You are on page 1of 8

Social Science & Medicine 72 (2011) 116e123

Contents lists available at ScienceDirect

Social Science & Medicine


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

A longitudinal analysis of the relationship between changes in socio-economic


status and changes in health
Bjrn Hallerd a, *, Jan-Eric Gustafsson b
a
b

Department of Sociology, University of Gothenburg, Box 720, 405 30 Gothenburg, Sweden


Department of Education, University of Gothenburg, Sweden

a r t i c l e i n f o

a b s t r a c t

Article history:
Available online 21 October 2010

In this paper we contribute to discussion on the relationship between different aspects of socioeconomic status (SES) and health. Separating different aspects of SES facilitates the specication of
a structural relationship between SES indicators and morbidity. Longitudinal data and the utilization of
growth curve modelling enable an empirical analysis of the direct relationship between changes over
time in SES indicators and changes in morbidity. Our empirical analysis is based on panel data (N 2976)
derived from the annual Swedish Survey of Living Conditions. The panel, which consists of respondents
that at the rst panel wave were between 31 and 47 years old, is followed for 16 years, starting in 1979.
Data are gathered at three points of time. A growth curve model is set up using structural equation
modelling. The structural relationship and changes over time are simultaneously estimated. It is shown
that in relation to health occupational position is crucial, canalising the effects of class of origin and
education. More prestigious jobs are related to initially good health and to a less rapid deterioration in
health. At the same time initial health affects occupational mobility, conrming a health selection into
less prestigious jobs. It is also shown that change of occupation and income are related to change in
health. The analysis conrms a strong relationship between SES and morbidity and shows that initial SES
affects later changes in morbidity, i.e., a causal relationship exists between SES and morbidity. But, the
analysis also demonstrates the existence of selection effects, meaning that initial morbidity causes less
favourable changes in SES. It is nally revealed that changes in occupational prestige and income changes
co-vary with changes in morbidity. Hence, the analysis provides basic information necessary to make any
assumption about causality and selection in relation to SES and health.
2010 Elsevier Ltd. All rights reserved.

Keywords:
Socio-economic status
Latent growth curves
Morbidity
Incomes
Sweden
Causality
Health selectivity
Longitudinal

Introduction
In this paper, longitudinal data are used to unveil the relationship between socio-economic status (SES) and morbidity among
Swedish adults. Our aim is to separate different central aspects of
SES from each other, enabling specication of a structural model,
while simultaneously estimating the relation between different SES
indicators and morbidity. We will utilize longitudinal data to estimate changes over time in SES and morbidity, thereby elucidating if
and how individual changes in SES are related to changes of
morbidity. That is, we wish to empirically analyse the degree to
which changes in a given SES indicator is directly related to changes
in health status. This type of analysis is essential for our understanding of the mechanisms that generate inequity in health. For
example, if we think that income affects health and that a more

* Corresponding author. Tel.: 46 0 31 786 4758.


E-mail address: bjorn.hallerod@sociology.gu.se (B. Hallerd).
0277-9536/$ e see front matter 2010 Elsevier Ltd. All rights reserved.
doi:10.1016/j.socscimed.2010.09.036

equal income distribution will decrease inequity in health, then we


need to pursue a policy that results in a more equal income
distribution. However, the rst step needed to make sure that this
policy actually will have any effect is to know if there is a direct link
between changes in income and changes in health.
The paper is organized in the following way. A discussion about
the relation and causality between different aspects of SES and
health is held in the following section. Thereafter the data, operationalizations and method are presented. The empirical analysis is
undertaken in the penultimate section, followed by a discussion of
the results.
Class of origin, occupational position, education, and income
Looking at sociological theories that have evolved around the
concept class, i.e., the individuals positions on the labour market,
occupational position can be seen as important from three
perspectives. First, different occupations are related to income
differences, disparate career prospects, uneven exposure to

B. Hallerd, J.-E. Gustafsson / Social Science & Medicine 72 (2011) 116e123

unemployment risks, etcetera. These differences make it possible to


distinguish between economical classes (Bihagen & Hallerd, 2000;
Scott, 1996), which determine the ability to consume goods and
services, for example health care or high quality food, which in
turn can be linked to differences in health status (Davey Smith &
Brunner, 1997; Wagstaff & van Doorslaer, 2000). Relative economic
differences can also lead to social exclusion (e.g. Townsend, 1979)
and status differences, which in itself can generate health differences (Marmot, 2004; Wilkinson & Pickettt, 2010). An important
point is that the mechanism connecting economical classes to
health is based on an assumption about equal consumption preferences across classes, which leads to the central conclusion that
eradication of income differences would lead to an abolishing of
systematic consumption differences. From a second perspective
employment positions can be seen as the basis for behavioural
cleavages and, hence, constitute social classes (Bourdieu, 1984).
Here the assumption is that preferences are unequal and that
people in different occupational positions consume differently at
any given level of economic standard, which means that class
differences in health can be relatively unrelated to economic
differences. A third reason to expect health to correlate with
occupation is the differences in health hazards across various
workplaces (Chandola & Jenkinson, 2000; Karasek, 1979).
Likewise when discussing the impact of parents SES on the
health situation of their offspring (Halldrsson, Kunst, Khler, &
Mackenbach, 2000; van de Mheen, Stronks, Looman, & Mackenbach,
1998; van de Mheen, Stronks, & Mackenbach, 1998; Remes,
Martikainen, & Valkonen, 2010) different mechanisms related to
economic and social class can be distinguished. Thus, children that
are exposed to economic hardship might develop health problems
because they lack basic consumption items, are relative deprivation
and lack status. At the same time, parents class-specic behaviour
can expose children to different types of health hazards during
childhood and also, via socialisation, affect behaviour in adult life as
well (Mayer, 1997; Sobolewski & Amato, 2005; Wagmiller, Lennon,
Kuang, Alberti, & Aber, 2006). Also education can be of importance
to health for a number of reasons. Educational attainment is related
to class of origin insofar that it reects both parents economic
constraints and class specic behaviour during upbringing and is
pivotal as determinant of class position in adult life (Erikson &
Jonsson, 1996; Mayer, 1997). However, it could also be that education provides knowledge about how to avoid unhealthy behaviour
and optimize use of health services. Willingness to invest in human
capital may also promote self-esteem and may itself be promoted by
factors that encourage a healthy way of life (Flouri, 2006; Ross & van
Willigen, 1997; Shawn, Dorling, Gordon, & Davey Smith, 1999).
Causation and selection
The causal relationship between class background, educational
attainment, occupational position and income is in most cases
more or less given: people are brought up, acquire an education, get
a job and earn an income. The causal relationship between SES and
health is less obvious as health can function as a selection mechanism in relation to SES (Blane, Davey Smith, & Bartley, 1993;
Mulatu & Schooler, 2002; van de Mheen, Stronks, Schrijvers, &
Mackenbach, 1999). Thus, gaining knowledge about the direction
of causation is of fundamental importance. Yet, it is unlikely that we
will ever achieve a nal settlement of the issue, as reciprocal
processes are most likely at work (Mulatu & Schooler, 2002; Smith,
1999). However, causation seems to be more or less given
depending on which aspects of SES are studied.
The relation between class of origin and childrens health is
most reasonably interpreted as a causal relationship, where the
causation goes from parent to child (Marmot, 2004). There is some

