thus a threat to self. Stress induced by this offence to self might result, over time, in a change in body weight. The current study investigated the impact of changing working conditions--specifically social stressors, demands, and control at work--on women's change in weighted Body-Mass-Index over the course of a year. Fifty-seven women in their first year of occupational life participated at baseline and thirty-eight at follow-up. Working conditions were assessed by self-reports and observer-ratings. Body-Mass-Index at baseline and change in Body-Mass-Index one year later were regressed on self-reported social stressors as well as observed work stressors, observed job control, and their interaction. Seen individually, social stressors at work predicted Body-Mass-Index. Moreover, increase in social stressors and decrease of job control during the first year of occupational life predicted increase in Body-Mass-Index. Work redesign that reduces social stressors at work and increases job control could help to prevent obesity epidemic.

Introduction

Being overweight and obesity have become an increasing problem of modern society1). Previous research has shown high levels of Body-Mass-Index (BMI) to be a risk factor for impaired well-being2,3,4) and health5,6,7,8). Additionally, being overweight and obesity are associated with considerable costs in the workplace9, 10), direct medical costs and indirect costs following disability retirement11, 12), or occupational injuries13).

Stress might be an important factor in the development of obesity1, 14,15,16). Research on work stress and BMI mainly focused on two stress models − the job demand-control (JDC) model17) and the effort-reward imbalance (ERI) model18) − with only limited evidence19,20,21,22,23). In general, a recent meta-analysis of longitudinal studies of stress and adiposity found work-related stress to be associated with increasing adiposity24). Moreover, several psychosocial work factors predict changes in BMI25, 26). Nevertheless, effects are rather small and results are heterogeneous24). Thus, future longitudinal research that includes change in body weight should rely on objective weight measurement and objective assessment of work conditions to reduce mono-method bias27).

So far, work-related social stressors (social animosities, conflicts with co-workers and supervisors, unfair behaviour, and negative group climate28) have largely been neglected in this context. Recently, Berset et al. highlighted the role of social stressors as a direct threat to an individual’s need of belongingness29). They analysed the effects of social stressors, JDC and ERI models, and found that social stressors at work negatively influence a person’s BMI, and a person’s control over work situation are positively associated with BMI two years later. However, both stressors and BMI measures relied on self-report, implying the danger of inflated effects due to common method variance27, 30,31,32).

Therefore, the present longitudinal pilot study focused on self-reports of social stressors at work as well as observer-rated work stressors and job control to predict changes in women’s BMI. We want to extend previous research by including: a) observer-rated job demands and control, as well as self-reported social stressors; b) a longitudinal design with a time lag of one year; c) change in BMI predicted by change in work stressors; and d) to avoid the healthy worker effect, include female employees in the first months of their occupational life33, 34). The healthy worker effect is especially important as persons who have developed impaired well-being might quit or be assigned to less stressful activities thus, underestimating the impact of work stress33, 34).

We expect change in BMI to be positively associated with change in social stressors (hypothesis 1). Additionally, we expect it to be positively associated with change in work stressors (hypothesis 2.1) and negatively associated with change in job control (hypothesis 2.2). According to the JDC model we expect a significant interaction between work stressors and job control showing a buffering effect of job control in the association between job stressors and BMI (hypothesis 2.3).

Subjects and Methods

Participants and design

The data refers to a large-scale questionnaire study on working conditions and quality of life of young workers in Switzerland35). In the core project at Time 1, 1,394 apprentices were recruited from vocational schools in the German and French speaking areas of Switzerland in the last months before final exams. For this particular subsample all German-speaking participants (n = 802, n = 395 female) were addressed. They received an information letter and consent form including a prepaid envelope. As incentive to participate in the current study a lottery with different prices was announced (e.g., small amounts of money, vouchers for clothes or compact discs etc.). In case of agreement to participate, subsequently, supervisors of employed participants received an information letter and were contacted by phone to give informed consent and to schedule data collection. In each wave, participants were observed by trained raters as well as asked to complete questionnaires and interviews. Measurements at work were conducted with the least impairment of daily working routine as possible. However, it is a very intense study design (from the raters as well as participants points of view) and due to work place observations, the agreement of the supervisors was essential. At Time 1, 93 entrants (61 women and 32 men) agreed to participate, thus 11.6% of the German speaking participants in the core project were willing to also participate in the present study. The present analyses refer to the female subsample due to the small number of male participants and the influence of gender on BMI reactivity21, 36). Altogether 57 women and their supervisors from four occupations within the service work sector (nurses, cooks, sales persons, and bank clerks) voluntarily participated in first BMI measurement in the first six months after finishing vocational training (M=3.96 months, SD=2.84). Two women were excluded from regression analyses due to missing observational data. At time of the first measurement, participants were employed in 35 different low-, medium- and large-size organizations located in rural and urban areas of the Swiss speaking parts of Switzerland. Mean age was 22.47 yr (SD=3.63). Of those, 38 (67%) took part in the follow-up survey one year later.

