American Journal of Epidemiology
© The Author 2015. Published by Oxford University Press on behalf of the Johns Hopkins Bloomberg School of
Public Health. All rights reserved. For permissions, please e-mail: [email protected].
Vol. 181, No. 10
DOI: 10.1093/aje/kwu355
Advance Access publication:
April 3, 2015
Original Contribution
Sickness Absence and Psychosocial Job Quality: An Analysis From a
Longitudinal Survey of Working Australians, 2005–2012
Allison Milner*, Peter Butterworth, Rebecca Bentley, Anne M. Kavanagh, and
Anthony D. LaMontagne
* Correspondence to Dr. Allison Milner, McCaughey VicHealth Centre for Community Wellbeing, Melbourne School of Population and Global
Health, Level 5, 207 Bouverie Street, University of Melbourne, Melbourne, VIC 3010, Australia (e-mail: [email protected]).
Initially submitted June 19, 2014; accepted for publication December 1, 2014.
Sickness absence is associated with adverse health, organizational, and societal outcomes. Using data from a
longitudinal cohort study of working Australians (the Household, Income and Labour Dynamics in Australia (HILDA)
Survey), we examined the relationship between changes in individuals’ overall psychosocial job quality and variation in sickness absence. The outcome variables were paid sickness absence (yes/no) and number of days of paid
sickness absence in the past year (2005–2012). The main exposure variable was psychosocial job quality, measured using a psychosocial job quality index (levels of job control, demands and complexity, insecurity, and perceptions of unfair pay). Analysis was conducted using longitudinal fixed-effects logistic regression models and negative
binomial regression models. There was a dose-response relationship between the number of psychosocial job
stressors reported by an individual and the odds of paid sickness absence (1 adversity: odds ratio (OR) = 1.26,
95% confidence interval (CI): 1.09, 1.45 (P = 0.002); 2 adversities: OR = 1.28, 95% CI: 1.09, 1.51 (P = 0.002); ≥3
adversities: OR = 1.58, 95% CI: 1.29, 1.94 (P < 0.001)). The negative binomial regression models also indicated
that respondents reported a greater number of days of sickness absence in response to worsening psychosocial
job quality. These results suggest that workplace interventions aiming to improve the quality of work could help reduce sickness absence.
cohort studies; employment; job quality; job stressors; psychosocial factors; sickness absence
Abbreviations: CI, confidence interval; HILDA, Household, Income and Labour Dynamics in Australia; IRR, incidence rate ratio; OR,
odds ratio.
rather than examining measures of the overall psychosocial
quality of a job. Aside from this, past studies in this area
have been based on specific occupational or organizational
contexts (2, 6, 8), and results may not have been generalizable
across skill levels or occupational groups in different working
populations. Further, we are not aware of other studies of the
job stressor–sickness absence association based on the experience of the same individual over time using fixed-effects
regression models—a more causally robust approach than
comparing people with one another in between-person analysis, which is more prone to confounding due to stable
within-person characteristics (9).
We sought to examine the relationship between changes in
an individual’s overall psychosocial job quality and variation
in sickness absence using data from a longitudinal cohort
Absence from work due to sickness results in millions of
lost working days and high health- and productivity-related
expenditures each year (1), as well as being associated with
greater risk of ill health and mortality (2, 3). The main predictors of sickness absence reflect both individual (e.g., relationship status, health conditions) and workplace (e.g., low job
control, lack of perceived fairness at work) factors (4, 5). From
a population health perspective, identifying the modifiable
factors that contribute to sickness absence and intervening
in these factors is likely to have benefits for both employers
and society.
Despite a number of previous studies on sickness absence,
there are still gaps in current knowledge, particularly with regard to the work environment. This is because previous studies have usually examined job stressors individually (6, 7)
781
Am J Epidemiol. 2015;181(10):781–788
782 Milner et al.
study of working Australians. We used the psychosocial job
quality measure developed by Butterworth et al. (10–12) and
extended 1- or 2-stressor models, such as job strain (the combination of high demands and low control), by including
other prominent psychosocial working conditions in contemporary employment, such as job insecurity and perceptions of
unfair pay relative to effort/contribution. The measure was
developed and validated for use in the Australian context
and has been studied in relation to mental health (10) and
physical functioning (11). The current study represents an
important and novel extension of this work by examining
whether this scale of psychosocial job quality is also a predictor of sickness absence. The specific questions we addressed
in this study were: 1) Does the psychosocial quality of a job
influence whether a person takes paid sickness leave? and
2) Does the psychosocial quality of a job influence the amount
of paid sickness leave a person takes? Further, because a
number of studies have found different relationships between
psychosocial job stressors and sickness absence for males
and females (6, 13), we also investigated whether the relationship between psychosocial job quality and sickness absence was modified by sex.
