Social Science & Medicine 153 (2016) 20e26

Contents lists available at ScienceDirect

Social Science & Medicine

journal homepage: www.elsevier.com/locate/socscimed

The role of social support in protecting mental health when employed

and unemployed: A longitudinal fixed-effects analysis using 12 annual
waves of the HILDA cohort
Allison Milner a, b, *, Lauren Krnjacki c, Peter Butterworth d, e, Anthony D. LaMontagne a, b
a
Work, Health and Wellbeing Unit, Population Health Strategic Research Centre, School of Health & Social Development, Deakin University, Melbourne,
Australia
b
McCaughey VicHealth Community Wellbeing Unit, Melbourne School of Population and Global Health, The University of Melbourne, Melbourne, Australia
c
Centre for Health Equity, Melbourne School of Population and Global Health, The University of Melbourne, Victoria 3010, Australia
d
Centre for Mental Health, Melbourne School of Population and Global Health, The University of Melbourne, Melbourne, Australia
e
Melbourne Institute of Applied Economic and Social Research, The University of Melbourne, Melbourne, Australia

a r t i c l e i n f o a b s t r a c t

Article history: Perceived social support is associated with overall better mental health. There is also evidence that
Received 29 October 2015 unemployed workers with higher social support cope better psychologically than those without such
Received in revised form support. However, there has been limited research about the effect of social support among people who
14 January 2016
have experienced both unemployment and employment. We assessed this topic using 12 years of
Accepted 27 January 2016
Available online 30 January 2016
annually collected cohort data.
The sample included 3190 people who had experienced both unemployment and employment. We
used longitudinal fixed-effects modelling to investigate within-person changes in mental health
Keywords:
Mental health
comparing the role of social support when a person was unemployed to when they were employed.
Employed Compared to when a person reported low social support, a change to medium (6.35, 95% 5.66 to 7.04,
Unemployed p < 0.001) or high social support (11.58, 95%, 95% CI 10.81 to 12.36, p < 0.001) was associated with a large
Social support increase in mental health (measured on an 100 point scale, with higher scores representing better
Longitudinal mental health). When a person was unemployed but had high levels of social support, their mental
Fixed effect health was 2.89 points (95% CI 1.67 to 4.11, p < 0.001) higher than when they were employed but had
lower social support. The buffering effect of social support was confirmed in stratified analysis.
There was a strong direct effect of social support on mental health. The magnitude of these differences
could be considered clinically meaningful. Our results also suggest that social support has a significant
buffering effect on mental health when a person is unemployed.
© 2016 Elsevier Ltd. All rights reserved.

1. Background
1.1. Social support and mental health
There is substantial evidence that those with greater social
support (defined as the social resources that persons perceive to be
available or that are actually provided to them by non-professionals
(Cohen and Gottlieb, 2000 cited in Gottlieb and Bergen, 2010)) have
better mental health than those with less social support (Kawachi
* Corresponding author. Work, Health & Wellbeing, Centre for Population Health
Research, School of Health & Social Development, Deakin University, Building
BC3.213, Burwood, VIC 3125, Australia.
E-mail address: allison.milner@deakin.edu.au (A. Milner).
URL: http://www.deakin.edu.au
and Berkman, 2001; Barrera, 1986; Lakey and Orehek, 2011;
Cohen and Wills, 1985). Social support has been conceptualised in
many ways e but has commonly been measured through the
experience of perceived emotional, practical, or informational
support (Gottlieb and Bergen, 2010). Social support may affect
health through direct and/or indirect pathways (Cohen and Wills,
1985). The indirect relationship suggests a “buffering” effect, in
that the link between life stresses and poor mental health is
stronger for people with low social support than for people with
high social support (Lakey and Orehek, 2011). The direct relation-
ship suggests that social support promotes mental health by
providing persons with regular positive experiences and a set of
stable, socially rewarded roles. Cohen and Wills (1985) argue that
this promotes positive affect, a sense of predictability, stability and