Ethics

The study was performed in consensus with all requirements defined by the Swiss Society of Psychology, including participants information about their rights and guarantee of anonymity. Informed consent of participants (and supervisors in the case of observation at work) was obtained.

Measures

Job demand-control model. According to the JDC model17) job strain arises when employees simultaneously have high job demands and a low level of control over work. We used the observer version of the Instrument for Stress Oriented Task Analysis37). Raters were trained intensively. Ratings were based on an observation period of 60 min plus three additional time-samples of 15 min each. Rated items were related to frequency or intensity of job characteristics on 5-point Likert-scales. In occupational health research, the demands component of the model most often included time pressure or role conflict38). We therefore conceptualised job demands by time pressure (e.g., high work pace; three items). Cronbach’s alpha ranged between 0.74 and 0.86. Job control was measured by rating method control (e.g., independently plan and organize their own work; four items); Cronbach’s alpha ranged between 0.79 and 0.86. Observation was complemented by an interview with the jobholder (30–45 min), questions to supervisor(s) and colleague(s), and consultation of organizational documents, if necessary. The interviews were conducted using an interview guide with key questions for each dimension and served the purpose of obtaining an overview of the sequence of task conduct, verification of observer impressions and enhancement of comprehension. When observations were incomplete missing observational data were inputted with information provided by observation interview data and self-report (16%). Observations and interviews were conducted and completed during one working day (Tuesday, Wednesday, or Thursday). In order to reduce potential effects from the observation itself that might affect those who are being observed, raters were present in the first half of the working day but true observations started in the second half of working day.

Social stressors. Social stressors were measured with seven (out of eight; due to item-total correlation) items of the scale developed by Frese and Zapf39); e.g., “With some colleagues there is often conflict”). Items were answered on a 5-point Likert-scale that ranged from 1 (strongly disagree) to 5 (strongly agree); Cronbach’s alpha ranged between 0.75 and 0.76. The reliability of difference scores of social stressors, observed work stressors and job control ranged between 0.55 and 0.79.

Body-Mass-Index. BMI (kg/m2) was calculated using self-reported height measurements and body weight by use of an electronic scale (SOEHNLE Omega 7240 PW30). Weight measurement was done by raters before observation started. Participants were weighed with clothes and without shoes.

Control variables. Regarding BMI, there are numerous confounding variables. We collected some of these potential influences like age40), education (proxy for social class,41,42,43,44)), and health related behaviour (e.g., smoking, alcohol intake, staying on a diet, exercising15, 45)). To keep predictors in the analysis to a reasonable number, we only included control variables if they were associated with BMI (either at the first or second time or the difference scores). Therefore, we only controlled for age. Furthermore, no mean differences concerning these control variables between responders and non-responders (dropout) appeared (p between 0.12 and 0.94).

Statistical analysis

First, correlations between all the included variables were calculated. Second, separately, three (both A) cross-sectional and B) prospective) multiple regression analyses (hierarchical regression; Ordinary Least Squares method (OLS)) predicting A) BMI and B) change in BMI were computed. We include variables in different blocks (step one to three). The method of regression was enter. Age (step 1), predictor variables (step 2; social stressors at work, job stress and control with respect to the model a), b) or c)), and the interaction term (step 3; which is indicated by b) JDC model) were entered block by block. Components of interaction were grand mean centred. Third, a joint model of all significant predictors was calculated.

Response bias

Regarding the rather small number of respondents with respect to the main study, there might be potential response bias. We therefore compared self-reported data of responders (n=57) and non-responders (n=395) within the main study (Time 1 and Time 2 were assessed six months earlier than the data that is used in this subsample). However, there is of course a delay in time and with respect to stressors and work related control different methods were used. There were no differences in employment rate between responders (M=98.00) and non-responders (M=96.84) and the number of children (M =0.12, M =0.01) a year later. Neither at the beginning of the study nor a year later did differences for social stressors and work related control occur. Nevertheless, contrary to expectations, responders seem to have slightly more time pressure at the beginning of the study (M=3.71) than non-responders (M=3.27; t (448)=−3.53, p=0.001, two-tailed) but not a year later (M =3.55, M =3.38). In addition self-reported BMI two years later was different in that responders (M=23.69; n=40) reported higher BMI values than non-responders (n=165; M=21.93; t (203)=−2.67, p=0.008, two-tailed).

Results

Nineteen participants (out of 57) did not participate at follow-up. A test for potential bias compared whether dropouts differed from others in baseline measurement. There was no significant difference in study variables (p was between 0.22 and 0.95).

