Content uploaded by Lawrence Whalley
Author content
All content in this area was uploaded by Lawrence Whalley on Dec 09, 2014
Content may be subject to copyright.
Papers
Longitudinal cohort study of childhood IQ and survival
up to age 76
Lawrence J Whalley, Ian J Deary
Abstract
Objectives To test the association between childhood
IQ and mortality over the normal human lifespan.
Design Longitudinal cohort study.
Setting Aberdeen.
Subjects All 2792 children in Aberdeen born in 1921
and attending school on 1 June 1932 who sat a
mental ability test as part of the Scottish mental
survey 1932.
Main outcome measure Survival at 1 January 1997.
Results 79.9% (2230) of the sample was traced.
Childhood mental ability was positively related to
survival to age 76 years in women (P < 0.0001) and
men (P < 0.0001). A 15 point disadvantage in mental
ability at age 11 conferred a relative risk of 0.79 of
being alive 65 years later (95% confidence interval
0.75 to 0.84); a 30 point disadvantage reduced this to
0.63 (0.56 to 0.71). However, men who died during
active service in the second world war had a relatively
high IQ. Overcrowding in the school catchment area
was weakly related to death. Controlling for this factor
did not alter the association between mental ability
and mortality.
Conclusion Childhood mental ability is a significant
factor among the variables that predict age at death.
Introduction
Inequalities in health and mortality exist among differ-
ent socioeconomic groups. People living in deprived
conditions generally suffer more illness and die
younger,
1-5
and socioeconomic circumstances in child-
hood are related to mortality from several illnesses.
67
Educational level also contributes to differences in
mortality between socioeconomic groups,
8-10
although
the size of this effect varies nationally.
11
Higher mental
ability, as assessed by psychometric tests, is associated
with more favourable educational and occupational
life outcomes.
12
Socioeconomic status, educational level, and
mental ability are closely related. However, there is lit-
tle information about the link between mental ability
and morbidity and mortality. Mental ability was signifi-
cantly associated with longevity in a longitudinal study
of Australian Vietnam veterans after discharge,
13
and in
old people whose mental functions were declining.
14 15
We examined the effects of childhood mental ability on
survival up to 76 years in a year of birth cohort.
Subjects and methods
Scottish mental survey 1932
Under the auspices of the Scottish Council for
Research in Education, an intelligence test (the Moray
house test No 12) was given to all Scottish children
who were born in 1921 and were attending school on
1 June 1932.
16
This was called the Scottish mental sur-
vey 1932 and provided a unique record of intelligence
test scores for a complete age group of school
children.
17
Test data were obtained for 87 498 children
(44 210 boys and 43 288 girls).
In reports of the Scottish mental survey 1932 the
cognitive ability test was termed the verbal test. It has
71 items with a maximum score of 76. It contains sev-
eral general, spatial, and numerical reasoning items.
We will refer to it as the Moray house test. The test has
criterion validity. Its correlation with the Stanford Binet
test was 0.81 for boys (n = 500) and 0.78 for girls
(n = 500).
16
Follow up
We obtained survey data from the Scottish Council for
Research in Education. For each subject, these data
comprised family name, given name, date of birth,
name of school, and raw Moray house test score. We
identified Aberdeen city as our target area and
searched for all subjects (n = 2792) who had attended
schools within its boundaries using public and health
records in the United Kingdom. We began a forward
search in the Register of Deaths from 1932 until 1997.
Untraced subjects were then sought in the Scottish
Community Health Index, which records everyone
registered with a family doctor (more than 99% of the
population). For untraced women, we next examined
the Register of Marriages in Scotland from 1937
onwards.When we discovered that an untraced woman
had married in Scotland, we repeated searches using
her married name in the death register and the
community health index. Subjects remaining untraced
were sought by computerised and hand searching of
the NHS Central Register in Southport. We obtained
ethical approval for the study from multicentre
research ethics committee (Scotland), and Grampian
local research ethics committee, and the privacy
committee of the NHS Central Register.