http://dx.doi.org/10.1016/j.socscimed.2016.01.050
0277-9536/© 2016 Elsevier Ltd. All rights reserved.
 A. Milner et al. / Social Science & Medicine 153 (2016) 20e26
self-worth. Evidence from meta-analytic studies suggests that the
main effects of social support on mental health are more consis-
tently observed than buffering (indirect) effects (Lakey and Cronin,
2008, Lakey and Orehek, 2011). There is also evidence to suggest
that perceived social support has a greater effect on mental health
than received social support (Thoits, 2011; Turner and Marino,
1994).
1.2. Mental health, unemployment and social support
Considerable evidence from longitudinal surveys suggests that
people have worse mental health when they are unemployed
compared to when they are employed (Paul and Batinic, 2010;
Milner et al. 2013b). Further, repeated spells of unemployment
lead to further declines in mental health; that is, with additional
bouts of unemployment, there is a continual decline in mental
health (Milner et al. 2013b). At the same time, research has also
shown that poor mental health is both a consequence of and risk
factor for unemployment (Olesen et al. 2013). Thus, policy and
practice addressing unemployment may benefit from considering
strategies to enhance factors that could improve mental health, as
well as mitigating the negative impacts of unemployment.
There has been some past research indicating that social support
has an important influence on the mental health of those who are
unemployed. For example, a population based case control study on
young people set in Sweden found that mental health was gener-
ally worse among unemployed persons with low social support
from family and friends than among unemployed persons with
higher social support (Axelsson and Ejlertsson, 2002). The authors
of this study suggest that high support decreased the risk of having
mental health problems among unemployed persons. A much older
study (Gore, 1978) followed up males who lost their jobs due to a
factory closure and demonstrated the role of social support in
protecting against a range of physical and mental health outcomes.
This aligns with evidence from a 2009 meta analysis, which found
that unemployed workers with higher social support felt better
psychologically than those without such support (Pinquart and
Sorensen, 2000; Mckee-Ryan et al., 2005).
A limitation of much of the research discussed above is that it
tends to be cross-sectional and based on between-person com-
parisons, and may therefore be subject to a range of biases con-
nected to unmeasured or unadjusted differences in the groups
being compared. For example, there may be selection effects in that
those persons who become unemployed may already have worse
health and social support than those who are employed; such dif-
ferences would reduce the exchangeability of the groups being
compared, thus reducing causal inference. A within-person
approach allows identification of changes in mental health when
a person has low social support, compared to when that same
person has higher social support; thus, a person is able to act as
their own control, removing bias that may be present due to time-
invariant differences between people (Gunasekara et al. 2014).
Related to this, there is some concern that perceived-support
measures may be confounded with either personality factors or
depression (Finch and Zautra, 1992; Henderson, 1984), thus there
maybe the possibility that people with poorer mental health attract
less social support. However, past prospective studies have evi-
denced functional differences between social support, loneliness
and mental health problems (Finch and Zautra, 1992), e.g., social
support and loneliness predict subsequent depression longitudi-
nally, even after controlling for initial depression levels (Cacioppo
et al. 2010, 2006; Green et al. 1992; Hagerty and Williams, 1999;
Heikkinen and Kauppinen, 2004; Tiikkainen and Heikkinen,
2005), which suggests some support for this causal pathway.
Further, a number of other studies have found that common mental
21
illnesses, such as depressive symptoms, represent clearly different
statistical constructs from loneliness and social support (Weeks
et al. 1980; Cacioppo et al. 2006; Russell, 1996). Regardless, this
emphasises the importance of controlling for person-related char-
acteristics that could influence the relationship between social
support and mental health. The fixed-effect regression approach
described above eliminates this source of bias by inherently con-
trolling for time-invariant person-related and other characteristics.
The main aim of the current paper is to assess the relationship
between mental health and social support among persons who
have experienced both unemployment and employment. We
hypothesise that social support will have a beneficial effect on
mental health regardless of employment status (in line with the
main effects hypothesis). However, we would also hypothesise a
significant buffering effect of high social support for people who are
unemployed. We base this hypothesis on the research discussed
above which suggests that social support may be particularly
important when people are going through stressful life events. We
would argue that unemployment represents one such life stressor
that could lead to a lowering of mental health. Thus, we seek to test
the buffering and main effects hypotheses (discussed above) using
a within-person analytic approach.
2. Methods
2.1. 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 in-
formation annually from over 13,000 individuals within over 7000
households (Wilkins, 2013). The response rate to wave 1 was 66%
(Wilkins, 2013). The survey covers a range of dimensions including
social, demographic, health and economic conditions using a
combination of face-to-face interviews with trained interviewers
and a self-completion questionnaire. Although data are collected
on each member of the household, interviews are only conducted
with those older than 15 years of age.
The initial wave of the survey began with a large national
probability sample of Australian households occupying private
dwellings (Wilkins, 2013). Interviews were sought in later waves
with all persons in sample households who turned 15 years of
age. Additional persons have been added to the sample as a
result of changes in household composition. For example, if a
household member left his or her original household (e.g. chil-
dren left home, or a couple separated), he/she formed an entirely
new household including all persons living with the original
sample member. Inclusion of these new households is the main
way in which the HILDA survey maintains sample representa-
tiveness. A top-up sample of 2000 people was added to the
cohort in 2011 to allow better representation of the Australian
population using the same methodology as the original sample
(i.e., a three-stage area-based design) (Watson, 2011). The
response rates for the HILDA survey are above 90% for re-
spondents who have continued in the survey and above 70% for
new respondents being invited into the study (Wilkins, 2013).
The main variables examined in this study were available in all
annual waves of HILDA (2001e2012).
2.2. Outcome variable
Mental health was assessed using the five-item Mental Health
Inventory (MHI), a subscale from the SF-36 general health survey.
The MHI assesses symptoms of depression and anxiety (nervous-
ness, depressed affect) and positive aspects of mental health
22 A. Milner et al. / Social Science & Medicine 153 (2016) 20e26
(feeling calm, happy) in the past 4 weeks. The MHI has reasonable
validity and is an effective screening instrument for mood disorders
or severe depressive symptomatology in the general population
(Rumpf et al. 2001; Yamazaki et al. 2005; Gill et al. 2006). The
current analyses use the continuous MHI score, with higher scores
representing better mental health. Although there is no universally
accepted translation of MHI score difference to clinical meaning-
fulness, a difference of three points on the norm based scale (T-
score) has been suggested to reflect a minimally important differ-
ence (Ware, 2000), and a difference of four or more on the
unstandardised scale has been characterised as indicating a
moderately clinically significant effect (Contopoulos-Ioannidis et al.
2009).
2.3. Exposure variable
The social support measure was designed to assess an in-
dividual's perception of the social support they receive from friends
and family. This measure has been used in previous studies (Crosier
et al. 2007; Hewitt et al. 2010). This measure primarily taps into
emotional support. We are particularly focused on perceived social
support as this has been recognised as having a more important
effect on mental health (Mcdowell and Serovich, 2007) and mor-
tality (Lyyra and Heikkinen, 2006) than received social support. The
10-item scale included the following items:

I have no one to lean on in times of trouble (reverse coded);

I often feel very lonely (reverse coded);

I enjoy the time I spend with the people that are important to
me;

I seem to have a lot of friends;

People don't come and visit as much as I would like (reverse
coded);

I often need help from other people but can't get it (reverse
coded);

I don't have anyone that I can confide in (reverse coded);

There is someone who can always cheer me up when I am
down;

When I need someone to help me out, I can usually find
someone; and

When something's on my mind, just talking with the people I
know can make me feel better.
Items are rated on an 8-point Likert scale ranging from
strongly disagree (1) to strongly agree (7) and the Cronbach's
alpha, for the current data, was 0.84 across all waves. The social
support scale was created by averaging the ten items in the scale,
with lower scores representing lower social support and higher
scores representing higher social support (the items we reverse
coded above). In its original form, the scale was strongly posi-
tively skewed. Hence, we tested the analytic relationships of
interest with social support coded in three levels (Low (1e4.69),
Medium (4.70e5.59), High (5.6e7)) based on the observed dis-
tribution of the variable. As a sensitivity test, we also created a
binary measure where 1e5.59 were classified as “low” social
support, and 5.60 to 7 was classified as “high” social support
based on a median split.
2.4. Other covariates
The model also included a number of likely confounders. These
included age (measured continuously), education (postgraduate,
bachelor degree, diploma or certificate, year 12, not completed year
12), long-term health condition (yes or no) and household structure
(couple or lone adult residing with dependents, couple without
dependents, lone person without dependents, and a group or
multiple person household). We also controlled for marital status
(Marital status, Married, Separated, Divorced, Widowed, Defacto,
Single), recognising that the perceptions of social support may
differ according to whether someone is in a relationship or not (i.e.,
differences between received and perceived social support) (Haber
et al. 2007).
2.5. Analytic strategy
Longitudinal linear fixed-effects regression models were used
to estimate the association between social support (as a three
level variable) and MHI score within individuals. Fixed effects
regression models describe differences in MHI score associated
with lower social support compared with that individual's mean
MHI score when they had medium or high social support. These
models provide an indication of within-person effects. Fixed-
effects models are particularly useful where time-invariant con-
founding is likely to create bias in causal estimates (Gunasekara et
al., 2014).
We controlled for time-varying factors by including a number of
relevant covariates into the fixed-effects models. Including these in
fixed effect models accounted for individual change in these vari-
ables that might influence the relationship between social support
and mental health. We assessed the effect of social support on
mental health contemporaneously, on the premise that perceived
social support would have an immediate effect on a scaled measure
of mental health. We tested the hypothesis that the association
between social support and mental health differed depending on
whether a person was employed or unemployed using an interac-
tion term in fixed-effects models, and examined this using the
likelihood ratio test and inspection of significance values of inter-
action terms in the model.
2.6. Analytic sample and missing data
In order to test both the social support-mental health rela-
tionship and effect modification of that relationship by employ-
ment status within-persons, those persons eligible for the analytic
sample had to experience waves of both employment and un-
employment. Within the eligible sample, we noted about 4% of
data was excluded due to missingness on variables included in the
analysis (Fig. 1). We would note that this analysis is on a selected
sample which excludes cohort participants who were continu-
ously employed or unemployed (or not in the labour force) for all
of their contributed waves. While internally consistent and
maximising causal inference, this is at the expense of generaliz-
ability. For example, those persons who were excluded from the
analytic sample had slightly lower scores on our mental health
measure (mean score of 69.33, 95% CI 68.47 to 70.20) than those
who were included from the sample (mean score of 71.30, 95% CI
71.05 to 71.56). Although we would note these differences were
minimal.
3. Results
The characteristics of the analytic sample can be seen in Table 1.
The minimum number of observations for respondents included in
the study was 2 and the maximum was 12; on average respondents
contributed 5.6 waves of data. The mean scores on the MHI mea-
sure was similar at baseline and in the final wave of the sample. As
can be seen there was an equal distribution of males and females.
About 41% of the sample were unemployed in their first contrib-
uted wave, while 26.43% were unemployed in their last contributed
wave. Despite the inclusion of new survey entrants (children
 A. Milner et al. / Social Science & Medicine 153 (2016) 20e26
Fig. 1. Sample selection.
reaching 15/new households), over the time of the study, the an-