METHODS
Data source
The Household, Income and Labour Dynamics in Australia
(HILDA) Survey is a longitudinal, nationally representative
study of Australian households established in 2001. It collects detailed information annually from over 13,000 individuals within over 7,000 households (14). The survey covers a
range of dimensions, including social, demographic, health,
and economic conditions, using a combination of face-toface interviews with trained interviewers and a self-completion
questionnaire. Additional information on the HILDA cohort
can be seen in the report by Wilkins (14) and in the Web Appendix (available at http://aje.oxfordjournals.org/).
Outcome variable
The primary outcome in this study was a self-reported measure of whether the respondent had had any paid sickness absences in the 12 months prior to the interview and previous to
the last HILDA survey, as identified through the question,
“During the last 12 months, have you taken any paid sick
leave?” The secondary outcome was a self-reported measure
of the number of days of paid sick leave a person had taken
in the 12 months prior to the interview and previous to the
last HILDA survey. This was ascertained through the question,
“How many days did you spend on paid sick leave (in the past
12 months)?” This variable was used to calculate an incidence
rate ratio (number of sick days in the past year/total number of
days in the past year). Information on paid sickness absence is
available in the HILDA data set from 2005 onwards.
Exposure variable
Using the measures of psychosocial job characteristics
available in the HILDA Survey ( job control, demands and
complexity, job insecurity, and perceptions of unfair pay),
we constructed a multidimensional indicator of overall job
quality. Full details on the construction and validation of the
job quality measure are presented elsewhere (10–12) and can
be seen in the Web Appendix.
Other variables
Variables controlled for in the analytical models were selected on the basis of past research (4, 5). Only variables
identified as possible causes of both psychosocial job quality
and sickness absence were included in analysis (confounders); possible mediators lying on the causal pathway were excluded (see Web Figure 1). The variables included in the
analysis were age (measured continuously in years as a timevarying factor) and household structure (couple or lone adult
residing with dependents, couple without dependents, lone
individual without dependents, or group or multiple-person
household). The latter variable provided an indication of both
relationship status and the presence of children within a
household. We also included employment arrangement (permanent employment vs. fixed-term contract), occupational
skill level (low, low-medium, medium-high, or high, according to Australian and New Zealand Standard Classification
of Occupations groupings (15)), educational attainment (no
completion of high school, completion of high school, vocational certificate/diploma, or bachelor’s degree or above),
equivalized annual household income (in tertiles), and presence of a long-term health condition, disability, or impairment
(yes/no).
Analytical sample
In HILDA, about 70% of permanent employees and 63.1%
of fixed-term contract employees reported having a paid sickness absence in the past year (fixed-term contract workers
tend to have working conditions similar to those of permanent
workers, with the exception of a specified end date) (16), as
compared with 7.2% of casual or temporary workers and
6.8% of self-employed persons. This reflects the fact that most
casual employees and self-employed persons in Australia are
not entitled to the same paid employment benefits (such as
paid sick leave) as permanent employees (17). Because the
main outcome in this study was paid sickness absence, casual
employees and self-employed persons were excluded from
the analysis. A flow chart outlining the steps used to select the
final sample is presented in Figure 1. Information on missing
data can be seen in the Web Appendix.
Analysis
Because items assessing paid sickness absence were first
introduced to the HILDA Survey in 2005, the analysis included 8 waves of data collected annually between 2005 and
2012. Longitudinal fixed-effects logistic regression models
were used to estimate the association between psychosocial
job quality and the likelihood that a respondent had had at
least 1 paid sickness absence (yes or no) in the previous 12
months. Coefficients were transformed into odds ratios to
aid interpretability. Our analytical focus on within-person
Am J Epidemiol. 2015;181(10):781–788
Sickness Absence and Psychosocial Job Quality 783
Entire HILDA Cohort
26,965 persons
165,299 observations
Employed, With Information on Job
Quality Measure and Data on Paid
Sickness Absence
15,558 persons
55,106 observations
Had Data on Paid Sickness
Absence (Yes/No) in Past Year
15,588 persons
55,106 observations
Logistic Regression Analysis
2,991 persons
13,895 observations
Had Data on No. of Days of Paid
Sickness Absence in Past Year
13,944 persons
47,617 observations
Negative Binomial
Regression Analysis
10,091 persons
31,224 observations
Figure 1. Selection of the study sample into different logistic regression models, Household, Income and Labour Dynamics in Australia
(HILDA) Survey, 2005–2012. Longitudinal fixed-effects logistic regression models were used to estimate the association between psychosocial job quality and the likelihood that a respondent had had at least
1 day of paid sickness absence (yes or no) in the previous 12 months.
