1. Introduction and Background
Despite a rise in the level of public health, work absenteeism (also referred to as sickness absence) remains very high in Norway, with a current annual cost of 120 billion NOK. This total annual cost includes: sickness benefits, disability benefits, vocational rehabilitation allowance, work assessment allowance, and rehabilitation allowance (SSB 2017
), which since 1990 has amounted to around 5% of GDP (OECD 2010
). This is high compared with the OECD average of 2–2.5% during this time period. In this study, the term work absenteeism includes only work absences due to illnesses longer than one week. Absences of this nature require a doctor’s approval. Women have a much higher rate of work absenteeism than men, and there are many theories offered to explain this phenomenon (Ose et al. 2014
). The largest diagnosis categories for both men and women are muscle/skeletal complaints and mental illness. Aside from pregnancy-related reasons, these two illness categories also have the largest gender difference (NAV 2014
). One theory for the gender difference is the ‘double burden theory’, which states that because women work full-time jobs and do the majority of care work at home; they then become fatigued/burned out and more likely to take sick leave (Kostøl 2010
). Another theory for the gender difference in work absenteeism rates is that men and women are susceptible to different illnesses, i.e., women generally have higher rates of anxiety and depression (Eaton et al. 2007
) and skeletal and muscle complaints (Gjesdal et al. 2011
). Yet another theory explains that the gender disparity in work absenteeism rates is related to gender differences in profession categories (Campos-Serna et al. 2013
Despite government policy that encourages women to enter male-dominated fields (and vice versa), the gender distribution in the Norwegian labor market follows traditional gender norms. Women make up the vast majority of what are called ‘care workers’. In this study, care workers include: nurses, teachers (up to and including middle school—‘ungdomsskole’), and elder and childcare workers. Care work is shown to be not only much more physically demanding than typical male-dominated professions (e.g., engineering), but it also has a unique psychological strain on the employees. Besides the emotional toll that care work has on the employee, the psychological strain in care work derives largely from understaffing/over-work and low employee involvement in decision-making (Mitchie and Williams 2003
; Elstad and Vabo 2008
; Magnusson Hanson et al. 2008
). In addition, care work is characterized by a high rate of part-time employees (Yerkes 2009
There are approximately 5.2 million people in Norway and circa 2.77 million people in the labor force. Unemployment in Norway is very low (see discussion), and there are circa 2.64 million employed persons in Norway (SSB 2016a
). Care work employs around 250,000 people with 86% of them women. The care work absenteeism rate is about 10%, which is high compared to the overall work absenteeism rates for both men (ca. 4.5%) and women (ca. 8%) (SSB 2016b
). Care work is a large contributor to high overall work absenteeism rates for women. In addition to this, the two highest diagnosis categories are muscle/skeletal complaints and mental illness. Care work has been cited as both very physically and psychologically demanding work (Barford and Whelton 2010
; Ose et al. 2014
Because of the large economic burden to the state, more research focusing on how women attain such a high rate of work absenteeism is vital for the economic and social sustainability of the Norwegian welfare state. In addition to this, varied methods have the potential to shed new light on the problem (Ferragina and Seeleib-Kaiser 2011
). To understand how women experience higher rates of work absenteeism, this study investigates operationally the working life of women using system dynamics modeling. To do this, the model focuses on where women are highly represented in the workplace: care professions.
This research is not a comparative study between men and women’s work absenteeism. This is a case study, where care work is taken as the representative profession for female employees because 86% of care workers are women, representing 21% of the total female labor force (Karlsen 2012
; SSB 2010
; Bakken 2009
). There are two research questions that this study addresses: (1) How does the nature of care work lead to higher rates of work absenteeism? (2) What does this mean in terms of cost to the state? This study does not attempt to build or contribute to theory as is typical in social policy studies, but instead focuses on empirically testing theories of the high female work absenteeism rate in Norway. Work absenteeism literature is rich and includes many concepts than are not covered in this study. This study investigates only work absenteeism due to sicknesses longer than one week (which require a doctor’s approval). Because of this, this study does not include motivational absenteeism, presenteeism, seasonal influenza, and work absences due to illnesses of children.
An additional aim of this study is contribute new methodological approaches to the study of absenteeism. Absenteeism is complex, and in many countries, is a persistent problem. No single research methodology can tackle it alone. Although there are many system dynamics studies on public health issues (Homer and Hirsch 2006
) and decision-making for occupational health and safety (Nikolaou 2016
), absenteeism is a new domain for system dynamics modeling. In addition, system dynamics has made limited inroads in the social sciences in general (Palmer 2017a
), and this study seeks not only to contribute to work absenteeism literature, but also to illustrate system dynamics as a method for absenteeism research.
