Content uploaded by Karen Evelyn Hauge
Author content
All content in this area was uploaded by Karen Evelyn Hauge on Oct 12, 2017
Content may be subject to copyright.
Hauge and Ulvestad IZA Journal of Labor Policy (2017) 6:11
DOI 10.1186/s40173-017-0088-y
ORIGINAL ARTICLE Open Access
Having a bad attitude? The relationship
between attitudes and sickness absence
Karen Evelyn Hauge*† and Marte Eline Ulvestad†
*Correspondence:
k.e.hauge@frisch.uio.no
†Equal contributors
The Ragnar Frisch Centre for
Economic Research, Gaustadalleen
21, 0349 Oslo, Norway
Abstract
Is sickness absence related to attitudes? Several studies point to attitudes as an
important factor for sickness absence. We study the relation between sickness absence
and attitudes towards possible reasons for sick leave, towards cheating and towards
work, by linking a survey among Norwegian healthcare workers, aimed at identifying
attitudes, to detailed data on sickness absence from the employers. We find that there
is an association between sickness absence and certain attitudes but mainly for
self-certified sick leave. Employees with more lenient attitudes towards sick leave have
more self-certified sick leave, but not more GP-certified sick leave. Furthermore, we find
no evidence of attitudes being able to explain the persistently observed differences is
absenteeism between different demographic groups.
JEL Classification: I1, I12, J01, J45
Keywords: Norway, Sickness absence, Attitudes, Absenteeism, Demographic groups,
Gender, Survey, Healthcare sector
1 Introduction
Paid sick leave and sickness benefits are central parts of the social security systems of the
European welfare states. The first preliminary outline of a European pillar of social rights
states that, “All workers, regardless of contract type, shall be ensured adequately paid sick
leave during periods of illness” (European Commission 2016a, 13). Sick pay and sickness
benefits are important as protection of an employee’s income during periods of illness or
injury. Without this financial insurance, employees that cannot afford the loss of income
might be forced to work while sick. This can further deteriorate the employee’s health
and might also have other unfortunate consequences, such as the spreading of disease
and lower firm productivity (Scheil-Adlung and Sandner 2010; Hemp 2004; Hansen and
Andersen 2008).
The entitlement to sick leave, sick pay and sickness benefits, and the duration and
replacement level of the compensations, vary considerably between the welfare states.
Nevertheless, all the EU member countries provide rights to sick leave and to sickness
benefits and most of them also to paid sick leave. Sick leave is the right to be absent
from work, while paid sick leave is the payment of (part of ) the employee’s salary by the
employer during sickness. Sickness benefits are covered by the social protection system
(Spasova et al. 2016).
© The Author(s). 2017 Open Access This article is distributed under the terms of the Creative Commons Attribution 4.0
International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and
reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the
Creative Commons license, and indicate if changes were made.
Hauge and Ulvestad IZA Journal of Labor Policy (2017) 6:11 Page 2 of 27
The OECD countries spend on average around 0.8% of GDP on sickness benefits alone
(OECD 2010, 58). In Norway, where the sickness benefit scheme is particularly generous,
around 1.6–1.7% of GDP is allocated to sickness benefits annually (Bjørnstad 2013, 22).
How these spendings can be reduced, or whether these spendings can be utilised in a
better way, is currently being debated, both in Norway and in other developed countries
(OECD 2010).
While the necessity for short-term sick leave is often based on the employee’s own
assessment, long-term sick leave is usually certified by a general practitioner. But whether
an employee actually seeks medical help is usually dependent on the employee’s own
initiative. This means that the employee’s sick leave might be influenced by own health
assessment and the employee’s opinion of when sick leave is needed. Additionally, this
implies that the sickness benefit schemes depend on trust in the employees and their
assessments.
Several studies have pointed to attitudes as an important factor for sickness absence.
For instance, Dale-Olsen and Markussen (2010), who have studied trends in absenteeism
over time for specific diagnoses, find that the number of sickness absence spells due to
specific diagnoses has not changed much over time, but that the duration of each spell has
increased by 20% in the period under study (1972–2008). The authors point to changing
demands from employers and changing attitudes to explain why it would take longer to
heal a broken leg in 2008 compared to in 1972.
Henrekson and Persson (2004) find that increases in the generosity of sickness insur-
ance benefits tend to be associated with more permanent sick leave. Also, Askildsen et al.
(2005) show that the often seen negative correlation between sickness absence and unem-
ployment rates are mainly caused by established workers changing their behaviour, rather
than by the composition of the labour force. The fact that employees seem to exercise
some flexibility in their absence behaviour opens up the possibility that attitudes might
be a part of the explanation for variation in sickness absence.
Although several studies have pointed to attitudes as an important factor for sickness
absence, few studies address this relationship directly (Allebeck and Mastekaasa 2004).
Holmås and et al (2008) study attitudes towards sickness absence in the Nordic coun-
tries, but do not study the relation between attitudes and absence behaviour. Hansen and
Andersen (2008) analyse the relationship between attitudes and sickness presence (going
to work while sick). Their measure of sickness presence, as well as sickness absence,
is, however, based on self-reported data. A weakness with using self-reported data on
absence is the possibility of misreporting, unconsciously or consciously, due to either not
recalling correctly or a desire to present oneself in a positive way.
Furthermore, most studies on sickness absence within economics rely on data con-
cerning general practitioner (GP)-certified or long-term sickness absence, as Nationwide
registers often are available for GP-certified or long-term sickness absence (where the
employee is eligible for sickness benefits). Register data is an objective and reliable data
source, but does not include self-certified sickness absence nor information about atti-
tudes. This study is the first, to our knowledge, to use actual sickness absence data as
opposed to self-reported data, for both self-certified and GP-certified sickness absence,
and to match this with data on individual attitudes. Our study therefore contributes to
the literature on sickness absence by studying the relationship between attitudes and
sickness absence empirically and by using objective, employer-registered data on both
Hauge and Ulvestad IZA Journal of Labor Policy (2017) 6:11 Page 3 of 27
self-certified and GP-certified sickness absence. Our study includes attitudes towards
possible reasons for sick leave, towards cheating and towards work.
Understanding the mechanisms behind sickness absence is important for several rea-
sons. First of all, a possible relationship between attitudes and sickness absence is
interesting in its own right. Secondly, the sickness insurance system takes for granted
that self-certified sick leave is claimed by, and sick leave certified by a GP is provided to,
people with reduced work capacity and that this measure of work capacity can be deter-
mined fairly objectively. However, if it turns out that individual attitudes towards sick
leave are related to actual sickness absence behaviour, this might indicate that the term
“reduced work capacity” is perceived differently and that it is, to some extent, a subjective
perception that determines whether an individual is entitled to sickness insurance.
Furthermore, it is important to utilise the welfare state’s resources effectively, as sickness
benefits constitute a significant share of developed countries’ spendings. An important
precondition for reducing sickness absence, and thereby its costs, is to understand the
underlying mechanisms.
In this study, we analyse whether and to what extent sickness absence is related to atti-
tudes. To investigate this matter, we have collected data from two different sources. A
survey, aimed at identifying relevant attitudes, has been conducted among employees
within public healthcare in the city of Oslo, Norway. Additionally, detailed data on sick-
ness absence for the respondents was assembled from the employers. By linking these
two data sources, we obtain information on the actual absence behaviour, together with
subjective assessments.
The public healthcare sector, which is the largest sector within the public sector in
Norway, comprises around half of all full-time equivalents in the public sector (Gran-
Henriksen 2014). Furthermore, the public healthcare sector has a particularly high rate
of sickness absence compared to other sectors in Norway. The average sick leave rate for
all sectors in Norway in 2014 was 6.4%, while the average sick leave rate within public
healthcare was 9.1% in 2014 (SSB 2015), making this an interesting sector to study. The
sections of home care, assisted living and mental healthcare were chosen because these
sections have good digital registers on personnel data, including all sickness absence of
their employees.
The main finding of the paper is that there is an association between attitudes and
sickness absences but mainly for self-certified sick leave. The only attitude variable associ-
ated with GP-certified sick leave is low work satisfaction, which is associated with higher
GP-certified sick leave.
The level of sickness absence varies between different demographic groups (Allebeck
and Mastekaasa 2004; Markussen et al. 2011). Perhaps most studied is the large difference
in sickness absence observed between men and women. Additionally, other demographic
characteristics, such as age, the level of education, the sector of work and the aver-
age number of working hours, have been found to be associated with specific attitudes
towards sickness absence (Holmås and et al 2008). The data in this paper enables us to
study whether attitudes can explain some of the observed variation in sickness absence
between demographic groups, focusing on gaps in absenteeism between men and women,
between younger and older employees, between those with lower and higher education
and between immigrants and those born in the country of residence. In general, we do not
find that attitudes are able to explain the observed differences in sickness absence between
Hauge and Ulvestad IZA Journal of Labor Policy (2017) 6:11 Page 4 of 27
demographic groups. There is, however, some indication that attitudes can explain parts
of the difference in self-certified sickness absence between employees with high and low
education.
The paper is organised as follows: Section 2 provides details about the observed vari-
ations in sickness absence between demographic groups and the Norwegian sickness
insurance system. Section 3 describes our questionnaire about attitudes towards sickness
absence, the data and our empirical strategy. In Section 4, we present the results, and
Section 5 concludes the study.
