Gradient, risk factors, baseline health
Do risk factors explain more of the social gradient in self-reported health when
adjusting for baseline health?
Jacob Nielsen Arendt(*) and Jørgen Lauridsen(**)
(*) Associate Professor, Institute of Public Health – Health Economics, University of
Southern Denmark
(**) Professor, Institute of Public Health – Health Economics, University of Southern
Denmark
Correspondance: JN. Arendt, J. B. Winsløws vej 9B, 5000, Odense C, E-mail:
[email protected] Phone: +45 6550 3843. Fax: +45 6550 3880.
Word count: 3392
1
Gradient, risk factors, baseline health
Abstract
Background: It has often been debated the extent to which known risk factors explain socioeconomic differences in health. While common in mortality studies, few studies of morbidity
adjust for baseline health. In this study we argue that there are sound reasons to do so, and
examine whether a set of risk factors explain a larger part of social gradients in men’s and
women’s self-rated health (SRH) in Denmark when controlling for previous health.
Methods: We use interval regression models on longitudinal survey data from 1990 and
1995 separately for Danish male and female workers aged 18-59. Results: Large social
gradients are found in SRH for both men and women. The included risk factors (smoking,
body-mass index, high blood pressure and job satisfaction) reduce the educational gradient in
SRH by 40% (based on highest versus no education), the wage gradient by 18% and leaves
occupational gradients (based on no employment versus white collar workers) unaltered for
men. For women similar gradients are altered by 6% and 22% and 14% in cross-sectional
models. Controlling for base-line health 5 years earlier, the risk factors reduce the education,
occupation and wage gradients by 45%, -15% and 17% for men and by 5%, 25% and 15%
for women. Conclusion: The findings suggest that common risk factors do not explain a
larger fraction of social health inequalities in dynamic than in static models of self-reported
health.
Keywords: health, dynamics, socio-economic status, risk factors.
2
Gradient, risk factors, baseline health
INTRODUCTION
During the last couple of decades, a considerable amount of attention, academic as well as
public, has been directed towards differences in health and mortality that are related to social
status. In European countries, several trans-national working groups have addressed the issue
and many public authorities have set as a main health goal to reduce social inequalities in
health1-4. This study adds to our knowledge on how large a fraction of social inequalities that
is explained by a given set of risk factors using Danish data.
Some evidence suggests that social inequalities in morbidity and mortality are quite
substantial by international standards in Nordic European countries in spite of their
egalitarian policies5-7. This has though been critized8. Moreover, in many western countries,
including Denmark, there has been a rise in social inequality in mortality and morbidity over
recent decades9-11. There are therefore great incentives to try to understand these
relationships further to help improve policy.
It is often debated how large a fraction of social inequalities in health that is explained by
known risk factors. There is now general consensus that the causes of social inequalities in
health are multidimensional, emphasizing both material (e.g. poverty, financial problems),
behavioural (e.g. smoking, alcohol consumption, diet and physical activity), biological
(blood pressure, cholesterol, body mass index) and psychosocial pathways (e.g. stress, lack
of social support, control)12-20. Health knowledge and health-related behaviour often explain
a quarter and sometimes more than half of social inequalities in health3;13;16;17;21-27 while a
combination of various sets of risk factors may explain all social inequalities in health14;18;28.
3
Gradient, risk factors, baseline health
Studies that decompose inequalities into mediating pathways provide important insights on
which pathways that seem to be the most important mediators from socioeconomic position
to health. Comparison of the explanatory power of given risk factors in different studies is,
however, often complicated by the use of different study populations, health measures and
risk factors. Moreover results should be interpreted carefully in the sense that “explaining
inequalities” is not tantamount to being able to manipulate inequalities nor is it equivalent to
knowing why inequalities arise.
The outset of this study is hypotheses of the social health gradient that take the dynamic
nature of health into account. This holds for the various selection hypotheses as well as the
literature emphasizing a life course perspective. Selection may occur directly, i.e. when adult
health affects adult SES (when SES is measured by employment status, this is referred to as
the healthy worker effect) or indirectly, when unobserved factors influence both adult SES
and adult health, creating a spurious relation among the two. When health selection arises in
adulthood it is often called intragenerational selection, as opposed to intergenerational
selection, arising when childhood health prevents upward mobility in SES
29
. While there is
certainly evidence that all types of selection occurs, there is less consensus on the size of its
contribution to social inequalities in health. Most evidence suggests that direct selection on
occupation plays a minor role for occupation related health inequalities30-32, whereas
evidence on selection with respect to education33;34 and particularly income is more mixed3538
. Indirect selection has been suggested as a process of which unfavourable health events
accumulate over life30. This is exactly the dynamic aspect stressed by the accumulation
model, which is one out of several models with a life course perspective, that has received
growing interest during the recent decade1;39-45. Related models include the biological
programming model, where conditions in utero or early childhood “programme” the
4
Gradient, risk factors, baseline health
individual risk for later adult diseases39 and the pathway model (or chain-of-risk model),
where exposure to adverse health events increases the risk of later adverse exposures46.
