Working Conditions, Compensation and Absenteeism

Journal of Health Economics 24 (2005) 161–188
Working conditions, compensation and absenteeism

Solveig Osborg Ose∗
Department of Economics, Norwegian University of Science and Technology, N-7491 Trondheim, Norway
Received 11 September 2003; received in revised form 6 April 2004; accepted 8 July 2004
Available online 16 September 2004
Abstract
This paper examines absence behaviour in relation to the working environment. A theoretical model
is built in order to separate the effects of voluntary absences and absences related to ill health, where
health effects are assumed to be tied to working conditions. This model is based on the Shapiro and
Stiglitz efficiency wage model. In addition, work environment is introduced as a part of the compen-
sation package. The model gives a testable hypothesis of compensating wage differentials. A panel
of quarterly firm level data from 1990 to 1998 are used and the theoretical model is supported by the
empirical findings. The result indicates that the workers may not be fully compensated when experi-
encing high levels of noise in the work area, or when the job involves a high degree of monotonous
work, heavy or frequent lifting or poor work postures. Ill health, and thus increased long-term absence,
is not highly related to economic variables. However, long-term absence is relatively higher if the
firm is troubled with many accidents or near misses. In addition, disamenities for which workers are
not fully compensated cause ill health and increased long-term absence.
© 2004 Elsevier B.V. All rights reserved.
JEL classiﬁcation: J16; J28; J33; J41; J81
Keywords: Compensation; Working environment; Health; Absenteeism; Efficiency wages
1. Introduction
The present paper offers a theoretical and empirical analysis of the relation between
work environment and work absenteeism. The theoretical analysis modifies an efficiency
∗ Present address: SINTEF Health Research, N-7465 Trondheim, Norway. Tel.: +47 73 593000;
fax: +47 73 596361.