2 Background
2.1 Variations in sickness absence between demographic groups
Generally, women claim more sickness absence than men (Thorsen et al. 2015), and vari-
ous explanations have been suggested and explored in the research literature. One factor
that is found to be important is pregnancies. Myklebø (2007) find that as much as 60%
of the gender gap in sickness absence is related to pregnancies. However, pregnancies do
not seem to explain the entire gender gap.
Additional explanations for the gender gap in sickness absence which often are sug-
gested are that women hold less healthy occupations and the double burden hypothesis,
that employed women do a larger share of home- and childcare tasks than employed
men. The results of studies investigating these mechanisms, however, are not conclusive.
Mastekaasa and Dale-Olsen (2000), studying a representative population of Norwegian
employees, find that the gender gap in sickness absence is not due to women being in
less healthy jobs. A study of employees of the City of Helsinki, on the other hand, find
that “Differences between occupations held by women and men explain a substantial part
of the female excess in sickness absence” (Laaksonen and et al 2010, 394). Inconsistent
results are also the case for studies of the double burden hypothesis. Bratberg et al. (2002)
find support for the hypothesis, but Markussen and Strøm (2013) do not. Men and women
have also been found to have different coping strategies, where women who experience
being bullied at work have higher sickness absence, while men who experience the same
are more likely to leave the labour force (Eriksen et al. 2016).
Despite considerable research on why women claim more sickness absence than
men, our knowledge about the mechanisms behind the gender gap is still inadequate
(NOU 2010:13 2010).
As is the case for gender, age is generally found to be a strong predictor of sickness
absence behaviour (Allebeck and Mastekaasa 2004; Markussen et al. 2011). However, even
though sickness absence generally increases with age, younger workers tend to be absent
more frequently than older workers, who are associated with fewer, but longer, absence
spells (Thomson et al. 2000; Alavinia et al. 2009). In line with this, Markussen et al. (2011)
find that, up to the age of 45, the probability of entering into a sickness absence spell
declines sharply with age. They further argue that there might be two explanations for this
relationship: either younger employees have a lower threshold for claiming sick than older
employees or there are different norms regulating younger and older employees’ absence.
Socioeconomic status is also an important determinant of sickness absence, with a neg-
ative association between the two (Allebeck and Mastekaasa 2004; Markussen et al. 2011;
Piha and et al 2010). A part of this relationship is explained by unhealthy behaviour by
those with lower socioeconomic status, such as smoking and inactivity (Allebeck and
Hauge and Ulvestad IZA Journal of Labor Policy (2017) 6:11 Page 5 of 27
Mastekaasa 2004; Christensen et al. 2008; Thorsen et al. 2015). Markussen et al. (2011)
find that educational attainment reduces the entry into sick leave but that it is the level of
education, and not its type, that is relevant.
Holmås and et al (2008) analyse attitudes towards sickness absence in the Nordic coun-
tries. They find that women seem to be more restrictive than men in what they consider
to be acceptable reasons for calling in sick. The same finding applies to older compared
to younger employees: older employees seem to be more restrictive towards sickness
absence than younger employees. When comparing employees with high education to
those with lower education, Holmås and et al (2008) get ambiguous results, with higher
educated employees being more restrictive towards some causes of sickness absence but
less restrictive towards other causes.
There is a rising concern that increased immigration will cause a deterioration of exten-
sive welfare programmes, because more heterogeneity in the population might weaken
norms and support for such programmes (Bay et al. 2007). We study whether relevant
attitudes can explain some of the observed variation in sickness absence between dif-
ferent demographic groups, focusing on gaps in absenteeism between men and women,
between younger and older employees, between those with lower and higher education
and between immigrants and those born in the country of residence.
2.2 The Norwegian sickness insurance system
Norway has a universal sickness insurance system including all employees, where the
wage is normally 100% compensated from the first day of sickness and up to a year. The
first 16 days of an absence spell is covered by the employer. After that, the expenses are
paid by the Norwegian Labour and Welfare Service (NAV).
All employees are entitled to self-certified sick leave, meaning that the employee can
notify the employer that he or she is unable to work due to illness or injury without having
to present a medical certificate. Such self-certified sick leave can generally be used for up
to three calendar days at a time. However, medical certification is not required until the
ninth day for employees in firms participating in The Inclusive Workplace Agreement
(IWA), known as the “IA agreement” in Norway, between employers, employees and the
state. The population in this study is part of this agreement.
The employee must have been employed for at least two months to be entitled to self-
certified sick leave. Self-certified sick leave can be used four times in the course of a 12-
month period.
After the eighth calendar day of a sick leave spell, the employee must present a medical
certificate from a general practitioner (GP). The GP evaluates the work capacity of the
patient and issues the certificate if he or she finds sick leave necessary. If the employee is
able to perform some work, graded (partial) sick leave is prescribed.
3 Materials and methods
3.1 Data
The data in this study were obtained from a paper-based questionnaire and linked to per-
sonnel data provided by the city of Oslo for the same sample. The survey was conducted
in September and October 2014.
The questionnaire, a 12-page booklet, contained a cover page with the title, one page
informing about the study and a statement of content, one page about how to fill
Hauge and Ulvestad IZA Journal of Labor Policy (2017) 6:11 Page 6 of 27
out the questionnaire and ten pages consisting of 65 questions. The questionnaire was
constructed to measure attitudes and norms related to sickness absence. The complete
questionnaire can be found in the Additional file 1.
Personnel data was provided by the city of Oslo and contains information about
the employee’s contracts, position, department affiliation and shift plan. Furthermore,
there is detailed information about the employee’s sick leave: the start and end of
each absence spell, type of absence, grading of sick leave and which department the
employee was absent from. The data covers the period from 1 January 2013 to 1 Octo-
ber 2014. Descriptive statistics for all the variables used in the analysis is presented in
Table 1 .
3.1.1 Procedures
All employees within home care, assisted living and mental healthcare from two dis-
tricts, District A and District B, within the city of Oslo were invited to a meeting within
working hours. At these meetings, a representative from the research team informed the
participants about the study. The employees were given the opportunity of answering the
questionnaire during the meeting. Employees who were not on duty at the time of the
meeting received a copy of the questionnaire and an envelope with pre-paid postage. Par-
ticipants were told that the study was about sickness absence. The study was reported to
the Data Protection Officer. Participants were informed about, and consented to, the city
of Oslo providing information about them which would be matched with their answers to
the questionnaire. It was emphasised that all information would be treated confidentially
and that no individual answers would be given to the employer. Participants generally
completed the questionnaire within 20 min.
3.1.2 Sample and representativeness
In total 284 public healthcare employees answered the questionnaire. Participants who
handed in questionnaires lacking, or with errors in, the respondent’s employee ID number
were excluded from the study as they couldn’t be matched with the employee register.
On-call employees without a fixed number of working days were also excluded from the
sample. This left us with a total sample of 226 employees and a response rate of 50.4%.
The final sample consists of 32% men and 68% women between 21 and 67 years of age,
where the average age is 43 years. About 5% of the participants have primary education
only, while 52% have education at the university level. The sample is ethnically diverse
with around 27% with immigrant status. Almost 90% of the employees in our sample were
employed throughout the whole period.