117

empirical evidence that show that childhood health not only is


linked to health later in life but also has a more direct impact on
adult SES and childhood health has been related to cognitive skills,
educational achievements and, hence, labour market prospects in
adult life (Haas, 2008; van de Mheen, Stronks, & Mackenbach, 1998;
van de Mheen, Stronks, Looman, et al., 1998; Palloni, Milesi, White,
& Turner, 2009; Smith, 2009). Smith (2009) has also shown that
poor childhood health is associated with marrying a partner who
has lower earnings, which leads to a selection into low income
households. Some studies report a direct effect of childhood
health on the risk of downward inter-generational mobility (Manor,
Matthews, & Power, 2003; van de Mheen, Stronks, & Mackenbach,
1998). However, Lundberg (1991), did not nd any health selection
related to inter-generational mobility, and Kreholt (2000) on the
contrary found support for the healthy worker effect i.e., that
people with health problems are selected into less physically
demanding, white collar occupations. So, selection into educational
and occupational positions can be observed, but the result of the
selection is not always given. It also seems likely that selection
effects play a minor role for the general relationship between SES
and health (Cardano, Costa, & Demaria, 2004; Davey Smith, Blane, &
Bartely, 1994; Huurre, Rahkonen, Komulainen, & Aro, 2005; Manor
et al., 2003; Marmot, 2004; Power, Matthews, & Manor, 1996;
Shawn et al., 1999; van de Mheen et al., 1999).
We will, however, argue that health selection may be more
important for understanding the relationship between health and
income (van Doorslaer & Koolman, 2004; van de Mheen, Stronks, &
Mackenbach, 1998). Among the employed, health problems are
associated with income shortfall. The degree of income loss is
dependent on welfare state arrangement and varies between
countries and over time within countries (Korpi & Palme, 2004).
During the period studied here, that is the early 1980s to the late
1990s, several changes of the Swedish sickness benet took place
but compared to most other countries the income replacement rate
was fairly high throughout the whole period. Nevertheless, absence
from work was associated with an income shortfall and we can
expect that individual change in morbidity is closely connected to
changes of income. In the longer perspective, health can affect
income also because of a negative impact on career opportunities.
But, as discussed above, both low income and lack of occupational
career can have a negative impact on health. It is therefore likely
that what we see is reinforcing mechanisms not one way causality
(Smith, 1999). The uncertainty regarding causality is shown in
empirical studies. Thiede and Traub (1997) concluded that there is
a mutual relationship between poverty and health. van Agt,
Stronks, and Mackenbach (2000) found that persons with chronic
illness tend to be poor because their illness jeopardizes their
economy. Hallerd (2000) found an impact of health on economic
standard, but not the reverse relationship.
So, our interpretation of extant research is that the relationship
between income and health can best be described in terms of
vicious circles, i.e., that ill health can damage earning ability,
which further undermines the health situation. However, in order
to be sure that there is any kind of causal relationship between
income and health we need to know if individual changes in
income are related to individual changes in health. If not, both the
assumption of causal relationship and the idea of vicious circles are
wrong.
Data and operationalizations
The analyses are based on data from the annual Swedish Survey
of Living Conditions, which is based on a representative sample of
the Swedish population aged 16e84. In 1979, a panel approach was
introduced, and since then a sub-sample has been re-interviewed

118

B. Hallerd, J.-E. Gustafsson / Social Science & Medicine 72 (2011) 116e123

every eighth year. Here data from 1979, 1980, 1981 and 1982
constitute the rst panel wave. The panel was re-interviewed in
1986, 1987, 1988, 1989 and 1990 and in 1994, 1995, 1996, 1997 and
1998. The rst panel wave will henceforth be labelled t0, the second
t1 and third t2. The working sample is restricted to those who were
between 31 and 47 years old at t0, and accordingly between 47 and
63 years old at t2. The sample is further restricted to those who
were part of the labour force at t0, that is, those who were
employed, self employed or temporarily unemployed. The reason
for this restriction is that we need to link each individual to an
occupation. With these limitations in place, we are left with
a working sample of 2976 cases.
Our indicator of morbidity is based on self-reported diagnoses.
Respondents were asked whether they suffer from a long-standing
illness, injury, handicap or other weakness. If the answer was yes,
they were asked to specify what kind of illness or illnesses they
suffered from. Every respondent could report an innite number of
diagnoses. For the six rst reported diagnoses two follow up
questions was asked: First, how often are you affected by the
complaint? The respondent had to choose between four response
alternatives ranging from seldom to all the time. Second, how
serious is the complaint? Again four response alternatives were
offered, this time ranging from trivial to very severe. Given this
information we calculate a morbidity index in the following way:

M d1.6i fc
where d is the self-reported diagnoses given by respondent i, f
denotes how often i is affected by d and, c indicates how severe the
complaints are. Hence, the morbidity index M is the sum of selfreported diagnoses, where each diagnoses is weighted in relation
to how often the respondent is suffering from complaints and how
serious the complaints are. As can be seen from Table 1, there is, as
expected, an increase of morbidity prevalence over time.
Occupational position is measured with the Standard International Occupational Prestige Scale (SIOPS) (Ganzeboom & Treiman,
1996). Even though SIOPS is based on occupational prestige there
is a strong correlation with more commonly used categorical occupational classications such as for example the EGP-schema

Table 1
Descriptive statistics.
t1

t2

37.3

49.2

51.2
48.8

e
e

e
e

28.3
19.8
7.5
8.8

e
e
e
e

e
e
e
e

6.0
12.5
15.3

e
e
e

e
e
e

34.0
31.2
10.8
11.5
12.5

30.3
30.6
11.4
13.9
13.9

29.7
31.4
10.3
13.4
15.2

2.38(1.37)
40.9(12.3)
120,803
(38,411)
2.09(4.51)

2.52(1.39)
41.9(12.5)
129,804
(41,166)
3.37(6.71)

2.54(1.41)
44.0(13.6)
140,326
(73,077)
5.41(9.16)

t0
Percent
At least one reported symptom 27.1
Women
Men
Class of origin
Unskilled blue collar worker
Skilled blue collar worker
Lower white collar worker
Middle range white collar
Worker
Higher white collar worker
Self employed
Farmer
Education
Primary education
Lower secondary
Upper secondary
Post secondary
Tertiary education
Mean (standard deviation)
Ordinal scaled education index
Occupational prestige (SIOPS)
Equivalent disposable income
(SEK)
Morbidity index

(Erikson & Goldthorpe, 1993). Hence, blue collar workers are typically scoring low and white collar workers are scoring high on the
SIOPS scale. In relation to this study, SIOPS have two advantages.
First, there is a clear assumption about an occupational hierarchy
(Bihagen & Hallerd, 2000). Second, if we, for example, use the EGP
schema the only changes over time that can be observed are when
individuals are crossing class borders. In contrast, the SIOPS will also
capture changes, i.e., positive or negative careers, within occupational classes. We therefore nd SIOPS more suitable when estimating if change of occupational positions is related to morbidity. An
individual that becomes unemployed or leaves the labour force at t1
or t2 keeps their previous occupational status. The distribution of
SIOPS is shown in Table 1, and there is, as can be seen from the mean
value, a tendency towards upward mobility over time. Between t0
and t1, about 25% of the population makes an upward move but at the
same time 21% makes a downward shift. The upward movements
between t1 and t2 are somewhat larger, 29% while the proportion
that makes a downward move remains the same.
Class of origin refers to the respondents fathers, or stepfathers,
principal occupation during upbringing. It is a categorical variable
based on Statistics Swedens Socio-Economic Index. Educational
attainment is classied into ve groups and will be treated as an
ordinal variable in the analysis. Table 1 reveals that people tend to
be better and better educated over time, which is in line with the
ambition to provide adult education and lifelong learning. The
economic aspect of SES is indicated by the households equivalent
disposable income, i.e., the total household income (after taxes and
transfers) adjusted for household size in accordance with Statistics
Swedens equivalence scale (1st adult 1.16, 2nd adult 0.76,
additional adult 0.96, child 0e3 year 0.56, child 4e10
year 0.66, child 11e17 year 0.76) and recalculated into the 1998
price level. The data are gathered from the tax return register. The
reason why we have chosen household income, even though
individual income probably is more sensitive to individual changes
in health, is that household income relates more directly to the
individuals ability to consume goods and services. There is a slight
increase in mean equivalent disposable income (hereafter referred
to as income) over time, which is primarily explained by the ageing
of the panel and decreasing household sizes, i.e., children that are
leaving the nest (Hallerd, 1999). Age and sex are used as control
variables because of their well-documented relation to morbidity
(Hallerd, 2009; Wamala & gren, 2002).
Method
In order to estimate change over time we will begin by specifying a Latent Growth Curve model (LGC) (Hamaker, 2005; Llabre,
Spitzer, Siegel, Saab, & Schneiderman, 2004; McArdle & Epstein,
1987). The basic idea of LGC modelling is to describe developmental trajectories over time in terms of parsimonious models.
Using structural equation modelling (SEM), a latent variable is
identied through its relations to multiple observed variables. The
approach has several advantages. The latent variable is not inuenced by random errors of measurement, separating actual change
from measurement errors. It allows formulation of structural
models and can also be used for estimating models from incomplete data, so-called missing data modelling (Muthn, Kaplan, &
Hollis, 1987).
In the latent variable approach, assumptions are made about
distribution and the estimation problem is to determine the
parameters of the distribution, i.e., mean and variance for a normal
distribution (Muthn, 1996; Muthn & Khoo, 1998; Willet & Sayer,
1994). In a linear LGC model, one latent variable represents individual differences in intercept parameters, and another latent
variable represents individual differences in the slope parameters.

