In the single-exposure models, several of the studied factors at baseline, including the level of burnout, were of importance for the level of burnout at follow-up. In the multi-exposure model, among the occupational and personal factors, low self-efficacy and high job demands were the strongest explanatory variables. These two factors were weakly correlated; and thus appear to be independently of importance for burnout.
Changes in job demands and decision latitude scores between baseline and follow-up were associated with changes in the levels of burnout between baseline and follow-up. Increases in job demand scores was associated with an increase in burnout whereas an increase in decision latitude scores was associated with a decrease in burnout.
The content analysis of the individual interview responses, which were provided from a subgroup of the participants, identified that work demands could be classified into two major, but partly intertwined, categories: Too high workload and a sense of inadequacy. A majority of the teachers perceived that an increasing amount of administrative tasks contributed to an increased workload and had a negative impact on the available time for planning of lessons. Many teachers experienced that time pressure combined with a feeling of not being able to do a good job lead to a sense of inadequacy.
Regarding the underlying dimensions of the burnout measure, the mean values in the dimension exhaustion increased between baseline and follow-up, while the mean values in cynicism and professional efficacy did not differ significantly. The frequency of high burnout (level 2 + 3) was similar at baseline and at follow-up (14% vs. 15%, respectively). However, we observed a large fluctuation on the individual level. About one fourth of the teachers reported an increased level of burnout at follow-up, while another fourth reported a reduced level.
Additionally, among the drop-outs from our study, burnout at baseline was associated with being off duty/changing work at follow up. This may be an indicator of that teachers with high burnout were more likely than those with lower levels of burnout, to change job or find other alternatives in order to avoid unfavourable working conditions.
Strengths and weaknesses of the study
There are several strengths but also several limitations in this study that warrant attention before we reach to the conclusions. To begin with, a strength is the longitudinal study design and that we used a common measure of burnout (i.e., MBI-GS) as well as common and tested indicators for occupational, sociodemographic and life-style factors and self-efficacy. Another advantage is the use of the burnout-measure both in the standard way (i.e. calculating a mean score in the three dimensions exhaustion, cynicism and professional efficacy), and as a combined measure of the three dimension on an individual level. Since the original three dimensions were incrementally more unfavorable reported through the increasing levels of burnout, we considered the combined measure as a relevant outcome in the analysis, and a major strength of the study. Further, that we approached teachers in 50 schools across seven municipalities in the south of Sweden is also a strength of the study in that it increases the ecological validity of the study. Likewise, the individual interviews also contributes to the ecological validity of the study in that they provide additional details and insights behind the specific factors the teachers perceive as demanding.
In any event, and despite the longitudinal design, an important limitation of the study is that we only had two assessment rounds with a quite long time separation (i.e. on average 30 months). Observably, the study design is insensitive to finding, or tracking, potential changes and fluctuations that may occur between the two measurements. Further, the different scales and directions of the variables may to some extent make the interpretation of the explanatory variables (risk factors and protective factors) more difficult. Other limitations were that all data were self-reported, and that there was some statistical uncertainty with wide confidence intervals in the outcome measures.
There was only a minor difference in the frequencies of high burnout at baseline between the participants at follow-up and the total group of drop-outs. Thus, we do not believe that this overall selection of participation at follow-up have influenced the results to any major extent. However, a careful analysis in relation the various causes for non-participation suggest that a higher frequency of both the most affected and the healthiest teachers left the study, whereas the teachers with medium burnout levels at baseline remained at follow-up. For example, high burnout at baseline was more common among the group of drop-outs (n = 24) who were off duty/changed work, compared to those who participated at follow-up (37% vs. 14%). In contrast, among the teachers that were retired at follow-up (n = 43) only 4% reported high burnout at baseline. This selection may have influenced the results, but most likely towards less obvious patterns.
As shown when baseline data was stratified with respect to self-efficacy and job demands, which were the factors most strongly associated with burnout in the previously reported cross-sectional analysis , both low self-efficacy and high job demands co-occurred with a number of other factors. Thus, there is a risk of conceptual overlap between certain variables. Most pronounced, the three demand indicators (i.e. job demands, emotional demands and demands on hiding emotions) were correlated to an extent that only job demands were entered in the multivariate statistical analyses. The risk of potential confounding suggest that appropriate caution is warranted when interpreting the results.
The aim of the interviews was to complement the quantitative analysis with additional information, from many individuals, about their perception of the work environment. However, the interviews lasted for about ten minutes and cannot be considered as any in-depth interviews. Further, we did not conduct any recordings, but the interviewer took field notes. Thus, there is a need of a cautious interpretation of the results. Still, we had a large number of responders and the results gave a picture of underlying factors associated with the high job demands that many teachers experience. Large fractions of the teachers gave similar responses which made it possible to distinguish patterns and receive details of the exposure that was not captured by the questionnaire. Such information is valuable for guidance in preventive actions.
In analyses of longitudinal studies, aiming to identify causal relationships, the most common method is to select only the participants who were healthy at baseline and study the exposures in baseline in relation to the onset of disease at follow up. However, such selection not only decreases statistical power but is also inappropriate if the aim is to study fluctuating health conditions where the investigated factors may not only influence the onset but also recovery from the conditions.