alytic sample got older, and were more likely to become married.
Within-person regression results can be seen in Table 2, which
shows the adjusted main effects model (1), the unadjusted inter-
action model (2), and the adjusted interaction model (3) for those
persons who had experienced both unemployment and employ-
ment. There was a strong direct relationship between social sup-
port and mental health (model 1). Compared to when a person had
lower support, an increase to “medium” social support was asso-
ciated with approximately a 6.35-point improvement in mental
health (95% CI 5.66 to 7.04, p < 0.001) on the 100-point MHI scale.
There was a greater than 11-point increase in mental health when a
person changed from being in the low social support category to
the high social support category (95% CI 10.81 to 12.36, p < 0.001)
on the 100-point MHI scale. Model 1 also shows that, compared to
being employed, there was a 1.96-point decline in mental health
when unemployed (95% CI 2.56 to 1.37, p < 0.001).
Results between the unadjusted (2) and the adjusted interaction
model (3) were similar, so we will discuss the adjusted model (3)
only. Compared to when a person was employed and had the
lowest levels of social support (the reference category), an increase
in social support to medium levels was associated with 6.15 (95% CI
23
5.42 to 6.87, p < 0.001) rise in mental health (on the 100-point MHI
scale) and a 10.93 (95% CI 10.11 to 11.74, p < 0.001) rise in mental
health when a person had high levels of social support. When a
person was unemployed but had high levels of social support, their
mental health was 2.89 points (95% CI 1.67 to 4.11, p < 0.001) higher
than when they were employed but had lower social support, while
the presence of high social support does not make up for the
decline in mental health due to unemployment completely, it does
buffer the impact considerably, with those who are unemployed
with high social support showing a net difference in mental health
of 0.89 points on the 100-point MHI scale (3.78 plus 2.89).
We then stratified results by whether people had low, medium
or high social support (Table 3) to assess mental health–employ-
ment status relationships, while holding social support stable.
Among those people with low social support, being unemployed
was associated with a 3.48 point decline in mental health (95% CI-
4.79 to 2.18, p < 0.001). Among those with medium social support,
unemployment was associated with a 2.44 decline in mental
health (95% CI-3.78 to 1.11, p < 0.001), while among those with
high social support, being unemployed was associated with a very
small decline in mental health (0.49, 95% CI-1.32 to 0.35, <0.001).
4. Discussion
The results of this analysis showed that: 1) social support had a
strong association with mental health, independently of whether a
person was unemployed or employed; 2) compared to when a
person was employed, unemployment was associated with a
decline in mental health; 3) high social support improved mental
health among the unemployed, albeit not to the level a person had
when they were employed, and; 4) unemployment was associated
with the greatest decline in mental health among those persons
with low social support. Thus, our hypothesis one was supported
(the main effects hypothesis); hypothesis two was also supported
(the buffering hypothesis), although high social support does not
completely buffer the effects of unemployment on mental health.
This study builds on past research demonstrating the effects of
social support on mental health using cross-sectional designs
(Lakey and Cronin, 2008). There is also evidence from meta-
analyses, which have shown consistent main effects between low
perceived support and posttraumatic stress disorder symptoms
(Brewin et al. 2000) and nonspecific psychological distress (Finch
et al. 1999; Procidano, 1992). It is thought that social influence is
one of the main pathways through which social support influences
mental health, i.e., the way members of a social network obtain
normative guidance about health-relevant behaviours, such as help
seeking for mental distress. According to Kawachi and Berkman
(2001) it is also possible that social support directly produces
positive psychological states, including a sense of purpose,
belonging, and security, as well as recognition of self-worth. These
positive psychological states may indirectly benefit mental health
because of increased motivation for self-care (e.g., regular exercise,
moderation of alcohol intake), as well as the modulation of the
neuroendocrine response to stress.
Our study shows that when a person is unemployed, low social
support has an adverse influence on their mental health. Jahoda's
(1982) latent deprivation model of unemployment suggests that,
aside from the loss of income, there are five latent functions that
individuals are deprived of once they lose a job. These include loss
of time structure, social contact, collective purpose, social identity/
status, and activity (Jahoda, 1982, p.188). There has been a number
of studies showing that these latent benefits (particularly social
loneliness) are related to wellbeing among unemployed persons
(Creed and Macintyre, 2001a, Creed and Reynolds, 2001b), and that
their detrimental effects may actually be greater than loss of
24 A. Milner et al. / Social Science & Medicine 153 (2016) 20e26