Within the health economic literature, it has been pointed out how a version of the life course
hypothesis can be captured by an autoregressive dynamic model of health47, capturing the
idea that on average individuals recover from illnesses, but some events have lasting negative
health effects. Social status may matter for two reasons: either because recovery from
illnesses differs across social groups or because of different exposures to health threats across
social groups. A final strand of literature that emphasizes the dynamic nature of health
determinations is also found within health economics. This literature stresses health as a
capital good and therefore, that health today depends on previous health and health
investments34;48
The main message from this literature is that health is determined over potentially long time
periods. This has consequences when trying to explain health differences. Our a priori
expectation is that current risk factors may explain a larger fraction of inequalities in adult
health when adjusting for baseline health some years ago, than unadjusted health levels
where inequalities may be accumulated over a life time. Of course, it is possible that
controlling for baseline health reduces the explanatory power of risk factors, if they primarily
explain baseline rather than current health. It is also possible that adjusting for baseline
health may alter which pathways that acts as the most important mediators from social
position to health as well as the quantitative importance of each pathway. These issues
cannot be determined a priori and remain empirical questions.
The goal of this study is therefore to compare to what extent a given set of risk factors
explain social health gradients, with and without adjustment for baseline health. Various
5
Gradient, risk factors, baseline health
strands of research are related to this purpose. Numerous studies have been conducted that
examine the life course perspective. Most of this research examines the impact of socioeconomic position during the life-course on adult health1;40-43, This perspective differs from
the present study, where we aim at controlling for previous health rather than previous
socioeconomic conditions. Other types of studies more closely related to the present are
studies that control for general measures of baseline health. This includes numerous studies
of mortality16;18;25;49-54. However, even though many of these studies confirm the importance
of baseline health, only few of these allow us to infer the extent to which risk factors explain
more of the gradient when controlling for baseline health.
There is much more limited empirical work on morbidity that control for baseline health.
One study, which employs the same data as the present study, examines whether obesity,
smoking and work environment can account for SES differences in self-reported health using
probit models given either previous poor or good SRH55. As opposed to this study they use
an indicator of social class (based on employment grade, job title and education) as measure
of SES. They show that obesity, smoking and work environmental risk factors reduce the
social class differences in SRH by 57% when controlling for baseline health. No comparison
with static models is provided. A British cohort study found that adjusting for social class
and self-rated health at age 23, social class differences in self-rated health at age 33
vanished56. A few US studies have applied two or three-wave panels to deal with the
dynamics in health57;58. Indeed, it has been found that social gradients in a range of health
conditions are weakened when controlling for both health-risk factors and past health
conditions58. However, it is not possible to infer to what extent the risk factors or previous
health conditions account for the gradient.
6
Gradient, risk factors, baseline health
METHODS
The statistical analyses are conducted using an interval regression model, which is an
alternative to ordered probit/logit in the case where the threshold parameters among health
categories are known from an external source. Using such information, the estimates of the
coefficients for the individual characteristics are more efficient59 and are directly
interpretable as marginal effects on a health scale that is bounded between zero (dead) and
one (perfect health). We use two sets of external estimates of the thresholds between health
categories to obtain robustness. The first set was derived from applying the HUI-3 health
instrument in a 1994 Canadian survey60;61. The second set was derived applying the 15D
health instrument to a Finnish 1995/96 survey62. These thresholds of the health categories are
summarised in Table 1. Notice that the HUI index is multiplied by 100 so that it ranges from
0 to 100 rather than from 0 to 1, in order to avoid unnecessarily small coefficients in the
interval regressions to follow. As the two sets of thresholds yielded quite similar results only
those based upon HUI thresholds are shown. The results based on 15D are available from the
authors upon request.
Joint analyses of wage, occupation and education gradients are conducted. Four sets of
estimates are presented for each gender: one with only SES variables, one where we add the
risk factors and then these are repeated in models with baseline health. All estimations
control for age. The overall results are robust to different functional forms in age or wages.
References to significance levels will be at a five percent level, unless otherwise mentioned.
Data
The present study uses data from the Work Environment Cohort Study (WECS), which is a
survey of Danish employees, aged 18-59 in 1990 with a follow-up in 1995. The survey was
7
Gradient, risk factors, baseline health
collected by the Danish National Institute of Occupational Health (AMI) and the National
Institute of Social Research (SFI). Further details are found in the appendix.
The data set includes self-rated health (very poor, poor, fair, good and very good), which will
be our main health measure. Self-rated health has been found to be a good overall summary
measure of health, related e.g. to risks of functional disability and mortality63;64. Some studies
have been concerned that self-reported health measures are reported with error which, if
related to SES, may create a spurious gradient65-68.
We use three indicators of SES: education, occupation and wages. Occupation is divided into
white collar workers, skilled and unskilled blue collar workers and a residual group,
consisting mostly of people out of employment or self-employed. The following educational
levels are used: 9th form or less, semi-skilled, short education, vocational, short advanced,
medium advanced and long advanced degree. Short education covers those with 10th form,
another type of education beyond 9th form of at most a year of length and those with high
school. The vocational and the three advanced studies constitute the main types of educations
directly qualifying for the labour market. Finally, we use pre-tax hourly wages as SES
measure. Hourly wages are typically viewed as a measure of employee productivity in the
economic literature and may in broader terms reflect social status derived on the labour
market. Besides, using hourly wages rather than gross income cleans the measure for
differences in hours of work. The final data set consists of 1985 observations for men and
1702 observations for women.
Risk factors
The following binomial risk indicators are included. Physiological: being fat (body mass
index (BMI) between 25 and 30) or obese (BMI above 30). Biological: whether a doctor has
reported having high blood pressure. Behavioural: number of years having been smoking.