B. Hallerd, J.-E. Gustafsson / Social Science & Medicine 72 (2011) 116e123

Let us assume that we estimate an LGC model from an observed


measure of morbidity, collected at three points in time. The model
includes two latent variables: intercept and slope. The variance of
the intercept variable represents individual differences in
morbidity at t0. This latent variable has a xed relation of unity to all
three manifest variables (i.e., morbidity at t0, t1, and t2). The slope
variable represents individual differences, i.e., variance, in change
in morbidity over time. If we assume that change is a linear function
of time the slope variable has a relation of 0 with the t0 measure,
a xed relation of 1 with the t1 measure, and of 2 with the t2
measure. However, this assumption is not always realistic;
morbidity, for example, tends to accelerate as people get older. So,
in this particular analysis the linear assumption is relaxed and the
slope for t2 is estimated from data, allowing for a non-linear growth
function.
The means of the manifest variables are constrained to 0. This,
along with the xed relations between the latent variables and the
manifest variables, allows the latent variables to function as
a container, as it were, of parameters of the random coefcients
model. Thus, for the intercept latent variable, a mean and a variance
are estimated, where the variance represents individual differences
in the intercept of the LGC model. For the slope variable as well,
a mean and a variance are estimated, and here the mean represents
the general development in the population and the variance
parameter represents individual differences in change over time.
Empirical analysis and results
In the rst step of the analysis an LGC measurement model was
tted to data, i.e., to education, occupation, income, and morbidity
at t0, t1, and t2. For each one of these indicators intercept and slope
was estimated. The relationship between observed data and
intercepts are set to 1 at each point of observation. Table 2 shows
that each intercept has a signicant variance. The relationships for
the slopes are xed to 0 at t0 and 1 at t1 while the relationship
between slopes and observed data at t2 are estimated from data.
Looking rst at education slope we can see that the estimated
parameter for t2 is insignicant and also that the variance of the
slope is insignicant. Hence, we do not observe any signicant
variation and educational slope therefore is dropped from further
analyses. For occupational status there is an almost linear trend, the
estimated value for t2 being close to 2 and the slope variable has
signicant variance, which shows that there are statistically
signicant differences between individuals when it comes to
changes of occupational position. Also income and morbidity slopes
have signicant variances and for both of them the relationship
with time is curvilinear. The slope mean for income at t2 is 3.4 and
the slope mean for morbidity is 2.6, revealing a non-linear growth
of mean income and likewise a non-linear deterioration of mean
individual health over time. The t of the measurement model,
when allowing for covariation between latent variables is very
good. RMSEA is for example as low as 0.031. Separate models for
men and women have been estimated, but these results are not

119

Table 2
Latent growth curve measurement model.
Intercept

Slope

Education t0
Education t1
Education t2
Variance: latent variable

1
1
1
1.63***

0
1
1.09
0.08

Occupation t0
Occupation t1
Occupation t2
Variance: latent variable

1
1
1
123.82***

0
1
1.95***
13.91***

0.82
0.75
0.80

Income t0
Income t1
Income t2
Variance: latent variable

1
1
1
0.06***

0
1
3.40***
0.04***

0.42
0.48
0.85

Morbidity t0
Morbidity t1
Morbidity t2
Variance: latent variable

1
1
1
11.17***

0
1
2.60***
5.87***

0.55
0.43
0.65

Model t:
Chi2
Degrees of freedom
RMSEA
CFI
TLI

Explained variance
0.87
0.90
0.89

116.21
30
0.031
0.99
0.99

Signicance: p < 0.0001 ***.

reported, as they did not reveal any substantial difference regarding


the structure of the relationships.
In Table 3 bivariate correlations between the latent intercept
and slope variables are displayed. We see that people with higher
education intercept have higher occupation intercept, higher
income intercept and lower morbidity intercept. In other words,
a good education is, as expected, related to good jobs, high income
and a low level of morbidity. We can also see that education relates
to all three slopes, i.e., a good education promote occupational
careers, a positive income development and preservation of good
health. Disregarding the fact that there is a negative correlation
between occupation intercept and occupation slope, which reects
the fact that people who already at t0 have the most prestigious
occupational positions tend to stay at these positions, both occupation intercept and income intercept works in the same way as
education intercept. Table 3 also shows that all slope variables are
related to each other. For example, people that move into more
prestigious occupations tend to have a less rapid morbidity
increase. There is also an expected negative relation between
income slope and morbidity slope, i.e., increasing morbidity is
related to a less favourable income development.
Finally a structural model, specied in accordance with Fig. 1, is
estimated. The model is based on the assumption that age, sex, and
class of origin causally affect intercepts and slopes for all latent
variables. The intercept for educational attainment is assumed to
affect the intercepts for occupational position, income and
morbidity. Occupation intercept affects income and morbidity

Table 3
Correlations between intercepts and slope variables.

Occupation intercept
Income intercept
Morbidity intercept
Occupation slope
Income slope
Morbidity slope

Education intercept

Occupation intercept

Income intercept

Morbidity intercept

Occupation slope

Income slope

0.76***
0.37***
0.11***
0.14***
0.25***
0.18***

0.47***
0.15***
0.07***
0.23***
0.21***

0.10***
0.11***
0.24***
0.11***

0.14***
0.06**
0.52***

0.26***
0.25***

0.16***

Signicance: p < 0.001 **, p < 0.0001 ***.

120

B. Hallerd, J.-E. Gustafsson / Social Science & Medicine 72 (2011) 116e123

Age
Sex
Class of origin

Education
intercept

Occupation
intercept

Occupation
slope

Income
intercept

Income
slope

Morbidity
intercept

Morbidity
slope

Fig. 1. Structural model.

