The Social Determinants of Health

Health is where the sociology of everything else becomes visible in a body.

Class, race, gender, work, housing, neighbourhood and childhood all appear in mortality statistics — and they appear not as a difference between the poor and everyone else, but as a gradient running the entire length of the distribution.

That single feature is what makes this literature difficult and important, because it rules out the explanation everyone reaches for first.

A study of civil servants, and the finding nobody expected.

In the late 1960s, researchers began following more than seventeen thousand male British civil servants.

The population was chosen partly for convenience. It was stable, traceable, and had a clean administrative hierarchy. And it had a property the researchers understood only afterwards: nobody in it was poor.

All of them were employed. All were salaried. All had job security, occupational pensions, paid sick leave, and free health care at the point of use. They worked in the same city, in similar buildings, in offices.

They were followed for ten years, and their mortality was compared by employment grade.

Men in the lowest grade had roughly three times the mortality of men in the highest.

And the shape of the finding is what made it famous. It was not a gap between the bottom and everyone else. It was a step at every level.

Administrators had the lowest mortality. Professional and executive grades were higher. Clerical staff were higher again. And the messengers, doorkeepers and other support staff were highest of all. Each grade worse than the one above it, all the way up.

Then the researchers did the obvious thing and checked whether the known risk factors explained it.

Smoking, blood pressure, cholesterol, weight, physical activity, blood sugar — measured at baseline, and used to account for the difference.

They explained roughly a third of it. Two-thirds of a threefold mortality difference, among employed, pensioned, salaried men with free health care, was not accounted for by any of the things medicine was looking at.

A second study, begun in the 1980s and including women, was designed around that residual. It added measures of the psychosocial environment — and one in particular kept coming out significant.

Control over one's own work. How much say a person had over what they did, how they did it, and when. It was strongly graded by rank, and it accounted for a substantial part of the mortality difference that the physical risk factors had not.

And a coda that puts the scale in view. In several cities, life expectancy between small areas a short distance apart differs by figures that have been quoted at twenty years and more.

These particular small-area estimates deserve a caution — they are computed on small populations, have wide uncertainty, and the most-quoted single comparison has been disputed on those grounds. The general pattern is not in doubt: life expectancy differences of a decade or more between neighbourhoods within one city are found repeatedly, in many countries, using robust data.

A gradient is a much harder thing to explain than a threshold.

If the poor were sick and everyone else was healthy, the explanation would be straightforward: deprivation, malnutrition, cold housing, dangerous work, no medical care. That is a threshold effect, and it is what most people assume the finding is.

A gradient rules it out. Senior civil servants are not malnourished, and neither are the middle grades — and yet each step down the hierarchy carries higher mortality.

So whatever produces the pattern must be something that varies continuously with position, all the way to the top.

Two candidate families, and a prior question.

The prior question is direction. Does social position cause ill health, or does ill health cause a fall in social position? Both happen, and disentangling them requires longitudinal data with health measured early.

The finding, broadly, is that causation dominates for most outcomes. Studies following cohorts from childhood find that early social circumstances predict adult health after accounting for early health, and that adult position predicts subsequent health among people healthy at baseline. Selection matters more for some conditions than others — it is substantial for severe mental illness, where onset in early adulthood can derail education and employment — and it does not account for the gradient in cardiovascular disease, which is where the Whitehall finding sits.

Then the two families.

Materialist explanations: the gradient reflects differences in material conditions — housing quality, damp, cold, overcrowding, air quality, physical working conditions, exposure to hazards, nutrition, and the public infrastructure available where people live.

Psychosocial explanations: the gradient reflects the biological consequences of position itself — of low control, chronic demand without discretion, subordination, insecurity and status anxiety, operating through sustained stress responses.

The argument between them was long and bad-tempered, and the resolution is that both operate, in proportions that differ by outcome and by setting. The interesting question is what each can and cannot explain, which is answerable.

The mechanisms

What each family has going for it.