Negative binomial regression models were used to examine the number
of days of paid sickness absence in the past year. Only persons with
complete data on all relevant covariates were included in the analyses.
associations over time in the survey meant that each individual included in the HILDA data set acted as his or her own
control (e.g., 1, 2, and 3 or more adversities were compared
with having a psychosocially optimal job) (see Web Table 1).
This provided estimates that were not confounded by timeinvariant personal, demographic, and environmental factors,
which were statistically removed from the model (9). Hence,
to be included in the logistic regression analysis, respondents
had to have at least 1 wave each of “yes” and “no” responses
to the question, “During the last 12 months, have you taken
any paid sick leave?” We controlled for time-variant factors by
including relevant covariates in the fixed-effects models.
We tested the possibility that the relationship between psychosocial job stressors and sickness absence differed for
males and females by assessing the significance of the interaction terms and the results of a likelihood ratio test. We also
evaluated possible effect modification by age (assessed in
10-year age groups) as well as by whether a person worked
full-time or part-time, in order to assess the robustness of the
relationship between psychosocial job quality and sickness
absence. We did this by including an interaction term in the
logistic regression models and comparing this against a model
with main associations only. As above, we used the likelihood
ratio test and inspected individual interaction terms in assessing evidence of an interaction.
We then modeled the number of days of sickness absence
during the year. Negative binominal regression was used to
Am J Epidemiol. 2015;181(10):781–788
model the relationship between sickness absence and job
quality, since there was evidence of overdispersion in the outcome variable. Because of problems involved in fitting fixed
effects in negative binominal panel models in commercial
statistical software packages such as STATA (StataCorp LP,
College Station, Texas) (18), we generated the difference
scores manually (i.e., deviation from each person’s mean
score for all time-varying covariates (19)) and implemented
the model as a random-effects model. This produces estimates consistent with those of a fixed-effects model but avoids
the incidental parameters problem, as it does not require the
estimation of individual-specific parameters (19). Coefficients were transformed into incidence rate ratios to ease interpretation. As a sensitivity analysis, we characterized the
difference of the between-person (by including the mean
value of each variable in the regression analysis) and randomeffects models from the fixed-effects within-persons approach described above (Web Tables 2 and 3).
For both the logistic and negative binominal models, we
first derived results from a simple model representing the
association between the key exposure variable and the outcome, before adjusting for other relevant confounders. Analyses were performed using STATA, version 12.1 (18).
RESULTS
Table 1 provides information on the demographic characteristics of the analytical sample at entry into the HILDA Survey (i.e., baseline). As explained above, to be included in the
logistic regression analysis, respondents had to have experienced a change in sickness absence—that is, reported both
“yes” and “no” to the question, “During the last 12 months,
have you taken any paid sick leave?” There were slightly
more males in this sample than females, and most workers
were of a higher skill level. At baseline, 50% of this group
reported having 1 psychosocial job stressor, approximately
35% were employed in high-skill jobs, and the majority
were living with no dependents. Most people in this group
were employed permanently at baseline.
Table 1 also shows the characteristics of persons included
in the negative binomial analysis examining the number of
days of paid sickness absence in the past year. These models
were able to include persons with a sickness absence count of
zero, which explains why there were more individuals included in them than in the logistic regression models. The
baseline characteristics of this sample were similar to those
of persons in the logistic regression analysis. The sample
was much larger because the analysis could include persons
who might take sickness absences every year but vary in
terms of the number of days taken.
Table 2 shows results from the fixed-effects logistic regression model assessing the role of psychosocial job quality in
sickness absence, both before and after adjustment for other
covariates. In the multivariate model, there was evidence of a
dose-response relationship between the number of psychosocial job stressors a person reported and having a paid sickness absence. Compared with reporting no psychosocial job
stressors, reporting 1 adversity was associated with a 26% rise
in the odds of having a sickness absence (odds ratio (OR) =
1.26, 95% confidence interval (CI): 1.09, 1.45; P = 0.002).