intercept while income intercept and morbidity intercept are


assumed to co-vary, i.e., because of the indecisive relation between
income and morbidity, no assumption about causality is made. For
each pair of latent variables it is assumed that the situation at t0
impacts on the further development, which means that intercepts
are supposed to affect slopes. We do not assume any denite causal
order between slopes, only that they co-vary. Again, separate models
for men and women have been estimated and again no substantial
differences were found concerning the structure of the
relationships.
The structural model is presented in Table 4. All parameters are
simultaneously estimated in accordance with Fig. 1 and independent
variables are listed as rows and dependent variables as columns. So,
the second column show the impact of class of origin, sex, and age on
education intercept. As can be seen all estimates are signicant,
revealing the expected relation between class of origin and educational attainment. Women have slightly lower average education and
there is also an age difference that basically reects the expansion of
the educational system. The next column shows that the direct

impact of class of origin on occupational intercept is rather weak, in


fact close to nil. However, the impact of education intercept on
occupational intercept is at the same time very strong (0.76), which,
beside the fact that a higher education leads to more prestigious jobs,
means that there is a substantial indirect effect of class of origin on
occupational intercept. There is an admittedly weak but nevertheless
direct effect of class of origin on income intercept but the largest
coefcient is, not surprisingly, the one that describes the relationship
between occupational intercept and income intercept (0.44). Looking at morbidity, we can see that there is a negative effect of occupational intercept, that is, the more prestigious occupation, the less
morbidity. There is also a signicant negative covariation between
income intercept and morbidity intercept.
Basically, what we see in this part of the structural model is more
or less what we should expect to see in most multivariate analysis:
Class of origin to a large degree determine educational attainment,
education has a substantial impact on occupational position and
occupational position is an important predictor of income. Both
occupational position and income are negatively related to
morbidity, again conrming that a good job and good economy are
linked to good health. Class of origin and education have an impact
on morbidity but only indirectly via occupational position.
The next part of the structural model focuses on change over
time. As can be seen from column six, there is a slight, but nevertheless remaining, direct effect of class of origin on occupational
slope. People who have a white collar father or a self employed
father have, net of education-, occupation-, and income intercepts,
a slightly better career prospect. Education intercept has a strong
positive effect on occupation slope, that is, it is the well educated
that have the most positive career development. The strong negative effect of occupation intercept on occupation slope simply
means that people who, given their education, already at the outset
had a prestigious occupation are less likely to make additional
career steps. There is also a positive effect of income intercept on
occupation slope. Most interesting, from the perspective of this
paper, is the negative relationship between morbidity intercept and
occupation slope. Thus, people who at t0 suffered from health
problems are less likely to move into more prestigious occupations,
a result that clearly indicates an ongoing health selection into less
prestigious occupations.

Table 4
Structural equation model. Standardized regression coefcients and correlations (within brackets).
Intercepts:
Edu.
Class of origin:
Unsk. b. collare ref.
Skilled blue collar
Lower white collar
Middle white collar
Higher white collar
Self employed
Farmers

0.06**
0.13***
0.27***
0.32***
0.14***
0.04*

Intercepts:
Education
Occupation
Income
Morbidity

Slopes:
Occup.

Inc.

0.04**
0.03*
0.02
0.01
0.01
0.01

0.04*
0.04*
0.06***
0.07***
0.02
0.00

0.76***

0.03
0.44***

Morb.

0.03
0.03
0.03
0.02
0.02
0.02
0.03
0.15***
[0.05**]

Occup.

0.01
0.05**
0.03
0.04*
0.04*
0.01

0.01
0.04*
0.04
0.06**
0.01
0.00

0.41***
0.51***
0.19***
0.14***

0.15***
0.01
0.17***
0.02

Slopes:
Occupation
Income
Control variables
Sex
Age

Inc.

[0.21***]

0.04**
0.09***

0.07***
0.03**

Signicance: p < 0.05 *, p < 0.001 **, p < 0.0001 ***.

0.05**
0.22***

0.02
0.10***

0.03
0.10***

0.01
0.00

Morb.

0.02
0.01
0.02
0.00
0.02
0.00
0.04
0.10***
0.02
0.49***
[0.17***]
[0.10***]
0.02
0.07***

B. Hallerd, J.-E. Gustafsson / Social Science & Medicine 72 (2011) 116e123

Looking at the estimates for income slope we can see that there
is again a small effect of class of origin. Education intercept has
a positive effect and we can also see that those who had a good
income at t0 also tend to have an even better income later on in life.
There are no effects of occupation- or morbidity intercept on
income slope. However, there is a covariation between occupation
slope and income slope, which means that moves to more prestigious occupations tend to generate income growths that exceeds
the mean of the general growth curve.
The last column in Table 4 relates to morbidity slope. Class of
origin has no direct impact on morbidity slope, neither does
education intercept, nor income intercept. This is not to say that
these variables do not have any impact on morbidity slope only that
the associations are indirect, mediated mainly via occupation
intercept and morbidity intercept: The higher occupation intercept,
the lower morbidity growth and the higher morbidity intercept the
higher morbidity growth. Especially the latter effect is substantial
and shows that health inequality between individuals increases
over time. Finally, and for this study most importantly, we can
observe two things. First, changes of occupational prestige are
negatively related to changes in morbidity. Hence, health development is related to change of occupational position. People that
move upwards in the occupational hierarchy tend to have a health
development that deteriorates slower than what is indicated by the
general growth curve, or alternatively, people that have a favourable health development tend to get more prestigious jobs. Second,
there is also a negative covariance between income slope and
morbidity slope. So, again we can conrm that there actually is
a relationship and that people who, net of change in occupational
status, experience increasing morbidity are also exposed to an
unfavourable income development. But, important to underline, we
cannot say if the causality mainly works one way or the other.
Conclusions
In this study longitudinal data were used to link individual
change in SES to individual change in health status. By doing so, two
things were achieved. First, we could analyse to what degree
individual SES positions at one point of time were affecting individual health development later on in life. Second, we could, in
contrast to most existing studies, analyse to what degree individual
changes in SES over time were related to change of individual
health. This type of analysis provided an essential piece of knowledge: Regardless if we look upon the relationship between SES and
health as a causal impact of SES on health or a selection effect, i.e.,
that people are selected into certain SES positions because of their
health, we need to conrm that individual change in SES is related
to individual change in health.
Our empirical approach was based on Latent Growth Curve
(LGC) and Structural Equation Modelling (SEM). An LGC measurement model was used to construct eight latent variables that
separate initial SES (educational attainment, occupational position,
and income) and initial morbidity from changes over time in SES
and morbidity. The SEM model included the latent variables plus
information about class of origin, age, and sex. The working sample
consisted of a representative longitudinal sample of the Swedish
population that were interviewed at three points in time within
a time span of 16 years and that in the early 1980s were employed
or self employed and, at that point, were between 31 and 47 years
old.
Generally the SEM analyses conrmed the relationship between
SES and health. Occupational position and, to a lesser degree,
income were negatively related to morbidity, that is, the more
favourable occupational position and the higher income the less
morbidity. Class of origin did not affect morbidity directly but had