Inordertobeabletosaysomethingabouttherepresentativenessofoursample,the
sample should be compared to the total population. District A provided us with neces-
sary statistics for all employees within their district for 2014. If we concentrate on the
subsample recruited from District A, our sample consists of 48.7% of the total number
of employees in this district and 51% of the full-time equivalents. If this subsample was
perfectly representative in terms of sick leave, we should expect to have around 51% of
the employees’ sick leave days in our sample. The actual number of self-certified sick
leave days in this subsample is 52%, which is just above the expected share. However, the
share of GP-certified sick leave days is somewhat low (40%). The reason might be that the
employees having a sick leave spell during the period in which the survey was conducted,
Hauge and Ulvestad IZA Journal of Labor Policy (2017) 6:11 Page 7 of 27
Table 1 Descriptive statistics
Mean (SD) Definition
Personnel data
Permanent job contract (%) 88.9
Works shifts (%) 80.1
Director (%) 8.0
Employed within home care (%) 45.1
Employed within assisted living (%) 37.2
Employed within mental healthcare (%) 15.0
Employed by District A (%) 42.0
Employed by District B (%) 58.0
Employed throughout the whole period (%) 89.4 Employed throughout
the period from January
1 2013 to October 1 2014
Self-certified sick leave rate (%) 3.2 Number of self-certified sick
leave days/ number of
contracted workdays
GP-certified sick leave rate (%) 7.0 Number of GP-certified sick
leave days/ number of
contracted workdays
Survey questions
Female (%) 68.1
Age 43.2
Living in a single household (%) 39.5
Primary education only (%) 4.9
Education from a college/university (%) 51.8
Student (%) 10.3
Another job in addition to this one (%) 14.0
Immigrant (%) 27.2 Born outside Norway to
non-Norwegian parents
Norwegian-born to immigrant parents (%) 3.13
Child custody (%) 31.0
Attitude towards sick leave due to a cold (1 = wrong–5 = OK) 3.57 (1.37)
Attitude towards sick leave due to alcohol (1 = wrong–5 = OK) 1.37 (0.88)
Attitude towards sick leave due to lack of sleep (1 = wrong–5 = OK) 1.98 (1.10)
Attitude towards sick leave due to back pain (1 = wrong–5 = OK) 2.57 (1.24)
Attitude towards sick leave due to stress (1 = wrong–5 = OK) 2.42 (1.22)
Attitude towards sick leave due to bullying (1 = wrong–5 = OK) 3.10 (1.45)
Attitude towards sick leave due to pregnancy (1 = wrong–5 = OK) 3.29 (1.31)
Norm regarding sick leave due to a cold (1 = wrong–5 = OK) 3.58 (1.18)
Norm regarding sick leave due to alcohol (1 = wrong–5 = OK) 1.89 (1.13)
Norm regarding sick leave due to lack of sleep (1 = wrong–5 = OK) 2.36 (1.08)
Norm regarding sick leave due to back pain (1 = wrong–5 = OK) 3.08 (1.18)
Norm regarding sick leave due to stress (1 = wrong–5 = OK) 2.94 (1.21)
Norm regarding sick leave due to bullying (1 = wrong–5 = OK) 3.22 (1.32)
Norm regarding sick leave due to pregnancy (1 = wrong–5 = OK) 3.49 (1.16)
Work-family strain (1 = low–5 = high) 2.60 (1.31)
Family-work strain (1 = low–5 = high) 1.38 (0.74)
Stressed at work (1 = disagree–5 = agree) 3.12 (1.30)
Trouble unwinding (1 = disagree–5 = agree) 2.20 (1.26)
Enjoyable work tasks (1 = disagree–5 = agree) 4.19 (0.88)
Meaningful job (1 = disagree–5 = agree) 4.41 (0.90)
Inspiring job (1 = disagree–5 = agree) 2.97 (1.22)
Extrinsic motivation (1 = disagree–5 = agree) 2.64 (1.33)
Content with management (1 = disagree–5 = agree) 3.48 (1.30)
Enjoy colleagues (1 = disagree–5 = agree) 4.43 (0.70)
Not enough people at work (1 = disagree–5 = agree) 3.58 (1.27)
Having a job is important (1 = disagree–5 = agree) 4.28 (0.88)
Gender equal responsibility (1 = disagree–5 = agree) 4.60 (0.84)
Work is just for money (1 = disagree–5 = agree) 2.36 (1.30)
Homemaking is just as fulfilling (1 = disagree–5 = agree) 2.37 (1.29)
Work is a duty (1 = disagree–5 = agree) 4.15 (1.00)
Working mother also close with her children (1 = disagree–5 = agree) 4.19 (1.04)
People who do not work become lazy (1 = disagree–5 = agree) 3.51 (1.31)
Like taking chances (1 = not at all–5 = a lot) 3.04 (1.15)
Biking without helmet (1 = never–5 = always) 2.84 (1.57)
Like gambling (1 = not at all–5 = a lot) 2.14 (1.19)
Like competing (1 = not at all–5 = a lot) 3.04 (1.28)
Anxious (1 = never–5 = Always) 3.21 (1.04)
Receiving benefits, not looking for a job (1 = wrong–5 = OK) 1.37 (0.77)
Cheating on taxes (1 = wrong–5 = OK) 1.28 (0.67)
Black market services (1 = wrong–5 = OK) 1.73 (0.98)
Avoiding public transport fare (1 = wrong–5 = OK) 1.63 (1.00)
Too much tax (1 = disagree–5 = agree) 3.013 (1.44)
Many misuse social security benefits (1 = disagree–5 = agree) 3.57 (1.21)
Wage deduction of sick (1 = disagree–5 = agree) 1.69 (1.21)
Reduce living standard for environment (1 = disagree–5 = agree) 3.13 (1.34)
N226
Hauge and Ulvestad IZA Journal of Labor Policy (2017) 6:11 Page 8 of 27
were harder to reach. Nevertheless, the questionnaires were sent home to those on longer
sick leave spells, giving them a chance to participate.
3.1.3 Survey questions
The questionnaire included 45 questions attempting to measure attitudes, norms and
preferences of potential importance to sickness absence. The entire questionnaire is
provided in the Additional file 1.
The attitudes towards different justifications for being absent from work were measured
with questions such as “What is your opinion regarding calling in sick (self-certified sick-
ness absence) if one has not slept enough during the previous nights?” Answers were given
on a 5-point scale where 1 represented “Wrong” and 5 represented “OK”. Answers with
higher values thus represent more lenient attitudes towards sickness absence. The ques-
tions about attitudes towards different justification for being absent from work (question
1–7 in Table 2) were inspired by similar questions in Holmås and et al (2008) and Hansen
and Andersen (2008).
The questionnaire included claims regarding more general attitudes towards work and
work life, attitudes towards cheating or free-riding, attitudes towards gender equality
and gender roles and attitudes towards social welfare benefits. The questions about work
(questions 26–29) were inspired by the Norwegian Value Study (Holth 2010), and ques-
tion 27 was also inspired by Hansen and Andersen (2008). The questions about cheating
(questions 33–36) as well as gender roles (questions 30–32) were inspired by questions
in the Norwegian Value Study (Holth 2010). The questions about social welfare benefits
(questions 37–40) were our own.
Furthermore, the questionnaire included questions related to the social norms regard-
ing sickness absence in the workplace (questions 8–14), measured by the respondents’
beliefs concerning what attitudes most of their colleagues hold towards different justifi-
cations for being absent from work. These questions were our own.
Our questionnaire further included questions about stress and motivation. The ques-
tions about stress were inspired by the Norwegian Time Use Study (Rønning 2002) and
Thomas and Ganster (1995) (questions 15 and 16), and Hansen and Andersen (2008)
(questions 17 and 18). The questions about motivation included questions about intrin-
sic motivation (questions 19–21), taken from the Norwegian formulation used by Kuvaas
and Dysvik (2009), and extrinsic motivation (question 22) used by Kuvaas and Dysvik
(2011). In addition, there were questions about being content with one’s work and man-
agers (question 23), and whether the respondents enjoyed being around their colleagues
(question 24), both inspired by Hansen and Andersen (2008).
Finally, the questionnaire included questions related to the employees’ preferences and
personality, as these aspects might be related to the propensity to report in sick. The three
questions about risk preferences (questions 41–44) were inspired by the DOSPERT scale
(Blais 2006), while the questions about competitive appetite (question 44) and anxiety
(question 45) were our own. Table 1 presents the descriptive statistics for all variables.
3.2 Variables
3.2.1 Sickness absence
Our measures of sickness absence, self-certified sick leave and GP-certified sick leave
are the rates of sick leave in percent of contracted working days. The rates are found
Hauge and Ulvestad IZA Journal of Labor Policy (2017) 6:11 Page 9 of 27
Table 2 Construction of variables
Variable Question number Questions
What is your opinion on calling in sick (self-certified
sickness absence)...
1 ...if one has a cold and a low-grade fever?
2 ...if one feels bad after drinking alcohol the
evening before?
1: Lenient attitudes towardssick leave 3 ...if one has not slept enough during the
previous nights?
4 ...if one has back pain when waking up in the morning?
5 ...if one feels unwell due to stress at work?
6 ...if one is being bullied at work?
7 ...if one is pregnant and feeling nauseous?
What do you think is the opinion of most others at your
work place
(your department) on calling in sick (self-certified
sickness absence)...
8 ...if one has a cold and a low-grade fever?
9 ...if one feels bad after drinking alcohol the
evening before?
2: Lenient norms regarding sick leave 10 ...if one has not slept enough during the previous
nights?
11 ...if one has a sore back/back pain when waking up in
the morning?
12 ...if one feels unwell due to stress at work?
13 ...if one is being bullied at work?
14 ...if one is pregnant and feeling nauseous?
Do you agree or disagree with the following statements?
15 “I work so much at my job that I don’t have enough
time for everything that needs to be done at home.”
3: Being stressed 16 “I have so much to do at home that I don’t have
enough time for everything that needs to be done
at work.”
17 “I often feel stressed at work.”
18 “ I do not manage to unwind/relax when I a m off w ork .”
Do you agree or disagree with the following statements?
19 “The tasks that I do at work are enjoyable.”a
4: Low motivation 20 “My job is meaningful to me.”a
21 “Sometimes I become so inspired by my job that I
almost forget everything else around me.”a
22 “If I am supposed to put in extra effort in my job, I need
to receive extra pay.”
Do you agree or disagree with the following statements?
23 “I am content with the management at
my work place.”a
5: Low work satisfaction 24 “I enjoy being together with my colleagues.”a
25 “We are not enough people at work.”
Do you agree or disagree with the following statements?
26 “You need to have a job to have a good life.”a
6: Low work ethic 27 “Work is just a way of earning money.”
28 “Work is a duty towards society.”a
29 “People who do not work become lazy.”a
Hauge and Ulvestad IZA Journal of Labor Policy (2017) 6:11 Page 10 of 27
Table 2 Construction of variables (Continued)
Do you agree or disagree with the following statements?
30 “Men should take as much responsibility as women for their
house and children.”
7: Gender equal attitudes 31 “Being a homemaker is just as fulfilling as working for pay.”
32 “A working mother can have an equally close and good
relationship with her children as a mother who does not work.”
What is your opinion on the actions below?
33 “Not looking for a job when receiving unemployment benefits.”
8: Lenient towards cheating 34 “Cheating on taxes.”
35 “Buying services on the black market.”
36 “Avoiding a fare on public transport.”
Do you agree or disagree with the following statements?
37 “We pay too much tax.”
9: Negative towardswelfare state 38 “Too many are misusing our social security benefits.”
39 “Sickness absence should give a deduction from wage.”
40 “I am willing to reduce my standard of living in order to save the environment .”a
41 Do you like taking chances (risks)?a
42 Do you bike without a helmet?a
10: Risk-averse preferences 43 Do you like gambling?a
44 Do you like to compete?a
45 Do you worry?
aThe Likert scale has been reversed
by dividing an employee’s number of sick leave days in the period from 1 January 2013
to 1 October 2014, adjusted for possible grading of sick leave, by the total number of
contracted working days for the employee in the same time period.