784 Milner et al.
Table 1. Baseline Characteristics (%) of Respondents in the HILDA
Survey, 2005–2012a
Type of Analysis
Characteristic
Logistic
Regression
(n = 2,991)
Negative
Binomial
Regression
(n = 10,091)
Age group, years
15–24
22.47
24.36
25–34
22.73
23.43
35–44
24.37
23.11
45–54
22.10
19.60
55–64
7.79
8.41
≥65
0.53
1.09
Male
53.63
49.97
Female
46.37
50.03
Lowest
14.68
13.62
Medium-low
24.31
26.31
Medium-high
25.94
24.78
Highest
35.07
35.29
Optimal
13.77
14.48
1 adversity
50.15
50.48
2 adversities
26.15
25.49
9.93
9.55
Sex
Occupational skill level
Psychosocial job quality
≥3 adversities
Household structure
Couple with no dependent(s)
46.67
42.61
Couple/lone adult with
dependent(s)
33.80
36.38
Lone individual
16.78
15.96
2.74
5.04
Group/multiperson household
Employment arrangement
Permanent
86.73
85.70
Fixed-term
13.27
14.30
Abbreviation: HILDA, Household, Income and Labour Dynamics in
Australia.
a
The logistic regression analysis included only those persons who
reported a change in sickness absence in the past year (i.e., reported
both yes and no), as identified through the question, “During the last 12
months, have you taken any paid sick leave?” The negative binomial
regression analysis included persons who reported information on
days of paid sickness absence in the past year, as identified through
the question, “How many days did you spend on paid sick leave (in the
past 12 months)?”.
Reporting 2 adversities was associated with a 28% rise in the
odds of a sickness absence (OR = 1.28, 95% CI: 1.09, 1.51;
P = 0.002), while having 3 or more adversities was associated
with a 58% rise (OR = 1.58, 95% CI: 1.29, 1.94; P < 0.001).
When the job quality index was included as a continuous
measure, the odds ratio for the continuous job quality measure was 1.12 (95% CI: 1.06, 1.19; P < 0.001).
Results for other covariates indicated that respondents had
lower odds of a sickness absence during times when they
were working in fixed-term employment than when they were
permanently employed, and that older age and long-term
health conditions were related to greater odds of taking paid
sick leave. There was an observable gradient in occupational
skill level, showing that respondents were more likely to take
paid sick leave when they were working in higher-skill jobs
than when they were in lower-skill jobs.
We then assessed the possibility that sex modified the association between psychosocial job quality and sickness absence. Results of the likelihood ratio test (χ2 (3 df ) = 6.57,
P = 0.090) and inspection of interaction terms provided no
support for significant sex differences. We also found no evidence of effect modification by age (χ2 (15 df ) = 13.81,
P = 0.540) or by whether a person worked full-time or parttime (χ2 (3 df ) = 3.01, P = 0.390) in the relationship between
sickness absence and job quality.
Results from the negative binominal fixed-effects regression model (Table 3) indicated a dose-response relationship
between the number of job-related adversities experienced
by an individual and paid sickness absence. After adjustment
for other covariates, there was a marginally nonsignificant
5% increase in the number of days of paid sick leave in relation to 1 adversity (incidence rate ratio (IRR) = 1.05, 95% CI:
1.00, 1.10; P = 0.058), a 7% increase in days of sick leave in
relation to 2 adversities (IRR = 1.07, 95% CI: 1.02, 1.13;
P = 0.010), and an 11% increase in days of sick leave when
a person had 3 or more adversities (IRR = 1.11, 95% CI: 1.04,
1.19; P = 0.001). When the job quality index was included as
a continuous measure, the incidence rate ratio was 1.03 (95%
CI: 1.01, 1.05; P = 0.001). Results for other variables were
similar to those from the logistic regression models.
Findings from sensitivity analyses including between- and
random-effects estimates are shown in Web Tables 2 and 3.
Results were in the same direction as those from the fixedeffects analysis but were much greater in magnitude. Part
of the difference may be explained by residual confounding
from time-invariant factors (e.g., stable individual characteristics). The random-effects analyses also included betweenperson contributions to the coefficients, which potentially
accounts for some of the increase in magnitude.
