Longitudinal association between psychological demands and burnout for employees experiencing a high versus a low degree of job resources

A questionnaire was sent to 6289 employees in seven Swedish organizations in 2010–2012 with a follow-up after 2 years. Inclusion criteria were being employed in one of the participating organizations and understanding Swedish. Prior to the distribution of questionnaires, organizational schedules and lists of employees containing names, age and gender were retrieved from the organizations. For the follow-up, organizational schedules and lists of employees were again collected and manually checked for changes, where the follow-up questionnaires were only sent to employees who had participated at baseline. Employees in the organizations received a questionnaire on paper or electronically. The paper questionnaires were distributed in individually addressed and closed envelopes which included a coded survey, an informative letter concerning the project, and a pre-stamped response envelope. The electronic version was sent along with an informative letter to employees’ personal work e-mail. Employees were allowed to fill in the questionnaire during working hours. In total, n = 3551 responded at baseline (56%). At follow-up, 2696 of the respondents at baseline were still employed and received a second questionnaire to which n = 1722 responded (64%). In this final longitudinal sample, n = 295 were employed in a private production company, n = 183 in a private care company, n = 477 in a municipal organization including several different occupations, n = 121 in a different municipal care unit, n = 68 in a public care organization, n = 322 in a government authority, and n = 256 in a government organization. The mean age was 49 years at follow-up (standard deviation = 10.20 years), 66% were women and the majority had high school/vocational school education. A flow-chart of the data collection is shown in Fig. 1.

Outcome: The main outcome for this study was burnout measured at the time two survey. Burnout was measured at both time points using the generic subscale “Personal burnout” from The Copenhagen Burnout Inventory [2]. It consists of six items concerning physical and psychological fatigue (e.g. How often are you emotionally exhausted? How often do you feel worn out?). Internal consistency: Cronbach’s α .88. Reponses were standardized into a scale from 2 to 10, with higher scores indicating more burnout.

Main predictors: The main predictors were job psychological demands and burnout measured at the time one survey. The measurement of burnout is described above. Psychological demands was measured with The Swedish Demand Control Questionnaire [11, 26]. Five items measure psychological demands (e.g. Does your work demand that you work very hard? Is there enough time to perform work tasks?) Answers were given on a four-point Likert scale. Internal consistency for psychological demands: Cronbach’s α .79. Responses were standardized into a scale from 2 to 10, with higher scores indicating higher psychological demands.

Potential moderating variables: We used two measures to define supportive work environments. These were decision latitude and social capital. Decision latitude was measured with The Swedish Demand Control Questionnaire [11, 26], containing six questions (e.g. Do you have the opportunity to decide for yourself how to carry out your work? Do you have the opportunity to learn new things in your work?). Answers for decision latitude were given on a four-point Likert scale, with internal consistency of α .68. Social capital at work refers to the degree to which connections between individuals in the workgroup are trusting and enables collaboration and information sharing [27, 28], and was measured by eight items (e.g. Members of the work unit build on each other’s ideas in order to achieve the best possible outcome; People keep each other informed about work-related issues in the work unit). Answers were given on a five-point Likert scale. The internal consistency for social capital was Cronbach’s α .90. Responses for both dimensions were standardized into a scale from 2 to 10, with higher scores indicating higher decision latitude and social support.

To define supportive work environments over the study period we used both time one and time two measures of social capital and decision latitude, to reduce measurement error from only using one time point. Decision latitude and social capital at baseline and follow-up were dichotomized into high and low based on a median split approach. Based on categories and baseline and follow-up respondents were then grouped into the following four categories for each resource: consistently high (high at both time points); decreasing (high at baseline, low at follow-up); increasing (low at baseline, high at follow-up); and consistently low (low at both time points).

Covariates: Potential confounders for regression models included organization, gender, age (continuous), level of education, marital status and having children living at home.

Initial descriptive analyses examined the distribution of the sample across our main outcome, independent variable, moderating variables and covariates.

Linear regressions were used to examine the associations between burnout and psychological demands at baseline and burnout at follow-up. Initial analyses investigated how psychological demands and burnout at baseline were associated with burnout at follow-up, adjusted for decision latitude, social capital at work and demographics. Additional models examined if there were differences in the relationship between psychological demands and burnout at baseline on burnout at follow-up for employees who experienced a high versus a low degree of job resources over the follow-up period. For these models, we restricted our analyses to respondents who had either supportive environments at both time points (based on measures of social capital and decision latitude) or non-supportive environments at both time points (N = 1163 for analyses using decision latitude and N = 1079 for analyses using social capital). The restriction of the sample in this way was to better ensure the comparison of effects was between respondents with consistently high and consistently low job resources, given the increased accuracy gained by the use of two time points [25].