Correlations

Descriptive statistics are shown in Table 1. As expected, BMI at baseline is highly correlated with BMI at follow-up (r=0.92, p<0.01), indicating considerable stability of inter-individual differences in body weight.

Table 1.

Descriptive statistics and correlations among the study variables

Variables	M	SD	1	2	3	4	5	6	7	8	9	10	11	12	13	14	15	16	17
1. BMI T1	23.00	4.09
2. BMI T2	23.04	3.87	0.92**
3. ∆BMI	0.08	1.57	–0.16	0.25
4. Stressors T1	3.22	0.90	0.03	–0.13	–0.26
5. Stressors T2	3.39	0.68	0.13	–0.09	0.07	0.39**
6. ∆Stressors	0.20	0.82	0.04	0.06	0.30†	–0.64**	0.46**
7. Control T1	2.73	0.74	–0.02	0.19	0.42**	–0.01	–0.06	0.03
8. Control T2	2.75	0.65	–0.01	0.05	0.04	0.18	0.25	0.04	0.14
9. ∆Control	0.02	0.91	.07	–0.11	–0.31†	0.19	0.23	0.00	–0.71**	0.60**
10. Social stressors T1	1.64	0.46	0.28*	0.02	–0.21	0.29*	0.15	–0.09	–0.14	–0.16	0.01
11. Social stressors T2	1.71	0.51	0.31*	0.34*	0.15	0.25†	0.28†	–0.01	–0.16	0.01	0.13	0.46**
12. ∆Social stressors	0.06	0.51	0.15	0.31*	0.34*	0.04	0.14	0.07	–0.02	0.15	0.12	–0.46**	0.57**
13. Age	22.47	3.63	0.21	0.50**	–0.11	0.08	–0.03	–0.05	0.15	0.15	–0.19	–0.17	–0.21	0.06
14. Education	2.77	1.00	–0.04	0.15	0.12	–0.23	–0.07	0.16	0.21	0.20	–0.03	–0.28	–0.14	0.13	0.11
15. Dietary restraint	1.65	0.33	–0.02	–0.03	–0.23	0.26	–0.17	–0.39*	0.10	–0.12	–0.17	0.26	0.22	–0.02	0.09	0.15
16. No. of cigarettes	22.68	50.81	0.25	0.31	–0.04	–0.08	–0.08	0.09	–0.29*	–0.01	0.18	0.16	0.17	0.11	–0.14	0.07	0.00
17. Grams of alcohol	52.93	122.18	0.09	–0.01	–0.06	0.03	0.05	0.06	–0.15	0.02	0.14	0.12	–0.05	–0.10	–0.14	0.09	0.04	0.28*
18. Exercise	3.18	0.66	–0.07	–0.01	–0.13	0.06	0.15	–0.16	0.13	0.05	–0.18	0.02	0.10	–0.06	–0.09	0.03	0.31*	0.10	0.00

Pearson correlation coefficients. N<57. Working conditions refer to a five-point Likert-scale that ranged from 1 (strongly disagree) to 5 (strongly agree). Dietary restraint refer to the voluntary cognitively controlled effort to restrict food intake to control body weight by four items (1=no; 2=yes). Number of cigarettes and grams of pure alcohol refers to one week; data were assessed daily during examination week. Doing exercise refers to the frequency of sports by an index of three items on a five-point Likert scale, ranging from 1 (never) to 5 (daily). ∆ change t2–t1. †p<0.05 ; * p<0.05 ; ** p<0.01 (2-tailed).

Regression Analyses

Table 2 shows results of the regression analysis. At baseline, there is only a positive effect of social stressors on BMI.

Table 2.

Summary of multiple regression analysis for social stressor and the JDC model predicting BMI and change in BMI one year later

	A) BMI at baseline						B) Change in BMI one year later
Variables	b	SE b	β	ΔR2	R2		b	SE b	β	ΔR2	R2
Step 1					0.04						0.01
Age	0.24	0.15	0.21				–0.05	0.08	–0.11
a) Social stressors at work model
Step 2				0.11**	0.15					0.13†	0.14
Social stressors	2.90	1.14	0.33**				1.12	0.47	0.37*
b) JDC model
Step 2				0.00	0.05					0.18*	0.19
Observed stressors	0.05	0.62	0.01				0.58	0.30	0.30†
Observed control	–0.30	0.77	–0.05				–0.59	0.28	–0.31*
Step 3				0.03	0.07					0.02†	0.21
Demands × control	–0.99	0.82	–0.17				0.36	0.41	0.14
c) Joint model of job control and social stressors
Step 2										0.23*	0.24
Social stressors							1.13	0.45	0.37*
Observed control							–0.58	0.28	–0.31*

N = 55/ N = 38. A) cross-sectional and B) prospective multiple regression: Variables predicting BMI at baseline refer to baseline values (t1); variables predicting change in BMI refer to difference scores (t2 − t1). b = unstandardised regression coefficient. SE b = standard error of unstandardised regression coefficient. β = standardised regression coefficient. ΔR change in explained variance. R = explained variance. †p<0.10. * p<0.05. ** p<0.01 (2-tailed). Post hoc power analysis (2-tailed): main effects = 68−92%; interaction = 22%.

