Leaded petrol and the crime drop

American violent crime quadrupled between 1960 and 1991 and then fell by half, and nobody who was paid to predict it saw the fall coming. The lead hypothesis says the cause was in the fuel tank: leaded petrol poisoned a generation of infants, those infants grew up more impulsive and more violent, and when the petrol went away so did the crime. It is the most satisfying chart in criminology. It is also the chart that most rewards a hard look.

The crime wave

FBI violent crime rate per 100,000 people. Up from the early sixties, peaking in 1991, then a fall nobody predicted.

Crime line uses published UCR rates at five-year anchors. The petrol index is a schematic of the rise and phase-out, not a series from the paper.

The drop nobody ordered

American violent crime roughly quadrupled between 1960 and 1991, then fell by half over the next twenty years. Every theory of the nineties — policing, prisons, crack markets, abortion — is a theory about this shape.

Meanwhile, in the fuel tank

Tetraethyl lead went into petrol from the 1920s, peaked in the early seventies, and was legislated out between 1975 and 1986. Blood lead levels in American children followed it down.

Put them on the same axes

Side by side and unshifted, the two curves are not obviously about each other. One peaks around 1973, the other around 1991.

Now shift by a generation

Lead poisons infants. Infants become twenty-year-olds. Shift the petrol curve forward about twenty-one years and it lands almost exactly on the crime curve. This is the chart that launched the theory.

And that is the problem

Shift it by thirteen years, or seventeen, or twenty-five, and it still fits. Two smooth humps in the same century will correlate at almost any lag. The chart cannot tell you which lag is the real one, or whether any of them is.

Try it yourself

The serious evidence is not this chart at all. It comes from studies that exploit differences between states and cities in when the phase-out actually bit. Those are in the article.

The claim

Tetraethyl lead went into American petrol from the 1920s, peaked around the early 1970s, and was legislated out between 1975 and 1986. Childhood blood lead levels followed the same path down. Rick Nevin (2000, Environmental Research) noticed that if you shift a national lead exposure series forward by about two decades it tracks the violent crime rate closely, and that the same trick works in several other countries with their own phase-out timing.

Jessica Wolpaw Reyes (2007, B.E. Journal of Economic Analysis and Policy) did the harder version. States phased leaded petrol out at different speeds, so she used that variation as the source of identification rather than the national time series. Her headline estimate was that the removal of lead from petrol accounted for a bit over half of the decline in violent crime during the 1990s.

~50%

fall in US violent crime, 1991 to 2014

UCR rate per 100,000

1975–1986

leaded petrol phase-out

~56%

of the 1990s crime drop attributed to lead

Reyes (2007) headline estimate

How they found it

Three tiers of evidence, and they are not equally good.

The weakest is the national time series, which is the chart above. Two smooth humps in the same century will correlate at almost any lag you pick, and the scene shows that directly: thirteen years fits, seventeen fits, twenty-five fits. A correlation that survives every lag is not evidence for a particular lag.

The middle tier is the cross-sectional work: states, cities and countries that removed lead at different times, compared against each other. This is real identification, and Reyes is the best-known example. It is still observational. States that cleaned up their fuel early differ from states that did not in a long list of ways that also predict crime.

The strongest tier is individual-level. Cohort studies follow children with measured blood lead into adulthood and look at arrests. The Cincinnati Lead Study is the most cited of these, and it finds an association between prenatal and childhood blood lead and later arrests, including violent arrests. Small samples, one city, but actual people rather than aggregate curves.

The mechanism

This is the part that nobody disputes. Lead is a neurotoxin with no safe exposure level. It damages the developing brain, and the damage concentrates in exactly the places you would want intact if your plan for life involved impulse control: lower IQ, worse executive function, more attention problems, more aggression. That is settled toxicology, not a finding from the crime literature.

So the lead hypothesis is not a case of a spurious correlation with no story. The story is excellent. The question is entirely about magnitude.

How much to believe it

The meta-analysis is brutal about the literature, not the mechanism. Anthony Higney, Nick Hanley and Mirko Moro (2022, PLOS ONE) pulled together the published estimates linking lead and crime and tested for publication bias. They found a lot of it. Once you correct for it, the average effect is far smaller than the headline papers suggest, and the authors conclude that the effect on homicide specifically is not distinguishable from zero. Their summary is that lead probably explains a modest share of the crime decline rather than most of it.

Publication bias is the expected failure mode here. A paper reporting that lead explains half of the crime drop gets published and covered. A paper reporting a null coefficient on lead goes in a drawer. When a literature’s incentives all point one direction, the pooled estimate drifts that way, and you need exactly this kind of correction to see it.

Too many causes, one effect. The 1990s crime decline has been attributed to lead, to abortion legalisation, to policing tactics, to incarceration, to the end of the crack epidemic, to demography, and to the economy. The claimed shares add up to well over 100 percent. At most one of those literatures can be right about its own magnitude, and probably none of them are.

The international pattern helps and hurts. Other countries phased lead out on different schedules and also saw crime fall, which Nevin treats as replication. But crime fell in the 1990s across nearly all of the rich world, including in places where the timing does not line up. A common decline with different treatments is weak evidence for any one treatment.

What survives is worth keeping. The individual-level cohort work and the toxicology both point the same way: lead exposure in childhood makes worse outcomes more likely, including criminal ones. That is a real finding about a real poison. It just does not license the claim that lead explains the crime wave.

What it means for you

The lag slider in the scene above is the transferable lesson. Any two slow-moving series with one hump each will correlate at a suspicious number of offsets, and if you get to choose the offset after looking, you will find one. This shows up constantly in marketing attribution and in anything involving a “delayed effect”: spend now, revenue in six weeks, pick the lag that maximises the fit and report it. If your lag was chosen by search, the correlation it produces is not evidence. Fix the lag from theory first, or estimate it on data you then throw away.

Suggested chart

Two panels. Left, the national lead and crime series with a lag slider, so the reader can see the fit survive any choice. Right, the same correlation coefficient plotted against every lag from 0 to 30 years, which turns the first panel’s problem into a single flat-topped curve. The second panel is the honest version of the first.

Sources

Nevin, R., 2000, “How lead exposure relates to temporal changes in IQ, violent crime, and unwed pregnancy,” Environmental Research, 83(1), 1-22. sciencedirect.com.
Reyes, J. W., 2007, “Environmental Policy as Social Policy? The Impact of Childhood Lead Exposure on Crime,” B.E. Journal of Economic Analysis and Policy, 7(1). nber.org.
Higney, A., Hanley, N. and Moro, M., 2022, “The lead-crime hypothesis: A meta-analysis,” PLOS ONE / Regional Science and Urban Economics. journals.plos.org.
Wright, J. P. et al., 2008, “Association of prenatal and childhood blood lead concentrations with criminal arrests in early adulthood,” PLOS Medicine, 5(5), e101. journals.plos.org.