E-mail address: solveig.ose@sintef.no (S.O. Ose).
0167-6296/$ – see front matter © 2004 Elsevier B.V. All rights reserved.
doi:10.1016/j.jhealeco.2004.07.001
162 S.O. Ose / Journal of Health Economics 24 (2005) 161–188
wage model to include working conditions. This model forms the theoretical basis for the
empirical analysis that utilizes detailed survey data on working conditions for Norwegian
firms.
If a worker is tired from physical exhaustion or cannot bear the psychological pressure
at work, he or she is more likely to be absent from work than a person working in an
environment of better quality. Both the physical and psychological aspects of the work
situation are assumed to influence the decision regarding going to work on a given day, and
cause both voluntary and involuntary absences. The physical aspect can also induce injuries
and then sickness absence. In Norway, muscular–skeletal and mental illness are the most
common diagnoses for employees reported absent.1
Few economists have looked into the potential relationship between sickness absence and
the working environment. There is a substantial number of studies from other fields, such
as applied psychology, business administration and occupational medicine. However, these
are often case studies and are typically based on individual absence rates often specified by
medical diagnostics as lower back pain, respiratory illness, etc.
In this paper, I incorporate work environmental issues into an economic framework with
the aim of explaining differences in worker absence across firms. The working environment
includes issues such as temperature and comfort, lighting, noise and general surroundings.
When any one of these working conditions is substandard they can affect the health of the
workers.
It is often argued, especially in the field of occupational medicine, that a healthy work
environment gives marginal workers the possibility to participate in the working force. The
inclusion of new workers leads to potentially higher absence rates as these are marginal
workers. An improved work environment could also lead to unhealthy workers extending
their working career, thus giving rise to a higher absence rate in the firm. Worker absence
is often looked upon as a composition mechanism, as a higher participation rate is assumed
to result in higher absence. A less stringent view is that improvements in the work envi-
ronment will reduce the negative health effects on the current workers and therefore lower
absenteeism. The composition effect may partly offset this effect as new marginal workers
will drive absence rates up.
It seems important to disentangle these effects. If firms expect absenteeism to increase
when the work environment is improved, non-economic incentives have to be considered. If,
on the other hand, improvement in the work environment is followed by lower absenteeism,
this has to be factored into the decision on whether to invest in the work environment or
not. It is important to identify the aspects of the working environment that cause potential
health problems for workers. For all parties, the main goal is to improve workers’ health.
This is obviously beneficial to individuals, firms and society as a whole. Worker absence is
a good proxy for workers’ health (Marmot et al., 1995) and may be used to investigate the
effect of the working environment on workers’ health. For policy purposes, it is important
to distinguish between the different causes of worker absence and their relative roles in
explaining observed patterns.
1 Muscular–skeletal and mental illness account for 48 and 14% of all sickness benefit incidences, respectively
(National Insurance Administration, 2001).
S.O. Ose / Journal of Health Economics 24 (2005) 161–188 163
Absenteeism has been the focus of much research by economists and other social scien-
tists in recent years.2 However, as mentioned above, few economists have included work
environmental issues in their studies on absenteeism. An important exception is the work
by Johansson and Palme (1996). They explicitly model the everyday economic choice of
being absent from work in relation to occupation and personnel characteristics that may
influence this choice. Cross-sectional data from the 1981 Swedish Level of Living Survey
are used in combination with information on individual work absence from the National
Social Insurance Board register. The data set consists of 1967 individuals for the year 1981.
Two types of working condition variables are used: occupation-specific risk indices and
self-reported information from the survey. Johansson and Palme find that risk, i.e., acci-
dents at work and work-related diseases, will increase work absence. In addition, they find
significant effects related to different self-reported job characteristics, which indicate that
individuals with a work profile involving low stress and outdoor work have, on average, a
lower absence rate. Drago and Wooden (1992) use a cross-national data set to test different
theoretical frameworks for predicting self-reported employee absence rates. They regard
shift-work as a negative working condition and find that workers on shift-work are more
prone to be absent. In addition, they include indices for job satisfaction and work group
cohesion and find that work group cohesion leads to low absence where job satisfaction is
high and to high absence when job satisfaction is low. Another contribution that is relevant
for this study is by Brown et al. (1999), which investigates the effects on absenteeism of
two types of employer-sharing plans – profit sharing and employee share ownership – in
127 French firms over the years 1981–1991. They find that both plans were associated with
reductions in absenteeism. Absence is defined as the total number of absence events in the
firms divided by the total number of employees employed by the firms. To measure the actual
or perceived danger associated with the job, Brown et al. use the number of workdays lost
following workplace accidents and the amount of safety expenditure per employee. They
find positive effects from both, but do not discuss the endogeneity problems related to this.
In the field of applied industrial psychology, it is argued that major changes in orga-
nization may explain variations in worker absence. These changes include reductions in
personnel and general reorganization. It is common to study this in an individual-orientated
framework grounded on theories of Demand-Control and Effort Reward Imbalance. See
Karasek and Theorell (1990), Siegrist (2001) respectively for details. Variables such as job
control, job demand, support, and the attitude towards the job may for instance be measured
before and after an organizational change and be traced back to the process of change. Thus,
they explain variations in worker absence. For instance, Vahtera et al. (2000) finds that neg-
ative changes in the psychosocial work environment, as a result of the severe economic
decline Finland faced in the 1990s, have adverse effects on the health of employees. Those
working in an unfavorable psychosocial environment before changes are at greater risk.
More specifically, Vahtera et al. find that lowered job control causes a higher risk of sick
leave. In addition, they find that changes in job characteristics are stronger predictors of sick
2 Brown and Session (1996) concentrate mostly on the economic literature while Harrison and Martocchio
(1998) give a critical review of the time frame used in different studies put in a management context. In addition,
Alexanderson (1998) gives a systematic overview of the literature where the different contributions are placed at
national, workplace and individual level.
leave for employees with a high income than for the others. The highest risk of sick leave
is associated with combined effects related to negative changes in job control, job demands
and social support. However, although the effects of organizational changes on individual
absence behaviour are empirically verified, the aggregate effect on firm level behaviour
remains unknown. Knowledge of the effects of major organizational changes on a firm’s
absence rates will be important to the management, as precautions may be taken to offset
the negative effects. Looking into these aggregate effects is one of the topics of this paper.
Another paper from outside the field of economics is by Lund et al. (2001). They look
at the rates at which waste collectors and municipal workers leave their jobs, and identify
associated risk factors within the work environment and health, life-style and marital status.
Survey data are used, and their conclusion suggests that there is potential for preventing
people from leaving physically demanding occupations by improving the physical and
psychological work environment.
A major problem with the economic studies of worker absence is that they utilize
theoretical models developed to explain shirking or voluntary absence to study long-term
absence only. The labor supply model is often used without including health in the utility
function or anything else that could explain sickness absence or long-term absence. The
efficiency wage framework gives predictions about shirking, but does not, in its original
form, include variables meant to explain sickness absence. Therefore, an issue in this study
will be to establish a health effect of adverse working conditions in addition to the efficiency
wage theory. When looking at working conditions and shirking, elements of compensated
wage differentials theory have to be included. The model discussed in this paper will
establish this argument and result in a testable hypothesis about compensation from adverse
work conditions. Separating the theoretical effects on sickness absence and shirking will,
therefore, provide explanations of both the effect on health from the work environment
and economic incentives related to shirking. In addition, a common critique of traditional
absence models is met, as the demand side of the labor market is included. Firms differ with
respect to work environment, and this may explain different sickness absence rates across
firms.
Based on efficiency wage theory and wage compensating theory, predictions about
worker absence and working conditions given in the theoretical model will be tested
using a panel of firms from the Norwegian manufacturing industry from 1990 to 1998.
Information on absence, wage and other economic variables that are the background
material for the published quarterly reports by Confederation of Norwegian Business and
Industry (NHO) are obtained. In addition, a survey was carried out to collect data on
working environment issues (both physical and psychological), resulting in a sample of
326 firms. We included questions about time and the type of major changes in the firm
during the period. This included information on mergers and de-mergers, major changes
in the production processes, and so on. Time determination of these changes is necessary
to assess the effect of these variables on worker absence. These variables will also control
for specific events in the firm during the period.
The merge of these two datasources represents an unique database with detailed infor-
mation on work environment issues for over 300 firms. The employer-reported short- and
long-term absence rates, wage information and other economic variables, are available for
these firms for a period of 36 quarters.
The paper is organized as follows: Section 2 contains the underlying theory, where the
point of departure is the standard shirking model extended to include non-pecuniary rewards
from working. Data are described in Section 3. Section 4 specifies the empirical approach,
and the results are presented in Section 5, followed by the Conclusion.
2. Theoretical approach
Accepting a job offer implies accepting not only a wage rate, but also a number of
non-wage characteristics. The theory of compensating wage differentials suggests that the
worker may accept worse working conditions in exchange for a higher wage rate.3 Even
if the existence of wage compensation for dangerous work has been questioned (see for
instance Dorman and Hagstrom (1998), the concept may be applied to incorporate general
work environmental issues in absenteeism models. What is most important in this setting
is that a worker that is not properly compensated for working in a poor work environment,
may choose to shirk because he or she is dissatisfied with the pay.4 Given imperfect com-
pensation, we may talk about compensating shirking differentials instead of compensating
wage differentials.
If the worker is employed by a firm with a poor work environment, he or she is likely to
have a higher rate of work absence for two reasons. First, as the worker faces a higher level
of risk at the workplace, he or she will strive to minimize the exposure time to these risks,
i.e., he or she has, ceteris paribus, greater incentives to be absent from work (Johansson and
Palme, 1996). The fact that less risk-averse workers may self-select into higher risk jobs
may partially offset this effect. Secondly, workers who are exposed to higher risks also have
a higher probability of being absent due to work-related illness or injuries. Johansson and
Palme (1996) argue that omitting risk or wage compensation may lead to biased estimates
of the relationship between wage rates and absence.
Employers have different abilities to offer combinations of pay and working environ-
ments. Technology may dictate that some jobs will be less agreeable than others in some
respects, but employers can also vary the negative impact of the working environment by
devoting resources to the removal of these negative factors.
A large number of firms offering different bundles of working environments and wages
will give a smooth upward-sloping offer curve. This is because negative working condi-
tions are costly to remove and, when faced with a reduction in the quality of the working
environment, employees must be paid higher wages to keep utility constant.
In the theoretical model presented below, sickness absence is related to ill health, while
shirking is not.
An efficiency wage model, which is set in continuous time along the lines of the shirking
model by Shapiro and Stiglitz (1984), is considered. Basic assumptions in the shirking
model are that all workers are identical, identical firms so that firms pay the same wage
3 A substantial literature on work hazard and compensation exists (Cousineau et al., 1992; Martinello and Meng,
1992; Meng and Smith, 1999). The evidence indicates that the market does compensate workers, at least in part,
for the possibility of being injured on the job, by inducing employers whose workplaces are dangerous to offer
compensation above that offered in safer workplaces.
4 This is related to the fair wage-effort hypothesis, see Akerlof and Yellen (1986).
in equilibrium, firms do not monitor their workers perfectly, and firms fire workers when
they are caught shirking. These assumptions together are shown to lead to efficiency wages,
which is the same as a “non-shirking wage”, i.e., the wage required for the worker not to
shirk on the job. The assumption of a constant wage across firms is modified in my setting,
and it is assumed that rather the wage and work conditions bundle is constant across firms.
In line with wage compensating theory, firms with a healthy work environment may pay
lower wages than firms offering an unhealthy work environment.
There are N identical workers. By assuming that the workers are identical, the matching
aspect will be lost as individuals are assumed to have the same preferences regarding the
work environment.5 This will also simplify the firms’ adaptation.
The instantaneous utility of an employed worker is U(z, e) = z − e, where e is the level
of effort and z includes both pecuniary and non-pecuniary rewards from working, i.e.,
z = w + j, where w is the wage rate, and j the non-pecuniary reward. This will capture
the working conditions in the firm in which the worker is employed, including physical
working conditions and the psychosocial elements of the working environment. Assuming
that j > 0 if the worker evaluates the working environment to be of a better quality, and
j < 0 if the work environment is poor, with j → −∞ if the probability of dying at work
moves towards unity. Individuals value non-market (i.e., leisure) time more as j → −∞.
The worker is assumed to choose between no effort at all (e = 0) or a positive level of
effort (e > 0) and has a productivity equal to 1. If e > 0, the worker does not shirk, i.e.,
s = 0, where s refers to shirking. If e = 0, the worker shirks, i.e., s = 1. From this, the effort
choice and the shirking choice will be synonymous. In addition to the wage, the worker has
a non-wage component regarding working.
All workers maximize their expected discounted value of utility, discounted by a rate r.
The worker chooses a bundle z of pecuniary and non-pecuniary rewards from working. If
he or she chooses a poor working environment with a correspondingly higher wage, some
unavoidable sickness absence is likely to occur. In this manner, both shirking and sickness
absence are incorporated in the model.
2.1. The effort choice of workers
The worker selects an effort level to maximize his or her discounted utility stream.
When choosing the effort level, shirking is a direct outcome of a decision to offer no
effort. Comparing utility from shirking with the utility from not shirking is essential. As the
Shapiro–Stiglitz model is a value function approach, the value of being in a current state
is viewed as an asset with a particular return, which consists of two main parts: first, a net
current return, and second, an expected change in asset value. The expected value of the
lifetime utility of an employed worker who shirks is given by:
rVs = (w + j) + (b + q)(Vu − Vs ) (1)
where rVs is the return to the asset from being employed, but shirking. (w + j) is the current
return, what is gained from working. In addition, the work environment indicator has to be
5 The assumption of identical workers may be relaxed using efficiency models of heterogeneous workers (see
for instance (Strand, 1987)). As the main interest of this paper is to explain the effects of the work environment
on shirking and sickness absenteeism, this will be left for future research.
included in the first term because the worker will lose something when not going to work
if the work environment is good, but will gain more by shirking if the work environment is
poor. The second term is the expected value of the change in the asset, where (b + q) is the
total rate at which this change occurs for a shirker, where q is the probability of being caught
and fired while b is the exogenous separation probability. (Vu − Vs ) is the asset change once
it occurs and Vu is the expected lifetime utility of an unemployed worker. This change is
generally negative, i.e., there is a utility loss associated with losing one’s job.
If j < 0, the worker experiences a poor work environment. The expected value of lifetime
utility will be lower if the worker is not fully compensated by a higher wage. If, on the
other hand, the working environment is good, j > 0, the worker may be working in a less
demanding job, indicating a lower wage rate.6 So, if the worker is not fully compensated for
a poor working environment by receiving a higher wage, the current return from working
is lower.
For a non-shirker, the asset equation is given by:
rVn = (w + j − e) + b(Vu − Vn ) (2)
where effort e > 0 is used, and a gain or loss from the work environment, j, is obtained or
inflicted.
Solving (1) for Vs and (2) for Vn and assuming that the worker will choose not to shirk
if and only if Vn ≥ Vs , the non-shirking condition will be given by:
(r + b + q)e
w+j ≥ + rVu (3)
q
This is the same as the non-shirking condition (NSC) in the Shapiro–Stiglitz model except
for the additional term, j, on the left-hand side. The logical interpretation concerns proper
compensation. If the worker is fully compensated in a poor working environment or if there
is a proper wage reduction associated with a good working environment, this condition is
the same as in the Shapiro–Stiglitz model. That is, the bundle of wage and non-pecuniary
rewards must be higher than the right hand side of the Eq. (3) to avoid shirking. If the worker
is not fully compensated for a poor working environment, the current value of working
will be lower than needed to avoid shirking. If there was no penalty associated with being
unemployed, everyone would shirk, i.e., NSC could never be satisfied. As the effect depends
on whether the worker is compensated or not, these situations will be discussed separately.
2.1.1. The worker is fully compensated

In this case, the results will be similar to those of the Shapiro–Stiglitz model.
If the firm pays a sufficiently high wage in relation to the working conditions, the worker
will not shirk. The critical wage/working environment bundle is better when there is a higher
required effort (e), a higher expected utility of being unemployed (Vu ), a lower probability
of being detected shirking (q), a higher rate of interest (r), and a higher exogenous quit
rate (b).
6 j = 0 may be thought of as a situation where the work environment imposes neither positive nor negative
effects on the worker, i.e., it is a completely neutral work environment.
In addition, it is necessary to differentiate between good and poor working environments.