We also include a binomial indicator of whether being satisfied with ones job, to account for
8
Gradient, risk factors, baseline health
risk factors related to psychosocial work environment. Job satisfaction has been shown to be
closely related to sickness absenteeism and to be related to psychosocial problems such as
burnout, low self-esteem, depression and anxiety69;70. More typical measures of psychosocial
risk factors include locus of control, job strain or stress21;51, see also15;71. We refer broadly to
the set of variables describing BMI, blood pressure, smoking and job satisfaction as risk
factors.
Table 2 contains descriptive statistics; distributions of explanatory variables and the share
with poor SRH (the original three worst outcomes fair, poor or very poor health). The table
shows that SRH varies with type of education, age, occupation, wage level, the risk factors
and baseline health Finally, to apply health in 1990 to the regression, the categories of health
for this year were scaled by replacing with the mid-point of the intervals of the health
categories according to the HUI and 15D thresholds.
RESULTS
Social gradients for men
Table 3 contains results with HUI thresholds for men. The results show that on average all
men with any type of schooling beyond 9th form, besides the semi-skilled, report better health
than those with at most 9th form. However, only the semi-skilled education group is
significant, making the education variables jointly significant (see bottom of the table). It
should also be noted that if the dummies for semi-skilled and short education are included in
the reference group, all higher educations are significant. Therefore, it makes sense to
evaluate the size of the education gradient. Those with an advanced degree have the best
health (table 3, model (1)), on average reporting higher health corresponding to 0.004-0.0075
higher health index. To put this into a perspective, the standard deviation of the HUI based
health index is 0.054 in 1990.
9
Gradient, risk factors, baseline health
Only the SRH of the residual occupation group (mainly out of employment) differs
significantly from the other occupation groups. The occupation variables are however jointly
insignificant. Finally, hourly wages are strongly and significantly related to SRH. If wages
increase by one standard deviation, the health index increases by 0.005, corresponding to the
impact of 7.5 years of age.
Controlling for risk factors and baseline health
All the risk factors are significant and particularly low job satisfaction and high blood
pressure is associated with poor health. When the risk factors are included both the wage and
the educational gradient are reduced (table 3, model (1) versus model (2)), but significant
wage and education differences persist. Measured by the difference in coefficients for longadvanced education versus no schooling, the reduction is 40%. The differences between
white collar workers and the residual group are unaltered. The wage gradient is reduced by
18%. In model (3) we see that when baseline health status in 1990 is controlled for, but the
risk factors are excluded, there is a significant wage gradient similar to the one in model (2).
Controlling for the risk factors reduces this gradient in a similar manner as in the static
model, namely by 45% for the educational differences and by 17% for the wage gradient,
although widening occupational differences by 15%. The results reveal that SRH shows a
very large degree of persistence over the 5 year period.
Social gradients for women
The education related differences in SRH are substantial for women (table 4, model (1)).
Women with a vocational education have better health than those with a medium or long
advanced degree, but the best health is reported by those with a short advanced degree as was
the case for men (controlling for wages, occupation and age). They report better health
corresponding to a 0.0037 higher health index. For women there are significant health
differences between white collar and unskilled blue collar workers. Finally wages are
10
Gradient, risk factors, baseline health
significant, although the association is weaker than for men. One standard deviation of wages
alters the health index by 0.002 corresponding to an effect of 3.9 years of age.
Controlling for risk factors and baseline health
Three risk factors are highly significant: obesity, high blood pressure and job satisfaction.
Controlling for the risk factors only alter health advantages for the educated slightly and not
in a significant manner (table 4, model (1) versus model (2)).The occupational health
difference is reduced by 24% and the wage gradient by 22%. The latter gradient becomes
insignificant. Again, when controlling for baseline health, but excluding the risk factors, the
gradients are reduced compared to the static model, particularly because of reductions in
SRH differences between people with different wages (table 4, model (1) versus model (3)),
but educational and occupational differences persist. Finally, adding controls for the risk
factors again does not significantly alter the educational gradient (table 4, model (4)),
whereas both wage and occupational gradients are reduced and both become insignificant.
The persistence over time in SRH is substantial but less than the persistence for men.
DISCUSSION
The present study examines socioeconomic inequalities in health measured by education,
occupation and wage related differences in self-reported health, using a Danish survey of
individuals who have been working within two months of an interview in 1990, with a
follow-up interview in 1995. Besides providing new evidence on social gradients in SRH the
main contribution of this study is to infer the extent to which adjustment for baseline health
affects the explanatory power of given risk factors. Adjustment for baseline health can be
advocated theoretically, because it limits gradients due to social selection and also because
health is accumulated over time. Thus as stressed by the literature taking a life course
perspective, a large part of the variation in health has early determinants. This may alter the
scope for explaining social inequalities as well as which pathways are the predominant ones.
11
Gradient, risk factors, baseline health
In general, a sample of employees would be expected to be a group where health inequalities
are lower than in the total population11. Nevertheless, substantial social differences in SRH
among both men and women are found in this sample. As opposed to several other studies,
including Danish findings on mortality, education related differences in SRH are larger for
women than for men24;65;67;72. For both gender, non-monotone relationships between length
of education and SRH are found which correspond roughly to findings on mortality in
Denmark73.
For men, even for this sample we see large differences in health between employed and a
residual group consisting of individuals mainly out of work as well as large wage related
health differences that persist after controlling for baseline health and risk factors. For
women, occupational and wage related health differences vanish when controlling for
baseline health and risk factors. It should be stressed that particularly the findings with
respect to wage effects should be interpreted with care, as wages are only observed in 1995
and changes in health from 1990 to 1995 may affect wages in 1995.