121

an indirect effect mediated by, primarily, education. Our analysis


revealed the importance of education but, as opposed to several
other studies (e.g. Dupre, 2008; Gjonca, Tabassum, & Breeze, 2009;
Mackenbach et al., 2008), we could not verify any direct impact of
education, only indirect effects that worked via occupational
position and income development. There are two main reasons to
why our results at least partly diverge from earlier research. First,
our measure of occupational position is based on a ne graded
detailed scale, which means that we are able to capture more of the
occupational variation than usually is the case. That is, our analysis
is able to capture not only differences between broad occupational
categories but also variations within these categories. Second, since
our analysis separates initial occupational status and initial income
from change over time, we can also see that education affects
occupational carers and income development e the higher
education, the better occupational carer and income development.
This indicates that education is important for the career both
between and within occupational categories as well as for, net of
occupational status, income development. Since both change of
occupational status and income growth are important in relation to
health development the analysis helps us to understand how
education and health are related. What our analysis suggests is that
education does not affect morbidity per se, but it gives people
a valuable tool that can be translated into positive occupational and
income careers. This leads us to the more interesting ndings
related to individual change of SES and morbidity.
First, there was a negative impact of initial occupational position
on change in morbidity supporting the idea of a causal link between
occupational status and morbidity. Second, at the same time initial
morbidity had a negative impact on change of occupational position, which supports the idea that there is a health selection into
less favourable job positions. Hence, the result indicates that the
relationship between occupational position and health is caused
both by a causal impact of occupation on health and a selection of
less healthy people into less prestigious occupations. The degree to
which these two processes are mutually reinforcing each other
could not be empirically analysed in the present study but nevertheless gives indicative support to this highly plausible hypothesis
(e.g. Smith, 1999). Third, initial morbidity is strongly related to the
development of morbidity over time, indication that health
inequality within a given cohort is increasing over time, which is in
line with earlier results (e.g. Dupre, 2008; Eriksson, 2008). Fourth,
there are an increasing number of longitudinal studies of the
relationship between SES and health (Allanson, Gerdtham, & Petrie,
2010; Avendano, Jurges, & Mackenbach, 2009; Dupre, 2008;
Garbarski, 2010; Giordano, Lindstrm, 2010) and there are also
a small but growing number of studies that uses LGC models in
relation to health (Haas & Rohlfsen, 2010; Kim & Miech, 2009;
Roesch et al., 2009; Zajacova & Burgard, 2009). The special contribution of this study is that the LGC approach is used, not only to
analyse the correlations between specic SES positions at one point
of time and future health development, but to analyse to what
degree individual change of SES is related to individual changes in
health. For example, LGC allows us to estimate if and to what degree
actual advance to more prestigious occupation is directly correlated
with a more positive health development. Our results showed
a negative correlation between changes of occupational position
and changes in morbidity, which means that individuals that have
a favourable occupational career also tend to have a more positive
health development. Even though this result does not give any
denite information of the direction of the relationship it does
show that changes of occupational position and changes in health
status actually are linked to each other. In the same way, the
analysis veries that income change is related to morbidity change,
i.e., getting richer is associated with positive health development.

122

B. Hallerd, J.-E. Gustafsson / Social Science & Medicine 72 (2011) 116e123

In the introductory discussion, we identied three potential


mechanisms through which occupational position can affect
health. One of these was economic class. Our analysis shows that
both initial income and income development are of importance, net
of occupational position. Hence, economic differences do matter.
However, we can also draw the conclusion that the link between
occupation and health is not primarily related to income difference.
The searchlight should also be put on the other two mechanisms:
social class, emphasising behavioural differences, and differences in
work conditions. These two aspects of class position are not
distinguishable in the analysis presented here. In terms of the focus
of future research, the results nevertheless indicate the importance
of continuing to unpack the occupational class box.
The analytic strategy used in this article has helped to unveil the
relationship between different aspects of SES and health. It has
provided information that has conrmed the complex relationships
that are generated, on one hand, by a causal impact of SES on health
and, on the other hand, the reverse causation, i.e., health selection
into different SES positions. It has also provided essential ndings
that conrm that individual changes in SES are related to individual
changes of health, an empirical precondition to any assumptions
about causality. The results also indicate that we need to more
thoroughly investigate to what degree individual development of
SES and health are mutually reinforcing each other, creating either
good or vicious circles or, to use another terminology, cumulative
disadvantages (Dupre, 2008; Haas & Rohlfsen, 2010).
As always, the analysis has a number of limitations. We use
a global measure of morbidity, weighing together the occurrence of
symptoms, how often and how seriously the individual was
affected by the symptoms. We do not make any distinction
between different kinds of symptoms. Thus, our analysis was only
providing a general picture and overlooked the fact that different
symptoms relate differently to SES and possibly also differently to
different aspects of SES. Analysing more specic symptoms will
probably provide both a more insightful and diversied picture of
the links between SES and health (e.g. Erikson & Torssander, 2008).
Such an analysis is certainly a feasible next step but requires both
a larger data set and a further developed analytic strategy. The
analysis is also hampered by the fact that, even though our data
covers a period of 16 years, we only have data from three waves of
measurement. More frequent panel waves would open up for more
elaborated analyses and sophisticated tests of causality. Our model
is fairly parsimonious and can, like almost any model, be criticised
because of omitted confounders, our model does for example lack
an indicator of wealth (Avendano et al., 2009). This is a potential
problem and it is certainly possible to develop the model further,
not least so that we can reach a better understanding of the actual
mechanisms that link occupational position to health.
Acknowledgments
Financial support for this research was provided by the Swedish
Council for Working Life and Social Research. We have beneted
greatly from comments by Urban Janlert and his colleagues and the
reviewers remarks.
References
Allanson, P., Gerdtham, U. G., & Petrie, D. (2010). Longitudinal analysis of incomerelated health inequality. Journal of Health Economics, 29, 78e86.
Avendano, M., Jurges, H., & Mackenbach, J. P. (2009). Educational level and changes
in health across Europe: longitudinal results from SHARE. Journal of European
Social Policy, 19, 301e316.
Bihagen, E., & Hallerd, B. (2000). The crucial aspects of class: an assessment of the
relevance of class in Sweden from the 1970s to the 1990s. Work, Employment
and Society, 14, 307e330.