Measures of childhood IQ and social factors
We converted Moray house test scores to IQ-type scale
scores (with mean = 100, SD = 15) and corrected them
Department of
Mental Health,
University of
Aberdeen, Clinical
Research Centre,
Cornhill Hospital,
Aberdeen
AB24 2ZD
Lawrence J Whalley
professor of mental
health
Department of
Psychology,
University of
Edinburgh,
Edinburgh EH8 9JZ
Ian J Deary
professor of
differential psychology
Correspondence to:
I J Deary
I.Deary@ed.ac.uk
BMJ 2001;322:1–5
1BMJ VOLUME 322 7 APRIL 2001 bmj.com
for the subject’s age in days at the time of testing (Pear-
son’s r between age and raw test score = 0.142,
P < 0.0001). We used an estimate of overcrowding in
the childhood family home as an ecological measure
of social disadvantage. The 1931 UK census recorded
overcrowding in the residential accommodation of 12
districts in the city. Catchment areas for the schools
were mapped on to census districts to obtain estimates
of overcrowding for pupils attending each school. The
populations of the 12 districts ranged from 10 853 to
19 080 people (mean = 13 938, SD = 3306). The mean
number of people living per hundred rooms in the 12
districts ranged from 72.6 to 182.7 (mean = 129.8,
SD = 35.9). Paternal occupation was stated on 722 of
1084 death certificates, and we classified these into
Office of Population Censuses and Surveys’ categories.
9
Statistical analyses
We used one way analyses of variance with conserva-
tive post hoc (Scheffe) tests to compare the age
adjusted Moray house test scores for dead, living, and
untraced subjects and those known to have moved
away from Scotland. We tested the influence of
childhood IQ on subsequent survival using Cox’s pro-
portional hazards regression, including exhaustion
analysis. Men who died in combat during the second
world war were compared with other men in terms of
childhood mental ability using Student’s independent t
tests. For the subsample of deceased subjects, we tested
models of association among childhood IQ, social
factors, and age at death using partial correlation
and structural equation modelling (with the EQS
program).
Results
We traced 2230 (79.9%) of the 2792 subjects tested in
schools within Aberdeen city in 1932. On 1 January
1997, 646 men had died, 507 were alive, 247 could not
be traced, and 27 were known to have moved from
Scotland between 1987 and 1997; 438 women had
died, 594 were alive, 315 could not be traced, and 18
had moved from Scotland between 1987 and 1997.
Subjects who died before 1 January 1997 had a sig-
nificantly lower mean IQ at age 11 years than subjects
who were alive or untraced (table 1). This effect was
also seen when men and women were analysed
separately. Overall, untraced subjects had childhood
IQs similar to those of subjects who were still alive.
A Cox regression analysis including all traced sub-
jects (alive, dead, and moved out of Scotland) showed
that IQ at age 11 years on 1 June 1932 was significantly
related to survival up to age 76 years on 1 January
1997 (P < 0.0001, table 2). While gathering data for a
follow on report, we rechecked all deceased people to
obtain death certificate details. These checks discov-
ered that 39 people whom we had coded as dead
(based on community health index information or
notes available in the General Register Office) had no
death certificate available. Their mean IQ was 108.0
(SD = 10.3). Inclusion of these people as dead leads to
a small underestimate of the true effect of IQ on
survival. To illustrate this, we reassigned the 39 from
the dead to the untraced category and redid the main
univariate Cox proportional hazards regression analy-
ses. The change in survival expectancy showed slightly
stronger associations as follows: from 0.9847 to 0.9840
for all subjects, from 0.9887 to 0.9883 for men, and
from 0.9775 to 0.9765 for women. We prefer to let the
more conservative estimates stand.
The influence of childhood IQ on survival was
weaker in men than in women. This could be due to
the effect of the second world war on death rates in
men. Figure 1 shows that women with a high
childhood IQ had a consistently better average survival
expectancy than women with low childhood IQ. How-
ever, for men with a high IQ, survival suddenly drops
during the second world war and does not catch up
and improve on that in men with low childhood IQ
until later in life.
The implications of the Cox regression analyses
can be shown by comparing the mean probabilities of
Table 1 Mean (SD) IQ at age 11 years for subjects who were dead, alive, untraced, and
migrant on 1 January 1997
Total Dead Alive Untraced Moved away P value*
All subjects
No 2792 1084 1101 562 45
IQ score 100.0 (15.0) 97.7 (15.4) 102.0† (14.2) 100.8† (15.1) 98.9 (17.1) <0.0001
Men
No 1427 646 507 247 27
IQ score 100.5 (15.5) 98.9 (15.6) 102.5† (14.8) 101.1 (15.6) 99.0 (19.5) 0.001
Women
No 1365 438 594 315 18
IQ score 99.4 (14.5) 95.9 (14.8) 101.5† (13.6) 100.5† (14.7) 98.7 (13.3) <0.0001
*Analysis of variance.