Table 1
Sample characteristics (n ¼ 3190) on first and last contributed waves to the analysis.

First observed wave Last observed wave

MHI score (0e100), mean with std.dev. 70.79 (17.80) 70.86 (18.02)
Social support (%)
Low 28.28 32.11
Medium 27.78 25.34
High 43.98 42.55
Gender (%)
Male 50.27 50.27
Female 49.73 49.73
Employment status (%)
Employed 59.05 73.57
Unemployed 40.95 26.43
Age group (%)
Under 30 yrs 60.58 48.98
31e44 yrs 23.61 23.99
45e60 yrs 14.71 21.76
61 yrs pls 1.1 5.27
Marital status (%)
Married 22.89 29.6
Separated 3.17 3.45
Divorced 7.15 9.03
Widowed 0.97 1.35
Defacto 12.23 15.8
Never married/single 53.59 40.77
Long term health (%) conditions
Yes 17.84 22.92
No 82.16 77.08
Household structure (%)
Couple no children 15.21 22.14
Couple with children 49.89 38.79
Lone with children 17.72 13.64
Lone person 10.69 18.12
Other 6.49 7.31

Notes: std. dev. ¼ standard deviation.

Table 2
Fixed effect within-person regression results, the effect of social support on mental health by employment status, HILDA, 2001 to 2012, 3190 persons.

(1) Adjusted main effects model (2) Unadjusted interaction model (3) Adjusted interaction model

Coef L 95% CI U 95% CI p value Coef L 95% CI U 95% CI p value Coef L 95% CI U 95% CI p value

Social support
Low 1 1 1
Medium 6.35 5.66 7.04 <0.001 6.33 5.61 7.04 <0.001 6.15 5.42 6.87 <0.001
High 11.58 10.81 12.36 <0.001 11.29 10.49 12.09 <0.001 10.93 10.11 11.74 <0.001
Employment status
Employed 1 1 1
Unemployed 1.96 2.56 1.37 <0.001 3.00 3.97 2.03 <0.001 3.78 4.98 2.58 <0.001
Employment status*social support
Employed*lower social support 1
Unemployed* medium social support 0.73 0.59 2.06 0.278 0.63 0.78 2.04 0.38
Unemployed* high social support 2.76 1.58 3.94 <0.001 2.89 1.67 4.11 <0.001
Constant 61.10 58.30 63.89 <0.001 65.03 64.52 65.54 <0.001 61.10 58.30 63.89 <0.001

Notes: The MHI runs from 0 (low) to 100 (high). Coef ¼ model coefficient; L CI ¼ lower confidence intervals at 95% significance; U CI ¼ upper confidence intervals at 95%
significance; p value ¼ significance value at 95%. Adjusted models includes age group, marital status, long term health conditions, education, household structure. Reference
categories in the interaction models represent employed people with low social support.