If j > 0, the worker is in a healthy working environment and, according to the compensation
theory, the wage is lower. However, this lower wage will not induce the worker to shirk
because this is the efficient wage in the context of the working environment; that is, it is
part of the efficiency wage/working condition bundle. Therefore, shirking will be prevented
in this situation and involuntary absence (sickness absence) will be lower as a result of the
good working environment.7
If, on the other hand, j < 0, the worker is in a poor working environment, with physically
demanding work tasks and poor relationships with other workers and/or management. Then,
the wage needed for the worker to show up at work must be higher. In the case of full
compensation, the wage is set according to the current quality of the work environment, and
workers will not shirk. However, a poor working environment gives rise to higher sickness
absence.
2.1.2. The worker is not fully compensated

In this situation, the worker is not fully compensated for a poor work environment. In
Eq. (3), the left-hand side will increase or decrease depending on the quality of the working
environment. Whenever the bundle of wage/working conditions changes, the worker is not
paid according to preferences regarding the trade-off between wage and working conditions.
If the working environment is good (j > 0), the wage needed to avoid shirking is lower.
The potential cut in wages powered by a good working environment is not fully carried out
in this situation, so the workers are more than happy to work, indicating lower shirking. At
the same time, negative health effects arising out of poor working conditions are avoided,
indicating relatively lower sickness absenteeism.
If the working environment is poor (j < 0) and this is not offset by the wage being set
at a high rate, the wage necessary to avoid absence is too low and there will be shirking.
However, at the same time, the working environment will be causing bad health and thus
high sickness absence. Consequently, in this situation, both shirking and sickness absence
will be relatively high compared with the situation described above.
If the required effort, the expected utility of being unemployed or the exogenous quit
rate increases, the wage necessary to avoid shirking increases, given the current quality of
the work environment. However, if sickness absence is high and costly to the firm, it may
be difficult to increase wages, with the result that shirking will be even higher.8
Shapiro and Stiglitz (1984) define the expected value of the lifetime utility of an un-
employed worker in the same manner as in Eqs. (1) and (2). In the model presented here,
in the situation where the workers are not fully compensated, the value of working will
differ between individuals, making aggregation more complicated. In this setting, there is
7 A good working environment may give the worker the opportunity to work even if he or she is sick. Alternatively,
the worker may go to work just to be part of a good social environment or may feel that by staying home sick,
he or she is letting down colleagues or fellow workers. This is the problem of presenteeism, i.e., staying at work
even when feeling sick. Aronsson et al. (2000) find a positive correlation between absenteeism and presenteeism
but as pointed out by the authors, more detailed analysis and dynamics must be introduced to elaborate on this
relationship.
8 If shirking results in high turnover, there may be a snowball effect as this imposes higher costs on the firm,
leading to lower wages and more shirking.

no need to construct the aggregate NSC as I assume that the value of being unemployed is
an increasing function of the unemployment benefit (w̄) and a decreasing function of the
unemployment level (U).
This adds new predictions to the model. The critical wage/working conditions bundle is
higher if the unemployment benefit increases and if flows out of unemployment are higher.
If the unemployment rate increases, or if flows out of unemployment decrease, the penalty
of getting caught shirking is greater as it is harder to find a new job. The wage or the quality
of the working conditions necessary to avoid absence is then lower. When higher unemploy-
ment is not followed by a reduction in wages or worsening working conditions, it results in
reduced absenteeism. However, in a situation with high unemployment, resources within
the firm may become available for investing in the working environment or the resources
required to undertake this work may be cheaper. In addition, physical and psychological
stressors may be reduced in a situation with lower work pressure, thus lowering sickness
absence. Therefore, sickness absence may be reduced in periods of high unemployment
caused by an improved working environment. However, there is another possibility. When
the unemployment rate increases, it may be more difficult for employees to demand the
maintenance of, or improvements to, the work environment, as demanding workers are
easily replaced when there are many unemployed workers willing to work. If this mech-
anism is effective, the work environment is poor when the unemployment rate is high, so
that sickness absence is a negative function of the unemployment rate. Thus, both shirking
and sickness absence may be affected by changes in the unemployment rate, but for quite
different reasons.
Furthermore, assume that all firms have reduced the negative aspects of the working
environment as much as possible, given the available technology, production processes,
the behaviour of competing firms, availability of raw materials and so on. Firms with poor
working environments simply have to offer a higher wage than those with good quality
working environments. If a firm does not offer a sufficiently high wage given its work
environment, there will be shirking, i.e., the worker is not fully compensated.
2.2. Predictions from the theory
The model predicts that short-term absence or shirking is negatively related to wages,
the unemployment rate, the probability of being detected shirking and the interest rate.
A positive relation is predicted with unemployment benefits, the required effort and the
exogenous quit rate:
shirking (s) = f (w, U , q , b , r , w̄, e ) (4)

− − − + − + +
Sickness absence is a function of the work environment and the unemployment rate only,
as work pressure may be lower in economic downturns:
sickness absence (A) = f (j , U ) (5)

− ?
In addition, it is argued that resources may become available within the firm in economic
slowdowns, and hence sickness absence will be reduced when unemployment increases.
Table 1
Main predictions from the theoretical model on shirking (s) and sickness absence (A), when the quality of the
work environment is changed
Fully compensated Not fully compensated
Improved quality of the work environment s = 0, A < 0 s < 0, A < 0
Degenerated quality of the work environment s = 0, A > 0 s > 0, A > 0
The predictions that result from explicitly including the work environment in the shirking
model are illustrated in Table 1, where s refers to shirking and A is sickness absence or
health-related absence. If the worker is fully compensated for working in a poor work
environment, the quality of the work environment is not an important factor in the shirking
decision. However, if the worker does not receive a wage that reflects the work environment,
shirking will be affected. If a worker is not fully compensated (downward) for a good work
environment, this means that the wage is higher than necessary to prevent shirking, and the
corresponding wage in a poor work environment is too low to prevent shirking. However,
the effect on sickness absence is unambiguous as a good quality work environment implies
lower health-related absence.
The model offers an explanation as to why firms with similar work environments may
experience large differences in shirking. According to these predictions, it is possible to
test whether the workers are fully compensated from adverse working conditions. If a
relationship between voluntary absence and the work environment is found, the theoretical
model suggests that this may be the case because the worker is not fully compensated.
3. Data
Both employer-reported data and survey data are used in this analysis. Employer-reported
data include quarterly information about average absence days, number of employees, wages
and so on. Information about the work environments of these firms is collected in a survey.
Over 1000 different firms have reported both wage and sickness absence data separated
by gender to the Confederation of Norwegian Business and Industry (NHO) for the period
1990–1998.9 There is a large variation in the number of quarters for which these firms
delivered data. The requirement for a firm to be included in our sample was that they had
reported data for at least 20 quarters of the period, either for women or for men. This
criterion was satisfied by about half the firms, i.e., 493 firms.
Two questionnaires were sent to each of these firms, one copy to management and one to
a representative for the employees. The manager responsible for the day-to-day operation
of the firm represented the management. The safety delegate represented the employees and
his or hers role is stated in the Working Environmental Act of 1977. This is a person elected
by the employees and must be an employee of the firm. The duty of the safety delegate is to
safeguard the interest of the employees on the matters concerning their work environment.
9 This is assumed to be a representative sample of the member firms in the NHO, which is the largest employer
organization for firms in the private sector in Norway.
If there are 10 or more employees in the firm, a safety delegate must be elected.10 In firms
with a large number of employees, safety delegates are elected for various areas of work
and a head safety delegate is responsible of coordinating the work of the safety delegates.
In the form we are able to see if the respondent is a safety delegate or a head safety delegate
and we know from a question how many safety delegates the firm has.11
Almost 70% of the firms responded to the questionnaire.12 This resulted in a total sam-
ple of 331 firms.13 In this paper, answers from the safety delegates are used when they
responded, while information from the management is used when the safety delegate did
not respond. About 75% of the observations are based on answers from the safety delegates.
The data used in this paper are quarterly firm data from 326 different firms.14 In general,
larger firms seem to have better routines for reporting data to the NHO. As a result, this
sample is biased towards larger firms which, in general, have higher absence rates. The
average number of employees in the NHO sample (comprising approximately 1000 firms)
is 72.67 (141.44), where the standard deviation is given in parenthesis, compared with 88.79
(160.37) in this sample (326 firms).
In these data, short-term absence (STA) refers to absence spells lasting up to 3 days,
plus absence registered as shirking (see below). Long-term absence (LTA)is sickness spells
lasting 4 days or more. The NHO firms have, on average, 0.688 (0.578) STA days per quarter
for each employee, compared with 0.702 (0.499) in my sample. For LTA, the figures are
4.075 (4.360) for the NHO sample and 4.145 (4.835) in the current sample.
In the theoretical model, worker absence is separated into shirking and sickness absence.
One step on the way to operationalize this is to think of shirking as a voluntary act and
health related absences as an involuntary one. Voluntary absences are thus those incidents
that may be avoided and the theoretical model suggests that a better combination of pay and
working conditions may reduce this type of absence. Involuntary absence is health related
and is unavoidable. There is to some extent accepted that the best estimate of voluntary or
avoidable absence from work is to use absences with short duration and without medical
certification (Iverson and Deery, 2001) as very short absence spells can not be caused of
serious illness.
A certificate from a medical practitioner is required only for LTA. For shorter absence,
a self certification is delivered to the employer, and there is no need for specification of the
reasons for the sick leave. In the period of analysis, each employee was allowed to have
four self certifications of a maximum duration of 3 days within a 12-month-period. On
average, absence registered as shirking (no self certification or medical certificate) accounts
for about 8% of the total STA.15 Absences due to children’s sickness are not included.
10 In this sample, 39 of the firms employ less than 10 blue-collar workers but about half of these still have a safety
delegate.
11 Almost 30% of the firms report that they have more than 10 safety delegates. Among these, about 96% of the
response from the employees representatives comes from the head safety delegate.
12 For detailed information on the implementation of the survey, see Dyrstad and Ose, Notat 1, 2001, ISØ.
13 About 90% of the firms reported wage and absence data in 30 quarters or more.
14 Five firms were excluded as they delivered incomplete information on codes that we required to merge survey
data with data for absence and other factors.