DISCUSSION
The results derived from these fixed-effects analyses provide stronger evidence of a causal relationship between psychosocial job quality and sickness absence than previous
research. Compared with times when respondents were employed in jobs with optimal working conditions, experiencing a greater number of adversities at work was associated
with increased odds of sickness absence as well as a greater
number of days of sickness absence in the past year. This indicates that as the quality of a person’s work decreases, his or
her likelihood of being absent from the workplace increases.
Although there have been other studies on the relationship
between job stressors and sickness absence (2, 5, 6, 8, 20),
ours was the first (to our knowledge) to use a measure that
operationalizes a continuum of psychosocial job quality across
a working population and to evaluate the influence of combined
Am J Epidemiol. 2015;181(10):781–788
Sickness Absence and Psychosocial Job Quality 785
Table 2. Odds of Paid Sickness Absence in the Past Year According to Psychosocial Job Quality and
Sociodemographic Factors (Fixed-Effects Logistic Regression Model), HILDA Survey, 2005–2012
Bivariate Analysis
OR
95% CI
Multivariate Analysis
P Value
OR
95% CI
P Value
Psychosocial job quality
Optimal
1
1 adversity
1.25
1.09, 1.44
Referent
0.002
1.26
1
1.09, 1.45
Referent
2 adversities
1.29
1.10, 1.51
0.002
1.28
1.09, 1.51
0.002
≥3 adversities
1.58
1.29, 1.93
<0.001
1.58
1.29, 1.94
<0.001
0.002
Employment arrangement
Permanent
1
Fixed-term
0.83
Referent
0.72, 0.95
1
0.009
0.83
Referent
0.72, 0.96
0.011
Tertile of household income
Lowest
1
Medium
1.75
1.51, 2.03
<0.001
1.74
1.50, 2.02
<0.001
High
2.23
1.88, 2.65
<0.001
2.06
1.73, 2.46
<0.001
1.09
1.07, 1.11
<0.001
1.08
1.06, 1.10
<0.001
Age (per year)
Referent
1
Referent
Household structure
Couple with no dependent(s)
1
Couple/lone adult with dependent(s)
0.66
0.56, 0.76
Referent
<0.001
1
0.76
0.65, 0.88
Referent
<0.001
Lone individual
0.93
0.77, 1.13
0.488
1.06
0.87, 1.30
0.550
Group/multiperson household
0.77
0.57, 1.03
0.076
0.80
0.60, 1.08
0.152
Long-term health condition
No
1
Yes
1.24
Referent
1.07, 1.44
1
0.004
1.24
Referent
1.07, 1.44
0.004
Occupational skill level
Lowest
1
Medium-low
1.24
0.97, 1.57
Referent
0.081
1.22
1
0.96, 1.55
Referent
0.106
Medium-high
1.36
1.09, 1.69
0.007
1.36
1.09, 1.71
0.007
Highest
1.71
1.34, 2.18
<0.001
1.57
1.22, 2.00
<0.001
Abbreviations: CI, confidence interval; HILDA, Household, Income and Labour Dynamics in Australia; OR, odds ratio.
psychosocial working conditions rather than focusing on 1 or
2 conditions. Previous research using this measure has shown
a relationship between worsening job quality and mental (10)
and physical (11) health. Consistent with this past work, the
present study also found a gradient between the number of
adversities experienced at work and the likelihood of taking
sick leave. Our adoption of a fixed-effects approach enabled
us to focus on within-person change over time (rather than
between-person analysis as in previous studies), comparing
sickness absence under optimal job conditions with sickness
absence under varying degrees of poorer psychosocial job
quality.
There may be a number of different interpretations for
these results, depending on whether sickness absence is seen
as a health outcome, a health behavior, or a combination of
both. If sickness absence is viewed as a proxy for physical
and mental ill health (e.g., a health outcome), a possible interpretation of our result is that a low-quality work environment is related to a decline in health. A relationship between
measures of sickness absence and ill health and mortality has
Am J Epidemiol. 2015;181(10):781–788
been demonstrated in numerous studies (2, 5, 21, 22). The
consistency of these relationships is such that some researchers have suggested that sickness absence can be thought of as
an overall global measure of health status (2). Other researchers have preferred to view sickness absence as a health behavior and have conceptualized it as a form of coping (22). In this
case, workers may use sickness absence to avoid, reduce, or
control their exposure to work-related stress and gain the opportunity to recover from adverse working conditions (22–
24). Thus, this may be a mediator on the pathway between
psychosocial working conditions and health. The fact that
our data continued to show the same pattern of association between psychosocial job quality and sickness absence after we
controlled for long-term health conditions provides support
for the second explanatory framework, acknowledging that
these views are not mutually exclusive and that sickness
absence may represent both a health behavior and a health
outcome.