†Significant post hoc differences existed between groups (Scheffe tests): All subjects: dead<alive (P=0.0001;
95% CI for difference −5.6 to −1.7), dead<untraced (P<0.0001; −5.7 to −1.9); Men: dead<untraced
(P=0.018; −5.8 to −0.4); Women: dead<alive (P<0.0001; −8.3 to −2.8), dead<untraced (P<0.0001; −7.6 to
−2.3).
Age (years)
% alive
40
0 1020304050607080
60
70
80
90
100
50
Women - lowest IQ quarter
Women - highest IQ quarter
Age (years)
% alive
30
0 1020304050607080
50
60
70
80
100
90
40
Men - lowest IQ quarter
Men - highest IQ quarter
Fig 1 Probability of survival at ages 12-76 years for women and
men in highest and lowest quarters for IQ score at age 11
Papers
2 BMJ VOLUME 322 7 APRIL 2001 bmj.com
people of different childhood IQ levels being alive on 1
January 1997. When subjects with 1 SD difference in
childhood IQ are compared, the chances of those with
the lower IQ being alive on 1 January 1997 are 79% for
all subjects (95% confidence interval 75% to 84%), 71%
for women (64% to 78%), and 83% for men (76% to
89% including only those alive on 1 January 1950). If
the IQ difference is 2 SD
—
for example, 85 v 115
—
the
relative mean chances of survival for those with the
lower IQ compared with those with the higher IQ are
63% for all subjects (56% to 71%), 51% for women
(42% to 61%), and 68% for men (58% to 80%).
Overcrowding in the childhood school’s catchment
area was significantly related to survival when all
subjects were included (table 2). The effect was signifi-
cant in men but not in women when the sexes were
analysed separately. Overcrowding was significantly
correlated with childhood IQ score (r = − 0.22,
P < 0.001); children with higher ability scores tended to
live in catchment areas with less overcrowded homes.
Controlling for overcrowding hardly altered the
association between childhood mental ability and
survival. Details of these results are available from the
authors.
Complete data on IQ at age 11, father’s occupation,
overcrowding in the school’s catchment area, and age
at death were available for 722 subjects. IQ correlated
with age at death (0.18, P < 0.001), overcrowding
( −0.22, P < 0.001), and father’s occupational category
( −0.20, P < 0.001; more professional occupations have
lower numbers). Father’s occupational category corre-
lated significantly with overcrowding (0.09, P = 0.01)
but not age at death ( − 0.02, P > 0.05). Overcrowding
was not significantly correlated with age at death (0.02,
P > 0.05). The correlation between IQ at age 11 and
age at death after father’s occupation and overcrowd-
ing were controlled for was 0.19 (P < 0.001).
We used the EQS structural equation modelling
program competitively to test models of these data that
did and did not assume direct effects of IQ, father’s
occupation, and overcrowding on age at death. The
best fitting model conceptualises IQ at age 11 as a
mediating variable between social factors and age at
death (fig 2). Models which assumed direct effects of
the available social factors on age at death and those
which assumed social factors as mediators between IQ
and age at death had unacceptable fit statistics.
Discussion
Our data show that high mental ability in late
childhood reduces the chances of death up to age 76
years. The effect is not caused by a single factor and
may even be reversed, as was found for men during the
second world war. This result adds to our knowledge of
the personal traits in youth that contribute to survival
in subsequent decades. Studies of an unrepresentative
sample of children with high ability in the United
States found that conscientiousness, lack of cheerful-
ness, and permanency of mood (for men only) were
associated with living longer.
18
In our study, women
with a deficit in IQ of 15 points at age 11 had less than
75% survival and those with a deficit of 30 points were
about half as likely to survive.
The association between higher childhood IQ and
an increased risk of dying in the second world war
requires further investigation. Part of the effect might
be explained by some men being rejected for active
service because of low mental ability. More evidence is
needed on the roles fulfilled by people of higher men-
tal ability in the war and, indeed, whether the relation is
true beyond Aberdeen.