15 The Norwegian sickness benefit system provides 100% of covered earnings, payable from the first day of
incapacity for up to 52 weeks. The employer pays the total cost of cash sickness benefit for the first 16 days
From this, it is reasonable to state that LTA is absence related to ill health with a relatively
small share of shirking. However, STA is absence with a short duration, indicating that
the worker is not seriously ill.16 In addition, it is reasonable to believe that some of the
self-certified absence is not related to ill health and may be classified as shirking. Using
these arguments, STA may be used as a proxy for voluntary absence and LTA is absence
related to ill health. Long-term absence is thus related to the work environment due to the
effect of working conditions on health. We do not know from the data how much of these
absences that are related to the work environment, but other reasons are partly controlled
for using gender specific regressions and including seasonal dummies. In addition, it may
be argued that if health and family situation is spread equally across firms, the variation in
long-term absence will to a great extent be caused by the work environment provided by
the firm.
In addition to information on major organizational changes, such as mergers and re-
locations, questions concerning the physical and psychological work environment will be
included.17
4. Empirical model and estimation approach
The empirical analysis is based on the following model:

log Absi,t = α0 + α1 log Absi,t−1 + α2 log ni,t + α3 Wagei,t + α4 log Unems,t
+ α5 log Ohi,t + α6 Merger i,t−1 + α7 Movingi,t−1 + α8 Phci,t−1
+ α9 Layoff i,t−1 + α9 Workenvi + α9 Dummies + Errorterm (6)
Subscripts i and t refer to firm and time (year and quarter), respectively. Definitions of
variables in Eq. (6) are given in Table 2 and descriptive statistics for the economic variables
are presented in Table 318 while frequencies of the work environment variables are provided
in the Appendix.
From the theoretical model, voluntary and involuntary absence are affected differently by
differences in the work environment. Using the arguments given at the end of the previous
section, STA and LTA are therefore separated into different models, keeping the empirical
specification close to the presented theory. As all variables, except those from the survey,
are separate according to gender, separated models for men and women will be used. This
will result in four different models, all starting with variables included in Eq. (6), and
parsimonious models will be presented together with other specifications.
(extended from 14 to 16 days in 1998). In addition all employers in the private sector must insure the workers
against occupational injuries and illness (Workmans compensation insurance law of 1989).
16 If the absence incident is of a duration of 4 days or more, the first 3 days are of course not included in the
short-term absence figures.

17 In the survey, we included 32 initial questions, with several follow-up questions to capture changes in the
period. However, much of these data will be used in separate analyses.

18 As seen from the table, some of the absence values are zero (2 and 7%, respectively for short and long term
absence) resulting in missing values using logs. This is handled by setting them equal to 0.0001.
Table 2
Variable definitions
Variable Definitions
Economic variables
Abs Average number of days of absence per employee. Source: NHO
Wage Average wage rate for total working hours. This includes normal working hours, overtime work,
piece-work and other increments, deflated by the national consumer price index. Source: NHO
and SSB
Unem Gender specific municipality unemployment rate. Source: Municipality Database
Emp Number of workers in the firm. Source: NHO
Oh Average number of overtime hours worked per employee. Source: NHO
Dummies Set of time dummies each corresponding to the years 1991–1998, and one set of seasonal
dummies (quarters)
Dummies for organizational changes :
Merger During the period 1990–1998 has the company been involved in a merger or demerger? If Yes,
when? This variable is coded 1 the year the merger or demerger occurred and 0 otherwise.
Source: own survey
Moving During the period 1990–1998 has the company been geographically moved/relocated? If Yes,
when? This variable is coded 1 the year the moving occurred and 0 otherwise. Source: own
survey
Prod change During the period 1990–1998 has the company been exposed to major changes in production?
If Yes, when? This variable is coded 1 the year the changes occurred and 0 otherwise. Source:
own survey
Lay offs During the period 1990–1998 has there been lay-offs in the company? If Yes, when? This variable
is coded 1 the year the layoffs occurred and 0 otherwise. Source: own survey
Factors of work environment variables :
Surron Variable from a factor analysis that contains loadings from five different Health, Environment,
and Safety (HES) variables. They all say something about the quality of the physical surroundings
in the working area: accommodation room, ventilation system, lighting equipment, protective
equipment and general order and cleaning. The categories are good, average and bad. See Table
A1 for loadings and scores. Source: own survey
Work strain Variable from a factor analysis that contains loadings from three different HES variables. They all
say something about the physical strain of the work: Extent of monotonous work, heavy/frequent
lifting, and poor work postures. The categories are low, average and high. See Table A.1 for
loadings and scores. Source: own survey
Culture Variable from a factor analysis that contains loadings from three different variables. They all say
something about the firm culture: Evaluation of the co-operation among employees, relationship
of trust between employees and the closest superior, and the potential for employees to influence
their own work situation. The categories are Good, Average and Bad. See Table A for loadings
and scores. Source: own survey
Work environment dummies :
Noise Exposure to noise in the working area: Low, Average, High. Source: own survey
Dust Exposure to dust in the working area: Low, Average, High. Source: own survey
Chem Extent of use of chemicals/volatiles/solvent: Low, Average, High. Source: own source
Acci Extent of accidents and near misses (incidents): Low, Average, High. Source: own survey
In general, there are three types of variables in addition to the time dummies. The
first group includes firm, time and gender specific variables that will be referred to as the
“economic variables”, cf. Table 2, commonly included when investigating absence rates.
The second group is the dummies for organizational changes which all are firm and time
Table 3
Descriptive statistics
Variable Men Women
Obs Mean Std. dev. Min. Max. Obs Mean Std. dev. Min. Max.
Abs: short term 10812 0.68 0.39 0.00 6.24 7535 0.74 0.62 0.00 11.25
Abs: long term 10812 3.38 2.28 0.00 41.50 7535 5.25 5.12 0.00 66.00
Wage 10480 106.94 15.33 71.76 208.05 7218 101.19 14.74 30.67 220.28
Unem 10452 3.98 1.61 0.13 13.44 7342 2.88 0.94 0.00 7.82
Emp 10812 126.59 197.51 1.00 2508.00 7535 34.55 40.69 1.00 283.00
Oh 10431 24.40 69.22 0.00 4485.00 7124 10.25 12.76 0.00 221.00
(year) specific and coded one the year the change took place. These variables are however
included with a lag allowing for a delayed effect of the change on absence rates. The reason
for including these with a lag is partly because we have no information about which quarter
in the year the change occurred, and because potential effects on the work environment from
these changes may be more visible the year after the change when some stability has been
reestablished.19 However, when effects are detected of the lagged dummies, leads and level
dummies have been tried. The third group of variables is the firm specific work environment
variables, see Table 2.
As the work environment variables are firm-specific and time-invariant the standard
deviations in the OLS analysis will be downward-biased, often referred to as Moulton
bias (Moulton, 1987, 1990). This bias is corrected by assuming that the residuals are not
independent within firms, but between firms.20 As the results show these are important
corrections.
Furthermore, some of the environmental variables from the survey are highly correlated,
giving unstable results when many firm-specific dummies are included. A factor analysis
approach is carried out and principal component factors are retained and used in the regres-
sion analysis. Factor loadings are given in the Appendix. The respondents were asked, for
instance, to evaluate the level of fifteen different Health, Environment and Safety (HES)
indicators. Clearly, these will be correlated and have a common expected effect on absence.
The factor analysis groups these variables in an interpretable way, resulting in two factors.
The first is called Surron and contains loadings from five different HES variables, all of
which are indications of physical surroundings in the work area, see Table A.1. In addition,
the factor analysis returned a factor referred to as Workstrain, which is an index of vari-
ables giving the level of monotonous work, heavy/frequent lifting and bad work postures,
also in Table A.1. The other four variables from this part of the survey are included as
dummies with base “good” or “low”, and these are accidents or near misses (Acci), use
of chemicals/volatiles/solvents (Chem), the level of dust in the work area (Dust) and the
19 Obviously, only short-run effects are captured in this specification, and the reason for not using step dummies
has to do with the lack of time variation in the other work environment indicators. Giving the value one for a firm
with a change that occurred in 1990 for the whole period will be the same as including a firm specific dummy
(i.e., as the work environment dummies). However, if the change occurred late in the period no long run effects
would be obtainable either.
20 See the Stata reference manual, option “cluster”.
level of noise (Noise).21 In an attempt to capture psychosocial work environment indicators,