Controlling for the risk factors measured by being fat, obese, having high blood pressure,
years of smoking and job satisfaction reduces education, occupation and wage gradients in
static models by 0-40%, which is a commonly found explanatory power for such a limited set
of risk factors. Turning to dynamic models that control for baseline health shows that SRH is
very persistent over the five year period from 1990 to 1995. When the risk factors are
controlled for in addition to baseline health status, most education related differences in
health are insignificant for men, whereas this is not the case for women. In addition, the
impact of the risk factors on SES gradients is quantitatively alike in static and dynamic
models. Therefore, we do not find that the risk factors explain a larger share of the social
gradients in health in dynamic models than in static models. The converse could be expected
12
Gradient, risk factors, baseline health
because differences across individuals may be accumulated over long time periods thus to a
lesser degree affected by current behaviour.
Limitations
The empirical investigation is hampered by several data limitations. The study was
conducted on a rather small sample with only one follow up period and few risk factors. The
limited number of risk factors limits comparison with previous studies. It also limits to what
extend we may expect the risk factors to explain social inequalities in health. On the other
hand the limited number of risk factors may imply an overstatement of the impact of a single
risk factor. Moreover, the use of a sample of employees might introduce the healthy worker
effect, which is likely to downward bias the impact from the risk factors. The use of selfreported health and self-reported risk factors might also introduce biases. As mentioned
above, several studies have found measurement error in self-reported health to be limited and
that it has many advantages as a summary health measure. Clearly, the limited time period
used in this study, the high correlation between baseline and current health and the fact that
few previous studies have looked at the explanatory power of the risk factors in dynamic
health models, warrants caution in extrapolating the results to other populations.
Conclusion
When evaluating whether a set of risk factors “explain” socioeconomic inequalities in health
it is important be realistic about what to expect. In this study we hypothesized that common
risk factors may explain a larger part of health inequalities when adjusting for baseline
health, than differences in unadjusted health. We found this not to be the case. The study
showed the existence of particularly educational gradients in health for women and wage or
occupational gradients for men, which were present whether or not adjusting for baseline
health.
13
Gradient, risk factors, baseline health
ACKNOWLEDGEMENTS
The authors wish to thank participants at the Symposium in Applied Statistics in Odense
2005 and two anonymous referees.
Conflicts of interest: none
Key-points

This study examines to what extent a set of risk factors explain social inequalities in
self-reported health when controlling for base-line health

We highlight theoretical reasons for adjusting for baseline health and the potential
change this may have for interpretation of social gradients in health.

In a sample of Danish employees, we find large education, occupation and wage
gradients in self-reported health and that the risk factors explain 0-40% of the
gradients in static models.

The risk factors explain the same fractions of social inequalities in health in a
dynamic context when base-line health is controlled for.
14
Gradient, risk factors, baseline health
APPENDIX
Data construction
In 1990, 9,653 persons aged 18-59 were randomly chosen and 8,664 were interviewed (90%
response rate)74. In 1995, 7,532 were re-interviewed (86,9% response rate), 300 migrated or
died and 1,821 were not interviewed. We end up with a smaller sample mainly for two
reasons. First of all, self-reported health was only asked to people who were employed within
the last two months of the interview. Second, a number of individuals were deliberately only
interviewed once. These are deleted. Furthermore, observations with missing information on
education in both 1990 and 1995 are deleted, as well as the remaining individuals who are
still under education in 1995, since focus is on the effect of completed education. 10
individuals reported their age as more than 6 or less than 4 years in 1995 from the reporting
in 1990, so they are deleted. Among the remaining observations, 2,096 are deleted due to
missing information on SRH in 1990 or 1995. Finally 68 observations contained missing
information on wages are also deleted. The sample construction is illustrated in table A1,
showing the number of missing information for various variables and the remaining sample
size. We see that the largest relative drop occurs when removing missing SRH. Comparing
to the original sample, only 4,976 were employed among the 7,532 re-interviewed in 1995.
Since only employed individuals were asked about SRH, this corresponds to the drop in the
original sample.
Table A1. Sample selection rules and their impact on the sample size.
Original
intended
sampled
Dropped
Remaining sample
Rate remaining
9653
Not
Not
interviewe interviewe Missing
Under
Wrong
d in 1990 d in 1995 education education age
989
8664
90
1132
7532
87
15
949
6583
87
722
5861
89
10
5851
100
Missing
SRH
2096
3755
64
Missing
wages
68
3687
98
Gradient, risk factors, baseline health
The sample does not differ much with respect to age, education and occupational distribution
from the overall national representative sample of employed individuals. There is though a
significant geographical variation in attrition from the 1990 to the 1995 survey74, but
including controls for region does not alter the results.
The questions used to construct the hourly wage are: “What is your usual wage payment
before tax and deductions” (answer a1 in Danish kr.), “How is your wage usually paid”
(answer a2 = 1, 2, 3, 4, 5 or 6 for respectively pay per hour, day, week, second week, month,
or other) and “How many hours do you usually work per week in your main job” (answer a3
in hours). The following rules are used to convert the answers into one variable containing
hourly wages: w = a1 if a2 = 1, w = 5a1/a3 if a2 = 2, w = a1/a3 if a2 = 3, w = a1/2a3 if a2 = 4, w =
a1/4a3 if a2 = 5 and w = missing if a2 = 6.