3.2.2 Attitude variables
From the questionnaire, we use 45 questions related to attitudes, norms and pref-
erences. The answers to each of these questions have been standardised, so that
the mean of each question is zero and the standard deviation is one. In order to
reducethenumberofvariablesintheregressionanalysis,wehave,basedonour
own judgement, created attitude variables constructed of the sum of several question
values. This results in ten attitude variables, which we have called Lenient attitudes
towards sick leave,Lenient norms towards sick leave,Being stressed,Low motivation,
Low work satisfaction,Low work ethic,Gender equal attitudes,Lenient towards cheat-
ing,Negative towards welfare state and Risk-averse preferences. Table 2 illustrates how
the attitude variables have been constructed and which questions the variables are
comprised of.
To make sure that the way we construct the attitude variables does not affect our results,
a robustness test is presented in the Appendix where a factor analysis is used to group
questions into variables. In this alternative procedure, we use an exploratory factor anal-
ysis to simplify the data set based on the correlations between the variables. The factor
analysis results in five factors: Lenient sickness attitudes,Low intrinsic motivation,Lenient
towards cheating,Being stressed and High homeload. For details about the factor analysis,
and the results from it, see the Appendix.
Hauge and Ulvestad IZA Journal of Labor Policy (2017) 6:11 Page 11 of 27
3.2.3 Control variables
The control variables include gender, immigrant status, a linear variable for education
and age, a quadratic term for age and whether individual ilives in a single household,
has child custody, has more than one employer, is currently studying, is a director, holds
a permanent position or works shift work. We also control for whether the employee
works in assisted living or mental healthcare (compared to home care), and we control for
district.
3.3 The model
We use regression model (1) to analyse the relation between attitudes and sickness
absence, where Yiis the sick leave rate for individual i,Ail is the attitude of individual i
forattitudevariablel=1,...,10, Xik is the control variables kof individual iK is the total
number of control variables and εiis the error term
Yi=α0+
10
l=1
βlAil +
K
k=1
γkXik +εi(1)
The coefficients of interest are βlfor each of the attitudes l=1,...,10. These coefficients
will show the relationship between each attitude variable and sick leave. For instance,
the interpretation of βlwhere lfor instance is Lenient attitudes towards sick leave is that
increasing the leniency towards sick leave with one standard deviation is associated with
an increase in the sick leave rate of the value of βl.
The employees have contracts of various lengths. The information we have about sick
leave is more accurate for those individuals that we observe over a longer period than
for those we only observe for a week or two. We therefore include frequency weights in
the regressions, so that the employees with longer contracts count more than those with
shorter contracts. Mechanically, this means that the total number of observations, if we
take the frequency weights into account, is 104,229. We use cluster robust standard errors
to count for the serial correlation that occurs due to the frequency weights.
In order to investigate whether attitudes can explain some of the generally observed
variation in sickness absence between different demographic groups, we use regression
models (2)–(4). Models (2), (3) and (4) are essentially the same models as Model (1).
However, we are now interested in contrasting the difference in sick leave between two
dichotomous groups (men vs female, old vs young, high vs low educated and immigrants
vs non-immigrants) and next see whether this difference can be explained by attitudes,
that is whether significant differences in sickness absence vanishes or is reduced when the
attitude variables are included in the model. If, for instance, the gender gap in sickness
absence can be explained by women and men have different attitudes, then we should
expect the coefficient on Female to be reduced in size and significance when attitudes
are included in the model. By first excluding and then including the attitude variables in
the regression, we can see how the coefficient on the various characteristics of interest
changes. Because attitudes might affect self-certified sick leave and sick leave certified by
a GP differently, we run the regression separately for the two outcomes.
In Model (2) we now want to focus on two demographic characteristics: Female and
Immigrant. Whereas these two variables were included among the control variables in
Hauge and Ulvestad IZA Journal of Labor Policy (2017) 6:11 Page 12 of 27
Model (1) (in the vector Xik ), they have been moved out in separate terms in Model (2).
Yi=α0+α1Femalei+α2Immigranti+
10
l=1
βlAil +
K
k=3
γkXik +εi(2)
where Yiis the sick leave rate for individual i,Femaleiand Immigrantiare dummy vari-
ables, Ail is the attitude of individual ifor attitude land Xik is the remaining control
variables kof individual i.
Yi=α0+α3Low_educationi+
10
l=1
βlAil +
K
k=2
γkXik +εi(3)
In Model (3), we want to contrast employees with high and low education. While edu-
cation is a linear variable with four educational levels in Models (1)–(2), in Model (3),
Low education is a dummy variable taking the value of 1 for employees with no more than
mandatory education and 0 otherwise.
Yi=α0+α4Oldi+
10
l=1
βlAil +
K
k=2
γkXik +εi(4)
In Model (4), we want to contrast young and old employees. While age was a continuous
variable in Models (1)–(3), Model (4) includes the dummy variable Old, which takes the
value 1 if the employee is in the oldest half of the sample and 0 otherwise. The continuous
variable Age is excluded from this model. In this model, we use the categorical variable
for education (as in Models (1) and (2)), and not the dummy variable Low education as in
Model (3).
4 Results
In this section, we first present the results from Model (1) regarding the relationship
between attitudes and sickness absence. Then we go on to present the results of Models
(2), (3) and (4) on whether attitudes can explain the observed variation in sickness absence
between different demographic groups.
As can be seen from Table 1, the self-certified sick leave rate is 3.2% on average, while
the GP-certified sick leave rate is 7%. The correlation coefficient between GP-certified
sick leave and self-certified sick leave is, perhaps surprisingly, only 0.055. The low corre-
lation between the two sick leave variables is not due to there only being a few employees
with GP-certified sick leave. Out of the 226 employees, there are 136 employees (60%)
who have GP-certified sick leave in the period we analyse, 168 employees (74%) with self-
certified sick leave and 116 employees (51%) with both self- and GP-certified sick leave.
We do not find that the employees with a high level of GP-certified sick leave have a
high level of self-certified sick leave or the other way around. It is therefore interesting to
analyse the two sick leave variables separately.
4.1 The relationship between attitudes and sickness absence
The relationship between the attitude variables and sick leave (the results of Model (1))
is illustrated in Fig. 1. The figure shows the point estimates and the corresponding 90%
confidence intervals from regressions of self-certified sickness absence and GP-certified
sickness absence on the attitude variables, when including all the control variables. If
Hauge and Ulvestad IZA Journal of Labor Policy (2017) 6:11 Page 13 of 27
Lenient attitudes towards sick leave
Lenient norms regarding sick leave
Being stressed
Low motivation
Low work satisfaction
Low work ethic
Gender equal attitudes
Lenient towards cheating
Negative towards welfare state
Risk−averse preferences
−.5 0 .5 1 −2 0 2 4
Self−certified GP certified
Fig. 1 Point estimates and 90% confidence intervals for the attitude variables
the confidence interval overlaps the vertical line from zero, the attitude variable is not
significantly different from zero.
Figure 1 reveals that there are four attitude variables with a significant relation to self-
certified sick leave: Lenient attitudes towards sick leave,Low work ethics,Gender equal
attitudes and Lenient towards cheating. As expected, more lenient attitudes towards
sickness absence and lower work ethics are associated with more self-certified sick
leave, where a one-standard-deviation increase in Lenient attitudes towards sick leave
is associated with a 0.4 percentage point increase in the self-certified sick leave rate. A
one-standard-deviation increase in Low work ethics isassociatedwitha0.34percentage
point increase in the self-certified sick leave rate. Surprisingly, being more lenient towards
cheating is associated with lower self-certified sick leave. An increase in Lenient towards
cheating of one standard deviation is associated with a reduction in the self-certified
sick leave rate of 0.27 percentage points. Likewise, a one-standard-deviation increase in
reported gender equal attitudes is associated with a reduction in the self-certified sick
leave rate of 0.25 percentage points.
When considering GP-certified sickness absence, only Low work satisfaction is signifi-
cantly different from zero. Lower work satisfaction goes together with more GP-certified
sick leave. Increasing Low work satisfaction with one standard deviation (which implies a
reduction in work satisfaction) is associated with an increase in the GP-certified sick leave
rate of 2.1 percentage points.
Only low work satisfaction is associated with GP-certified sickness absence. Addition-
ally, looking at the R-squared in the upper panel of Table 5 reveals that the highest
R-squared among the four specifications is for the model of self-certified sick leave where
the attitude variables are included (specification (2)). Including the attitude variables in
the model of self-certified sick leave gives a larger percentage increase in R-square (a 36%
increase, from 0.157 to 0.214) relative to including the attitude variables in the model for
GP-certified sick leave (a 32% increase, from 0.133 to 0.176).
Hauge and Ulvestad IZA Journal of Labor Policy (2017) 6:11 Page 14 of 27
Taken together, we have seen that lenient attitudes towards sickness absence are related
to self-certified sick leave, but not GP-certified sick leave. This is not surprising as the
reasons for sickness absence which this variable is based on are most relevant for short-
term sickness absence (for instance, being absent from work due to a cold or due to having
back pain when waking up in the morning). The results are supported by the robustness
test presented in the Appendix, where all the analyses are repeated but where the variables
are constructed using a factor analysis (for details about the factor analysis and the results
of the robustness test analyses, see the Appendix).
4.2 Can attitudes explain the observed variation in sickness absence between
demographic groups?