cooperation among the workers, between workers and with their closest superiors, as well
as the possibility for workers to influence their own work situation, are included. These
variables are correlated, so an index is created using factor analysis, and will be referred
to as Culture as these variables may say something about the general tone and involvement
of the employees within the firm. See Table A.2 for more information on this factor and
Table A.3 for frequencies of the work environment variables.
Given the known weaknesses with factor analysis and the fact that the variables that
are grouped into factors are not continuous, the estimated effects are investigated in the
following way: the two factors Surron and Work strain are per definition not correlated.
In an attempt to determine the effect of the variables with high loadings in one factor, the
dummies have been included one by one, along with the other factor where the dummy has
low load. The results of this will be commented upon when discussing the estimated effects.
Fixed Effects models are estimated to account for unobserved heterogeneity. Time-
invariant variables will obviously be excluded in the within transformation, although this
may not be a good approach when trying to identify the effect of the work environment
variables. However, even if the effects will not be estimated, they are implicitly accounted
for in the fixed effects, together with all other firm-specific or time-invariant information.
Fixed Effects models will give the upper limit, or the benchmark specification, of explanatory
power using firm-specific dummies.
5. Results
Four different models are presented in Tables 4 and 7. Tables 4 and 5 give the results for
short-term absence for men (STAm) and women (STAw), respectively. Long-term absence
results are presented in Table 6 for men (LTAm) and Table 7 for women (LTAw). Each
of these tables reports eight different specifications. The first two specifications are OLS
regressions on economic variables only, based on different samples. Specification (I) is
based on all firms that have reported wage and absence data, and (II) is based on firms
that also gave information about the work environment. Specification (III) includes work
environment variables in addition to economic variables. Fixed Effects results are given in
(IV). The parsimonious model of specification (III) is presented in (V) and the same model
is corrected for Moulton bias in (VI). The results reported in (VII) are based on a larger
sample, allowing the sample to increase when reducing the model. This model is corrected
for Moulton bias in (VIII).
Before turning to the different types of absence models, I will give some general com-
ments on the differences in the results across the specifications given in (I)–(VIII). First,
the results given in (I) and (II) are similar, indicating robustness across sample size. When
including work environment variables in (III), the effects from the economic variables are
unaltered, except for the wage effect, which will be commented upon in relation to the
different models below. The Fixed Effects specification is included to show the upper limit
21 The factor analysis resulted in factors where “Acci”, “Dust”, “Noise” and “Chem” were included in one factor
each, thus giving the same result as including them as separated dummies.
176
Table 4
Short-term absence, men
OLS (I) OLS (II) OLS (III) FE (IV) OLS (V) OLS CORR (VI) OLS (VII) OLS CORR (VIII)
Abslag 0.501(39.81)∗∗ 0.465(27.24)∗∗ 0.445(25.77)∗∗ 0.107(7.63)∗∗ 0.449(26.02)∗∗ 0.449(16.72)∗∗ 0.483(35.31)∗∗ 0.483(20.89)∗∗
S.O. Ose / Journal of Health Economics 24 (2005) 161–188

Log empl 0.058(9.76)∗∗ 0.055(7.10)∗∗ 0.053(6.58)∗∗ 0.042(1.46) 0.054(7.01)∗∗ 0.054(3.65)∗∗ 0.060(9.40)∗∗ 0.060(4.79)∗∗
Log wage −0.434(9.28)∗∗ −0.255(4.26)∗∗ −0.171(2.73)∗∗ −0.093(0.68) −0.201(3.39)∗∗ −0.201(1.75) −0.344(7.16)∗∗ −0.344(3.42)∗∗
Unemp rate 0.003(0.70) 0.004(0.87) 0.004(0.81) −0.024(2.60)∗∗ 0.004(0.87) 0.004(0.54) 0.002(0.40) 0.002(0.24)
Log overtime 0.032(5.19)∗∗ 0.029(3.76)∗∗ 0.030(3.77)∗∗ 0.035(3.55)∗∗ 0.032(3.97)∗∗ 0.032(2.98)∗∗ 0.033(4.82)∗∗ 0.033(3.25)∗∗
Merger −0.014(0.37)
Moving 0.111(1.25)
Prodchange −0.008(0.15)
Lay offs −0.050(1.20)
Surron 0.022(2.89)∗∗ 0.017(2.43)∗ 0.017(1.12)
Work strain 0.047(6.36)∗∗ 0.045(6.05)∗∗ 0.045(2.98)∗∗ 0.033(5.77)∗∗ 0.033(2.66)∗∗
Culture −0.012(1.69) −0.014(1.92) −0.014(0.92)
Noise (average) 0.035(1.73) 0.033(1.88) 0.033(0.97) 0.033(2.33)∗ 0.033(1.20)
Noise (high) 0.074(3.23)∗∗ 0.053(2.46)∗ 0.053(1.21) 0.075(4.54)∗∗ 0.075(2.25)∗
Dust (average) −0.008(0.38)
Dust (high) −0.046(1.84)
Chem (average) 0.028(1.88)
Chem (high) 0.008(0.42)
Acci (average) −0.014(0.88)
Acci (high) −0.008(0.34)
Cons 1.560(7.52)∗∗ 0.794(2.93)∗∗ 0.377(1.33) 0.067(0.10) 0.504(1.87) 0.504(0.97) 1.084(5.07)∗∗ 1.084(2.42)∗
Nobs 9168 5312 5312 5312 5312 5312 8180 8180

adjusted R2 0.3891 0.3526 0.3613 0.4909 0.3597 0.3597 0.3885 0.3885
root MSE × 100 47.874 47.245 46.99 41.834 47.006 47.006 47.713 47.713
Dummies for year and quarter are included in all specifications. t-Values are reported in the parentheses based on robust standard errors.
Table 5
Short-term absence, women
Abslag 0.425(27.79)∗∗ 0.370(18.49)∗∗ 0.357(17.83)∗∗ 0.107(5.80)∗∗ 0.363(18.11)∗∗ 0.363(12.66)∗∗ 0.384(21.73)∗∗ 0.384(14.21)∗∗

Log empl 0.008(0.73) 0.016(1.25) 0.021(1.48) −0.180(3.69)∗∗ 0.021(1.55) 0.021(0.77) 0.025(1.98)∗ 0.025(1.01)
Log wage −0.243(3.48)∗∗ −0.206(2.14)∗ −0.101(0.97) −0.269(1.17) −0.123(1.24) −0.123(0.71) −0.198(2.25)∗ −0.198(1.31)
Unemp rate −0.033(3.26)∗∗ −0.062(4.68)∗∗ −0.058(4.33)∗∗ −0.048(1.99)∗ −0.057(4.30)∗∗ −0.057(2.66)∗∗ −0.046(3.72)∗∗ −0.046(2.08)∗
Log overtime 0.039(5.82)∗∗ 0.032(3.58)∗∗ 0.031(3.35)∗∗ 0.034(3.12)∗∗ 0.031(3.52)∗∗ 0.031(2.27)∗ 0.031(3.80)∗∗ 0.031(2.53)∗
Merger 0.028(0.55)
Moving 0.305(2.86)∗∗ 0.297(2.83)∗∗ 0.297(3.34)∗∗
Prodchange 0.007(0.11)
Lay offs 0.095(1.52) 0.132(2.34)∗ 0.132(2.15)∗
Surron −0.006(0.45)
Work strain 0.035(2.88)∗∗ 0.036(3.10)∗∗ 0.036(1.47) 0.028(2.57)∗ 0.028(1.26)
Culture 0.033(2.32)∗
Noise (average) 0.039(1.28) 0.070(2.77)∗∗ 0.070(1.58)
Noise (high) 0.006(0.17) 0.055(1.81) 0.055(1.05)
Dust (average) 0.011(0.39)
Dust (high) −0.045(1.07)
Chem (average) 0.004(0.15) 0.009(0.36) 0.009(0.21)
Chem (high) −0.066(1.98)∗ −0.063(2.09)∗ −0.063(1.24)
Acci (average) −0.026(0.95)
Acci (high) −0.052(0.88)
Cons 1.039(3.40)∗∗ 0.979(2.33)∗ 0.453(0.98) 2.036(1.84) 0.567(1.29) 0.567(0.76) 0.684(1.73) 0.684(1.03)
Nobs 5273 3057 3057 3057 3057 3057 3790 3790

adjusted R2 0.2655 0.2374 0.2451 0.3627 0.2419 0.2419 0.2520 0.2520
root MSE × 100 59.076 58.607 58.455 53.445 58.473 58.473 58.86 58.86
177
178
Table 6
Long-term absence, men
Abslag 0.614(50.23)∗∗ 0.604(37.27)∗∗ 0.597(36.67)∗∗ 0.362(27.31)∗∗ 0.599(36.54)∗∗ 0.599(20.46)∗∗ 0.600(44.35)∗∗ 0.600(23.76)∗∗

Log empl 0.039(5.25)∗∗ 0.031(3.27)∗∗ 0.030(2.97)∗∗ 0.004(0.11) 0.029(3.04)∗∗ 0.029(2.27)∗ 0.041(5.01)∗∗ 0.041(3.81)∗∗
Log wage −0.013(0.26) 0.085(1.36) 0.106(1.61) 0.374(2.19)∗ 0.084(1.33) 0.084(0.92) 0.010(0.18) 0.010(0.13)
Unemp rate 0.008(1.84) 0.012(2.14)∗ 0.014(2.36)∗ −0.003(0.27) 0.013(2.27)∗ 0.013(1.56) 0.008(1.75) 0.008(1.18)
Log overtime 0.009(1.28) 0.014(1.34) 0.010(0.98) −0.023(1.83) 0.010(1.00) 0.010(0.70) 0.012(1.39) 0.012(0.91)
Merger 0.003(0.07)
Moving 0.018(0.11)
Lay offs −0.016(0.37)
Surron 0.014(1.49)
Work strain 0.013(1.61) 0.010(1.23) 0.010(0.85) 0.024(3.52)∗∗ 0.024(2.43)∗
Culture 0.001(0.07)
Noise (average) 0.034(1.43) 0.016(0.75) 0.016(0.49) 0.031(1.72) 0.031(1.13)
Noise (high) 0.071(2.63)∗∗ 0.064(2.54)∗ 0.064(1.84) 0.065(3.15)∗∗ 0.065(2.20)∗
Dust (average) −0.039(1.79)
Dust (high) −0.036(1.29)
Chem. (average) −0.019(1.06) −0.030(2.08)∗ −0.030(1.25)
Chem (high) −0.032(1.41) −0.047(2.51)∗ −0.047(1.73)
Acci (average) 0.010(0.61) 0.005(0.30) 0.005(0.19) 0.009(0.66) 0.009(0.41)
Acci (high) 0.075(3.03)∗∗ 0.071(2.99)∗∗ 0.071(1.72) 0.060(2.82)∗∗ 0.060(1.60)
Cons 0.292(1.32) −0.201(0.72) −0.289(0.98) −0.981(1.20) −0.211(0.75) −0.211(0.52) 0.177(0.72) 0.177(0.50)
Nobs 8935 5207 5207 5207 5207 5207 7550 7550