The materialist case is strongest where exposure is direct. Respiratory disease and air quality, injury and dangerous work, infectious disease and overcrowding, cold-related mortality and housing — these are gradients with obvious physical pathways, and they are large.

And its most useful development shifts attention from individual material circumstances to collective provision: the observation that health differences track not only what individuals have but what is provided — transport, water, air, housing standards, safe streets, public services. Which relocates the question from behaviour to policy (see 8.8.1).

Its limitation is the civil servants. None of them was in damp housing or breathing industrial dust, and the gradient was threefold.

The psychosocial case is strongest where the biological pathway can be traced.

Sustained activation of the stress response has documented physiological consequences: elevated cortisol, raised inflammatory markers, higher blood pressure, impaired glucose regulation, and — over years — the cumulative wear described as allostatic load.

And the specific finding that low control at work predicts cardiovascular outcomes has been replicated in several countries, with the mechanism being demand combined with low discretion rather than demand as such. A demanding job with high control is not the harmful combination; a demanding job with low control is.

Supporting evidence comes from an unexpected direction. Long-term study of a wild primate troop found that subordinate animals showed the physiological stress profile the theory predicts — and, crucially, that the profile tracked position in the hierarchy rather than material deprivation, since all animals in the troop were adequately fed.

Its limitations are honest ones. The pathway from stress physiology to population mortality involves several inferential steps; the psychosocial measures are self-reported and correlate with everything; and causal identification at population level is weak — the strongest evidence is biological plausibility plus observational association, which is not the same as a natural experiment.

And the theoretical contribution that explains why the gradient survives everything.

Here is a puzzle the mechanism accounts above do not solve. Over the last century the leading causes of death changed completely — from infectious disease to cardiovascular disease to cancer. Sanitation transformed. Antibiotics arrived. Smoking rose and fell. Health systems were built.

Through all of it, the socioeconomic gradient in mortality persisted, and in several countries widened.

Why should a gradient survive the complete replacement of the diseases producing it?

The answer is that socioeconomic position is a fundamental cause, and the argument is worth stating precisely.

Position confers flexible resources: money, knowledge, power, prestige, and beneficial social connections. These are flexible because they can be deployed against whatever the current threat happens to be, whatever that threat is.

So when a new risk is identified or a new protection becomes available, the advantaged deploy their resources first — they hear about it sooner, understand it faster, can afford it, can access it, and can arrange their lives around it.

The prediction is sharp and testable: gradients should re-emerge or widen whenever new health-relevant knowledge or technology appears, and should be flattest for conditions about which nothing can be done.

And it has held repeatedly.

Smoking was initially more common among the better-off, and reversed as the evidence became known — the advantaged quit first, and the gradient inverted.

Screening programmes, preventive medication, and effective treatments for previously untreatable conditions have each shown the same pattern: uptake concentrated among the advantaged first, with the mortality gradient for that condition widening before it narrows.

And for conditions with no effective prevention or treatment, the socioeconomic gradient is much weaker — which is the theory's most striking confirmation, because it is the case where the resources have nothing to buy.

This is why the gradient cannot be eliminated by addressing any particular risk factor. Address one, and the resources redeploy to the next. It is a mechanism-level explanation of why health inequality behaves like 8.3.3's advantage relocating when a channel is closed — and it is the same finding, in a different institution.

The life course, and one clean natural experiment.

Health at fifty is produced over fifty years, and the life course framework (see 9.2.3) distinguishes three ways early circumstances matter.

Critical periods: exposures at particular developmental stages with lasting effects that later circumstances cannot undo.

Accumulation: the summing of exposures over a lifetime, so that duration of disadvantage matters rather than its presence at any moment.

And chains of risk: one exposure raising the probability of the next — poor early health affecting schooling, affecting employment, affecting exposure, affecting health.

The critical-period claim has one unusually clean piece of evidence.

A famine occurred in a well-documented country over a defined period, with clear start and end dates, in a population with excellent birth and health records. Women pregnant during it can be identified precisely, and — crucially — the stage of gestation at which each foetus was exposed can be determined from the birth date.