income (Paul and Batinic, 2010).
At the same time, having high social support while being un-
employed was shown to be associated with better mental health
than being employed with low social support. This finding aligns
with past research, which suggested that unemployed workers
with greater social support coped better psychologically than those
without such support (Mckee-Ryan et al., 2005; Paul and Batinic,
2010). In addition to improvements in mental health, the provi-
sion of social support to those who are unemployed has also been
shown to be associated with greater job seeking intentions
(Vinokur and Caplan, 1987), as well as lowered financial stress in
the general population (Aslund et al. 2014).
The strengths of this study include its large longitudinal design,
which enabled us to examine the relationship between social
support and mental health over 12 annual waves using a large
representative national sample. The fixed effects analytical
approach allowed us to examine causally-robust within-person
associations controlling for time-invariant confounders that may
have otherwise biased results, even though the estimates obtained,
strictly speaking, are generalizable only to those participants
reporting changes in exposure over their contributed waves (and
not to the entire source population). Further, our study provides a
novel contribution to research as it among the first to assess dif-
ferential effects in the relationship between social support and
mental health by employment status.
At the same time, the paper has a number of limitations. First is
 A. Milner et al. / Social Science & Medicine 153 (2016) 20e26
Table 3
Fixed effect within-person regression results, employment status and mental health,
stratified by low, medium and high social support, HILDA, 2001 to 2012.
Coef Low 95% CI Upper 95% CI p value
Those persons with low social support (1787 persons, 5300 observations)
Employed 1 mentation, and effectiveness research, and that this strategy could
Unemployed 3.48 4.79 2.18 <0.001
Constant 58.91 52.72 65.11 <0.001
Those persons with medium social support (2013 persons, 4970 observations)
Employed 1 ment. Future research needs to further assess this possibility
Unemployed 2.44 3.78 1.11 <0.001
Constant 72.37 65.52 79.23 <0.001
Those persons with high social support (2207 persons, 7637 observations)
Employed 1 Acknowledgments
Unemployed 0.49 1.32 0.35 <0.001
Constant 72.94 69.4 76.48 <0.001
Notes: The MHI runs from 0 (low) to 100 (high). Coef ¼ model coefficient; L
CI ¼ lower confidence intervals at 95% significance; U CI ¼ upper confidence in-
tervals at 95% significance; p value ¼ significance value at 95%. Adjusted models
includes age group, marital status, long term health conditions, education, house-
hold structure.
the fact that our outcome and exposure are self-reported, which
introduces the possibility for dependent misclassification, whereby
errors in the exposure and outcome are correlated. However, it is
important to note that we are primarily interested in measuring
perceived social support, hence self-report is appropriate. We
would also acknowledge that our analysis, which relates social
support and mental health contemporaneously, cannot formally
demonstrate the time-dependent directionality of the relationship
between social support and mental health. While it might appear
that modeling lagged exposuredoutcome relationships (e.g., to test
whether social support in the preceding wave influences mental
health in a later wave) would address this concern, this produces
the “incidental parameters problem” (the error term of the
outcome variable being correlated with the predictors (Allison,
2009)) resulting in biased results. Addressing this would require
an alternate modelling approach outside the scope of this study and
which we would suggest as a further research topic. There is also a
need to consider factors connected to the HILDA study, including
the fact that there is greater retention into the study by persons
with higher socio-economic status. Last, we would note that we
restricted our sample to people who had experienced both unem-
ployment and employment in order to maximise causal inference.
This affects external validity e as most people in HILDA have not
experienced both unemployment and employment. Most people
have been employed only. However, our analysis regarding the
overall measures of mental health among those persons excluded
versus included in the sample indicate relatively minimal mental
health differences between these two groups (less than a 2 point
difference on the 100-point MHI scale).
In conclusion, our study finds that social support is an important
resource associated with an improvement in mental health.
Compared to when a person was employed, being unemployed was
generally associated with a reduction in mental health. However,
when people who were unemployed reported higher levels of so-
cial support, their mental health substantially increased. These
findings have implications for the development of intervention and
prevention strategies. Our results suggest that improvements in
social support for the unemployed may reduce the mental health
impacts of job loss. There are relatively few interventions for the