Blane, D., Davey Smith, G., & Bartley, M. (1993). Social selection: what does it contribute
to social class differences in health? Sociology of Health and Illness, 15, 1e15.
Bourdieu, P. (1984). Distinction. Cambridge: Harvard University Press.
Cardano, M., Costa, G., & Demaria, M. (2004). Social mobility and health in the Turin
longitudinal study. Social Science & Medicine, 58, 1563e1574.
Chandola, T., & Jenkinson, C. (2000). The new UK National Statistics Socio-Economic
Classication (NS-SEC); investigation social class differences in self-reported
health status. Journal of Public Health Medicine, 22, 182e190.
Davey Smith, G., Blane, D., & Bartely, M. (1994). Explanations for socio.economic
differentials in mortality. European Journal of Public Health, 4, 132e144.
Davey Smith, G., & Brunner, E. (1997). Socio-economic differentials in health: the
role of nutrition. Proceedings of the Nutrition Society, 56, 75e90.
Dupre, M. E. (2008). Educational differences in health risks and illness over the life
course: a test of cumulative disadvantage theory. Social Science Research, 37,
1253e1266.
Erikson, R., & Goldthorpe, J. H. (1993). The constant ux: A study of class mobility in
industrial societies. Oxford: Clarendon Press.
Erikson, R., & Jonsson, J. O. (1996). Can education be equalized? The Swedish case in
comparative perspective. Boulder: Westview Press.
Erikson, R., & Torssander, J. (2008). Social class and cause of death. European Journal
of Public Health, 18, 473e478.
Eriksson, B. G. (2008). Dispersion of registered death causes as a function of age in
the 1999 US population. In S. H. Murdock, & D. Swanson (Eds.), Applied
demography in the 21st century (pp. 223e233). New York: Springer.
Flouri, E. (2006). Parental interest in childrens education, childrens self-esteem and
locus of control, and later educational attainment: twenty-six year follow-up of
the 1970 British Birth Cohort. British Journal of Educational Psychology, 76, 41e55.
Ganzeboom, H. B. G., & Treiman, D. J. (1996). Internationally comparable measures
of occupational status for the 1988 international standard classication of
occupations. Social Science Research, 25, 201e239.
Garbarski, D. (2010). Perceived social position and health: is there a reciprocal
relationship? Social Science & Medicine, 70, 692e699.
Giordano, G. N., & Lindstrm, M. (2010). The impact of changes in different aspects
of social capital and material conditions on self-rated health over time:
a longitudinal cohort study. Social Science & Medicine, 70, 700e710.
Gjonca, E., Tabassum, F., & Breeze, E. (2009). Socioeconomic differences in physical
disability at older age. Journal of Epidemiology and Community Health, 63, 928e935.
Haas, S. (2008). Trajectories of functional health: the long arm of childhood health
and socioeconomic factors. Social Science & Medicine, 66, 849e861.
Haas, S., & Rohlfsen, L. (2010). Life course determinants of racial and ethnic disparities
in functional health trajectories. Social Science & Medicine, 70, 240e250.
Halldrsson, M., Kunst, A. E., Khler, L., & Mackenbach, J. P. (2000). Socioeconomic
inequalities in the health of children and adolecents. European Journal of Public
Health, 10, 281e288.
Hallerd, B. (1999). Economic standard of living: a longitudinal analysis of the
economic standard among Swedes 1979e1995. European Societies, 1, 391e418.
Hallerd, B. (2000). Poverty, inequality and health. In D. Gordon, & P. Townsend
(Eds.), Breadline Europe (pp. 165e187). Bristol: The Policy Press.
Hallerd, B. (2009). Ill, worried or worried sick? Inter-relationships among indicators
of wellbeing among older people in Sweden. Ageing and Society, 29, 563e584.
Hamaker, E. L. (2005). Conditions for the equivalence of the autoregressive latent
trajectory model and a latent growth curve model with autoregressive disturbances. Sociological Methods & Research, 33, 404e416.
Huurre, T., Rahkonen, O., Komulainen, E., & Aro, H. (2005). Socioeconomic status as
a cause and consequence of psychosomatic symptoms from adolescence to
adulthood. Social Psychiatry and Psychiatric Epidemiology, 40, 580e587.
Kreholt, I. (2000). Social class and mortality risk. In Swedish institute for social
research. Stockholm: Stockholm University.
Karasek, R. A. (1979). Job demands, job decision latitude and mental strain:
implications for job redesign. Administrative Science Quarterly KOLLA 258e308.
Kim, J., & Miech, R. (2009). The BlackeWhite difference in age trajectories of
functional health over the life course. Social Science & Medicine, 68, 717e725.
Korpi, W., & Palme, J. (2004). New politics and class politics in the context of
austerity and globalization: welfare state regress in 18 countries 1975e1995.
American Political Science Review, 97, 425e446.
Llabre, M. M., Spitzer, S., Siegel, S., Saab, P. G., & Schneiderman, N. (2004). Applying
latent growth curve modeling to the investigation of individual differences in
cardiovascular recovery from stress.". Psychosomatic Medicine, 66, 29e41.
Lundberg, O. (1991). Childhood living conditions, health status and social mobility:
a contribution to the health selection debate. European Sociological Review, 7,
149e161.
Mackenbach, J. P., Stirbu, I., Roskam, A. J. R., Schaap, M. M., Menvielle, G.,
Leinsalu, M., et al. (2008). Socioeconomic inequalities in health in 22 European
countries. New England Journal of Medicine, 358, 2468e2481.
Manor, O., Matthews, S., & Power, C. (2003). Health selection: the role of inter- and
intra-generational mobility on social inequalities in health. Social Science &
Medicine, 57, 2217e2227.
Marmot, M. (2004). Status syndrome: How your social standing directly affects your
health and life expectancy. London: Bloomsburry Publishing.
Mayer, S. E. (1997). What money cant buy. London: Harvard University Press.
McArdle, J. J., & Epstein, D. (1987). Latent growth curves within developmental
structural equation models. Child Development 110e133.
Mulatu, M. S., & Schooler, C. (2002). Causal connections between socio-economic
status and health: reciprocal effects and mediating mechanisms. Journal of
Health and Social Behavior, 43, 22e41.