Adjusted R2 0.4095 0.3971 0.4001 0.4709 0.3992 0.3992 0.4095 0.4095
Root MSE × 100 55.22 54.80 54.743 51.261 54.73 54.73 54.227 54.227
Table 7
Long-term absence, women
Abslag 0.538(35.59)∗∗ 0.534(25.80)∗∗ 0.512(24.41)∗∗ 0.323(18.72)∗∗ 0.515(24.64)∗∗ 0.515(21.30)∗∗ 0.528(27.78)∗∗ 0.528(21.51)∗∗

Log empl −0.016(1.40) −0.013(0.89) −0.028(1.82) −0.205(3.52)∗∗ −0.035(2.34)∗ −0.035(1.59) −0.056(3.83)∗∗ −0.056(2.37)∗
Log wage −0.103(1.21) −0.294(2.50)∗ −0.260(2.04)∗ 0.300(1.09) −0.259(2.20)∗ −0.259(1.37) −0.118(1.09) −0.118(0.64)
Unemp rate −0.009(0.71) −0.015(0.90) −0.014(0.84) −0.073(2.48)∗ −0.018(1.07) −0.018(0.78) −0.018(1.11) −0.018(0.83)
Log overtime 0.014(1.65) 0.006(0.52) 0.001(0.06) 0.004(0.27) −0.001(0.11) −0.001(0.09) 0.012(1.06) 0.012(0.80)
Merger −0.003(0.05)
Moving 0.412(2.69)∗∗ 0.385(2.71)∗∗ 0.385(2.44)∗ 0.451(4.64)∗∗ 0.451(4.06)∗∗
Lay offs 0.127(1.57)
Surron −0.001(0.04)
Work strain −0.002(0.18)
Culture 0.017(0.95)
Noise (average) 0.144(3.91)∗∗ 0.147(4.09)∗∗ 0.147(3.18)∗∗ 0.129(3.93)∗∗ 0.129(2.89)∗∗
Noise (high) 0.055(1.31) 0.071(1.75) 0.071(1.16) 0.026(0.68) 0.026(0.40)
Dust (average) 0.014(0.45) 0.001(0.03) 0.001(0.02) −0.025(0.85) −0.025(0.61)
Dust (high) −0.114(2.23)∗ −0.121(2.65)∗∗ −0.121(1.59) −0.141(3.57)∗∗ −0.141(2.03)∗
Chem (average) −0.025(0.87)
Chem (high) −0.072(1.86)
Acci (average) 0.025(0.79) 0.038(1.24) 0.038(0.83) 0.070(2.36)∗ 0.070(1.62)
Acci (high) 0.159(2.56)∗ 0.165(2.67)∗∗ 0.165(1.89) 0.098(1.63) 0.098(0.88)
Cons 1.260(3.30)∗∗ 2.259(4.19)∗∗ 2.139(3.68)∗∗ 0.855(0.66) 2.157(4.00)∗∗ 2.157(2.54)∗ 1.461(2.97)∗∗ 1.461(1.77)
Nobs 5253 3038 3038 3038 3038 3038 3461 3461