Past results (13) have suggested that females have a higher
number of sick days than males. However, we found no
786 Milner et al.
Table 3. Incidence Rate Ratios for Number of Days of Paid Sickness Absence in the Past Year According to
Psychosocial Job Quality and Sociodemographic Factors (Negative Binomial Regression Model), HILDA Survey,
2005–2012
Bivariate Analysis
IRR
95% CI
Multivariate Analysis
P Value
IRR
95% CI
P Value
Psychosocial job quality
Optimal
1
Referent
1
Referent
1 adversity
1.05
1.00, 1.10
0.052
1.05
1.00, 1.10
0.058
2 adversities
1.07
1.02, 1.13
0.011
1.07
1.02, 1.13
0.010
≥3 adversities
1.12
1.05, 1.19
0.001
1.11
1.04, 1.19
0.001
Employment arrangement
Permanent
1
Fixed-term
0.87
Referent
0.84, 0.90
1
<0.001
Referent
0.85
0.81, 0.89
<0.001
Tertile of household income
Lowest
1
Medium
1.21
1.16, 1.26
<0.001
1.21
1.15, 1.27
<0.001
High
1.32
1.25, 1.38
<0.001
1.30
1.23, 1.38
<0.001
1.03
1.03, 1.04
<0.001
1.03
1.02, 1.03
<0.001
Age (per year)
Referent
1
Referent
Household structure
Couple with no dependent(s)
1
Referent
1
Referent
Couple/lone adult with dependent(s)
0.92
0.88, 0.95
<0.001
0.95
0.90, 1.00
0.031
Lone individual
0.97
0.92, 1.02
0.238
0.99
0.93, 1.06
0.812
Group/multiperson household
0.97
0.90, 1.05
0.455
0.93
0.85, 1.02
0.144
Long-term health condition
No
1
Yes
1.14
Referent
1
1.10, 1.18
Referent
1.18
1.13, 1.24
<0.001
Occupational skill level
Lowest
1
Referent
1
Referent
Medium-low
0.99
0.93, 1.06
0.803
1.00
0.92, 1.08
0.944
Medium-high
0.97
0.91, 1.03
0.336
1.01
0.94, 1.09
0.717
Highest
1.07
1.01, 1.15
0.031
1.11
1.02, 1.20
0.012
Abbreviations: CI, confidence interval; IRR, incidence rate ratio; HILDA, Household, Income and Labour Dynamics
in Australia.
evidence that sex modified the association between job quality and sickness absence. There has also been some suggestion that people who are socially isolated (6) or living alone
(5) are at greater risk of sickness absence than those living
with others. This aligns with our research, which found that
living with children or dependents was associated with lower
odds of sickness absence. In a French cohort study, employees in manual and clerical positions were found to have much
higher levels of sickness absence than managers (7). In contrast, our study found that people were more likely to have
more absences when they were employed in higher-skill
positions than when they were employed in lower-skill positions. This finding probably reflects differences in the composition of the sample, which was comprised of permanent
and fixed-term contract workers, who are more likely to be
employed in jobs involving higher occupational skill levels.
This result may also be explained by methodological differences, as our fixed-effects analysis focused on within-person
change, while previous studies have involved within- and/or
between-person analyses.
One possible limitation of this study was its self-report
measure of sickness absence. The use of this subjective outcome may have resulted in measurement error, as respondents may have failed to accurately report whether they had
had any sickness absences in the past year and/or may have
misreported how many days of sick leave they took. There is
also a possibility of dependent misclassification, whereby
people were more attentive to sickness absence during the
years when they reported poor job quality than during the
years in which their psychosocial job quality was better.
We also acknowledge that our research was restricted to permanent and fixed-term employees in the Australian context
(since these were the only people in Australia who were guaranteed to have access to paid sick leave—approximately
50%–60% of the working population). The observed relationships clearly are not generalizable to the remaining
Am J Epidemiol. 2015;181(10):781–788
Sickness Absence and Psychosocial Job Quality 787
40%–50% of the Australian labor force, who are working in
other employment arrangements (including temporary and
labor-hire workers and the self-employed) (16). From a policy perspective, the lack of paid sick leave for such a large
proportion of the labor force may indeed be a substantial
driver of presenteeism. Last, the measure of job quality was
also self-reported and based upon a specific set of constructs,
as determined through factor analysis. There are many other
important psychosocial aspects of the work environment
(e.g., social support and bullying at work) that were not included in this measure or in this panel study, as well as aspects of the general working environment that could also
have had an influence on our results, suggesting that this study
provided a conservative estimate of the influence of workplace psychosocial stressors on paid sickness absence. In addition to the number of stressors a person is exposed to, the
duration of exposure may be an important aspect of the relationship and warrants further study.