16
Gradient, risk factors, baseline health
Table 1. HUI and 15D thresholds of health.
Health category
HUI thresholds
15D thresholds
Very poor
0
42.8
0
67.8
Poor
42.8
75.6
67.8
79.8
Neutral
75.6
89.7
79.8
91.4
Good
89.7
94.7
91.4
96.3
Very good
94.7
100
96.3
100
Note. The HUI and 15D thresholds are multiplied by 100.
17
Gradient, risk factors, baseline health
Table 2. Distribution of study population, 1995.
Men
Number
Percentage with
poor health
Number
Education:
No education
161
16
173
Short education
158
11
261
Semi-skilled
90
28
70
Vocational
1008
13
571
Short advanced
138
8
249
Medium advanced
240
11
313
Long advanced
190
6
65
Occupational status:
Unskilled worker
397
15
248
Skilled worker
369
12
67
White collar
1158
10
1307
Other
61
12
49
Age:
18-25
47
2
30
26-35
576
8
387
36-45
626
12
597
46-55
526
17
530
56-65
210
15
158
Hourly wage:
< 10 pctile
181
19
168
>10 and <25 pctile
313
14
257
>25 and <50 pctile
488
16
425
>50 and <75 pctile
505
10
426
>75 pctile
498
7
426
Risk factors:
Fat
933
15
398
Obese
154
25
87
High blood
154
25
148
pressure
Satisfied with job
1280
9
1072
10+ years of
1046
15
729
smoking
Baseline Health:
Poor Health 1990
192
43
174
Total
1985
12
1702
Notes: Percentiles are: 10th=87, 25th=118, 75th=142 DKK/hour.
18
Women
Percentage with
poor health
28
14
19
10
13
10
9
25
12
11
25
10
9
13
14
22
22
22
20
13
9
18
29
20
11
16
48
13
Gradient, risk factors, baseline health
Table 3. Interval regressions with self-rated health in 1995 as dependent variable. HUI
thresholds applied. Men.
(1)
(2)
(3)
(4)
Education (ref=no education):
Short education
Semi-skilled
Vocational
Short advanced
Medium advanced
Long advanced
Occupation (ref=Unskilled):
White collar
Skilled worker
Other
Wages:
Ln(wage)
0.24
(0.53)
-2.00***
(0.61)
0.25
(0.43)
0.75
(0.57)
0.40
(0.52)
0.57
(0.56)
0.07
(0.52)
-1.85***
(0.60)
0.21
(0.42)
0.81
(0.56)
0.30
(0.51)
0.34
(0.55)
0.33
(0.50)
-1.58***
(0.58)
0.37
(0.41)
0.82
(0.54)
0.36
(0.49)
0.44
(0.53)
0.17
(0.49)
-1.48***
(0.57)
0.32
(0.40)
0.86
(0.53)
0.28
(0.49)
0.24
(0.52)
0.09
(0.32)
0.02
(0.37)
-1.48**
(0.66)
-0.04
(0.32)
-0.17
(0.37)
-1.63**
(0.65)
-0.19
(0.30)
0.03
(0.35)
-0.93
(0.63)
-0.27
(0.30)
-0.11
(0.35)
-1.11*
(0.61)
1.60***
(0.36)
1.32***
(0.36)
1.27***
(0.34)
1.05***
(0.34)
Risk factors:
Fat
-0.17
(0.22)
-1.01**
(0.40)
-2.21***
(0.39)
1.47***
(0.21)
0.005*
(0.003)
Obese
High blood pressure
Satisfied with job
Years of smoking
Baseline health:
Health in 1990
Other:
Age
-0.11
(0.21)
-0.96**
(0.38)
-2.16***
(0.37)
1.21***
(0.20)
0.004
(0.003)
30.73***
(1.99)
-0.07***
(0.01)
0.99
(0.72)
88.78***
(1.74)
-2467.90
20.89***
5.95
29.32***
(1.95)
-0.07***
-0.04***
-0.04***
(0.01)
(0.01)
(0.01)
Age less than 25
0.81
0.71
0.58
(0.70)
(0.68)
(0.67)
Constant
89.29***
60.49***
62.29***
(1.71)
(2.49)
(2.44)
LogL
-2419.18
-2344.85
-2302.79
Wald test for education
18.60***
16.94***
15.77**
Wald test for occupation
6.82*
2.86
3.64
Wald test for risk factors
99.05***
85.50***
Notes: Standard deviations are presented in parentheses. Significance indicated at 1% level (***), 5% level
(**), and 10% level (*). HUI thresholds are multiplied by 100. Number of observations=1985
19
Gradient, risk factors, baseline health
Table 4. Interval regressions with self-rated health in 1995 as dependent variable. HUI
thresholds applied. Women.