Following those cohorts for decades found outcomes differing by the timing of exposure, with distinct patterns for those exposed in early, mid and late gestation, on measures including metabolic and cardiovascular outcomes and, in some analyses, mental health.

This is a natural experiment with an exposure that was involuntary, sharply dated and unrelated to family characteristics (see 7.4.3) — which is why it carries weight that the general foetal-origins literature, where prenatal circumstances are confounded with everything postnatal, does not.

The income inequality hypothesis, reported at its actual strength.

The claim: inequality itself damages health — that living in a more unequal society is worse for you, at any given income, so that the whole population including the affluent does worse.

It became widely known through a book presenting cross-national scatter plots showing strong associations between inequality and a range of health and social outcomes.

The criticisms were extensive and several are substantial. The country selection was contested; the analyses were cross-sectional; the associations were sensitive to which countries were included; the outcomes were correlated with each other so the many plots were not independent evidence; and the individual-level mechanism was underspecified.

And the honest current state, from systematic meta-analysis rather than from either camp's advocacy.

A modest association survives careful multi-level analysis. Pooling studies that measure individual income and contextual inequality, a higher Gini coefficient is associated with slightly higher mortality — an increase of roughly the order of eight per cent in mortality risk for a substantial rise in the Gini — with considerable heterogeneity between studies and some evidence of a threshold below which no association appears.

So: real, small, much smaller than the headline presentation, and with a contested mechanism.

And the underlying interpretive problem is 7.6.4's. A contextual effect of inequality is difficult to distinguish from the aggregation of individual income effects, because the relationship between income and health is curved — so a more unequal distribution of the same total income mechanically produces worse average health without any contextual effect at all. Separating the two requires exactly the multi-level designs that find the smaller estimate.

Two further findings that this Part cannot omit.

One — behaviour is patterned, and treating it as a competing explanation is a bad control (see 7.6.2).

Smoking, diet, physical activity and alcohol are all graded by social position, and they account for a real share of the health gradient.

But they are not an alternative to the social explanation. They are one of its pathways. Controlling for them and reporting the residual as "the unexplained social effect" removes a mechanism from the estimate and reports the removal as a limit on social causation. The right question is why the behaviours are patterned — and the answers involve price, availability, stress, marketing concentration, time, and what is normal in the immediate environment.

Two — and the finding that connects this Topic to Part 8 most sharply.

In the United States, pregnancy-related mortality is several times higher for Black women than for white women — and the gradient does not behave as an income or education story would predict.

Black women with a college degree have higher pregnancy-related mortality than white women who did not complete high school.

Read that again, because it defeats the standard framing. Education, which is the strongest single predictor of health in most analyses, does not close the gap; it does not even reverse the ordering across an enormous educational distance.

The leading sociological interpretation is weathering: the hypothesis that cumulative exposure to disadvantage and to discrimination produces accelerated physiological deterioration, so that biological age exceeds chronological age. Supporting evidence comes from allostatic load measures showing higher cumulative physiological wear at a given age, with the gap widening through the reproductive years — which is precisely when the maternal mortality gradient operates.

And it is the clearest available demonstration of 8.4.3's argument in a body: an exposure operating independently of class position, not reducible to income or education, with a documented physiological pathway.

Because health is the outcome where the consequences of everything in Parts 8 and 9 become measurable, and where the standard policy lever is the wrong one.

Health care is a modest contributor to population health in rich countries. The commonly cited figure — that medical care accounts for something like a tenth to a fifth of variation in health outcomes — is a modelling estimate rather than a measurement, and should be quoted as such. The direction is not seriously disputed: a health system treats illness, and does not principally produce health, which is determined largely by conditions outside it.

Four questions.

Is it a gradient or a threshold? If a gradient, deprivation-based explanations are insufficient by construction.

Causation or selection? Both operate; causation dominates for most outcomes; selection matters most for conditions with early onset.

Has behaviour been controlled for, and reported as a limit on the social explanation? That is a bad control, and the finding it produces is an artefact.