We examined differences in two ways. We first ran models which included an interaction term between psychological demands and burnout, and high/low resources (based on decision latitude and social capital scores at time 1 and time 2). Separate models were run for each interaction term, with four models being run in total (psychological demands and high/low decision latitude; psychological demands and high/low social capital; burnout and high/low decision latitude; and burnout and high/low social capital) To compliment these analyses we also ran a series of models, stratified by level of resources, to examine the relationships between baseline burnout and psychological demands on time two burnout for respondents with high and low decision latitude and social capital environments. All analyses were conducted using SAS V 9.3.

Non-respondents at baseline were older (46 years) than respondents (45 years) (p = .013). There was no difference concerning gender. Compared to those respondents who responded to both surveys, those who did not respond to the follow-up had lower psychological demands (p = .028) and lower decision latitude (p = .022) and were younger (p = .006). There were no significant differences concerning burnout, gender or social capital at work. Given the high alpha estimates for the burnout, psychological demand, decision latitude and social capital scales, if respondents were missing responses on less than half of the questions in the scale, these values were imputed (up to two items for burnout, psychological demands and decision latitude and up to three items for social capital) [29]. This resulted in 9 respondents having imputed values for some burnout items at time one and 4 respondents having imputed values for items at time two; 42 respondents with some imputed values for psychological demands at time one; 25 respondents with some imputed values for decision latitude at time one and 28 respondents at time two; and 49 respondents with some imputed values for social capital items at time one, and 33 at time two. Of the sample of 1722 respondents who responded to the follow-up survey, 107 respondents (6.2%) had missing information on more than half of the items assessing study exposures, or information on outcomes or covariates. The majority of these missing responses were missing values for age (N = 59, 55% of missing responses), with another 32 respondents missing information on more than half of the items for a given work exposure. This left a final analytical sample of 1615 respondents. A logistic regression analysis examined the probability of missing information on age across organization and demographic variables with complete information. This analysis observed male respondents and those who were single were more likely not to have reported age. No relationship was found between demographic factors or organizational factors and the likelihood of missing responses on work exposures.

Decision latitude and social capital have previously been found to predict work-related flow in the same sample [37] and are hence by definition job resources with buffering potentials [14]. It could however be argued that the hypothesis was rejected due to the variables that were used. Based on their review, Häusser et al. [18] suggest that in order for control or support to buffer the effect of high demands, these job resources must be based on qualitatively identical dimensions and measure the same aspect of work. In the present study, psychological demands concern the task level of work, while social capital at work concerns the collective dimension beyond the individual’s work tasks, reflecting the trust that is experienced between individuals at work and the collaborative capacities within groups [27, 32, 38]. Decision latitude and psychological demands are however both derived from the same instrument and are to be considered to concern the same task level of work as psychological demands [26].

A limitation of the study is that variables were only measured at two points in time. This constitutes a better test of the buffer hypothesis than cross-sectional studies and makes it possible to draw conclusions concerning across-time changes, but not concerning possible fluctuations in the variables between measurements [39]. Future studies should aim to replicate these findings with more measurement points. Further, it is possible that employees with increasing symptoms of burnout left their jobs, making the remaining sample more resilient to the development of burnout. We examined longitudinal attrition in the sample and observed that employees who did not respond to the follow-up survey did not differ for those that did in terms of burnout scores at baseline.

Another potential bias is negative affectivity, where conditions such as depression or neuroticism may impact reporting of personal resources or distress. This can be considered a common method bias, i.e., that the instrument itself causes variations in responses that does not correspond to actual differences. While this may be true for this study, the effect is somewhat limited by a longitudinal design, as outcomes at follow-up cannot influence factors at baseline.

The generalizability of the study is strengthened by the heterogeneous sample, containing respondents from seven organizations from various occupational groups and sectors, e.g., industry, health care, white collar workers, etc., at different levels in the organizations. This variation in the sample imply that there is a variety also in what demands and resources that the respondents meet in their work, which makes it more likely that the results are generalizable to a broader population, compared to studies focusing on single occupational groups. The mean values of psychological demands and burnout in this study are also similar to samples in other studies (cf. [2, 40‐42]; cf. [43] for unstandardized measures from the same dataset as the current study). Women are over-represented in the data since the sample contains public organizations where women are in majority. It also reflects the gender distribution in sick leave due to burnout in Sweden where women are over-represented, although it has been noted that women and men with similar work conditions develop symptoms on burnout to the same extent [44]. Still, the over-representation may imply that the work conditions in female-dominated jobs dominates the results, which may limit the generalizability to male-dominated sectors. The drop-out rate at follow-up (36%) may further impact the generalizability of the findings. The dropout analysis shows that dropouts were slightly younger and had lower demands and decision latitude, while there were no significant differences for gender, burnout or social capital. This may imply that the follow-up results reflect slightly more demanding jobs with a higher amount of decision latitude, potentially limiting the generalizability to jobs with other work characteristics.