Frequently Asked Questions

That was my question, too, when I first started seeing these results. I assumed that any effect worth reporting would be measured in years, because I was used to newspaper accounts and journal articles that talk about period life expectancy and the major changes in it over time. Period life expectancy gains over the twentieth century were on the order of thirty years, and today we worry about gains and losses in the one-to-two-year range — COVID, for example, caused temporary losses of around two years, and bigger ones in some groups and places. So I anticipated that notable estimates would look something like that.

Looking back, I don’t think any one particular in-utero or early-life condition would be as big as COVID. What this really taught me is the difference between period life expectancy and what this project actually measures: longevity in old age — how long you live conditional on surviving to a relatively old age, typically past 65. In those terms, the gains the United States has made past age 65 over a hundred years are more like two or three years. Against that yardstick, an effect of two or three months from a single early-life exposure is consequential — a meaningful share of what an entire century accomplished.

Ideally, we would run a randomized controlled trial for in-utero exposures and track people for ninety or a hundred years to measure the effects on longevity. Without the time and resources to conduct such a trial — we would have to wait a century for an answer — our strategy is what we call “moving forward by looking backward.” We start with recent deaths and trace the potential early-life determinants of those deaths back ninety years, rather than waiting ninety years.

And we are very attuned to finding instances in the world, and in the data, that come close to a randomized trial: where a policy, a natural disaster, or some other exposure strikes some people and not others, some places and times and not others. We are really tracing out the effects of being lucky or unlucky in the exposures you face. The closer we get to that ideal, the closer we are to estimating something truly causal.

We basically don’t know yet. It’s a much more challenging question, because if we want to keep relying on causal estimates, we need people to be lucky or unlucky twice to measure the combined effect of two exposures — and three times for three. Instances where we can ask whether two bouts of bad luck have synergistic effects — effects bigger than the sum of their parts — are genuinely rare in the data. It’s one of the frontiers of this work.

Two things. First, how many exposures seem to matter, and how similar their effects are — again and again we find effects of two or three months, across very different kinds of events. Second, how big some of the effects are for racial differences in longevity that may have their roots in violence and animus in the environments of ninety or a hundred years ago.

It’s actually hard to show that something has a true zero effect — which is different from saying the evidence is inconclusive. There are many instances where the data are simply too small to detect an effect of a month or two, and in those cases the results are inconclusive rather than zero. I don’t think we have cases of exposures that are truly zero, but we do have cases where the effect is a month or less — close enough to zero, especially for one-off exposures that affected few people and that we don’t expect to see again.

It’s also worth saying that we aren’t picking exposures at random. We focus first on exposures with a plausible pathway to lifespan — ones with documented effects on important mechanisms like educational attainment, occupation, or income.

I think this pathway reminds us how important education is in structuring the rest of our lives — and, in particular, in structuring many of the exposures we either face or avoid for the rest of our lives: a low-paying versus a high-paying job, work that exposes you to chemicals or physical strain, or a job with no control over what you do each day. Education can work through both resource mechanisms and stress mechanisms. Others have pointed to education as a “fundamental cause” of health, and I think our results are quite consistent with that framework. We’ve also added to that literature with some of the best evidence available on college attendance and longevity, using college openings from a hundred years ago.

The main lesson is that policies and events have many effects, and we happen to focus on longevity. Neither Prohibition nor the boll weevil was a longevity policy. Prohibition wasn’t a failed longevity policy — in a lot of ways it probably improved public health — but prohibiting alcohol has its own costs, starting with the fact that people didn’t like it.

One way to frame this: people aren’t health maximizers or longevity maximizers. Economists would call them utility maximizers — happiness maximizers. If alcohol makes you happy, you may be willing to trade off some health for that happiness. So Prohibition may have failed in terms of happiness while succeeding in terms of longevity. Judging any policy requires deciding which ledger you’re reading.