(1)
(2)
(3)
(4)
Education (ref=no education):
Short education
Semi-skilled
Vocational
Short advanced
Medium advanced
Long advanced
Occupation (ref=Unskilled):
White collar
Skilled worker
Other
Wages:
Ln(wage)
0.55
(0.57)
0.79
(0.78)
1.47***
(0.53)
0.37
(0.61)
1.27**
(0.60)
1.42
(0.87)
0.70
(0.57)
0.86
(0.78)
1.53***
(0.53)
0.50
(0.60)
1.39**
(0.59)
1.33
(0.86)
0.34
(0.56)
1.06
(0.76)
1.38***
(0.52)
0.39
(0.59)
1.14**
(0.58)
1.44*
(0.84)
0.49
(0.56)
1.12
(0.76)
1.44***
(0.52)
0.50
(0.59)
1.25**
(0.58)
1.37*
(0.84)
1.00**
(0.43)
0.37
(0.77)
-0.03
(0.72)
0.76*
(0.43)
0.25
(0.76)
-0.23
(0.72)
0.88**
(0.42)
0.69
(0.75)
-0.23
(0.70)
0.66
(0.42)
0.57
(0.74)
-0.41
(0.69)
0.97*
(0.51)
0.76
(0.51)
0.63
(0.50)
0.44
(0.50)
Risk factors:
Fat
-0.26
(0.34)
-1.60**
(0.66)
-1.12**
(0.47)
1.21***
(0.27)
0.001
(0.003)
Obese
High blood pressure
Satisfied with job
Years of smoking
Baseline health:
Health in 1990
-0.21
(0.33)
-1.52**
(0.64)
-1.10**
(0.46)
1.06***
(0.26)
0.001
(0.003)
24.73***
(2.21)
-0.05***
(0.02)
-0.64
(1.01)
67.98***
(3.06)
-2281.91
14.78**
6.35*
24.07***
(2.20)
Age
-0.07***
-0.06***
-0.04***
(0.02)
(0.02)
(0.02)
Age less than 25
-0.72
-0.53
-0.45
(1.04)
(1.03)
(1.01)
Constant
90.03***
90.23***
68.79***
(2.38)
(2.37)
(3.05)
LogL
-2341.66
-2323.31
-2265.90
Wald test for education
14.71**
14.31**
14.12**
Wald test for occupation
7.22*
4.97
4.67
Wald test for risk factors
36.94***
32.20***
Notes: Standard deviations are presented in parentheses. Significance indicated at 1% level (***), 5% level
(**), and 10% level (*). HUI thresholds are multiplied by 100. Number of observations=1702.
20
Gradient, risk factors, baseline health
Reference List
1. Sigriest J. Social variations in health expectancy in Europe. An ESF Scientific
Programme 1999-2003 - Final Report. Duesseldorf: University of Duesseldorf,
Medical Faculty, 2004.
2. Sundhedsministeriet. Regeringens Folkesundhedsprogram 1999-2008. København:
Sundhedsministeriet, 1999.
3. Sundhedsministeriet. Social ulighed i sundhed. Forskelle i helbred, livsstil og brug af
sundhedsvæsenet - 2. delrapport fra Middellevetidsudvalget. København: Nyt
Nordisk Forlag Arnold Busck, 2000.
4. Acheson D. Independent Inquiry into inequalities in health report. Norwich: The
Stationary Office, 1998.
5. Mackenbach J, Kunst A, Cavelaars A, Geurts J, Groenhof F, the EU Working group
on socioeconomic inequalities in health. Socioeconomic inequalities in morbidity and
mortality in Western Europe. The Lancet 1997;349:1655-59.
6. Kunst A, Groenhof F, Mackenbach J, the EU Working group on socioeconomic
inequalities in health. Occupational class and cause specific mortality in middle aged
men in 11 European countries: Comparison of population based studies. British
Medical Journal 1998;316:1636-41.
21
Gradient, risk factors, baseline health
7. Cavelaars A, Kunst A, Geurts J et al. Differences in self reported morbidity by
educational level: A comparison of 11 European countries. Journal of Epidemiology
and Community Health 1998;52:219-27.
8. Vågerö E, Erikson E. Socioeconomic inequality in morbidity and mortality in
Western Europe. The Lancet 1997;350(9076):516-18.
9. Pappas G, Queen S, Hadden W, Fisher G. The increasing disparity in mortality
between socioeconomic groups in the United States, 1960 and 1986. New England
Journal of Medicine 1993;329:103-9.
10. Mackenbach JP, Bos V, Andersen O et al. Widening inequalities in mortality in
Western Europe. International Journal of Epidemiology 2003;32:830-837.
11. Andersen O, Laursen L, Pedersen JK. Dødelighed og erhverv 1981-1995.
København: Danmarks Statistik, 2001.
12. Wilkinson R. Unhealthy societies: the afflictions of inequality. London: Routledge,
1996.
13. van Rossum CTM, Shipley MJ, van de Mheen H, Grobbee DE, Marmot MG.
Employment grade differences in cause specific mortality. A 25 year follow up of
civil servants from the first Whitehall study. Journal of Epidemiology and
Community Health 2000;54:178-84.
14. Strand BH, Tverdal Aa. Can cardiovascular risk factors and lifestyle explain the
educational inequalities in mortality from ishaemic heart disease and from other heart
22
Gradient, risk factors, baseline health
diseases? 26 year follow up from 50000 Norwegian men and women. Journal of
Epidemiology and Community Health 2004;58:705-9.
15. Siegrist J, Marmot MG. Health inequalities and the psychosocial environment: two
scientific challenges. Social Science and Medicine 2004;58(8):1463-73.
16. Pekkanen J, Tuomilehto J, Uutela A, Vartiainen E, Nissinen A. Social class, health
behavoiur, and mortality among men and women in eastern Finland. British Medical
Journal 1995;311:589-93.
17. Marmot MG, Fuhrer R, Ettner SL, Bumpass LL, Ryff CD. Contribution of
Psychosocial factors to socioeconomic differences in health. The Milbank Quarterly
1998;76(3):403-48.