The results above suggest that attitudes might be able to explain some of the variation in
sickness absence, in particular the variation in self-certified sick leave. Previous research
has documented that the level of sickness absence varies substantially between different
demographic groups. Specifically, the large difference in sickness absence between men
and women has attracted a lot of attention. However, research has so far not managed to
fully explain such differences in sickness absence. If attitudes should be able to explain
parts of the sick leave gaps between men and women, old and young, employees with low
and high levels of education and between immigrants and non-immigrants, there should
also be attitude differences between these groups. And we find that there are indeed dif-
ferences in our attitude variables between the groups of interest. Table 3 compares the
reported attitudes for the four demographic groups.
We find no significant differences between men and women in their leniency towards
sick leave. However, men in our sample score higher on low motivation and on low work
ethic, implying that men have lower motivation and work ethic than women. Men also
report lower gender equal attitudes and more lenient attitudes towards cheating than
women. Furthermore, Table 4 shows that there are sickness absence differences across the
demographic groups also in our sample, where women have on average more GP-certified
sick leave than men.
To investigate whether attitudes can explain the difference in sick leave between men
and women, we regress Model (2). The results from this regression is presented in the
upper panel of Table 5, where self-certified sick leave is presented in column (1) and GP-
certified sick leave in column (3). Secondly, we include our attitude variables in columns
(2) and (4). If attitudes can explain some of the differences in sickness absence between
men and women, the size of the coefficient on each of the dummy variables should be
reduced when we control for attitudes related to sickness absence. We see from column
(1) that there is no difference in self-certified sick leave between men and women. From
column (2), we see that when we include attitudes in the model, the difference between
men and women is still insignificant.
For GP-certified sick leave, there is a significant gap in sickness absence between men
and women in our sample. When we further include attitudes in our regression, in col-
umn (4), we see that the coefficient on Female does not decrease. Actually the coefficient
increases a little. This means that we find no evidence of attitudes being able to explain
the gender gap in sickness absence.
Regarding immigrants compared to non-immigrants, we can see from Table 3 that
immigrants in our sample are more stressed and report lower gender equal attitudes. We
Hauge and Ulvestad IZA Journal of Labor Policy (2017) 6:11 Page 15 of 27
Table 3 Differences in attitudes
Gender Age Education Immigration
Men Women Diff. Young Old Diff. Low High Diff. Non-immigrant Immigrant Diff.
Lenient attitudes towards sick leave 0.065 −0.030 1.07 0.062 −0.056 1.43 0.084 −0.080 2.00** −0.040 0.109 −1.61
Lenient norms regarding sick leave 0.039 −0.018 0.61 −0.014 0.013 −0.31 0.060 −0.057 1.34 0.019 −0.052 0.72
Being stressed −0.007 0.003 −0.10 0.019 −0.017 0.39 0.115 −0.109 2.42** −0.062 0.168 −2.19**
Low motivation 0.180 −0.084 2.76*** −0.002 0.002 −0.05 0.020 −0.019 0.44 0.010 −0.027 0.37
Low work satisfaction 0.054 −0.025 0.86 −0.041 0.037 −0.90 0.007 −0.007 0.16 0.037 −0.101 1.43
Low work ethic 0.141 −0.066 2.61*** −0.036 0.032 −0.92 −0.055 0.052 −1.44 0.036 −0.098 1.60
Gender equal attitudes −0.215 0.101 −3.68*** −0.043 0.038 −0.98 −0.085 0.081 −2.03** 0.068 −0.185 2.78***
Lenient towards cheating 0.174 −0.081 2.62*** 0.139 −0.125 2.90*** −0.010 0.010 −0.22 0.041 −0.112 1.48
Negative towards welfare state −0.035 0.016 −0.61 −0.016 0.014 −0.39 0.087 −0.083 2.21** −0.042 0.115 −1.82*
Risk-averse preferences −0.097 0.045 −2.09** −0.048 0.043 −1.44 0.001 −0.001 0.03 −0.002 0.006 −0.11
N226 226 226 226
*p<0.10, **p<0.05, ***p<0.01
Hauge and Ulvestad IZA Journal of Labor Policy (2017) 6:11 Page 16 of 27
Table 4 Sickness absence among the groups of interest
Men Women Young Old Low High Non- Immigrant
education education immigrant
Mean Mean Mean Mean Mean Mean Mean Mean
Sick leave rate (%) 6.42 8.18 6.37 8.27 8.86 6.16 7.30 8.26
Self-certified sick leave rate (%) 1.55 1.24 1.43 1.29 1.57 1.09 1.34 1.36
GP-certified sick leave rate (%) 4.90 6.95 4.97 6.97 7.30 5.09 5.97 6.90
N72 154 107 119 110 116 165 61
do not find any significant difference in sickness absence between immigrants and non-
immigrants in our sample. Furthermore, including the attitude variables does not change
this fact. This result also holds if we separate the immigrants into continent of origin
(results not shown).
When it comes to employees with different levels of education, and their attitudes, we
find that employees with low education are more lenient towards sick leave and are more
stressed than those with higher education. Those with higher education, on the other
hand, report somewhat more gender equal attitudes and report being more positive to
the welfare state. To investigate whether attitudes can explain the variation in sick leave
between employees with high and low education, we regress Model (3). The results are
presented in the middle panel of Table 5. Employees with low education have significantly
more self-certified sick leave than employees with higher education, while there is no
significant difference in GP-certified sick leave. When we control for attitudes in column
(2), the coefficient on Low education reduces slightly. This might reflect that attitudes are
part of the explanation of why employees with lower education have higher self-certified
sickness absence than highly educated employees.
Comparing young and old employees, we see that younger employees in our sample are
more lenient towards cheating. Regarding the relation between age and sick leave, it is not
clear what to expect. Older workers are associated with fewer but longer spells of absence,
while younger workers tend to be absent more frequently (Thomson et al. 2000; Alavinia
et al. 2009). To study whether attitudes can explain the differences in sick leave between
old and young employees, we regress Model (4). The results are presented in the bottom
panel of Table 5. Table 5 shows that while the oldest half of our sample has a significantly
higher GP-certified sick leave rate, there is no statistically significant difference in the self-
certified sick leave of young and old. Comparing the coefficient on Old in columns (3) and
(4), we get the same effect as we did for Female: The coefficient on Old for GP-certified
sick leave increases when controlling for attitudes. Despite having stricter attitudes, the
older employees have more GP-certified sick leave than the younger. This might indicate
that older employees would have had an even higher GP-certified sick leave rate, relative
to younger employees, if attitudes were equalised across the two groups.
To explore the difference in self-certified sick leave between employees with high
and low education further, we have performed a Blinder-Oaxaca decomposition. The
Blinder-Oaxaca decomposition decomposes differences between groups to differences in
endowments and coefficients. The results are presented in Table 6. Columns (1) and (2)
show the results for self-certified sick leave. In column (1), all the control variables are
used as explanatory variables, while in column (2), the attitude variables are included
in addition as explanatory variables. The difference in self-certified sick leave between
Hauge and Ulvestad IZA Journal of Labor Policy (2017) 6:11 Page 17 of 27
Table 5 Regression results
(1) (2) (3) (4)
Self-certified Self-certified GP certified GP certified
coef. coef. coef. coef.