B. Hallerd, J.-E. Gustafsson / Social Science & Medicine 72 (2011) 116e123


Muthn, B. (1996). Growth modeling with binary responses. In A. V. Eye, & C. Clogg (Eds.),
Categorical variables in developmental research: Methods of analysis (pp. 37e54).
San Diego: Academic Press.
Muthn, B., Kaplan, D., & Hollis, M. (1987). "On structural equation modeling with
data that are not missing completely at random. Psychometrika, 42, 431e462.
Muthn, B., & Khoo, S. T. (1998). Longitudinal studies of achievement growth using
latent variable modeling. Learning and individual differences. Special Issue:
Latent Growth Curve Analysis, 10, 73e101.
Palloni, A., Milesi, C., White, R. G., & Turner, A. (2009). Early childhood health,
reproduction of economic inequalities and the persistence of health and
mortality differentials. Social Science & Medicine, 68, 1574e1582.
Power, C., Matthews, S., & Manor, O. (1996). "Inequalities in self rated health in the
1958 birth cohort: lifetime social circumstances or social mobility. British
Medical Journal, 313, 449e453.
Remes, H., Martikainen, P., & Valkonen, T. (2010). Mortality inequalities by parental
education among children and young adults in Finland 1990e2004. Journal of
Epidemiology and Community Health, 64, 136e141.
Roesch, S. C., Norman, G. J., Adams, M. A., Kerr, J., Sallis, J. F., Ryan, S., et al. (2009).
Latent growth curve modeling of adolescent physical activity testing parallel
process and mediation models. Journal of Health Psychology, 14, 313e325.
Ross, C. E., & van Willigen, M. (1997). Education and the subjective quality of life.
Journal of Health and Social Behavior, 38, 275e297.
Scott, J. (1996). Stratication and Power: Structures of Class, Status and Command.
Cambridge: Polity Press.
Shawn, M., Dorling, D., Gordon, D., & Davey Smith, G. (1999). The widening gap.
Bristol: The Policy Press.
Smith, J. P. (1999). Healthy bodies and thick wallets: the dual relation between
health and economic status. Journal of Economic Perspectives, 13, 145e166.
Smith, J. P. (2009). The Impact of childhood health on adult labour market
outcomes. Review of Economics and Statistics, 91, 478e489.
Sobolewski, J. M., & Amato, P. R. (2005). Economic hardship in the family of origin
and childrens psychological well-being in adulthood. Journal of Marriage and
the Family, 67, 141e156.

123

Thiede, M., & Traub, S. (1997). Mutual inuence of helath and poverty. Evidence
from german panel data. Social Science & Medicine, 45, 867e877.
Townsend, P. (1979). Poverty in the United Kingdom. Harmondsworth: Penguin
Books Ltd.
van Agt, H. M. E., Stronks, K., & Mackenbach, J. P. (2000). Chronic illness and poverty
in The Netherlands. European Journal of Public Health, 10, 197e200.
van de Mheen, H., Stronks, K., Looman, C. W. N., & Mackenbach, J. P. (1998). Does
childhood socioeconomic status inuence adult health through behavioural
factors? International Journal of Epidemiology, 27, 431e437.
van de Mheen, H. D., Stronks, K., & Mackenbach, J. P. (1998). A lifecourse perspective on
socio-economic inequalities in health: the inuence of childhood socio-economic
conditions and selection processes. Sociology of Health & Illness, 20, 754e777.
van de Mheen, H., Stronks, K., Schrijvers, C. T. M., & Mackenbach, J. P. (1999). The
inuence of adult ill health on occupational class mobility and mobility out of and
into employment in The Netherlands. Social Science & Medicine, 49, 509e518.
van Doorslaer, E., & Koolman, X. (2004). Explaining the differences in incomerelated health inequalities across European countries. Health Economics, 13,
609e628.
Wagmiller, R. L., Lennon, M. C., Kuang, L., Alberti, P. M., & Aber, J. L. (2006). The
dynamics of economic disadvantage and childrens life chances. American
Sociological Review, 71, 847e866.
Wagstaff, A., & van Doorslaer, E. (2000). Income inequality and health: what does
the literature tell Us. Annual Review of Public Health 543e567.
Wamala, S., & gren, G. (2002). Gender inequity and public health. European Journal
of Public Health, 12, 163e165.
Wilkinson, R., & Pickettt, K. (2010). The spirit level: Why equality is better for
everyone. London: Penguin Books.
Willet, J. B., & Sayer, A. G. (1994). Using covariance structure analysis to detect
correlates and predictors of individual change over time. Psychological Bulletin,
116, 363e381.
Zajacova, A., & Burgard, S. A. (2010). Body weight and health from early to midadulthood: a longitudinal analysis. Journal of Health and Social Behavior, 51,
92e107.

You might also like