Despite these limitations, there were also a number of
strengths in this study. These included the ability to examine
the relationships between psychosocial working conditions
and sickness absence over time using a large representative
national sample. We were able to use a previously validated
measure of psychosocial job quality, shown to be associated
with mental and physical health and more objective aspects
of work in the Australian context. The fixed-effects analytical
approach allowed us to examine causally robust withinperson associations while controlling for time-invariant confounders that may have otherwise biased the results, even
though the estimates obtained are (strictly speaking) generalizable only to those respondents who reported changes in
both exposure and outcome during their contributed waves
(and not to the entire source population). Selection bias is
likely to have been limited, as we only had a small proportion
of missingness in the data set.
Based on these findings, we conclude that psychosocial
job quality is a significant predictor of sickness absence in
the Australian context. These results suggest that improving
the overall psychosocial quality of work may lead to a reduction in sickness absence. Such an approach aligns with the
growing recognition that workplace-based interventions will
have the largest population benefit when they systematically
address risk and protective factors at the primary, secondary,
and tertiary levels (25).
ACKNOWLEDGMENTS
Author affiliations: McCaughey VicHealth Centre for
Community Wellbeing, Melbourne School of Population and
Global Health, University of Melbourne, Melbourne, Victoria,
Australia (Allison Milner, Anthony D. LaMontagne); Centre
for Research on Ageing, Health and Wellbeing, ANU College
of Medicine, Biology and Environment, Australian National
University, Canberra, Australian Capital Territory, Australia
(Peter Butterworth); Gender and Women’s Health Unit, Centre
for Health Equity, Melbourne School of Population and
Global Health, University of Melbourne, Melbourne, Victoria,
Australia (Rebecca Bentley, Anne M. Kavanagh); and PopulaAm J Epidemiol. 2015;181(10):781–788
tion Health Strategic Research Centre, School of Health and
Social Development, Deakin University, Melbourne, Victoria,
Australia (Anthony D. LaMontagne).
This work was supported by the Australian National
Health and Medical Research Council (NHMRC) through
both project (grant 375196) and postdoctoral research fellowship (to A.M.) (NHMRC Capacity Building Grant 546248)
funding. This study also received center grant funding
(grant 15732) from the Victorian Health Promotion Foundation (Melbourne, Victoria, Australia).
The data used in this paper were extracted by means of the
add-on package PanelWhiz for Stata. PanelWhiz (http://www.
PanelWhiz.eu) was created by Dr. John P. Haisken-DeNew
( [email protected]).
This study used unit record data from the Household, Income and Labour Dynamics in Australia (HILDA) Survey.
The HILDA Survey was initiated and is funded by the Australian Government Department of Social Services (DSS) and
is managed by the Melbourne Institute of Applied Economic
and Social Research (Melbourne Institute). The findings and
views reported in this paper, however, are those of the authors
and should not be attributed to either DSS or the Melbourne
Institute. The funding sources had no involvement in the
study design; in the collection, analysis, and interpretation
of the data; in the writing of the report; or in the decision
to submit the article for publication.
Conflict of interest: none declared.
REFERENCES
1. Henderson M, Glozier N, Holland Elliott K. Long term sickness
absence. BMJ. 2005;330(7495):802–803.
2. Kivimäki M, Head J, Ferrie JE, et al. Sickness absence as a global
measure of health: evidence from mortality in the Whitehall II
prospective cohort study. BMJ. 2003;327(7411):364.
3. Roelen CA, Koopmans PC, Hoedeman R, et al. Trends in the
incidence of sickness absence due to common mental disorders
between 2001 and 2007 in the Netherlands. Eur J Public
Health. 2009;19(6):625–630.
4. Allebeck P, Mastekaasa A. Chapter 5. Risk factors for
sick leave—general studies. Scand J Public Health. 2004;
32(63 suppl):49–108.
5. Duijts SF, Kant I, Swaen GM, et al. A meta-analysis of
observational studies identifies predictors of sickness absence.