unemployed, as identified in a recent review on the topic (Audhoe
et al. 2010), but previous interventions aimed at unemployed per-
sons have mainly focused on work participation and mental
distress rather than focusing on social support (Audhoe et al. 2010).
However, given that unemployment is reliably associated with
increased rates of mental disorder (Paul and Moser, 2009) and
25
suicide (Milner et al. 2013a), we would suggest that interventions in
this area could also address social support as a strategy for pro-
tecting mental health during periods of unemployment In sum-
mary, our findings suggest that the improvement of social support
is a promising new strategy for intervention development, imple-
complement and extend previous efforts to devise the best possible
interventions to protect mental health during times of unemploy-
through both further observational studies and controlled trials.
This paper uses unit record data from the Household, Income
and Labour Dynamics in Australia HILDA) Survey. The HILDA Project
was initiated and is funded by the Australian Government
Department of Social Services DSS) and is managed by the Mel-
bourne Institute of Applied Economic and Social Research Mel-
bourne Institute). The findings and views reported in this paper,
however, are those of the author and should not be attributed to
either DSS or the Melbourne Institute. The data used in this paper
was extracted using the Add-On Package PanelWhiz for Stata.
PanelWhiz (http://www.PanelWhiz.eu) was written by Dr. John P.
Haisken-DeNew (john@PanelWhiz.eu).
References
Allison, P., 2009. Fixed Effects Regression Methods for Longitudinal Data Using SAS.
Cary, NC. SAS Institute Inc, USA.
Aslund, C., Larm, P., Starrin, B., Nilsson, K., 2014. The buffering effect of tangible
social support on financial stress: influence on psychological well-being and
psychosomatic symptoms in a large sample of the adult general population. Int.
J. Equity Health 13, 85.
Audhoe, S.S., Hoving, J.L., Sluiter, J.K., Frings-Dresen, M.H., 2010. Vocational in-
terventions for unemployed: effects on work participation and mental distress.
A systematic review. J. Occup. Rehabil. 20, 1e13.
Axelsson, L., Ejlertsson, G., 2002. Self-reported health, self-esteem and social sup-
port among young unemployed people: a population-based study. Int. J. Soc.
Welf. 11, 111e119.
Barrera Jr., M., 1986. Distinctions between social support concepts, measures, and
models. Am. J. Community Psychol. 14, 413e445.
Brewin, C.R., Andrews, B., Valentine, J.D., 2000. Meta-analysis of risk factors for
posttraumatic stress disorder in trauma-exposed adults. J. Consult Clin. Psychol.
68, 748e766.
Cacioppo, J.T., Hawkley, L.C., Thisted, R.A., 2010. Perceived social isolation makes me
sad: five year cross-lagged analyses of loneliness and depressive symptom-
atology in the Chicago health, aging, and social relations study. Psychol. aging
25, 453e463.
Cacioppo, J.T., Hughes, M.E., Waite, L.J., Hawkley, L.C., Thisted, R.A., 2006. Loneliness
as a specific risk factor for depressive symptoms: cross-sectional and longitu-
dinal analyses. Psychol. Aging 21, 140e151.
Cohen, S., Wills, T.A., 1985. Stress, social support, and the buffering hypothesis.
Psychol. Bull. 98, 310e357.
Contopoulos-Ioannidis, D.G., Karvouni, A., Kouri, I., Ioannidis, J.P.A., 2009. Reporting
and interpretation of SF-36 outcomes in randomised trials: systematic review.
BMJ Br. Med. J. 338, a3006.
Creed, P.A., Macintyre, S.R., 2001a. The relative effects of deprivation of the latent
and manifest benefits of employment on the well-being of unemployed people.
J. Occup. Health Psychol. 6, 324e331.
Creed, P.A., Reynolds, J., 2001b. Economic deprivation, experiential deprivation and
social loneliness in unemployed and employed youth. J. Community & Appl.
Soc. Psychol. 11, 167e178.
Crosier, T., Butterworth, P., Rodgers, B., 2007. Mental health problems among single
and partnered mothers. The role of financial hardship and social support. Soc.
Psychiatry Psychiatr. Epidemiol. 42, 6e13.
Finch, J.F., Okun, M.A., Pool, G.J., Ruehlman, L.S., 1999. A comparison of the influence
of conflictual and supportive social interactions on psychological distress.
J. Pers. 67, 581e621.
Finch, J.F., Zautra, A.J., 1992. Testing latent longitudinal models of social ties and
depression among the elderly: a comparison of distribution-free and maximum
likelihood estimates with nonnormal data. Psychol. Aging 7, 107e118.
Gill, S.C., Butterworth, P., Rodgers, B., Anstey, K.J., Villamil, E., Melzer, D., 2006.
Mental health and the timing of men's retirement. Soc. Psychiatry Psychiatr.
Epidemiol. 41, 515e522.
Gore, S., 1978. The effect of social support in moderating the health consequences of
unemployment. J. Health Soc. Behav. 19, 157e165.
26 A. Milner et al. / Social Science & Medicine 153 (2016) 20e26
Gottlieb, B.H., Bergen, A.E., 2010. Social support concepts and measures.
J. Psychosom. Res. 69, 511e520.
Green, B.H., Copeland, J.R., Dewey, M.E., Sharma, V., Saunders, P.A., Davidson, I.A.,
Sullivan, C., Mcwilliam, C., 1992. Risk factors for depression in elderly people: a
prospective study. Acta Psychiatr. Scand. 86, 213e217.
Gunasekara, F.I., Richardson, K., Carter, K., Blakely, T., 2014. Fixed effects analysis of