We found a weak association between estimated
overcrowding in the area of the childhood family
home and survival. However, the association between
childhood IQ and survival was not affected by control-
ling for overcrowding. The association was also
unaffected by controlling for overcrowding and father’s
occupational category in the subsample of people who
were dead by January 1997. These analyses were
Table 2 Results of Cox proportional hazards regression used to predict age at death from IQ scores at age 11 and overcrowding
Predictor variable No alive* No dead† Regression coefficient (SE) P value
Change in survival expectancy
(95% CI)‡
Moray house test
All subjects 1146 1071 −0.0155 (0.0020) <0.0001 0.9847 (0.9807 to 0.9886)
Women 612 438 −0.0228 (0.0032) <0.0001 0.9775 (0.9713 to 0.9837)
Men: 534 633 −0.0114 (0.0026) <0.0001 0.9887 (0.9837 to 0.9937)
Excluding deaths in second world war 532 586 −0.0139 (0.0027) <0.0001 0.9862 (0.9810 to 0.9913)
Including only those alive on 1 January 1950 534 560 −0.0128 (0.0027) <0.0001 0.9873 (0.9820 to 0.9925)
Overcrowding index
All subjects 1103 1024 0.0119 (0.0053) 0.026 1.0119 (1.0014 to 1.0226)
Women 589 417 0.0112 (0.0080) 0.158 1.0113 (0.9956 to 1.0272)
Men: 514 607 0.0142 (0.0073) 0.053 1.0143 (0.9998 to 1.0290)
Excluding deaths in second world war 512 561 0.0167 (0.0076) 0.028 1.0169 (1.0018 to 1.0322)
Including only those alive on 1 January 1950 514 537 0.0190 (0.0078) 0.014 1.0192 (1.0038 to 1.0348)
*Subjects alive on 1 January 1997 and those who moved out of the area but were alive at a known date before 1 January 1997.
†All people with a known date of death.
‡Change in survival expectancy for a unit change in the predictor variable.
Father's occupation
-0.181
0.181
-0.202
0.093
IQ at age 11 Age at death
Overcrowding
Fig 2 Best fitting structural equation model of associations among
paternal occupation, overcrowding, IQ at age 11, and age at death up
to 76 years. All parameter estimates are significantly greater than
zero. Squaring the parameter weights gives the variance shared by
adjacent variables; they are comparable to partial â weights in linear
regression. Average off diagonal standardised residuals=0.013; ÷
2
(df=2)=3.29, P1=0.19; Bentler Bonett normed fit index=0.965, Bentler
Bonett non-normed fit index=0.956; comparative fit index=0.985. All
of these indices are indicative of a well fitting model
Papers
3BMJ VOLUME 322 7 APRIL 2001 bmj.com
conducted on a smaller and less representative sample
than our main results.
Mechanisms for association
Various, non-exclusive, explanations exist for the
association between childhood IQ and survival. These
include genetic factors, environment before and after
birth, childhood illness, and nutrition and other
privation.
Childhood IQ as record of bodily insults—The mental
ability test was taken in 1932, at a time when poverty
was far greater and health standards were lower than at
present. IQ at age 11 years could therefore reflect the
effect of multiple factors on the developing brain.
These might include the quality of antenatal care, pre-
natal and postnatal nutrition, and the disabling effects
of chronic childhood physical illnesses. In this scenario,
childhood IQ in part represents a record of the
subject’s neurological tribulations before age 11. As
such, childhood IQ might be seen partly as a mediator
between physical and social disadvantage and survival.
These effects could be cohort specific.
Childhood IQ as an indicator of system integrity—
Childhood IQ might also act as a general, moderately
stable, indicator of system integrity within the body by
indexing the efficiency of information processing in
the nervous system. IQ, as tested by the Moray house
test, has a high stability coefficient (r = 0.63; 0.73 when
corrected for attenuation of the sample’s range of
scores) between age 11 years and 77 years.
19
Any
mechanism relating IQ to survival might be stronger at
lower levels of mental ability, when learning problems
are accompanied by physical disorders. People with
higher IQs are said to have more cerebral reserve
capacity
—
for example, lower IQ and linguistic ability in
children and young adults is associated with cognitive
decline and Alzheimer’s disease in late life.
20 21
Childhood IQ as predictor of healthy behaviours—
Childhood IQ might be related to the subsequent
acquisition of behaviours conducive to good health.
These include adopting healthy diets, sensible alcohol
consumption, avoidance of injury, and not smoking. A
similar set of factors was hypothesised to account
for the association between conscientiousness and
survival.
18
Childhood IQ as predictor of entry to safer
environments—Higher childhood IQ in men, especially
in the early and middle decades of the 20th century,
may have allowed entry into relatively safe employ-
ment (with wartime an important exception). In
women the effect of a higher childhood IQ was possi-
bly more indirect. Women with higher childhood men-
tal ability might have married higher ability men and
benefited indirectly from reduced exposure to occupa-
tional hazards, material privation, and, critically, the
impact on family life of the husband’s premature death
because of dangerous work.