J Clin Epidemiol. 2007;60(11):1105–1115.
6. Melchior M, Niedhammer I, Berkman LF, et al. Do
psychosocial work factors and social relations exert
independent effects on sickness absence? A six year
prospective study of the GAZEL cohort. J Epidemiol
Community Health. 2003;57(4):285–293.
7. Melchior M, Krieger N, Kawachi I, et al. Work factors and
occupational class disparities in sickness absence: findings
from the GAZEL cohort study. Am J Public Health. 2005;
95(7):1206–1212.
8. Vahtera J, Kivimäki M, Pentti J, et al. Effect of change in the
psychosocial work environment on sickness absence: a seven
year follow up of initially healthy employees. J Epidemiol
Community Health. 2000;54(7):484–493.
9. Gunasekara FI, Richardson K, Carter K, et al. Fixed effects
analysis of repeated measures data. Int J Epidemiol. 2014;43(1):
264–269.
788 Milner et al.
10. Butterworth P, Leach LS, Strazdins L, et al. The psychosocial
quality of work determines whether employment has benefits for
mental health: results from a longitudinal national household
panel survey. Occup Environ Med. 2011;68(11):806–812.
11. Butterworth P, Leach LS, Rodgers B, et al. Psychosocial job
adversity and health in Australia: analysis of data from the
HILDA Survey. Aust N Z J Public Health. 2011;35(6):
564–571.
12. Leach L, Rodgers B, Butterworth P, et al. Deriving an
evidence-based measure of job quality from the HILDA
Survey. Aust Soc Policy. 2010;9:67–86.
13. Laaksonen M, Martikainen P, Rahkonen O, et al. Explanations
for gender differences in sickness absence: evidence from
middle-aged municipal employees from Finland. Occup
Environ Med. 2008;65(5):325–330.
14. Wilkins R. Families, Incomes and Jobs. Vol. 8. A Statistical
Report on Waves 1 to 10 of the Household, Income and Labour
Dynamics in Australia Survey. Melbourne, Victoria, Australia:
Faculty of Business and Economics, Melbourne Institute of
Applied Economic and Social Research; 2013.
15. Australian Bureau of Statistics. Australian and New Zealand
Standard Classification of Occupations (ANZSCO). (Catalog
no. 1220.0). Canberra, Australian Capital Territory, Australia:
Australian Bureau of Statistics; 2013.
16. LaMontagne AD, Milner A, Krnjacki L, et al. Employment
arrangements and mental health in a cohort of working
Australians: are transitions from permanent to temporary
employment associated with changes in mental health?
Am J Epidemiol. 2014;179(12):1467–1476.
17. Australian Bureau of Statistics. Part-time and casual
employment [feature article]. In: Australian Labour Market
Statistics. (Catalog no. 6105.0). Canberra, Australian Capital
Territory, Australia: Australian Bureau of Statistics; 2013.
18. StataCorp LP. Stata Statistical Software, Release 12.1. College
Station, TX: StataCorp LP; 2012.
19. Allison PD. Fixed effects regression methods for count data. In:
Fixed Effects Regression Methods for Longitudinal Data Using
SAS. Cary, NC: SAS Institute, Inc.; 2005:79–104.
20. Lesuffleur T, Chastang JF, Sandret N, et al. Psychosocial factors
at work and sickness absence: results from the French National
SUMER Survey. Am J Ind Med. 2014;57(6):695–708.
21. D’Souza RM, Strazdins L, Broom DH, et al. Work demands,
job insecurity and sickness absence from work. How productive
is the new, flexible labour force? Aust N Z J Public Health.
2006;30(3):205–212.
22. Allebeck P, Mastekaasa A. Chapter 3. Causes of sickness
absence: research approaches and explanatory models. Scand J
Public Health. 2004;32(63 suppl):36–43.
23. Rael EG, Stansfeld SA, Shipley M, et al. Sickness absence in
the Whitehall II study, London: the role of social support and
material problems. J Epidemiol Community Health. 1995;
49(5):474–481.
24. Kristensen TS. Sickness absence and work strain among Danish
slaughterhouse workers: an analysis of absence from work
regarded as coping behaviour. Soc Sci Med. 1991;32(1):15–27.
25. LaMontagne AD, Martin A, Page KM, et al. Workplace mental
health: developing an integrated intervention approach. BMC
Psychiatry. 2014;14:131.
Am J Epidemiol. 2015;181(10):781–788