18. Kilander L, Berglund L, Boberg M, Vessby B, Lithell H. Education, lifestyle factors
and mortality from cardiovascular disease and cancer. A 25-year follow-up of
Swedish 50-year-old men. International Journal of Epidemiology 2001;30:1119-26.
19. Adler N, Boyce T, Chesney M. Socioeconomic status and health, the challenge of the
gradient. American Psychologist 1994;49(1):15-24.
20. Marmot MG. Social determinants of health inequalities. The Lancet 2005;365:1099104.
21. van Oort F, Van Lenthe F, Mackenbach J. Material, psychosocial and behavoiural
factors in the explanation of educational inequalities in mortality in the Netherlands.
Journal of Epidemiology and Community Health 2005;59(214):220.
23
Gradient, risk factors, baseline health
22. Schrijvers CTM, Bosma H, Manor O, Mackenbach JP. Hostility and the educational
gradient in health. European Journal of Public Health 2002;12:110-116.
23. Marmot MG. Social differences in health within and between populations. Daedelus
1994;123(4):197-216.
24. Macintyre S. The Black Report and beyond, what are the issues? Social Science and
Medicine 1997;44(6):723-45.
25. Lantz PM, House JS, Lepkowski JM, Williams DR, Mero PM, Chen J.
Socioeconomic factors, health behavior, and mortality: Results from anationally
representative prospective study of US adults. Journal of the American Medical
Assocation 1998;279(21):1703-8.
26. Laaksonen M, Roos E, Rahkonen O, Martikainen P, Lahelma E. Influence of material
and behavoiural factors on occupational class differences in health. Journal of
Epidemiology and Community Health 2005;59(163):169.
27. Kenkel D. Health Behavior, Healht Knowledge and Schooling. Journal of Political
Economy 1991;99(2):287-305.
28. Lynch JW, Kaplan GA, Cohen RD, Tuomilehto J, Salonen JT. Do cardiovascular risk
factors explain the relation between socioeconomic status, risk of all-cause mortality,
cardiovascular mortality and acute myocardial infarction=. American Journal of
Epidemiology 1996;144(10):934-42.
24
Gradient, risk factors, baseline health
29. van de Mheen H, Stronks K, Mackenbach JP. A lifecourse perspective on socioeconomic inequalities in health: the influence of childhood socio-economic
conditions and selection processes. Journal of health and social behaiour
1998;20(5):754-77.
30. Blane D, Smith GD, Bartley M. Social selection: What does it contribute to social
class differences in health. Sociology of health and illness 1993;15(1):1-15.
31. Manor O, Matthews S, Power C. Health selection: the role of inter- and intra
generational mobility on social inequalities in health. Social Science and Medicine
1993;57:2217-27.
32. Bartley M, Plewis I. Does health-selective mobility account for socioeconomic
differences in health? Evidence from England and Wales, 1971 to 1991. Journal of
Health and Social Behaviour 1997;38(4):376-86.
33. Arendt JN. In sickness and in health - till education do us part: education effects on
hospitalization. Economics of Education Review 2007;in press.
34. Grossman M. The human capital model of the demand for health. In: Culyer AJ,
Newhouse JP, editors. Handbook of Health Economics. Vol. 1A. Amsterdam:
Elsevier Science, 2000:347-408.
35. Ettner SL. New evidence on the relationship between income and health. Journal of
Health Economics 1996;15(1):67-85.
25
Gradient, risk factors, baseline health
36. Kington RS, Smith JP. Socioeconomic status and racial and ethnic differences in
functional status associated with chronic diseases. American Journal of Public Health
1997;87(5):805-10.
37. Smith JP, Kington R. Demographic and economic correlates of health in old age.
Demography 1997;34(1):159-70.
38. Smith JP. Healthy bodies and thick wallets: The dual relationship between health and
economic status. The Journal of Economic Perspectives 1999;13(2):145-66.
39. Barker D. The fetal and infanct originis of adult disease. London: MBJ publications,
1992.
40. Elstad JI. Chilhood adversities and health variations among middle-aged men: a
retrospective lifecourse study. European Journal of Public Health 2005;15(1):51-58.
41. Graham H. Building and inter-disciplinary science of health inequalities: the
examples of lifecourse research. Social Science and Medicine 2002;55:2005-16.
42. Kuh D, Hardy R, Langenberg C, Richards M, Wadsworth MJ. Mortality in adults
aged 26-64 years related to socioeconomic conditions in childhood and adulthood:
post war birth cohort study. British Medical Journal 2002;235:1076-80.
43. Sacker A, Clarke P, Wiggings RD, Bartley M. Social dynamics of health inequalities:
a growth curve analysis of aging and self assessed health in the British household
panel survey 1991-2001. Journal of Epidemiology and Community Health
2005;59:495-501.
26
Gradient, risk factors, baseline health
44. Singer B, Ryff CD. Racial and ethnic inequalities in health: Environmental,
psychological, and physiological pathways. Ch. 5, part II. In: Devlin B, Fienberg SE,
Resnick DP, Roeder K, editors. Intelligence, genes, success. Scientists respond to the
Bell curve. New York: Springer Verlag, 1997:89-122.
45. Wadsworth MJ. Health inequalities in the life course perspective. Social Science and
Medicine 1997;44(6):859-69.
46. Kuh D, Ben-Shlomo Y, Lynch J, Hallqvist J, Power C. Life course epidemiology.
Journal of Epidemiology and Community Health 2003;57:778-83.