Adjusted R2 0.3145 0.3153 0.3274 0.3905 0.3256 0.3256 0.3411 0.3411
Root MSE × 100 69.05 68.341 67.905 64.320 67.904 67.904 68.315 68.315
Dummies for year and quarter are included in all specifications. t-Values reported in the parentheses are based on robust standard errors.
179
of explanatory power given that the firm-specific effects are allowed to contribute to R2 -adj.
For STAm, the OLS model including work environment variables explains about 74% of
the Fixed Effects specification. For STAw, it explains 68%. For long term absence, OLS
explains 85% and 84% of the Fixed Effects model for men and women, respectively. How-
ever, the explanatory power does not increase much when including work environment
variables in the OLS specification, as shown by a comparison of (II) and (III). Reducing the
models by removing insignificant variables sequentially returns model (VII) when keeping
the sample constant. Allowing the sample to increase when omitting insignificant variables
returns model (VIII). These specifications result in similar results, again indicating robust-
ness. In models (VI) and (VIII), the bias in standard errors in the OLS specifications is
corrected. This correction is conservative in the way that much of the variation within firms
is removed, as shown by comparisons of specification (V) with (VI) and (VII) with (VIII).
In general, the results seem to be fairly robust across specifications.
The interpretation of the coefficients is straightforward. All economic variables are in
logs and their coefficients are therefore elasticities. The work environmental dummies are
also easy to interpret. For instance, the effect on absence of the Noise variable in Table 4
has the following effect: Noise = 1 is the base, indicating low exposure to noise in the
firm. Average noise exposure, Noise (Average) is set to unity when average level and zero
otherwise. The coefficient is estimated to 0.035 so that firms with an average exposure have,
on average, 3.5% higher short-term absence for men compared to firms with low exposure.
The dummy Noise (High) indicates much noise exposure in the firm, and these firms have
on average 7.4% higher short-term absence for men compared to firms with low exposure
to noise.
From the theoretical model, short-term absence is assumed to be affected by economic
variables to a larger extent than absence of longer duration. The effects of the working
environment on short-term absence will, according to the theoretical model, suggest that
the worker is not fully compensated.
Long-term absence, according to the theoretical model, is mainly determined by working
conditions and the unemployment rate. However, the effect of the unemployment rate is
originated from quite different arguments than for absence not related to ill health.
The estimation results for short-term absence for men are given in Table 4. Following the
theory, an increase in wages will reduce absence that is not related to ill health. The estimated
wage effect is negative and statistically significant. When the work environment variables
are included in the model, the wage effect is lower, indicating that it is over-estimated in
absence models that do not control for working conditions. This is shown by a comparison
of models (II) and (III).
The effect of firm size is positive in STAm, indicating that larger firms have a higher
absence rate per employee than smaller firms. These results are commonly found in other
empirical studies (Allen, 1981; Barmby and Stephan, 2000). Barmby and Stephan (2000)
argue that larger firms have lower costs of insuring against absenteeism, so that larger firms
may have higher absence rates in equilibrium compared with smaller firms. This effect may
however be an indication of a different system of monitoring between larger and smaller
firms. It may be more difficult to monitor a large staff, while small firms may have more
“informal monitoring”. If this is the case, it is in line with the theoretical model’s implication
that increased monitoring reduces absence. Overtime per employee increases STAm, but
this may be an endogenous effect as increased absence may lead to more overtime work.
The overtime effect is robust through specifications.22 The unemployment rate is a measure
of local labor market tightness and should work as a disciplinary device. According to this,
STAm is not disciplined by the unemployment rate.
Major changes, such as mergers, relocation, changes in production and layoffs, do not
affect short-term absence for men.
There are indications that the variable Surron, which measures the physical surroundings
of the workplace (see Table A.1), has an effect on STAm.23 However, the effect if this
variable is not statistically significant when correcting for Moulton bias. Work strain is the
index of monotonous work, heavy or frequent lifting and poor work postures. The effect
is highly significant and positive, as expected. Increased work strain results in increased
short-term absence.24
According to the theoretical model, this effect is a result of insufficient compensation for
workers. The same effect is present in noisy work environments, as firms with high levels
of noise do not compensate their workers sufficiently and experience a higher short-term
absence for men. None of the other working conditions have statistically significant effects
on this type of absence, indicating perfect compensation for these disamenities.25
Turning to short-term absence for women, (STAw), different effects are found. First, less
of the variation in STAw is explained by economic variables when these are judged on the
basis of their explanatory power. The wage elasticity is negative and the estimate is reduced
both in power and size when working conditions are included. This might be an effect
of compensating wage differentials, as some of the variation in wages is correlated with
elements from the working environment. As mentioned above, this supports the statement
that omitting working conditions would give biased wage effects. As opposed to STAm, the
unemployment effect is negative, statistically significant and robust across specifications.
This indicates that women are more disciplined by fluctuations in the labor market than
men.
The firm size effect is positive, indicating higher short-term absence for women in larger
firms. A specification involving the share of women in the firm was undertaken, but was
not found to affect short-term absence for either women or men.26 The overtime elasticity
is similar to STAm.
Looking at the effect of the organizational changes, geographical relocation of the firm
implies higher short-term absence for women. However, only 19 of the firms have been
22 When omitting the overtime variable, none of the other effects are altered.
23 This is a factor with five variables, and I have included these separately, one by one, in model (IV) in order to
identify the relevant effects. I find positive estimates for all variables, but only the variable indicating the level of
general order and cleaning is found to be statistically significant.
24 The result of including the variables in the factor Work strain one by one, is that all are found to be positive
and statistically significant. This is also the case when correcting for Moulton bias.
25 As explained in the data section, the safety delegate has responded for 75% of the firms while the management
provided the rest of the cases. To test the robustness against this possible bias, all regressions have been rerun
including only the firms where the safety delegates responded. Deviation from the presented results was found
only for short-term absence for men, where also accidents (Acci = 2) increased absence. Otherwise, the results
were robust also against this sample reduction.
26 The average share of female workers is 21%.
relocated which may explain non-robustness across sample size. When including more
observations, the results indicate that periods of layoffs increase this type of absence.27
This result is also robust against corrections of the standard deviations. STAw is higher
the quarter after the layoffs occurred (included with a four quarter/year lag). Testing this
more thoroughly, findings indicate effects in both the period of the layoff and the following
periods. While it is difficult to say what causes this effect, one explanation is that women
are more stressed by the uncertainty related to layoffs. Another interpretation is that women
work extra hard in order to avoid being laid off, resulting in the need for a few days off.
As for men, the effect of work strain is strong and significant in STAw, as is the noise
exposure. Including one by one of the variables in the factor Work strain gives that the
result is dominated by the effect of monotonous work. The Culture variable is found to
be positive and borderline significant. Investigating this further indicate that the important
variable is the cooperation among the employees. In addition, frequent use of chemicals,
volatile substances and solvents may reduce short-term absence. This may, according to the
theoretical model, indicate that women are overcompensated for such hazards. However,
the effect is only borderline significant. It diminishes when the standard deviations are
corrected and disappears altogether when the sample size is increased.
Table 6 gives the results of the estimation of long-term absence for men (LTAm). In
line with the theoretical model, most of the economic variables are insignificant. This
prevails through different specifications. Firm size is however estimated to be positive
and robust across specifications. According to the theory, long-term absence is related to
work environment and to the unemployment rate. Firm size may be related to the work
environment, as it is argued that larger firms have greater work environment challenges
than smaller firms. However, the firm size effect may indicate the presence of composition
effects as larger firms have more heterogeneous workforces, which have a higher probability
of including so-called marginal workers.
The unemployment elasticity, which is borderline significant, is positive. As unemploy-
ment increases, long-term absence for men increases. In the theoretical model this effect
was ambiguous. This result supports the hypothesis that periods of increased unemploy-
ment will reduce the workers’ demands for improved working conditions because they are
afraid of increasing the firm’s costs during economic downturns. If the firm’s costs rise, the
probability that workers will be made redundant may increase. In extreme cases, workers
may accept a decline in the maintenance of the working environment in difficult times, thus
becoming exposed to a higher risk of illness. The sickpay scheme may also be important in
this case. As the employer period in Norway in the period was 14 days, a reimbursement
scheme/system provides repayments to the firm if the worker was sick longer than 14 days.
This may affect the behaviour of both the worker and the medical doctor who provides the
worker with medical certificate (when sick more than 4 days). The worker and the medical
doctor may in this way together contribute to help the local employer (high unemployment
equal bad times) to overcome economic downturns as it would be better for the firm to re-
ceive refundment for sickpay than paying wage for redundant workers (the dismissal rules
are rather strict). The effect does not differ greatly across specifications.
27 A total of 86 firms were exposed to layoffs in the period and the occurrence is not clustered in any year. Only
one firm had both layoffs and relocation.
Non of the organizational changes seem to affect the male absence the year after the
change took place. However, when including leads and level, major changes in the pro-
duction process seem to increase LTAm the year the change took place. This may be due
to changes in the work environment, a feeling of not coping the new job tasks, general
frustration etc. The effect is however not present the year after, indicating relative short
lasting adjustment problems.
Several of the work environment indicators have an effect on LTAm. Work strain and
noise level are important in this absence type. According to the theory, these elements of the
work environment will cause short-term absence if workers are not compensated. They also
cause sickness absence due to work hazards. Furthermore, firms that make extensive use of
chemicals, volatile substances and solvents experience a lower LTAm. A work environment
where these elements are present would probably have high standards for safety and detailed
routines for the use of protective equipment, which reduce the hazards associated with the
use of chemicals and so on. This may also be an indication of sorting effects because these
firms may attract workers who are well suited to working in firms characterized by this type
of work environment. In addition, the factor Surron include variables that have effect on
absence without resulting in a significant effect from the factor itself, indicating weaknesses
with the method. The important variable is the quality of the lightning equipment, giving
that firms with good lightning equipment have, on average, 4.6% lower long term absence
for men compared to firms with average equipment.
If the incidence of accidents is high, LTAm is relatively higher. This strengthens the
arguments from the theoretical model. Work hazard may cause work-related injuries and
then sickness absence. Absence of a shorter duration did not experience any effect from the
incidence of accidents.
Table 7 gives the estimation results of long-term absence for women (LTAw). As for men,
the economic variables seem to have little influence on long-term absence. However, the firm
size effect is different in comparison with other types of absence as the elasticity is negative.
An explanation may be that larger firms put in a greater effort to reduce the highest type of
absence, which is often long-term absence for women. A specification based on the share of
female employees has been tried, and the effect is estimated to be statistically negative. This
indicates that firms with a relatively greater proportion of women reduce long-term absence
for female workers. However, when including work environment variables, the effect is no
longer present. One could argue that the share of female workers may be related to the
quality of the work environment. This is because males typically dominate the occupations
associated with hard manual work and high levels of noise and dust. Thus, this would indicate
that there are better physical work environments in firms with a high share of female workers.
Turning to the work environment variables, several of these affect LTAw. If the firms had
been exposed to major changes, such as geographical relocation, LTAw was higher in the
period before moving. However, including the dummy in the same period that the moving
occurred also result in higher LTAw. Including the dummy with a lead gives a negative
effect, indicating lower absence the period before the moving took place. This may be a
disciplinary effect as the workers fear for their jobs in the process and therefore goes to
work even if they are sick (presenteeism). The increased absence the year of the change
and the year after, may support this hypothesis. It is interesting to note that no effect from
work strain is found in this type of absence. This effect is highly robust and the variable is
not correlated with any of the other included variables. This supports the above explanation
regarding the negative effect on long-term absence relating to the share of female workers
in the firm. Furthermore, the exposure to noise increases LTAw, while use of chemicals
and other hazards reduce the absence level. This effect is the same as for men and may be
interpreted in the same way. In addition, the degree of accidents or near misses seem to
affect LTAw, as found for men.
Several important variables concerning the workforce, such as average age of the em-
ployees, education level, tenure and unionization are not included because information is
not available. However, arguing that the different industries includes rather homogenous
groups according to skills and educational level, unionization and so forth, including indus-
try dummies may be a way of incorporating these variables.
This has been done and 12 different industry dummies have been used. The results
indicate that including these does not explain much more of the variation in short and
long-term absence than the previously included variables. However, some changes in the
results are found and the effects are different when including only the economic variables
compared to including also the work environment variables.
When including industry dummies, the wage effect is the only one that changes. For
STA the wage effect is no longer significant and close to zero for both men and women. For
LTAm, the wage effect becomes strong and significant when including industry dummies
together with the economic and work environment variables. The opposite is observed for
women as the wage effect is no longer statistically significant and the elasticity is close to
zero. The other results are generally not altered. Including industry dummies in the absence
regressions along with economic variables only, does not change the elasticities, indicating
no serious correlation between wage and industry dummies.
However, the results indicate correlation between industry dummies, wages and the qual-
ity of the work environment. This may support the results presented earlier as the relationship
between the quality of the work environment and wage may be compensation effects.28
In total, the theoretical model is supported, as the economic variables are found to have
little influence on long-term absence. The work environment seems to affect both short
and long-term absence, but indicated by the effect of the economic variables, for different
reasons.
6. Conclusions
The theoretical model given in this paper offers different predictions on voluntary absence
and absence related to ill health. It indicates that failing to fully compensate workers for
adverse working conditions will result in increased voluntary absence. The model is based
on the Shapiro and Stiglitz shirking model with working conditions explicitly included. By
including working conditions, the model also includes absence that is related to ill health
due to a poor work environment. Furthermore, the theoretical model suggests that although
28 Simple wage regressions where industry dummies and the work environment variables are used as explanatory
variables support the results that the workers are not fully compensated for poor quality of the work environment
as these are found to be related to wages in a negative way. For instance, higher work strain indicates lower pay.
This is however related to another paper in progress.
the economic variables influence voluntary absence, they do not influence absence of a
longer duration or sickness absence to the same extent.
Utilizing quarterly firm level data from the period 1990–1998, the empirical results
support the theoretical predictions. The findings indicate that neither women nor men
are fully compensated from working in areas with high levels of noise. Moreover, firms
with a high degree of monotonous work, heavy or frequent lifting and poor work postures
experience higher short-term absence per employee, indicating that the workers are not
fully compensated for these disamenities. In addition, presence of these work environment
characteristics also indicate higher absence of a longer duration for men, which we interpret
as an ill-health effect. Long-term absence for female workers is not related to physical
work strain, but increases with the level of noise. Accidents and near misses increase
long-term absence for both men and women.
Incidents of layoffs increase women’s short-term absence and geographical moving has
a similar effect. Relocations of firms are believed to lead to work environment challenges.
Female long-term absence is the only type of absence that is affected by relocations, indi-
cating the measures that should be taken when planning relocation of a firm. Male absence
does not seem to be affected by organizational changes to a wide extent.
Including work environment variables reduces the size of the wage elasticity on short-
term absence for both women and men, and the effect is no longer statistically significant
for female absence. This means that the wage effect is overestimated when not controlling
for working conditions.
In general, the theoretical model is supported by the empirical findings. The predicted
negative wage effects are found for short-term absence. No wage effect is present for the
long-term absence, which is also in accordance with the theoretical model. Increased un-
employment rate reduces short-term absenteeism, which is also in line with the theoretical
predictions. Increased use of overtime increases short-term absence. Use of overtime may be
a measure of required work-effort and thus in accordance with the theoretical model. Further-
more, the theoretical model predicts that the economic variables should not affect the long-
term absence or health related absence. The empirical results support this prediction. How-
ever, it is important to emphasize that I have only looked at private firms in the manufacturing
industry. The public sector and the health sector are not included, but it is reasonable to be-
lieve that the composition mechanisms outlined in this study may be strong in these sectors.
Future research on worker absence should involve work environment issues. Identifying
these effects, both theoretically and empirically, is of major interest to firms considering
investment in work environments and to authorities regulating working conditions. For
instance, the results indicate that firms characterized with much noise and high physical
strain should pay the workers more to avoid shirking. In periods of layoffs, special attention
should be put on the situation for female workers as their short-term absence rates tend to
increase in periods after layoffs. In addition, if firms plan to relocate, they should be aware
of the risk for higher health related absenteeism for their female workers. High physical
strain also indicates higher long-term absence, and obviously, effort to reduce the strain
may improve the workers health.
Overall, the results tell us that investing in the work environment may affect absence
rates and given the costs of absenteeism it may be profitable to pay the worker higher wage
to compensate for poor working conditions.
Acknowledgments
I am grateful to the Confederation of Norwegian Business and Industry (NHO) for
financial support and to Kåre Johansen and Jan Morten Dyrstad for useful discussions and
comments. In addition, two anonymous referees contributed to improvements.
Appendix A
Table A.1
Loadings for the factors Surron and Work strain
Variable Factor loadings
Surron Work strain
Accomodation facilities 0.653 0.181
Ventilation system 0.665 0.151
Lighting equipment 0.568 0.113
Protective equipment 0.655 −0.039
General order and cleaning 0.737 0.064
Monotonious work −0.101 0.655
Heavy/frequent lifting 0.213 0.688
Poor work postures 0.101 0.810
Variance/total variance (%) 31 17
Source: own survey. Every question is given with three alternatives, coded 1, 2, and 3 where 1 indicates high
quality or low exposure.
Table A.2
Loadings for the factor Culture
Variable Factor loadings (Culture)
Cooperation among the employees 0.715
Relationship of trust between employees and the closest superior 0.821
Potential for employees to influence their own work situation 0.726
Variance/total variance (%) 57
Source: own survey. Every questions is given with three alternatives: Good, Average, and Bad.
Table A.3
Frequencies of the work environment variables
Low exposure/ Average exposure/ High exposure/ Number of
high quality average quality poor quality firms
Noise in the working area 66 172 76 314
Dust in the working area 87 133 77 297
Chemicals/volatiles/solvents 157 102 42 301
Accidents or near misses 185 112 18 315
Accomodation facilities 161 142 16 319
Ventilation system 91 152 69 312
Lighting equipment 188 125 7 320
Protective equipment 254 63 4 321
General order and cleaning 124 169 28 321
Monotonious work 72 160 83 315
Heavy/frequent lifting 94 160 63 317
Poor work postures 69 177 68 314
Cooperation among the employees 240 85 1 326
Relationship of trust between 165 153 7 325
employees and the closest superior
Potential for employees to influence their 114 174 37 325
own work situation
References
Akerlof, G.A., Yellen, J.L., 1986. Efficiency Wage Models of the Labor Market. Cambridge University Press,
Cambridge.
Alexanderson, K., 1998. Sickness absence: a review of performed studies with focus on level of exposures and
theories utilized. Scandinavian Journal of Social Medicine 26 (4), 241–249.
Allen, S.G., 1981. Compensation, safety and absenteeism: evidence from the paper industry. Industrial and Labor
Relations Review 37, 207–218.
Aronsson, G., Gustafsson, K., Margareta, D., 2000. Sick but yet at work. An empirical study of sickness presen-
teeism. Journal of Eidemiology and Community Health 54, 502–509.
Barmby, T., Stephan, G., 2000. Worker absenteeism: why firm size may matter. The Manchester School 68 (5),
568–577.
Brown, S., Session, J.G., 1996. The economics of absence: theory and evidence. Journal of Economic Surveys 10
(1), 23–53.
Brown, S., Fakhfakh, F., Session, J.G., 1999. Absenteeism and employee sharing: an empirical analysis based on
French panel data, 1981–1991. Industrial and Labor Relation Review 52 (2), 234–251.
Cousineau, J., Lacroix, R., Girard, A., 1992. Occupational hazard and wage compensating differentials. Review
of Economics and Statistics LXXIII, 166–169.
Dorman, P., Hagstrom, P., 1998. Wage compensation for dangerous work revisited. Industrial and Labor Relations
Review 52 (1), 116–135.
Drago, R., Wooden, M., 1992. The determinants of labor absence: economic factors and work group norms across
countries. Industrial and Labor Relation Review 45, 764–778.
Harrison, D.A., Martocchio, J.J., 1998. Time for absenteeism: a 20-year review of origins, offshoots, and outcomes.
Journal of Management 24 (3), 305–350.
Iverson, R.D., Deery, S.J., 2001. Understanding the “personological” basis of employee withdrawal: the influence
of affective dispositon on employee tardiness, early departure, and absenteeism. Journal of Applied Psychology
86 (5), 856–866.
Johansson, P., Palme, M., 1996. Do economic incentives affect work absence? Empirical evidence using Swedish
micro data. Journal of Public Economics 59, 195–218.
Karasek, R., Theorell, T., 1990. Jobs of the Future and the Global Economy. In Healthy Work, Stress, Productivity,
and the Reconstruction of Working Life. Basic Books, New York, pp. 302–334.
Lund, T., Iversen, L., Poulsen, K.B., 2001. Work environment factors, health, lifestyle and marital status as
predictors of job change and early retirement in physically heavy occupations. American Journal of Industrial
Medicine 40, 161–169.
Marmot, M.M., Feeney, A., Shipley, M., 1995. Sickness absence as a measure of health status and functioning:
from the UK Whitehall II study. Journal of Epidemiology Community Health 49, 124–30.
Martinello, F., Meng, R., 1992. Workplace risk and the value of hazard avoidance. Canadian Journal of Economics
25, 333–345.
Meng, R., Smith, D.A., 1999. The impact of workers’ compensation on wage premiums for job hazards. Applied
Economics 31, 1101–1108.
Moulton, B.R., 1987. Diagnostics for group effects in regression analysis. Journal of Business and Economic
Statistics 5, 275–282.
Moulton, B.R., 1990. An illustration of a pitfall in estimating the effects of aggregate variables on micro units.
The Review of Economics and Statistics 72 (2), 334–338.
National Insurance Administration, 2001. Basisrapport, 2001 (Basis-report), Rikstrygdeverket.
Shapiro, C., Stiglitz, J., 1984. Equilibrium unemployment as a worker discipline device. The American Economic
Review 74 (3), 433–444.
Siegrist, J., 2001. A theory of occupational stress. In: Dunham, J. (Ed.), Stress in the Workplace: Past Present and
Future. Whurr Publishers, London, pp. 52–56.
Strand, J., 1987. Unemployment as a discipline device with heterogeneous labor. American Economic Review 77,
489–493.
Vahtera, J., Kivimäki, M., Pentti, J., Theorell, T., 2000. Effect of change in the psychosocial work environment
on sickness absence: a seven year follow-up of initially healthy employees. Journal of Epidemiology and
Community Health 54, 484–493.
Weiss, A., 1985. Absenteeism and wages. Economic Letters 19, 277–279.