The following section explains what system dynamics is and why it was chosen to investigate the research questions. The system dynamics model in this paper utilizes data from Statistics Norway (SSB) and the Norwegian Labour and Welfare Administration (NAV), and relationships modeled endogenously (not using data) are based on published literature. The results section explains how system dynamics was used to address these questions, and the discussion puts the results in a wider context. The Appendix A
provides further information about system dynamics modeling and its limitations, including the documentation of the model.
2. Methods: System Dynamics
2.1. Basic Elements
System dynamics is a term used when analyzing problems as a system in order to understand the system feedback (Meadows 2008
). In this case, system dynamics is used to investigate how absenteeism in care work develops over time, and how system feedback contributes to this system behavior. System dynamics involves identifying elements, subsystems, and the systems’ context, boundaries, and properties. In this way, system dynamics systematically gathers what is known about how a system operates. However, system dynamics also systemically investigate relationships within that system with the analysis of feedback in the system structure (Haskins 2008
A system dynamics model is a set of ordinary differential equations (ODEs), which are used to simulate system behavior (Sterman 2000
). These models are represented with a stock and flow diagram (SFD). A SFD contains stocks, flows, and variables, which represent either data, equations, or parameter values. Stocks are accumulations over time, where the flows add and subtract from the stock. Variables affect the flows and each other through relationships represented as ‘instantaneous causal links’. The SFD is not the model itself, but a simplified representation of the model structure used to communicate the model. A system dynamics model is developed and simulated with software, and this study uses Stella Architect 1.5.1 by isee systems (please see Appendix A
for more information on the model and model building).
Working backwards from the stock in question (in this case ‘care worker absentees’), system dynamics methods require the modeler to understand what is influencing the behavior in the stock. This process is aided by literature review and talking with experts in the field. The goal of developing the SFD is to replicate actual system behavior (called the ‘reference mode’). There are various reference points in the model in which the model is validated, and the main reference mode acts as the focal point in which to validate the model.
The model presented in this study is one part of a larger model. The SFD given in the results section is a graphical representation of this part of the model structure, the Appendix A
provides the equations that form the basis of the simulation. The remaining sectors of the larger model are part of other published works (see Palmer 2017a
2.2. Exogenous vs. Endogenous
System dynamics models have both endogenous and exogenous relationships. The relationships chosen for endogenous representation depend on the research boundary. It is also possible for a variable to be partially exogenous, where part of a variable’s value is determined by relationships in the system structure and part of its value is determined by data. In this case study, for example, cost is partially exogenous. The cost per absentee is not studied endogenously because the intent of this study is not to understand how the cost per absentee develops over time. To be represented in the model endogenously, this variable would need to be developed into several relationships involving variables such as: average absentee salary, medical cost variables, and rehabilitation cost variables. Because this is outside the boundary of the model, exogenous data is used in the model for the variable ‘cost per absentee’. What is of interest in this study is the number of female work absentees, and the relationships for calculating this are endogenous, represented with equations.
The concept of feedback is a fundamental part of system dynamics models (Lane 1999
). When evaluating endogenous relationships, very often there are relationships that feedback onto itself through the interconnection of other variables (i.e., a closed-loop system). Feedback loops can be either reinforcing (behavior continually increasing/decreasing over time) or balancing (behavior reaching an equilibrium). A causal loop diagram (CLD) is a simplified version of the feedback loops in the model and is used as the centrepiece of the discussion. It is essential to understand the feedback structures in a system when developing policy, and this is one reason that system dynamics was chosen to investigate Norwegian work absenteeism rates for women. It is important to note that CLDs, as with SFDs, are graphical representations used to communicate system dynamics models and are not the model itself (which is given in the Appendix A
2.4. Using System Dynamics to Understand Work Absenteeism
Although there are many statistical modeling studies, there is currently no study using dynamic modeling to understand female work absenteeism rates in Norway. In addition to this, in an overview of the theories explaining this phenomenon, Ose et al.
) highlights the need for more empirical evidence for these theories (Ose et al. 2014
). This study helps answer this call.