And is a new technology or piece of knowledge involved? If so, expect the gradient to widen before it narrows — which is the fundamental cause prediction, and it should change how new health interventions are designed.

One closing observation about the civil servants.

They all had free health care. They were all employed, salaried, pensioned and secure. On every variable that public debate treats as the determinant of health, they were equal — and the man at the bottom of the hierarchy was three times as likely to die.

Whatever produced that was not medicine, and it was not poverty. It was the position itself — and that is the finding this whole literature exists to explain.

Seventeen thousand male civil servants, none of them poor, all employed, salaried, pensioned and with free health care, showed a threefold mortality difference from the lowest grade to the highest — as a continuous step at every level. The standard risk factors explained roughly a third of it. A later study including women found control over one's own work to be a major mediator of the remainder.

A gradient rules out the threshold explanation by construction. Whatever produces it must vary continuously with position all the way to the top.

Causation dominates selection for most outcomes — early circumstances predict adult health net of early health — with selection mattering most for conditions with early onset.

Materialist explanations are strongest where exposure is direct and shift attention to collective provision rather than individual possession; their limitation is the civil servants. Psychosocial explanations are strongest where the biological pathway is traceable — cortisol, inflammation, allostatic load — with demand combined with low control the harmful combination, and primate evidence showing the stress profile tracking hierarchical position among adequately fed animals. Their limitation is weak causal identification at population level.

Fundamental cause theory explains why the gradient survived the complete replacement of the diseases producing it. Position confers flexible resources deployable against whatever the current threat is, so gradients re-emerge or widen whenever new knowledge or technology appears — as with smoking's reversal, screening, and effective treatments — and are weakest for conditions about which nothing can be done, which is the theory's sharpest confirmation.

The life course distinguishes critical periods, accumulation and chains of risk — and a sharply dated famine in a population with excellent records, where gestational stage at exposure is determinable, provides the natural experiment the general foetal-origins literature lacks.

The income inequality hypothesis survives careful meta-analysis at a much smaller magnitude than its popular presentation — roughly an eight per cent mortality increase for a substantial Gini rise, with heterogeneity and a possible threshold — and the contextual effect is hard to separate from the curvature of the individual income–health relationship.

Behaviour is a pathway, not a competing explanation; controlling for it and reporting the residual is a bad control.

And Black women with college degrees have higher pregnancy-related mortality than white women who did not complete high school — which no income or education account can produce, and which the weathering hypothesis addresses through cumulative physiological wear.

Health care is a modest contributor to population health: a system treats illness and does not principally produce health.

Social gradient in health — a continuous relationship between position and health across the whole distribution.

Social causation versus health selection — position affecting health; ill health affecting position.

Materialist and psychosocial explanations — differences in exposures and provision; consequences of position, control and status.

Job control / demand–control model — discretion over one's own work; harm concentrated where demand is high and control low.

Allostatic load — cumulative physiological wear from sustained stress responses.

Fundamental cause — a social condition conferring flexible resources deployable against whatever the current health threat is.

Critical period / accumulation / chains of risk — three life course mechanisms linking early circumstances to later health.

Income inequality hypothesis — the claim that inequality itself damages health; a small association survives careful analysis.

Weathering — accelerated physiological deterioration from cumulative exposure to disadvantage and discrimination.

Behavioural pathway — health behaviours as a channel through which social position operates, not an alternative to it.

One — check the shape. For any health inequality statistic you meet, find out whether it is a gradient or a comparison of two groups. The first requires a different explanation from the second.

Two — test the fundamental cause prediction. Take a recent health technology or piece of health advice. Ask who adopted it first, and what that implies for the gradient over the next decade.

Three — find the bad control. Locate a study reporting that social differences in health "are largely explained by behaviour". Then ask why the behaviour is patterned.

Four — look up two neighbourhoods. Find life expectancy for two areas near you. Note the difference — and then note the sample size behind each estimate.

Five — apply the control question. For your own work or a job you know, rate demand and discretion separately. The combination, not the demand, is what the evidence implicates.