47. Deaton A, Paxson C. Health, inequality and inequality over the life cycle. Ch. 10, part
IV. In: Wise D, editor. Frontiers of the economics of aging. Chicago: University of
Chicago Press, 1998:431-57.
48. Grossman M. On the concept of health capital and the demand for health. Journal of
Political Economy 1972;80(2):223-55.
49. Heistaro S, Jousilahti P, Lahelma E, Vartiainen E, Puska P. Self rate health and
mortality: a long term prospective study in eastern Finland. Journal of Epidemiology
and Community Health 2001;55:227-32.
50. Hurd MD, McFadden D, Merrill A. Predictors of mortality among the elderly. In:
Wise D, editor. Themes in the Economics of Aging. Chicago: NBER, 1998.
27
Gradient, risk factors, baseline health
51. Khang Y-H, Kim HR. Explaining socioeconomic inequality in mortality among south
koreans: an examination of multiple pathways in a nationally representative
longitudinal study. International Journal of Epidemiology 2005;34:630-637.
52. Kiuila O, Mieszkowski P. The effects of income, education and age on health. Health
Economics 2007;16(8):781-98.
53. Luoto MD, Prättälä R, Uutela A, Puska P. Impact of unhealthy behaviors on
cardiovascular mortality in Finland, 1978-1993. Preventive Medicine 1998;27:93100.
54. Mete C. Predictors of elderly mortality: health status, socioeconomic characteristics
and social determinants of health. Health Economics 2005;14:135-48.
55. Borg V, Kristensen TS. Social class and self-rate health: can the gradient be
explained by differences in life style or work environment. Social Science and
Medicine 2000;52:1019-30.
56. Power C, Matthews S, Manor O. Inequalities in self rated health in the 1958 birth
cohort: lifetime social circumstances or social mobility? British Medical Journal
1996;313:449-53.
57. Taubman P, Rosen S. Healthiness, education and marital status. In: Fuchs V, editor.
Economic aspects of health. Chicago: University of Chicago Press, 1982.
28
Gradient, risk factors, baseline health
58. Adams P, Hurd MD, McFadden D, Merrill A, Ribeiro T. Healthy, wealthy and wise?
Tests for direct causal paths between health and socioeconomic status. Journal of
Econometrics 2003;112:3-56.
59. Jones A. Health Econometrics. In: Culyer AJ, Newhouse JP, editors. Handbook of
Health Economics. Amsterdam: Elsevier Science, 2000.
60. van Doorslaer E, Wagstaff A, Bleichrodt H. Income-related inequalities in health:
Some international comparisons. Journal of Health Economics 1997;16:93-112.
61. van Doorslaer E, Koolman X. Explaining the differences in income-related health
inequalities in European countries. Journal of Health Economics 2004;13:609-28.
62. Lauridsen J, Christiansen T, Häkkinen U. Measuring inequality in self-reported
health. Discussion of recently suggested approach using Finnish data. Health
Economics 2004;13:725-32.
63. Idler E. Self-ratings of health: Mortality, morbidity, and meaning. In: Schlechter S,
editor. Proceedings of the 1993 NCHS conference on the cognitive aspects of selfrated health status. Hyattsville, Maryland: US Department of Health and Human
Services, Public Health Service, CDC, NCHS, 1994:36-59.
64. Idler E, Benyamin Y. Self-rated health and mortality: A review of twenty-seven
community studies. Journal of Health and Social Behaviour 1997;38:21-37.
65. Benítez-Silva H, Buchinsky M, Chan HM, Cheidvasser S, Rust J. How large is the
bias in self-rated disability? Journal of Applied Econometrics 2004;19(6):649-70.
29
Gradient, risk factors, baseline health
66. Butler JS, Burkhauser RV, Mitchell JM, Pincus TP. Measurement error in self-rated
health variables. The Review of Economics and Statistics 1987;69(4):644-50.
67. Crossley T, Kennedy S. The reliability of self-assessed health. Journal of Health
Economics 2002;21(4):643-58.
68. Dwyer S, Mitchell JM. Health problems as determinants of retirement: Are self-rated
measures endogenous? Journal of Health Economics 1999;18(2):173-93.
69. Faragher EB, Cass M, Cooper CL. The relationship between job satisfaction and
health: a meta-analysis. Occupational and Environmental Medicine 2005;62:105-12.
70. Hoogendoorn WE, BOngers PM, de Vet HCW, Ariëns GAM, van Mechelen W,
Bouter LM. High physical work load and low job satisfaction increase the risk of
sickness absence due to back pain: results of a prospective cohort study. Occupational
and Environmental Medicine 2002;59:323-28.
71. Siegrist J, Starke D, Chandola T et al. The measurement of effort-reward imbalance at
work: European comparisons. Social Science and Medicine 2004;58:1483-99.
72. Preston S, Taubman P. Socioeconomic differences in aduly mortality and health
status. In: Martin L, Preston S, editors. Demography of Aging. Washington DC:
National Academy Press, 1994:279-318.
73. Munch JR, Svarer M. Mortality and socioeconomic differences in Denmark: A
competings risks proportional hazard model. Economics and Human Biology
2005;3:17-32.
30
Gradient, risk factors, baseline health
74. Borg V, Burr H. Danske lønmodtageres arbejdsmiljø og helbred 1990-95.
København: Arbejdsmiljøinstituttet, 1997.
31