(se) (se) (se) (se)
Female −0.311 −0.142 2.330* 2.460*
(0.19) (.018) (1.28) (1.48)
Immigrant 0.001 −0.096 0.736 1.087
(0.20) (0.21) (1.25) (1.30)
Lenient attitudes towards sick leave 0.406** 0.657
(0.16) (1.10)
Lenient norms regarding sick leave −0.045 −0.030
(0.12) (0.99)
Being stressed −0.082 1.037
(0.13) (0.88)
Low motivation 0.178 −1.117
(0.14) (0.91)
Low work satisfaction 0.084 2.127*
(0.12) (1.18)
Low work ethic 0.339** 0.609
(0.14) (1.16)
Gender equal attitudes −0.253* 0.064
(0.14) (1.16)
Lenient towards cheating −0.267** 1.854
(0.13) (1.63)
Negative towards welfare state 0.223 0.760
(0.16) (1.17)
Risk-averse preferences −0.230 1.028
(0.18) (1.34)
_cons 5.402*** 5.825*** 18.001* 22.818**
ControlsaYes Yes Yes Yes
Attitudes No Yes No Yes
R-squared 0.157 0.214 0.133 0.176
Low education 0.703*** 0.674*** 1.211 0.649
(0.22) (0.21) (1.20) (1.22)
ControlsbYes Yes Yes Yes
Attitudes No Yes No Yes
R-squared 0.154 0.213 0.135 0.176
Old −0.075 −0.032 2.617* 2.776*
(0.22) (0.23) (1.37) (1.60)
ControlscYes Yes Yes Yes
Attitudes No Yes No Yes
R-squared 0.141 0.194 0.107 0.140
Nd226 226 226 226
*p<0.10, **p<0.05, ***p<0.01
aage, age2, education, na_educ, single_hh, na_singlehh, child_cust, na_custody, immig_sec, na_immig, student, secondjob,
director, shifts, district_a, ass_living, mental_health
bfemale, age, age2, na_educ, single_hh, na_singlehh, child_cust, na_custody, immig_sec, immigrant, na_immig, student,
secondjob, director, shifts, district_a, ass_living, mental_health
cfemale, education, na_educ, single_hh, na_singlehh, child_cust, na_custody, immig_sec, immigrant, na_immig, student,
secondjob, director, shifts, district_a, ass_living, mental_health
dWe use frequency weights in the regressions due to the fact that we observe the employees for varying amounts of time,
putting more weight on the employees that we have more information about. The frequency weights duplicate the observations
according to the employees’ number of working days. This means that the total number of observations, if we take the frequency
weights into account, is 104,229. We use cluster robust standard errors to count for the serial correlation that occurs due to the
frequency weights
Hauge and Ulvestad IZA Journal of Labor Policy (2017) 6:11 Page 18 of 27
Table 6 Results from Blinder-Oaxaco decomposition of difference in self-certified sickleave between
employers with high and low education
(1) (2)
Self-certified Self-certified
coef coef
(se) (se)
Overall
High education 1.089*** 1.089***
(0.12) (0.12)
Low education 1.571*** 1.571***
(0.13) (0.14)
Difference −0.483*** −0.483***
(0.18) (0.18)
Endowments 0.425* 0.374
(0.23) (0.24)
Coefficients −0.464** −0.327
(0.22) (0.21)
Interaction −0.443* −0.530**
(0.26) (0.27)
Endowments
Female 0.011 0.001
(0.02) (0.01)
Age 2.006* 2.932**
(1.09) (1.22)
Age2−1.656* −2.493**
(0.98) (1.11)
Immigrant 0.025 0.028
(0.04) (0.04)
Coefficients
Female −0.124 −0.191
(0.28) (0.28)
Age 15.085 19.152**
(9.38) (8.86)
Age2−7.604 −9.542*
(5.29) (4.95)
Immigrant 0.150 0.060
(0.15) (0.15)
_cons −7.466* −9.284**
(4.14) (3.92)
Interaction
Female 0.008 0.013
(0.02) (0.03)
Age −1.758 −2.232*
(1.19) (1.19)
Age21.442 1.809*
(1.08) (1.07)
Immigrant −0.051 −0.020
(0.06) (0.05)
ControlsaYes Yes
Attitudes No Yes
N226 226
*p<0.10, **p<0.05, ***p<0.01
aControl variables: single_hh, child_cust, immig_sec, student, secondjob, director, shifts, district_a, ass_living, mental_health
Hauge and Ulvestad IZA Journal of Labor Policy (2017) 6:11 Page 19 of 27
employees with high and low education which we want to decompose is −0.483, mean-
ing that the low educated have higher sick leave than the high educated (1.089 −1.571 =
−0.483). The reference group for the decomposition in Table 6 is employees with low
education. The difference is decomposed into an explained part, which is due to differ-
ences in endowments or explanatory variables (labelled “endowments” in the table), and
an unexplained part, which is due to different behavioural effects or coefficients between
the two groups (labelled “coefficients” in the table) and an interaction effect between
the endowment and coefficient effects. If low educated employees had the same endow-
ments, that is if they had the same composition of background conditions (gender, age,
type of jobs, etc.) as the high educated, the model estimates that they would have had
even higher self-certified sick leave. At first sight, this might seem a bit surprising. In our
sample, the low educated on average are older than the high educated. This might reflect
that in for instance home care, the requirements for education has increased over time.
Also, the highest self-certified sick leave in our sample is among the young and low edu-
cated. The endowment effect can be interpreted as how much self-certified sick leave the
low educated would have if the group had the same age profile (and other characteristics)
as the high educated, and the coefficient (or slope) of the low educated group. The coef-
ficient effect, on the other hand, is the difference in self-certified sick leave estimated if
the low educated had their own characteristics but the coefficients of the high educated.
The model estimates that if the low educated had their own characteristics, but the coef-
ficients of the high educated, their self-certified sick leave would have been significantly
lower. The interaction effect is also negative, upweighing the increase in sick leave which
was predicted by the endowment effect.
If attitudes are important for self-certified sick leave, this would be captured in the
coefficient effect. From this Blinder-Oaxaca decomposition in specification (1), both the
endowment effect and the coefficient effect are significant and the coefficient effect
slightly larger. The results from the Blinder-Oaxaca decomposition in specification (1)
thus supports our result that attitudes are relevant for explaining the difference in self-
certified sick leave between those with high and low education. From specification (2),
we see that when including the attitude variables in the Blinder-Oaxaca decomposition,
the endowment effect and coefficient effect no longer are significant. When including
the attitude variables in the decomposition analysis, our measures of attitudes are moved
from the coefficients into the endowment effect. While we know from specification (1)
that the model predicts that self-certified sick leave would go up if the low educated had
the characteristics of the high educated, it seems as though having the attitudes of the
high educated would pull the sick leave of the low educated in the opposite direction,
resulting in a non-significant effect.
To sum up, we find some evidence of attitudes being related to the higher self-
certified absence rate for employees with low education, relative to those with higher
education. However, attitudes explain far from all of the differences in the self-certified
sick leave rate between these two groups, and most of the differences in our sample
remains unexplained. When considering gender and age, there are significant differ-
ences in GP-certified sick leave and there are significant differences in various attitudes
between the groups. However, if we look more closely at these attitudes, we see that
they are not the same variables as the ones we found actually having an association with
GP-certified sick leave in Section 4.1. Therefore, it is perhaps not surprising that the
Hauge and Ulvestad IZA Journal of Labor Policy (2017) 6:11 Page 20 of 27
attitude variables are not able to explain the gender and age differences in GP-certified
sick leave.
5Conclusions
In the present study, we examine the relation between attitudes and sickness absence. To
study this, a survey aimed at identifying attitudes towards sick leave, work and cheating
was conducted among employees within the public healthcare in Norway. The answers
from the questionnaire were linked to detailed data on sickness absence, both self-
certified and GP-certified, gathered by the employer. The study is therefore based on
a rather unique data set, as it combines objectively measured data on sick leave with
self-reported attitudes, and includes data on both GP-certified and self-certified sick
leave.
In general, one might expect employee attitudes to relate to self-certified sick leave to a
greater extent than GP-certified sick leave, as self-certified sick leave is determined by the
judgement of the employee alone. Nevertheless, whether an employee visits a doctor may
be influenced by the employee’s attitudes and preferences, as may the strength and power
of the employees’ arguments used towards the GP. Indeed, several previous studies have
suggested that attitudes might be important for understanding the reasons behind the
high levels of GP-certified sick leave (Allebeck and Mastekaasa 2004). We therefore have
argued that in principle, attitudes towards sickness absence can be important for both
self-certified and GP-certified sick leave and that this boils down to being an empirical
question.
The first main result of our study is that several of the attitude variables are associated
with sickness absence and mainly in the expected direction. It is especially interesting
that the variable measuring attitudes towards reasons for sickness absence is associated
with self-certified sick leave, but not GP-certified sick leave. This variable consists of
questions constructed to capture attitudes towards short-term sickness absence, and it
is precisely the self-certified and thus short-term absence where there is an association.
Low work satisfaction, on the other hand, is a state which can be draining in the long run,
and it is reasonable that this variable is associated with GP-certified sick leave. Summing
up, it seems as though the attitude variables are most important for self-certified sick
leave.
The second main result from our study is that attitudes generally are not able to explain
the observed differences in sickness absence between demographic groups. For exam-
ple, if attitudes towards sick leave should be able to explain the gender gap in sick leave,
women should have more lenient attitudes towards sick leave than men. Although we find
the commonly observed gender gap in GP-certified sick leave, we do not find a signifi-
cant difference in attitudes towards sick leave between men and women. We do, however,
find gender differences in some other attitude variables. In our sample, women have sig-
nificantly higher motivation and work ethics than men, have more gender equal attitudes
and are less lenient towards cheating than men. Nevertheless, the attitude variables were
not able to reduce the observed gender gap in GP-certified sick leave. If anything, women
would have had more sick leave had they had the same attitudes as men.
We do see an indication of attitudes being relevant for explaining why employees with
lower education have higher self-certified sickness absence than those with higher edu-
cation. However, controlling for attitudes in our regression only slightly reduces the
Hauge and Ulvestad IZA Journal of Labor Policy (2017) 6:11 Page 21 of 27
difference between low and high education, and the difference in sick leave remains sig-
nificant. We do not find any evidence of attitudes being important in explaining variation
in GP-certified sickness absence between other demographic groups.
There are some concerns related to this study. First, one might be concerned that
our sample is not large enough to estimate the relationship between attitudes and sick-
ness absence precisely. Although this might be the case, we do find large variations in
reported attitudes and we do reveal significant relationships. Also, the differences in sick
leave within our demographic groups are as expected, although the gender difference
in GP-certified sick leave is only significant at the 10% level. Furthermore, as discussed
in Section 3.1, our sample seems to be fairly representative compared to the population
under study.
A second concern might be the external validity of this study. The study was conducted
among healthcare workers within home care, assisted living and mental healthcare. We
know that there is a large fraction of women working in this field, and the sector might
attract workers with certain attitudes, or certain attitudes might arise in such a sector.
However, the public healthcare sector is a large employer in Norway. The results found in
this study might thus be relevant, at least, for a relatively large part of the work force. Nev-
ertheless, it is quite possible that one would find a larger variation in attitudes towards
sickness absence with a sample from several different sectors and that this study under-
estimates the relation between attitudes and sickness absence. Future research should
pursue studying attitudes across sectors.
Thirdly, we might not have been able to identify all relevant attitudes in our ques-
tionnaire. As there is a trade-off between including many questions and response rates,
including additional questions comes at a cost. Although trying to keep the questionnaire
as short as possible, the questionnaire includes the attitudes we considered most relevant.