One of the major goals of system dynamics is to understand the structure of the system that is generating a particular problematic behavior. Another reason for using system dynamics as the methodology for this study is because of the dearth of literature addressing structural relationships in the work absenteeism problem in Norway. In addition to this, there is value in researching social problems with varied methodology because of the potential for new and unique insight (Esping-Andersen 2009
; Ferragina and Seeleib-Kaiser 2011
). Using dynamic modeling can serve to complement other methodological approaches in social policy research.
System dynamics does not attempt to explain ‘why’ work absenteeism rates are different, but instead it explains the operational ‘how’. The model explains how the nature of care work brings about its absenteeism rate, but it does not explain why various elements of care work make people sick. This is not meant to suggest that system dynamics modeling does not rely on scientific data that attempts to explain ‘the why’. System dynamics modeling is a valuable addition to the body of work absenteeism literature because of its ability to bring together what is known about the system from the scientific literature in order to understand the problem operationally.
Models evolve over time through many iterations in published literature. This study presents a first iteration model, meaning that the model has not been built based on a previously published model. Because of this, the model scope is narrow and relies on model assumptions (though supported by literature and tested—see Appendix A
). It should also be noted that system dynamics modeling is not only a new method in researching the female work absenteeism rate in Norway, but it has also made few inroads into the analysis of social policy in general. One of the goals of this study is to illustrate how system dynamics can be used in the social sciences.
The results of this study include the system dynamics model (illustrated as an SFD) and the simulations it produces (the behavior graphs). The starting point for the model development is the problematic reference mode behavior showing the gender difference in work absenteeism rates (Figure 1
). Figure 1
shows the total male (green) and total female (pink) work absentees divided by the number of employed males and females, respectively. These are the actual (data, not simulated) Norwegian male and female work absenteeism rates. The goal of the model is to reproduce this actual system behavior through model simulation. This means that if the model closely simulates the actual system behavior, the model structure is a supported hypothesis of how the actual system works operationally. In this study, the model reasonably reproduces the male (red) and female (blue) work absenteeism rate behavior as shown in Figure 1
. Male work absenteeism behavior is largely exogenous in the model (aside from population dynamics—which is why male absenteeism does not exactly fit the data) and shown only as a comparison to female work absenteeism. Only female work absenteeism is investigated in this study, which is modeled endogenously for care work with exogenous data used for other professions. For simplicity purposes, and noted as a limitation of the model, all care workers in the model are assumed female. The male work absenteeism simulation is for males in all other professions. The simulated behavior is generated by the system dynamics model represented by the SFD shown in Figure 2
, and this section explains what this part of the model includes and how it was developed. Further technical information about the model is provided in the Appendix A
3.1. Stock and Flow Diagram of Care Work Absenteeism
This section will explain all the parts of the system dynamics model that are represented by the SFD in Figure 2
. Please note the SFD is a graphical representation of part of the system dynamics model given in the Appendix A
and is not a comprehensive picture of the entire model structure.
There are two stocks in the model: active care workers and care worker absentees. The number of people employed in care work are in the ‘active care workers’ stock. ‘Active’ means they are employed as care workers and are not on sick leave. The sum of these two stocks is the total number of people employed in care work.
They come into the active care worker stock when they are hired, and they leave when they move onto another industry, retire, die, or are fired. When care workers are on sick leave, they leave the ‘active care workers’ stock and enter the ‘care worker absentees’ stock. The sick leave rate is 0.0052, which is the average time it takes for a worker to get sick (in years, called ‘sick rate’ in the SFD). This average sick rate is based on the male sick rate and is used as the baseline to understand how it becomes higher due to the nature of care work.
Once the care worker recovers from their illness, they return to work and are an ‘active care worker’. The average number of days for sick leave is 42. The reason this number is high is that the model is normalized for seasonal influenza, leaving shorter sick leave instances out of the average. Hiring is based on a desired number of care workers (at the start of the simulation, desired equals the number of active care workers: 250,000). The ratio of active and desired care workers shows how understaffed care facilities are. This affects the attrition rate: the more overworked the employees are, the more likely they are to find employment elsewhere.
There are two main influences in care work discussed in psychological literature leading to high rates of sick leave in care work: understaffing and low involvement in work place decision-making (Mitchie and Williams 2003
). Understaffing indicates that care workers are overworked. Being overworked leads to fatigue, and fatigue leads to higher rates of sick leave. Involvement in decision-making in the workplace affects absenteeism because of the loss of control over one’s daily working life.