In contrast to traditional analyses of to which extent the exposure at baseline is of importance for the health status at follow-up, the conclusions regarding causal relationships from analyses of changes in exposure versus changes in outcome are weaker (the changes are measured simultaneously and by the study participants themselves). Still, in the light of the fluctuation among the teachers between the levels of burnout, the analyses give some interesting information of the factors associated with a changing work situation (in this case job demands and decision latitude) and changes in the teachers’ wellbeing.
The results in relation to previous studies
The personal resources in terms of general self-efficacy turned out to be the strongest explanatory factor for burnout at follow-up, also after adjustment of the levels of burnout at baseline. Similar results have been reported earlier: Shoji et al.  found associations between job burnout and self-efficacy in a meta-analysis of studies in different occupations, Dicke et al.  detected direct effects of self-efficacy on emotional exhaustion in a longitudinal study among teachers, and Lauermann and König  found negative correlations between high self-efficacy and burnout (all three underlying dimensions in our burnout measure). Also, in the latter study  a specific teacher self-efficacy was identified, which was strongly associated with burnout. According to Schwarzer and Hallum  teacher self-efficacy is a personal resource that may protect from the experience of job strain and thus make an escalation of burnout less likely.
Although not a static concept, general self-efficacy is sometimes regarded as an inherent, or long lasting, quality that may differ among individuals. On the group level there were only minor differences in self efficacy between baseline and follow-up. However, on the individual level there was a substantial variation – in both directions - between baseline and follow-up. This probably reflects, as originally theorized by Bandura , that self-efficacy is dependent on the interplay between external and internal factors. Still, the perceived changes in general self-efficacy scores seem to be less important for determining changes in levels of burnout, compared to variations in job demands and decision latitude scores. To what extent this reflects that people in general are more likely to attribute changes as due to external conditions as opposed to attributing changes to alterations in one’s own personality or self-image is not known. One may suspect that individuals with low self-efficacy perceive higher job demands compared to those with a high self-efficacy. However, the correlation between the continuous variables job demand and self-efficacy was rather weak.
Perceived high job demands at baseline was of importance for burnout at follow up. However, in the last step when adjusting the multi-exposure model for the level of burnout at baseline, job demands was no longer a significant explanatory factor. This may be explained by the fact that there was a strong association between job demands and burnout already at baseline , and that there was no further increase of the association at follow up.
Our finding that high job demands was of importance for burnout is in line with several other studies (e.g. Aronsson et al. ). Further, increased job demands scores between baseline and follow-up was associated with an increased level of burnout. However, it was somewhat unexpected that we neither found an association between a low decision latitude and burnout in the cross-sectional study at baseline , and nor as a explanatory factor in the present follow-up study. Compared to other occupational groups such as nurses and sonographers , most of the teachers generally perceived rather high job control, and thus job control may not be the most crucial risk factor for burnout among them. However, in the analysis of changes in decision latitude versus burnout at follow-up, a decrease of decision latitude was associated with an increased level of burnout. More extensive explanations to these observations may be found in the interview-responses: many teachers perceive a continuous increase of new demands and work tasks, which may result in increased time pressure, reduced influence and less freedom to determine how the work is to be performed. Further, the teachers’ perception of not being able to do a good job and achieve their own pedagogical goals may contribute to increased burnout.
Fortunately, a major fraction of those with high burnout at baseline reported a better health at follow-up. The fluctuation between the levels of burnout indicate that for most of the teachers the level of burnout is not a static condition. Only one third of the teachers were without any burnout signs (level 0) at both baseline and follow-up and only 5% reported high burnout at both occasions. The remaining part of the study sample reported either a better or a worse level of burnout at follow up. However, in spite of the fluctuation in burnout on the individual level, at group level the burnout-status at baseline was of importance for the level of burnout at follow-up.
Beyond the observed associations with changes in job demands and decision latitude there may be other possible explanations to the changes of the teachers’ burnout, at both work and in private life, which were not captured in our study. For example, a previous study showed that imbalance between work and private life, i.e. too much work and too little free time for recovery and pleasure, predicted stress-related disorders .
Our finding of a low self-efficacy as an explanatory factor for burnout indicate that actions that strengthen both individuals and the team/collective (collective efficacy [38, 39]), may have beneficial effects for the teacher’s well-being. However, to influence an inherent quality such as self-efficacy by organisational changes or political decisions is difficult. Further, the perceived changes in general self-efficacy scores seem to be less important for determining changes in levels of burnout, compared to variations in job demands and decision latitude scores.
There should be greater opportunities of preventing actions aiming to reduce the job demands. A contributing factor to the high job demands that teachers experience may be uncertainties in responsibilities and capacity/power. Thus, there is a need of clearer goals, both at national and local level, and a distribution of responsibilities that are in line with the goals. Support from school leaders in prioritizing between tasks and in assessing when a job is done well enough, may be other measures that reduce the workload. Further, the amount of different work tasks should be reduced, e g by increased, or better use of resources together with support from administrative staff.
A decrease in decision latitude may be a consequence of the high demands: failure to handle all work tasks due to high pressure may lead to a reduced opportunity to influence how work should be done, which in turn might lead to loss of control. Thus, measures to reduce the job demands may also have an impact on the perception of control.