One might also be concerned about whether the respondents have reported their atti-
tudes truthfully. There is an assumption in economics that people will lie if it is to
their material benefit. Evidence from lab experiments nevertheless suggest that peo-
ple have an aversion towards lying even when lying gives a material benefit (Abeler
et al. 2014; Fischbacher and Föllmi-Heusi 2013; Lundquist et al. 2009). This evidence,
taken together with the fact that there is no material benefit from lying in the sur-
vey and that the survey is anonymous, makes us less worried about respondents
not answering truthfully in the survey. If the inclination to lie varies systematically
between the demographic groups of which we study, however, there might still be rea-
son to worry. There does exist some evidence regarding lying aversion among men
and women. If women had a lower aversion towards lying than men, it could be the
case that the real attitudes of women were more lenient than that of men. How-
ever, as the evidence on the gender difference in lying if anything points towards
women being less averse to lying than men (Childs 2012; Dreber and Johannesson 2008;
Muehlheusser et al. 2015), we are more confident that lying does not confound our
results.
Finally, this study is not able to distinguish between whether attitudes cause sick
leave or whether sick leave causes certain attitudes. Nevertheless, an empirical study
of the relation between attitudes and absence behaviour is an important first attempt,
as a correlation between attitudes and sickness absence is a precondition for a causal
relationship.
Hauge and Ulvestad IZA Journal of Labor Policy (2017) 6:11 Page 22 of 27
In conclusion, this is the first paper studying the relationship between attitudes and
sickness absence using objectively measured data on sickness absence. Previous studies
have suggested that attitudes might be important for understanding variation in sickness
absence. We find that there is an association between attitudes and sickness absence but
mainly for self-certified sick leave. The results in our study do not support the notion that
attitudes are able to explain differences in sickness absence between demographic groups.
Appendix
Robustness test: analysis with attitude variables based on factor analysis
In the main analysis, we have grouped the survey questions into attitude variables based
on our own judgement. In this section, we perform an exploratory factor analysis to create
theattitudevariables.Thisistomakesurethatthewayweconstructtheattitudevariables
is not affecting our results.
Factor analysis
As in the main analysis, we want to reduce the number of questions from the ques-
tionnaire to a smaller number of variables describing attitudes and preferences. Factor
analysis is used to uncover possible latent structures in the data, and we can therefore use
it as a tool to group variables that pick up some of the same variation. We choose to per-
form an exploratory, as opposed to confirmatory, factor analysis because we do not want
to assume anything about the relationship between the survey questions.
First, we keep the 38 main items from our questionnaire, describing attitudes, pref-
erences and personality, and use the Kaiser-Meyer-Olkin measure to see if our data are
suited for factor analysis. We find an overall value of 0.6237, which is above the require-
ment of 0.5 (Frohlich and Westbrook 2001). Thereafter, we run a principal factor analysis
with promax rotation. Promax rotation is chosen because we want to allow our factors to
be correlated. The factors created are based on the eigenvalues of the covariance matrix
of the standardised attitude variables and presented in the scree plot of the eigenvalues
in Fig. 2. Based on the scree plot, we choose to keep five factors for the analysis. The
0 1 2 3 4
Eigenvalues
0 10 20 30 40
Number
Fig. 2 Scree plot of the eigenvalues
Hauge and Ulvestad IZA Journal of Labor Policy (2017) 6:11 Page 23 of 27
Lenient attitudes towards sick leave
Low intrinsic motivation
Lenient towards cheating
Being stressed
High homeload
−.5 0 .5 −2 0 2 4
Self−certified GP certified
Fig. 3 Point estimates and 90% confidence intervals for the attitude factors
eigenvalues measure the variance in the variables that are grouped into that factor, and
only factors with eigenvalues above one are retained.
Table 7 shows the factors resulting from the factor analysis, together with the
questions from the questionnaire with the highest loadings on this factor. The table
also gives our a priori expectations of the factors’ association with sickness absence.
Factor loadings are the correlations between each variable and the factor. Loadings
above 0.5 are characterised as large, moderate loadings are between 0.3 and 0.5, and
small loadings below 0.3 (Della Giusta et al. 2009). In Table 7, only loadings above
0.3 are shown, but all items loading on a factor are included when predicting the
factors.
The direction of all the factors are such that the expected relation with sickness absence
is positive; for instance, we expect that lenient attitudes towards self-caused sick leave
cohere with more sick leave.
Comparing the variables created by the factor analysis in Table 7 with the variables
presented in the main analysis (Table 2 in the main part of the paper), we see that the two
sets of variables are quite similar. The factor analysis resulted in fewer factors than in the
main analysis (five factors from the factor analysis compared to ten variables in the main
analysis). The two sets of variables are, in substance, quite similar.
Results from analysis using attitude factors
In this section, we will repeat the analysis done in the main part of the paper, but instead
of using the attitude variables from the main analysis, we will use the factors created by
the factor analysis as variables in the regressions. As can be seen from Table 8, the factor
Lenient attitudes towards sick leave is positive and significant for self-certified sick leave.
None of the factors have a significant relation to GP-certified sick leave.
The coefficient of the variable Lenient towards sick leave was significant for self-certified
sick leave also in the main analysis. In the main analysis, two additional variables were
significant: Low work ethics and Lenient towards cheating. In the regressions based on
Hauge and Ulvestad IZA Journal of Labor Policy (2017) 6:11 Page 24 of 27
Table 7 Factor analysis results
Factors Factor loadings Survey question A priori
expectation
What is your opinion on calling in sick (self-certified
sickness absence)...
0.4281 ...if one has a cold and a low-grade fever?
0.4336 ...if one has not slept enough during the previous
nights?
1: Lenient attitudes towards
sick leave
0.6520 ...if one has back pain when waking up in the
morning? +
0.5753 ...if one feels unwell due to stress at work?
0.4623 ...if one is being bullied at work?
0.6526 ...if one is pregnant and feeling nauseous?
Do you agree or disagree with the following
statements?
2: Low intrinsic motivation 0.8151 “The tasks that I do at work are enjoyable.”a+
0.7646 “My job job is meaningful to me.”a
What is you opinion on the actions below?
0.7423 “Cheating on taxes.”
3: Lenient towards cheating 0.6202 “Buying services on the black market.” +
0.7217 “Avoiding a fare on public transport.”
Do you agree or disagree with the following
statements?
4: Being stressed 0.8661 “I often feel stressed at work.” +
0.5185 “I do not manage to unwind/relax when I am
off work.”
Do you agree or disagree with the following
statements?
5: High homeload 0.6886 “I have so much to do at home that I don’t have
enough time for everything that needs to be done
at work.” +
aThe Likert scale has been reversed
the factor analysis, Low work ethics is not included as a variable, while Lenient towards
cheating is not significant. For GP-certified sick leave, there was one significant coefficient
in the main analysis: Low work satisfaction. However, the factor analysis did not provide
such a variable. Although some differences, the two methods of reducing the questions in
the questionnaire to fewer variables give quite similar results. The variables created based
on our own judgement are quite similar to those resulting from the factor analysis, and
the regression results show qualitatively similar patterns.
The regression results in Table 8 show the same patterns as our main results. Women
have significantly higher GP-certified sick leave rate than men, and including the attitude
factors in the regression increases the Female coefficient, implying that attitudes cannot
explain the higher sick leave rate among women. If anything, women would have had
more sick leave had they had the same attitudes as men.
Employees with lower education have significantly higher self-certified sick leave rate
than employees with higher education. As in the main analysis, adding the attitude factors
reduces the difference in sickness absence between the two groups, implying that atti-
tudes might explain some of the difference in sick leave between low and high educated
employees.
Hauge and Ulvestad IZA Journal of Labor Policy (2017) 6:11 Page 25 of 27
Table 8 Regression results using attitude variables based on factor analysis
(1) (2) (3) (4)
Self-certified Self-certified GP-certified GP-certified
coef. coef. coef. coef.
(se) (se) (se) (se)
Female −0.311 −0.219 2.330* 2.439*
(0.19) (.018) (1.28) (1.46)
Immigrant 0.001 0.020 0.736 0.302
(0.20) (0.19) (1.25) (1.30)
Factor 1: Lenient attitudes towards sick leave 0.314*** 0.339
(0.10) (0.73)
Factor 2: Low intrinsic motivation 0.192* −0.232
(0.11) (0.81)
Factor 3: Lenient towards cheating −0.076 0.834
(0.11) (0.78)
Factor 4: Being stressed 0.103 0.326
(0.14) (0.89)
Factor 5: High homeload −0.240 1.239
(0.15) (1.07)
_cons 5.402*** 5.918*** 18.001* 18.920*
ControlsaYes Yes Yes Yes
Attitudes No Yes No Yes
R-squared 0.157 0.188 0.133 0.150
Low education 0.703*** 0.668*** 1.211 0.642
(0.22) (0.21) (1.20) (1.19)
ControlsbYes Yes Yes Yes
Attitudes No Yes No Yes
R-squared 0.154 0.184 0.135 0.150
Old −0.075 0.046 2.617* 3.052*
(0.22) (0.23) (1.37) (1.58)
ControlscYes Yes Yes Yes
Attitudes No Yes No Yes
R-squared 0.141 0.170 0.107 0.119
Nd226 226 226 226
*p<0.10, **p<0.05, ***p<0.01
aage, age2, education, na_educ, single_hh, na_singlehh, child_cust, na_custody, immig_sec, na_immig, student, secondjob,
director, shifts, district_a, ass_living, mental_health
bfemale, age, age2, na_educ, single_hh, na_singlehh, child_cust, na_custody, immig_sec, immigrant, na_immig, student,
secondjob, director, shifts, district_a, ass_living, mental_health
cfemale, education, na_educ, single_hh, na_singlehh, child_cust, na_custody, immig_sec, immigrant, na_immig, student,
secondjob, director, shifts, district_a, ass_living, mental_health
dWe use frequency weights in the regressions due to the fact that we observe the employees for varying amounts of time,
putting more weight on the employees that we have more information about. The frequency weights duplicate the observations
according to the employees’ number of working days. This means that the total number of observations, if we take the frequency
weights into account, is 104,229. We use cluster robust standard errors to count for the serial correlation that occurs due to the
frequency weights
Older employees have significantly higher GP-certified sick leave rate than younger
employees and equivalently as in the main analysis; including the attitude factors amplify
this, again supporting the result from the main analysis.