Thus childhood mental ability is, arguably, a
conveniently measured, relatively reliable, and valid
indicator for several disparate antecedents and
outcomes.
12 22-25
The effect of IQ is difficult to separate
from the effects of social class and education. These
variables are moderately highly correlated, and one
can act as a surrogate for one or more of the others in
causing associations. For example, personality traits
have been found partly to explain associations between
childhood social class and poor health in adulthood.
26
The US national longitudinal study of youth showed
that, within the white American population, both
parental social class and cognitive ability in the late
teens were associated with multiple indices of social,
educational, and occupational outcomes many years
later, although the effects were often small.
22 27
Social
class and mental ability would often retain their
influence on outcomes after the covariation was statis-
tically controlled for. This indicates that mental ability
is not entirely a surrogate for social class and vice versa.
We found that overcrowding was a less powerful
predictor of survival than childhood IQ and that
controlling for overcrowding had little effect on the
association between childhood mental ability and
survival. Other, more reliable and valid ecological
measures or personal measures of social disadvantage
might prove more powerful.
6
Our data do not allow us
to tease apart the causal influences among IQ (and
other personal variables such as personality traits and
coping styles), education, and social class. This would
require larger and more homogeneous samples.
Possible sources of bias
We traced nearly four fifths of our target sample after a
gap of about 65 years. This is higher than the percent-
age traced in landmark studies examining the
influences of early life on health in old age. For exam-
ple, Barker and colleagues’ study of infant weight and
death from ischaemic heart disease was based on a
tracing of 71% of men.
28
Tracing in other influential
studies is lower.
6
Nevertheless, it is possible that the
association we found could be nullified or reversed by
data from the 20% of people we could not trace. One
reason for not being able to track subjects was
migration. Migrants are a relatively healthy group
29
and
had an average mental ability in our study. Although
such evidence is not definitive, it suggests that severe
bias in the opposite direction to our association is
unlikely.
Conclusions
In our cohort childhood IQ was a significant predictor
of human survival. We do not know, however, whether
this effect is cohort specific. Other possible mecha-
nisms for the effect include previous childhood
privation, the adoption of healthy behaviours in adult-
hood, and access to safer environments. Future studies
What is already known on this topic
People in deprived conditions tend to have more
illness and die younger
The reasons for this inequality in health are not
fully established
What this study adds
IQ at age 11 years was significantly associated with
survival up to 76 years in an Aberdeen cohort
The association was unaffected by adjustment for
overcrowding
Men with high IQ were more likely to die in active
service in the second world war
Papers
4 BMJ VOLUME 322 7 APRIL 2001 bmj.com
on the causes of inequalities in health and mortality
should investigate childhood mental ability as one of
the factors.
We thank the Scottish Council for Research in Education, espe-
cially Graham Thorpe, for providing data from the Scottish
Mental Survey 1932. Elizabeth and Patricia Whalley traced sub-
jects in public records. David Hunter assisted in compiling and
checking data. Steven Leaper assisted with census data.
Professor Gordon Murray provided statistical advice and
suggestions.
Contributors: LJW and IJD had the original idea for the
study and contributed to its design. LJW coordinated the collec-
tion, compilation and checking of data, discussed the analyses,
and made critical revisions to the paper. IJD contributed to the
coordination of data compilation, designed and conducted the
analyses, and wrote the first draft of the paper. The authors are
joint guarantors of the paper.
Funding: The chief scientist’s office of the Scottish Executive
and Henry Smith’s Charities supported this research.
Competing interests: None declared.
1 Fox AJ, Goldblatt PO, Jones DR. Social class mortality differentials: arte-
fact, selection or life circumstances? J Epidemiol Community Health
1985;39:1-18.
2 Black D. Inequalities in health: report of a working group chaired by Sir Doug-
las Black. London: Department of Health and Social Security, 1980.
3 McLoone P, Boddy FA. Deprivation and mortality in Scotland, 1981 and
1991. BMJ 1994;309:1465-70.
4 Duijkers TJ, Kromhout D, Spruit IP, Doornbos G. Inter-mediating risk
factors in the relation between socioeconomic status and 25-year
mortality (the Zutphen study). Int J Epidemiol 1989;18:658-62.
