1 The Causal Question
Does the environment in the womb leave a permanent economic imprint? The fetal origins hypothesis, associated with the epidemiologist David Barker, holds that conditions during gestation programme adult health, productivity, and longevity. If true, the implications for human-capital policy are enormous: the most consequential interventions might occur before a child is even born. But testing the hypothesis is fiendishly hard. Prenatal conditions correlate with everything that comes after family income, nutrition, parental education so any cross-sectional association between birth circumstances and adult outcomes is drenched in confounding.
Almond [2006] found a way around the problem in one of the cleanest natural experiments in economics. The 1918 influenza pandemic struck the United States suddenly in the autumn of 1918, infected roughly a third of women of childbearing age, and receded almost as quickly by early 1919. Children who happened to be in utero during that brief, severe window were exposed to a major maternal health shock through no choice of their own or their families'. Decades later, the U.S. Census recorded how their lives turned out. The pandemic, in effect, randomly assigned a prenatal insult to one birth cohort and spared its neighbours.
2 The Identification Strategy
The design is a cohort-based difference, exploiting the sharp timing of the pandemic. Children conceived before mid-1918 were born and largely past the most vulnerable gestational stages before the virus arrived; children born in 1919 were in the womb at the pandemic's peak; children conceived in 1919 escaped it. The treated group is therefore the cohort in utero during the autumn 1918 wave roughly those born in 1919 and the comparison groups are the adjacent, unexposed cohorts born just before and just after.
The key identifying assumption is that, absent the pandemic, adult outcomes would have evolved smoothly across these closely spaced birth cohorts. Formally, if Yc is the average adult outcome of birth cohort c and Dc indicates in-utero exposure, the estimand is
identified under the claim that cohort identity near 1919 is, but for the virus, unrelated to later outcomes. Two features make this credible. First, the shock was sudden and unanticipated families could not have timed conception in response to a pandemic no one saw coming. Second, it was transient, so the exposed cohort is sandwiched tightly between unexposed cohorts, leaving little room for slow-moving secular trends to masquerade as a treatment effect. Almond [2006] reinforces the design by examining the within-1919 pattern by quarter of birth, since the gestational timing of exposure varied across the cohort.
3 Data and Setting
The analysis draws on the long-form U.S. Census samples (and related administrative sources) that recorded the 1918-1919 cohorts in adulthood, from 1960 onward. Because the Census is population-scale, Almond [2006] could measure outcomes for cohorts defined at quarterly resolution and still command ample statistical power a luxury survey data rarely afford for a single birth year. The outcomes span the human-capital spectrum: educational attainment, physical disability, employment, income, and socioeconomic status, all observed when the affected cohort was middle-aged.
4 Key Findings
The fingerprints of a prenatal shock were unmistakable, four decades on. Relative to the adjacent unexposed cohorts, men and women who had been in utero during the pandemic showed:
- Lower educational attainment. The exposed cohort was markedly less likely to graduate from high school on the order of 15 percent lower graduation propensity relative to trend.
- Worse health. Rates of physical disability in adulthood were elevated for the in-utero-exposed cohort.
- Lower earnings and status. Adult wages and income were depressed by roughly 5 to 9 percent, with corresponding declines in occupational socioeconomic standing and increased reliance on public transfers.
The magnitudes are economically large for a health shock that lasted a single season and that the affected individuals could not consciously remember. They constitute some of the most credible evidence for the fetal origins hypothesis, precisely because the timing of exposure was driven by an exogenous epidemiological event rather than by family circumstances.
5 Limitations and What We Learn
The design's elegance does not exempt it from scrutiny, and the literature has pressed on two issues. The first is selection into the cohort. A pandemic that affects pregnant women may alter who is conceived and who survives gestation if the most fragile pregnancies were lost, the surviving exposed cohort could be positively selected, biasing the estimated harm
toward zero (a "culling" or mortality-selection concern). That such selection would, if anything, attenuate the findings makes the observed negative effects more, not less, convincing. A second concern is that the pandemic coincided with the end of the First World War and with economic disruption; Almond [2006] addresses confounding shocks through the tight cohort windows and the within-cohort timing analysis, but disentangling overlapping historical events is never fully clean.
The broader lesson is twofold. Methodologically, the study is a model of how a sharp, exogenous, transient shock converts the passage of birth cohorts into a quasi-experiment-a template later applied to famines, radiation events, Ramadan fasting during pregnancy, and pollution episodes. Substantively, it reframed early-childhood policy: if a few months in the womb can move adult earnings by several percent, the returns to protecting maternal and prenatal health may dwarf those of many later interventions. The 1918 pandemic, in Almond [2006]'s memorable framing, was not fully "over" even in the late twentieth century-its economic shadow still fell on the people who had been carried through it before they were born.
References
- Almond, D. (2006). Is the 1918 influenza pandemic over? Long-term effects of in utero influenza exposure in the post-1940 U.S. population. Journal of Political Economy, 114(4), 672-712.