repeated measures data. Int. J. Epidemiol. 43, 264e269.
Haber, M.G., Cohen, J.L., Lucas, T., Baltes, B.B., 2007. The relationship between self-
reported received and perceived social support: a meta-analytic review. Am.
J. Community Psychol. 39, 133e144.
Hagerty, B.M., Williams, R.A., 1999. The effects of sense of belonging, social support,
conflict, and loneliness on depression. Nurs. Res. 48, 215e219.
Heikkinen, R.L., Kauppinen, M., 2004. Depressive symptoms in late life: a 10-year
follow-up. Arch. Gerontol. Geriatr. 38, 239e250.
Henderson, A.S., 1984. Interpreting the evidence on social support. Soc. psychiatry
19, 49e52.
Hewitt, B., Turrell, G., Giskes, K., 2010. Marital loss, mental health and the role of
perceived social support: findings from six waves of an Australian population
based panel study. J. Epidemiol. Community Health 308e314.
Jahoda, M., 1982. Employment and Unemployment: A Social-psychological Analysis.
Cambridge University Press, Cambridge, England.
Kawachi, I., Berkman, L.F., 2001. Social ties and mental health. J. Urban Health 78,
458e467.
Lakey, B., Cronin, A., 2008. Low social support and major depression: research,
theory and methodological issues. In: DOBSON, K.S., DOZOIS, D. (Eds.), Risk
Factors for Depression. Academic Press, San Diego, CA, pp. 385e408.
Lakey, B., Orehek, E., 2011. Relational regulation theory: a new approach to explain
the link between perceived social support and mental health. Psychol. Rev. 118,
482e495.
Lyyra, T.-M., Heikkinen, R.-L., 2006. Perceived social support and mortality in older
people. J. Gerontology Ser. B Psychol. Sci. Soc. Sci. 61, S147eS152.
Mcdowell, T.L., Serovich, J.M., 2007. The effect of perceived and actual social support
on the mental health of HIV-positive persons. AIDS care 19, 1223e1229.
Mckee-Ryan, F., Song, Z., Wanberg, C.R., Kinicki, A.J., 2005. Psychological and
physical well-being during unemployment: a meta-analytic study. J. Appl.
Psychol. 90, 53e76.
Milner, A., Page, A., Lamontagne, A.D., 2013a. Duration of unemployment and sui-
cide in Australia over the period 1985-2006: an ecological investigation by sex
and age during rising versus declining national unemployment rates.
J. Epidemiol. Community Health 67, 237e244.
Milner, A., Spittal, M.J., Page, A., Lamontagne, A.D., 2013b. The effect of leaving
employment on mental health: testing ‘adaptation’ versus ‘sensitisation’ in a
cohort of working-age Australians. Occup. Environ. Med.
Olesen, S.C., Butterworth, P., Leach, L.S., Kelaher, M., Pirkis, J., 2013. Mental health
affects future employment as job loss affects mental health: findings from a
longitudinal population study. BMC Psychiatry 13. http://dx.doi.org/10.1136/
oemed-2013-101759, 144e144.
Paul, K.I., Batinic, B., 2010. The need for work: Jahoda's latent functions of
employment in a representative sample of the German population. J. Organ.
Behav. 31, 45e64.
Paul, K.I., Moser, K., 2009. Unemployment impairs mental health: meta-analyses.
J. Vocat. Behav. 74, 264e282.
Pinquart, M., Sorensen, S., 2000. Influences of socioeconomic status, social network,
and competence on subjective well-being in later life: a meta-analysis. Psychol.
Aging 15, 187e224.
Procidano, M.E., 1992. The nature of perceived social support: findings of meta-
analytic studies. In: Spielberger, C.D., Butcher, J.N. (Eds.), Advances in person-
ality assessment, 9. Erlbaum, Hillsdale, NJ, pp. 1e26.
Rumpf, H.J., Meyer, C., Hapke, U., John, U., 2001. Screening for mental health: val-
idity of the MHI-5 using DSM-IV Axis I psychiatric disorders as gold standard.
Psychiatry Res. 105, 243e253.
Russell, D.W., 1996. UCLA Loneliness Scale (Version 3): reliability, validity, and factor
structure. J. Pers. Assess. 66, 20e40.
Thoits, P.A., 2011. Mechanisms linking social ties and support to physical and mental
health. J. Health Soc. Behav. 52, 145e161.
Tiikkainen, P., Heikkinen, R.L., 2005. Associations between loneliness, depressive
symptoms and perceived togetherness in older people. Aging Ment. Health 9,
526e534.
Turner, R.J., Marino, F., 1994. Social support and social structure: a descriptive
epidemiology. J. Health Soc. Behav. 35, 193e212.
Vinokur, A., Caplan, R.D., 1987. Attitudes and social support: determinants of job-
seeking behavior and well-being among the unemployed1. J. Appl. Soc. Psy-
chol. 17, 1007e1024.
Ware Jr., J.E., 2000. SF-36 health survey update. Spine (Phila Pa 1976) 25,
3130e3139.
Watson, N., 2011. Methodology for the HILDA Top-up Sample. Hilda Project Tech-
nical Paper Series No 1/11. Melbourne Institute, Melbourne.
Weeks, D.G., Michela, J.L., Peplau, L.A., Bragg, M.E., 1980. Relation between loneli-
ness and depression: a structural equation analysis. J. Pers. Soc. Psychol. 39,
1238e1244.
Wilkins, R., 2013. Families, Incomes and Jobs, Volume 8: A Statistical Report on
Waves 1 to 10 of the Household, Income and Labour Dynamics in Australia
Survey. Melbourne Institute of Applied Economic and Social Research Faculty of
Business and Economics, Melbourne.
Yamazaki, S., Fukuhara, S., Green, J., 2005. Usefulness of five-item and three-item
Mental Health Inventories to screen for depressive symptoms in the general
population of Japan. Health Qual. Life Outcomes 3, 48e48.