Working Conditions, Compensation and Absenteeism

Uploaded by

Copyright:

Available Formats

Working Conditions, Compensation and Absenteeism

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Working Conditions, Compensation and Absenteeism

Uploaded by

Copyright:

Available Formats

Journal of Health Economics 24 (2005) 161–188

Working conditions, compensation and absenteeism

Keywords: Compensation; Working environment; Health; Absenteeism; Efficiency wages

fax: +47 73 596361.

2.1. The effort choice of workers

2.1.1. The worker is fully compensated

In addition, it is necessary to differentiate between good and poor working environments.

2.1.2. The worker is not fully compensated

leading to lower wages and more shirking.

2.2. Predictions from the theory

shirking (s) = f (w, U , q , b , r , w̄, e ) (4)

sickness absence (A) = f (j , U ) (5)

data with data for absence and other factors.

4. Empirical model and estimation approach

The empirical analysis is based on the following model:

short-term absence figures.

period. However, much of these data will be used in separate analyses.

level of noise (Noise).21 In an attempt to capture psychosocial work environment indicators,

S.O. Ose / Journal of Health Economics 24 (2005) 161–188

Nobs 9168 5312 5312 5312 5312 5312 8180 8180

S.O. Ose / Journal of Health Economics 24 (2005) 161–188

Nobs 5273 3057 3057 3057 3057 3057 3790 3790

S.O. Ose / Journal of Health Economics 24 (2005) 161–188

Nobs 8935 5207 5207 5207 5207 5207 7550 7550

S.O. Ose / Journal of Health Economics 24 (2005) 161–188

Nobs 5253 3038 3038 3038 3038 3038 3461 3461

You might also like