Longitudinally, an increase in social stressors is associated with an increase in BMI too (Fig. 1). Concerning the JDC model, decreasing control (and only marginally significantly increasing demands) predicts increasing BMI (Fig. 2). Thus, job control, while not correlated with baseline and follow-up BMI, did predict change in BMI, however, no significant interaction emerged. Both social stressors at work and control could explain a significant amount of variance in change of BMI − confirming hypotheses 1 and 2.2.

Fig. 1.

Difference in BMI as a function of social stressors at work.

Fig. 2.

Difference in BMI as a function of observed control at work.

Discussion

The current study investigates associations between working conditions and BMI. Thereby, social stressors at work and participant’s height were the only self-report measures included. In addition, we concentrated on female employees in the first months of their professional life to preclude the healthy worker effect33, 34). Change in social stressors and job control predict change in BMI over one year − confirming the importance of work stress in the development of obesity. Across the year of observation, work stress might have changed energy intake (e.g., increasing drive to eat;16, 45,46,47)), decreased physical activity45, 48), and stimulated the organism to accumulate fat to face stressful demands29).

Results are in line with previous research underlining the importance of social stressors within the scope of stress as an offence to self49,50,51). People strive to maintain a positive sense of self52). However, social stressors at work are common in job beginners and include attacks, making fun at the expense of other people, or a disproportional response to mistakes35). Thereby, they imply a lack of appreciation or even disrespect and thus a threat to self53). Especially stressful situations that involve social-evaluative threats tend to evoke an increased cortisol response54) and, therefore, they should be of particular importance. The release of the stress hormone cortisol might be an important physiological link between stress and body weight. Roberts et al.55) found daily changes in cortisol levels to predict changes in BMI, and this relationship was mediated by aspects of eating behaviour.

Regarding the JDC model, results reveal a main effect of observed control on decreasing BMI; observed demands are marginally significant. From a cross-sectional perspective, neither control nor demands were significant, which stresses the importance of investigating intra-individual shifts in variables over time. No interaction emerged reconfirming previous research29, 56). However, taking into account the lack of power due to the small sample size an existing interaction effect would be difficult to find57, 58). Therefore some authors58) suggest a liberal significance criterion of 0.10 to test an interaction. Thus, future research is needed.

In the joint model of social stressors and control, observed control and self-reported social stressors are associated with change in BMI. Thus, results are less subject to common method variance. Additionally, there are several reasons (e.g., timing or the abundance of potential influences) for empirical research − in cross-sectional as well as longitudinally findings − to rarely find effects of work-related stressors on well-being or health-related outcomes explaining more than 5 to 10% of the variance59). The joint model explained 23% of variance in change in BMI across one year. Thus, effects seem to be quite remarkable.

However, even though we measured body weight by scale we use self-reported data for height to calculate the body mass index. In consequence of social desirability, this might result in a potential bias. On the other hand, Donaldson and Grant-Vallone60) showed that agreement among self-reports and measures of height was quite high. In addition, using BMI as a continuous and not a categorical variable should lower the impact of potential bias on the analyses61). However, any biasing influence might especially tend towards the mean62), restricting the variance in the BMI measures and therefore causing effects to be underestimated, rather than overestimated. Nevertheless, future research should include an objective measure of height to reduce any kind of bias.

We longitudinally investigated the effects of working conditions on change in BMI using self-report and ratings by trained observers. However, due to the complex study design, the sample size is small: observations increase the effort involved and require permission of the management, and longitudinal studies involve some dropout. Sample size implies limited power and therefore extends the risk of missing a significant effect. As a consequence, effects might have been underestimated especially regarding observed demands and the interaction of demands and control. Moreover, the power to detect interaction in regression is low57, 58). As we investigated only young women and only controlled for age, generalisation was restrained as well.

Confidence in difference scores has been discussed63, 64). However, there is also a great deal of support for their use under certain conditions: difference scores are appropriate when test measures are reliable65, 66) and reliability of scores exceeds 0.5, which is true for all included psychological variables67). Future research should raise the number of measurements and estimate growth in BMI over larger periods of time.

In conclusion, results highlight the importance of social stressors and control at work in the context of BMI and thus, of population health.