Our results are really a test of a theory that says they should be common. The nine months before birth are the most consequential of our lives — the stage of development when all our organs are forming, when the developing body is, in a sense, trying to figure out how to survive. Shifts in resources, stress levels, and other exposures during that window can be particularly consequential.

Most of our evidence supports this: when we compare in-utero exposure with exposure at age eight or ten, the in-utero effects are often bigger. That’s consistent with the idea of fetal programming — the fetus is learning from its environment during those essential nine months. Learning doesn’t stop at birth; but some of what is locked in during those nine months is very hard to undo at age eight, or ten, or thirty.

It comes from that same backward-looking approach. A death record gives us a name, an age at death, and usually something about where the person was born or lived. We then go to the historical census and look for someone with the same name, age, and place. That key step — matching on name — works for men because they keep their surnames across their lives. Women in these historical periods generally did not, so a death record with a married surname usually can’t be traced back to a maiden name with most of our data. Methods for linking women are a very active area of work.

The teams that build these matching algorithms put in safeguards so that only very likely matches are kept. The price of that caution is that we lose people with common names. Anyone who died named John Smith is unlikely to be in our data — there are so many John Smiths that name, age, and birthplace alone can’t tell us which one he is.

We always worry that people with common names might have different experiences related to longevity, but we think that pathway is limited enough that it doesn’t threaten the results. Still, it’s an honest limitation: we lose the John Smiths.

That’s exactly what the logic of a randomized trial is meant to do — isolate one thing, with a treatment group that is just like the control group except for the treatment. For Prohibition, we rely on the fact that before federal Prohibition, localities, counties, and states adopted their own prohibitions at different times — and some counties were forced dry by their states. We call that involuntary prohibition, and people born in those involuntarily dry counties, just before versus just after the law changed, make a very good contrast group.

So the worry isn’t “other things happening in the 1920s” in general. The worry would be something else changing in those same counties, at those same moments, systematically across all the counties becoming involuntarily dry — something that also affects longevity. We run statistical tests designed to detect processes like that, and we don’t find evidence of them.

We would love to measure all of those. But death records contain one measure that matters here: when you died, and at what age. Nothing else. We are limited by the data — which is also why lifespan is the one outcome we can measure this completely, for this many people, this far back.

Because it tests the big conceptual idea: that fetal programming can last your whole life — that circumstances around birth, whether from policy, infectious disease, resources, or stress, can echo across ninety years. That’s the main headline.

And although time has passed and the specific exposures differ, the classes of exposure are similar. We have much less infectious disease today — though COVID is a recent counterexample. But we still have stress. We still have variation in family resources. We still have exposure to alcohol and other drugs during pregnancy. And we have environmental contaminants whose effects we don’t yet know. We can show that lead mattered and that clean water mattered; with that evidence in hand, it’s reasonable to expect that exposures like PFAS today also matter.

Heavy investment before and during pregnancy — reducing family stress and increasing family resources. Think of it as a pre-child tax credit. It should be large, and it could phase out a few years after birth. Our evidence suggests that investments concentrated in that window could really pay off across an entire lifetime.

We have received generous funding from the National Institute on Aging across several grants, and I received a Guggenheim Fellowship related to this project. Neither organization shapes the findings in any way.

We keep a very long list. We have roughly 25 to 30 projects done or near done — and probably another fifty to a hundred that could be done. We are also looking into building a website where people can suggest topics, with AI assisting in finding preliminary answers. Watch this space.

In the United States, it’s a story of gradual improvement that now seems to be stalling compared with other rich countries. The sources of both — the improvement and the stall — are what we’re trying to document. Our evidence suggests a lot of these patterns are tied to early-life circumstances. It adds to other evidence that education matters, smoking matters, access to health care matters — and that the reduction in infectious-disease exposure achieved by sanitation and public health works over a hundred years ago really mattered. Many of these elements seem to matter more than most pharmaceuticals, even blockbuster ones. A lot of the story of longevity is a story about early life.