5 Eames M, Ben-Shlomo Y, Marmot MG.Social deprivation and premature
mortality: regional comparison across England. BMJ 1993;307:1097-102.
6 Joseph KS, Kramer MS. Review of the evidence on fetal and early child-
hood antecedents of adult chronic disease. Epidemiologic Rev
1996;18:158-74.
7 Davey Smith G, Hart C, Blane D, Hole D. Adverse socioeconomic condi-
tions in childhood and cause specific adult mortality: prospective obser-
vational study. BMJ 1998;316:1631-5.
8 Valkonen T. Adult mortality and level of education: a comparison of six
countries. In: Fox J, ed. Health inequalities in European countries. London:
Gower, 1989:142-60.
9 Office of Population Censuses and Surveys. Occupational mortality series.
London: HMSO, 1986: table 4.8. (DS No 6.)
10 Doornbos G, Kromhout D. Educational level and mortality in a 32-year
follow-up study of 18-year old men in the Netherlands. Int J Epidemiol
1990;19:374-9.
11 Kunst AE, Mackenbach JP. The size of mortality differences associated
with educational level in nine industrialised countries. Am J Pub Health
1994;84:932-7.
12 Neisser U, Boodoo G, Bouchard TJ, Boykin AW, Brody N, Ceci SJ, et al.
Intelligence: knowns and unknowns. American Psychologist 1996;51:77-
101.
13 O’Toole BI, Stankov L. Ultimate validity of psychological tests. Personality
and Individual Differences 1992;13:699-716.
14 Deeg DJH, Hofman A, van Zonneveld J. The association between change
in cognitive function and longevity in Dutch elderly. Am J Epidemiology
1990;132:973-82.
15 Korten AE, Jorm AF, Jiao Z, Letenneeur L, Jacomb PA, Henderson AS, et
al. Health, cognitive and psychosocial factors as predictors of mortality in
an elderly community sample. J Epidemiol Community Health
1999;53:83-8.
16 Scottish Council for Research in Education. The intelligence of Scottish chil-
dren: a national survey of an age-group. London: University of London
Press, 1933.
17 Maxwell J. The level and trend of national intelligence. London: University of
London Press, 1961.
18 Schwartz JE, Friedman HS, Tucker JS, Tomlinson-Keasey C, Wingard DL,
Criqui MH. Sociodemographic and psychosocial factors in childhood as
predictors of adult mortality? Am J Public Health 1995;85:1237-45.
19 Deary IJ, Whalley LJ, Lemmon H, Crawford JR, Starr JM. The stability of
individual differences in mental ability from childhood to old age:
follow-up of the 1932 Scottish Mental Survey. Intelligence 2000;28:49-55.
20 Whalley LJ, Starr JM, Athawes R, Hunter D, Pattie A, Deary IJ. Childhood
mental ability and dementia. Neurology 2000;55:1455-9.
21 Snowdon DA, Kemper SJ, Mortimer JA, Greiner LH, Wekstein DR,
Markesbery WR. Linguistic ability in early life and cognitive function and
Alzheimer’s disease in late life. JAMA 1996;275:528-32.
22 Hernstein RJ, Murray C. The bell curve: intelligence and class structure in
American life. New York: Free Press, 1994.
23 Carroll JB. Human cognitive abilities: a survey of factor-analytic studies. Cam-
bridge: Cambridge University Press, 1993.
24 Mackintosh NJ. IQ and human intelligence. Oxford: Oxford University
Press, 1998.
25 Deary IJ. Looking down on human intelligence: from psychometrics to the brain.
Oxford: Oxford University Press, 2000.
26 Bosma H, Dike van de Mheen H, Mackenbach JP. Social class in
childhood and general health in adulthood: questionnaire study of con-
tribution of psychological attributes. BMJ 1999;318:18-22.
27 Fischer CS, Hout M, Jankowsky MS, Lucas, SR, Swidler A, Voss K. Inequal-
ity by design: cracking the bell curve myth. Princeton, NJ: Princeton University
Press, 1996.
28 Barker DJP, Winter PD, Osmond C, Margetts B, Simmonds SJ. Weight in
infancy and death from ischaemic heart disease. Lancet 1989;ii:577-80.
29 Bentham G. Migration and morbidity: implications for geographical
studies of disease. Soc Sci Med 1988;26:49-54.
(Accepted 16 January 2001)
Papers
5BMJ VOLUME 322 7 APRIL 2001 bmj.com