The variables: ‘effect of understaffing on attrition’, ‘fatigue’, and ‘decision-making effect’ are represented as non-linear graphical functions in the model (see Appendix A
for more information). In short, this means that as understaffing increases, the effect of understaffing on attrition and fatigue increases to a certain level, where the effect stabilizes. Part-time work is in a non-linear relationship with the sick-rate, meaning that as part-time work increases (position percent decreases), the involvement in decision-making decreases. This increases the sick rate to a certain level, where the effect stabilizes.
A reduced position percent (100% is full-time; less than 100% is part-time) affects the level of employee involvement in decision-making. Women have a weakened position in the labor market and in the work place due to higher rates of part-time work in childbearing/rearing years (Zanier and Crespi 2015
). This is modeled endogenously in another part of the model and ‘female position percent’ is shown as a ghost variable where it links to this model section (dotted circle in Figure 2
). This is modeled endogenously using the total number of childcare hours needed during childrearing/bearing years and the societal norm of women taking the vast majority of the hours that are not covered by state-provided childcare services. This decreases the female position percent (more part-time work).
Care work is also affected by part-time work because of the high rate of absenteeism leading to the need for either part-time or temporary workers. This leads to a low involvement in decision-making. This (along with fatigue) decreases the amount of time it takes for an active care worker to become sick, thereby making the care worker absentee stock increase.
The double burden of women was tested with this model design and did not show an effect on care work absenteeism rates. The reason for this is that, in this model design, women work part-time in childbearing/rearing years, and in addition to this, care work has a large number of part-time workers due to the understaffing issues caused by high turnover (see Section 4.1
). The women’s double burden does not materialize into higher rates of work absenteeism even though women are taking responsibility for the majority of unpaid care work at home. This is because part-time work prevents burnout in this model design. Although this model design does not lend support to the double burden theory for work absenteeism, this does not mean that the double burden theory would not be supported with a different model design (in the Norwegian care work context or others) or with this model design in a different context. There are many different model designs that need to test the double burden theory in a variety of contexts. This model design focuses only on care work, and there needs to be further modeling/research on part-time work regardless of profession type and other psychological factors and societal norms in relation to burnout absenteeism (and other absenteeism types) related to the double burden.
3.2. Cost to the State
The second research question concerns the cost of work absenteeism to the state. The average cost per absentee per year is almost 1 million NOK. This includes not only the employee’s salary, but also medical rehabilitation and benefits for vocational rehabilitation. The loss of productivity is not included in the cost per absentee per year.
Cost per year is calculated by the average cost per absentee multiplied by the total number of male and female absentees. To understand the cost of low involvement in decision-making and understaffing on work absenteeism costs, the work absentee expenditures were calculated with these effects (referred to as ‘care work effects’) until 2017 and then were taken out of the simulation with an extended time horizon (see Figure 3
By turning off the care work effects in the model, the fatigue and part-time work reinforcing loops (see Figure 4
) are cut, and the potential cost reduction of policy addressing these effects amounts to approximately 10 billion NOK in annual cost to the state by 2025. The future cost of care work absenteeism does not take into account the future increased need for care workers as the elderly population increases (this is the next step in model development.) This is an important area of future research, as care work is identified (for better or worse) as a sector to strengthen youth employment (Montgomery et al. 2017
). The extended time horizon should not be seen as a prediction of absenteeism expenditures if care work effects are eliminated because other factors may influence total expenditures. Figure 3
is meant to highlight that eliminating care work effects could reduce total work absenteeism expenditures by approximately 10 billion NOK annually with all else being equal. Cost is discussed further in the following section.
Empirical support for theories of the gender disparity in the work absenteeism rate is lacking, especially those using dynamic modeling. Despite this, politicians need to make policy decisions addressing the problem without having much in the way of empirical evidence or a comprehensive dynamic understanding of work absenteeism. This study contributes in addressing this situation, using care work absenteeism behavior as a case study. Further research must be done to find out how other profession categories experience their work absenteeism rates, and how this influences the gender disparity and overall cost. In addition to this, central variables in this model, such as part-time work and its balancing feedback relationship with burnout itself, need to be further researched. It is also important to look outside the boundaries of the model. For example, reducing the absenteeism rate overall could increase presenteeism. Failing to investigate the systemic root of high female work absenteeism will lead to policy decisions with a high potential for failure or at worst with unintentional consequences for the labor force. More studies addressing the systemic forces leading to work absenteeism must be conducted to investigate not only gender disparity, but also the high total rate of work absenteeism in Norway.