Immigrants do not have significantly different sick leave than non-immigrants.
Hauge and Ulvestad IZA Journal of Labor Policy (2017) 6:11 Page 26 of 27
Additional file
Additional file 1: Questionnaire. (PDF 775 kb)
Acknowledgements
Thanks to Knut Arne Hagtvet for advice regarding the factor analysis, to Simen Markussen and Oddbjørn Raaum for their
suggestions and comments and to Michael Soukup for his help with the processing of data. All remaining errors remain
the responsibility of the authors. We would also like to thank the editor for the useful comments.
Responsible editor: Juan Jimeno
Funding
The development and execution of the survey was financed by The Norwegian Ministry of Labour and Social Affairs. Data
made available by the city of Oslo have been essential for the research project. This article is part of the project
“Disentangling absence patterns”, financed by the Norwegian Research Council (grant # 227103/H20), and the project
“Work Life Challenges—workforce management and worker involvement solutions” (grant # 202647/S20).
Ethics approval and consent to participate
The research project is reported to the Data Protection Official at the Frisch Centre. Since no sensitive data was used in
the project, additional permission from the Norwegian Data Inspectorate was not required. The research is based on
information collected through a survey and linked to information from administrative registers from the city of Oslo, with
informed consent from all participants. The project complies with ethical standards set down in Norwegian law and
overseen by the Norwegian Data Inspectorate.
Competing interests
The IZA Journal of Labor Policy is committed to the IZA Guiding Principles of Research Integrity. The authors declare that
they have observed these principles.
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Received: 28 June 2017 Accepted: 25 August 2017
References
Abeler J, Becker A, Falk A (2014) Representative evidence on lying costs. J Public Econ 113:96–104
Alavinia SM, Van Den Berg TI, Van Duivenbooden C, Elders LA, Burdorf A (2009) Impact of work-related factors, lifestyle,
and work ability on sickness absence among Dutch construction workers. Scand J Work Environ Health:325–333
Allebeck P, Mastekaasa A (2004) Chapter 5. Risk factors for sick leave—general studies. Scand J Public Health 32(63
suppl):49–108
Askildsen JE, Bratberg E, Nilsen ØA (2005) Unemployment, labor force composition and sickness absence: a panel data
study. Health Econ 14(11):1087–1101
Bay A-H, Hellevik O, Hellevik T (2007) Svekker innvandring oppslutningen om velferdsstaten? Tidsskr Samfunnsforskning
48(3):377–408
Bjørnstad AF (2013) Utbetalingene av trygdeytelser siste 10 år. Arb og velferd 3:13–23
Blais A-R (2006) A domain-specific risk-taking (dospert) scale for adult populations. Judgment Decis Making 1(1):33–47
Bratberg E, Dahl S-Å, Risa AE (2002) ‘The double burden’: do combinations of career and family obligations increase
sickness absence among women? Eur Sociol Rev 18(2):233–249
Childs J (2012) Gender differences in lying. Econ Lett 114(2):147–149
Christensen KB, Labriola M, Lund T, Kivimäki M (2008) Explaining the social gradient in long-term sickness absence: a
prospective study of Danish employees. Journal of epidemiology and community health 62(2):181–183
Dale-Olsen H, Markussen S (2010) Økende sykefravær over tid? –Sykefravær, arbeid og trygd 1972–2008. Søkelys på
arbeidslivet 27(01-02):105–121
Della Giusta M, Di Tommaso ML, Shima I, Strøm S (2009) What money buys: clients of street sex workers in the US.
Applied Economics 41(18):2261–2277
Dreber A, Johannesson M (2008) Gender differences in deception. Econ Lett 99(1):197–199
Eriksen TLM, Hogh A, Hansen ÅM (2016) Long-term consequences of workplace bullying on sickness absence. Labour
Econ 43:129–150
European Commission (2016a) First preliminary outline of a European Pillar of Social Rights, Annex 1 of Launching a
consultation on a European Pillar of Social Rights. Communication from the Commission to the European Parliament,
the Council, the European Economic and Social Committee and the Committee of the Regions. SWD (2016) 51 final.
Strasbourg, France
Spasova S, Bouget D, Vanhercke B (2016) Sick pay and sickness benefit schemes in the European Union, Background
report for the Social Protection Committee’s In-depth Review on sickness benefits. European Social Policy Network
(ESPN), Brussels, European Commission
Fischbacher U, Föllmi-Heusi F (2013) Lies in disguise—an experimental study on cheating. J Eur Econ Assoc 11(3):525–547
Frohlich MT, Westbrook R (2001) Arcs of integration: an international study of supply chain strategies. J Oper Manag
19(2):185–200
Gran-Henriksen B (2014) Antall offentlige årsverk nesten tredoblet siden 1970, Statistics Norway. http://www.ssb.no/
offentlig-sektor/artikler-og-publikasjoner/offentlig- forvaltning-storrelse
Hauge and Ulvestad IZA Journal of Labor Policy (2017) 6:11 Page 27 of 27
Hansen CD, Andersen JH (2008) Going ill to work—what personal circumstances, attitudes and work-related factors are
associated with sickness presenteeism? Soc Sci Med 67(6):956–964
Hemp P (2004) Presenteeism: at work-but out of it. Harv Bus Rev 82(10):49–58
Henrekson M, Persson M (2004) The effects on sick leave of changes in the sickness insurance system. J Labor Econ
22(1):87–113
Holmås TH, et al (2008) Attitudes towards sickness absence in the nordic countries. Working Paper 18-2008. Stein Rokkan
Centre for Social Studies. Bergen, Norway
Holth BA (2010) Verdiundersøkelsen 2008. Dokumentasjonsrapport. Notater 47/2010 Satistisk Norway. Oslo-Kongsvinger,
Norway
Kuvaas B, Dysvik A (2009) Perceived investment in employee development, intrinsic motivation and work performance.
Hum Resour Manag J 19(3):217–236
Kuvaas, B, Dysvik A (2011) Permanent employee investment and social exchange and psychological cooperative climate
among temporary employees. Econ Ind Democr 32(2):261–283
Laaksonen M, et al (2010) Gender differences in sickness absence - the contribution of occupation and workplace. Scand
J Work Environ Health 36(5):394–403
Lundquist T, Ellingsen T, Gribbe E, Johannesson M (2009) The aversion to lying. J Econ Behav Organ 70(1):81–92
Markussen S, Røed K, Røgeberg OJ, Gaure S (2011) The anatomy of absenteeism. J Health Econ 30(2):277–292
Markussen S, Strøm M (2013) The effects of motherhood, Memorandum, Department of Economics. University of Oslo,
No. 19/2015
Mastekaasa A, Dale-Olsen H (2000) Do women or men have the less healthy jobs? An analysis of gender differences in
sickness absence. Eur Sociol Rev 16(3):267–286
Muehlheusser G, Roider A, Wallmeier N (2015) Gender differences in honesty: groups versus individuals. Econ Lett
128:25–29
Myklebø S (2007) Sykefravær og svangerskap. Arbeid og velferd, Nr 3-2007:21–27. Arbeids- og velferdsdirektoratet
NOU 2010:13 (2010) Arbeid for helse. Sykefravær og utstøting i helse- og omsorgssektoren. Departementenes
servicesenter Informasjonsforvaltning, Oslo
OECD (2010) Sickness, disability and work: breaking the barriers, A Synthesis of Findings across OECD Countries. OECD
Publishing, Paris. http://dx.doi.org/10.1787/9789264088856-en
Piha K, et al (2010) Interrelationships between education, occupational class, income and sickness absence. Eur J Public
Health 20(3):276–280. doi:https://doi.org/10.1093/eurpub/ckp162
Rønning E (2002) Statistisk sentralbyrås tidsbruksundersøkelse 2000/01. dokumentasjon og resultater fra intervjuet.
Technical report
Scheil-Adlung X, Sandner L (2010) The case for paid sick leave. World Health Report Background Paper, 9
SSB (2015) Statistikkbanken: sykefravær. https://www.ssb.no/statistikkbanken/selecttable/hovedtabellHjem.asp?
KortNavnWeb=sykefratot&CMSSubjectArea=arbeidog-lonn&checked=true. Accessed 17 July 2015
Thomas LT, Ganster DC (1995) Impact of family-supportive work variables on work-family conflict and strain: a control
perspective. J Appl Psychol 80(1):6
Thomson L, Griffiths A, Davison S (2000) Employee absence, age and tenure: a study of nonlinear effects and trivariate
models. Work Stress 14(1):16–34
Thorsen SV, Friborg C, Lundstrøm B, Kausto J, Örnelius K, Sundell T, Kalstø AM, Thune O, Gross B-O, Petersen H, Haram O
(2015) Sickness in the Nordic Countries. Nord Soc Stat Comm 59:1–77
