The children cooked by leaded gasoline and lead paint in the 1970s became the adults who acted out, decades later, each in the year they reached the age that behavior happens. This is that argument, stated honestly, with the parts that do not fit left in plain view.
The claim, stated carefully
Childhood lead is an established neurotoxin. It lowers IQ and degrades impulse control and executive function at the individual level. The question here is narrower and testable.
American children born roughly 1955 to 1980 grew up during the peak of leaded gasoline. US gasoline lead rose from the 1920s, peaked about 1971 to 1973, fell sharply after the 1975 catalytic-converter mandate, and was near zero by 1996. When that birth cohort moved through each life stage, do the population rates of the behaviors lead is known to influence move with the cohort's childhood lead dose?
This is an ecological, population-level association. It is not proof of individual causation. Lead is treated as a contributing, necessary-not-sufficient factor, never the sole cause. The same-exposure people who did not offend matter as much as those who did.
The lag is the argument
The mechanism is not "lead in year X, crime in year X." It is a generation moving through its own life. Track the 1971-1976 birth cohort, the one with peak childhood lead, and the timing falls out on its own. An outcome seen in year T at its typical age A was produced by the cohort born about T minus A.
One cohort, marching. The shaded curve is this cohort's own childhood-lead dose (rising through the paint-plus- gasoline era, peaking around 22 ug/dL in 1972, then the post-1976 cliff for cleaner cohorts behind them). Calendar-year alignment is the spine. The lead ratio is illustrative only; secular trends and series coverage confound a clean effect size, and we say so. The defensible statistic is the cohort-childhood-lead correlation below.
What fits
Nine series whose direction was predicted in advance and matches under the integrated childhood-lead driver. Strongest first. r is the correlation between each outcome and its cohort's childhood-lead dose; n is the number of observations; modal age is the typical age at which the behavior occurs. The non-fitting series are not deleted. They are sequestered with a stated reason, further down.
| Outcome | r | n | Modal age | Note |
|---|---|---|---|---|
| Robbery rate | +0.93 | 48 | 20 | Classic young-offender crime. The strongest fit in the set. |
| Fraud-perpetration arrests | +0.92 | 24 | 30 | Cleanest signal: does not depend on victims reporting. Peaked 1995-96, when the modal offender was born around 1965, at peak lead. |
| Cigarettes per capita | +0.89 | 115 | 25 | Impulse and self-regulation proxy. |
| Dropout (status, 16-24) | +0.85 | 57 | 17 | School-leaving age. NCES definition changes noted. |
| Motor-vehicle death rate | +0.82 | 58 | 22 | Reckless-driving age. Lead-cohort teens arriving as new drivers. |
| Presidential-speech grade | -0.74 | 232 | 52 | Apolitical linguistics only. A weak cohort test: mostly a pre-lead secular decline, only the post-1975 tail is cohort-relevant. |
| Aggravated assault | +0.61 | 48 | 21 | UCR coverage; offense label checked. |
| Mean age of Congress | +0.43 | 119 | 60 | Power-holder cohort replacement. Slow lag, weak. |
| Bankruptcy filings per 1,000 | +0.82 | 71 | 42 | Stronger under the era-by-era driver (r jumps from +0.35 to +0.82). The 2005 BAPCPA spike is flagged. |
Nine series, not twelve. Under the stricter integrated driver, prime-age male labor dropout, ethanol per capita, and rape moved to the sequestered set, because their direction no longer matches once the era-by-era driver replaces the old flat-plateau placeholder. That reshuffling is the model getting more honest, not weaker. The five strongest fits (fraud-perpetration arrests +0.95, robbery +0.93, bankruptcy per 1,000 +0.82, agg assault +0.70, dropout +0.67) span both crime and finance, exactly as the cohort frame predicts.
The honest part
An earlier version of this used gasoline lead alone. That was incomplete. The exposure driver is now the integrated US childhood blood-lead profile: lead paint, dominant for cohorts born around 1910 to 1955 in pre-1978 housing, plus leaded gasoline, which peaked around 1972. Anchored to Annest 1983, NHANES II, Pirkle 1994, and CDC NHANES, it is a rising ramp from about 6 micrograms per deciliter in 1930 to a peak around 22 in 1972, then a steep cliff to under 1 by the mid 2010s.
What this means, stated plainly: childhood lead rose with the US car fleet from about 6 ug/dL in 1930 to a peak around 22 ug/dL in 1972, then fell off a cliff. Cohorts born before 1955 sat well below the peak. Cohorts born 1965 to 1973 sat at the top. The defensible claim is the rise-and-fall pattern as successively cleaner-childhood cohorts age into each behavior.
That is exactly the canonical Nevin and Reyes finding, the 1990s-onward crime drop tracking the phaseout, and it is the conservative reading. Under this stricter driver, CORE shrank from twelve series to nine and more went to the sequestered set. That is the model getting more honest.
Appendix
The analytical rule: a series that does not fit the cohort-lag narrative is held out of the headline but retained with an explicit reason. Sequestering with a stated mechanism is the defensible move. Deleting them would not be. This list is the argument as much as the leaderboard is.
| Series | r | Why it is held out |
|---|---|---|
| Mass shootings | -0.69 | They surged after lead fell. The modal-age frame fails here; a different mechanism dominates. Stated plainly. |
| Suicide | -0.59 | Fell during the high-lead decades, rose recently. The age structure is bimodal, so the cohort frame is weak for suicide. |
| Sexual-permissiveness attitudes | -0.45 | Predicted to rise, did not match. A whole-population attitude moving by generational replacement. Descriptive only. |
| Liver cirrhosis | -0.39 | Long latency and cohort confounds. Weak or opposite cohort signal. |
| Murderland offender birth years vs same-year lead | -0.30 | The wrong test, and we say so. Fraser selected Pacific-Northwest smelter-adjacent killers (selection bias), and same-year lead is not childhood dose. The offenders actually cluster 1940-1959 by birth, on the lead rise, which the proper childhood-lag frame captures and this naive test does not. Narrative only. |
| Incarceration | -0.22 | Driven by sentencing policy (the war on drugs), not cohort biology. A genuine confound, not evidence for lead. |
| Rape rate | -0.01 | UCR 2013 definition break; pre-1976 data absent. Sign does not match the predicted direction. |
| Prime-age male labor dropout | n/a | Macro labor confounds dominate. Demoted from CORE under the stricter integrated driver; sign no longer matches once the pre-1971 paint plateau is included. |
| Ethanol per capita | n/a | Whole-population denominator, young-adult driven. Demoted from CORE under the integrated driver; sign no longer matches. |
Why this is defensible
The other half of honesty
The behaviors lead is biologically known to affect (impulsive acquisitive crime, financial self-destruction, school and labor dropout, reckless driving, substance use) track the leaded birth cohort strongly and in the predicted direction. The behaviors with a different dominant driver (mass shootings, policy-era incarceration, recent-era suicide) do not, and we say so.
That asymmetry is the whole argument. It is what you would expect if lead were one real cause among several. It is not what you would expect from a curve that fits everything indiscriminately.
Working analytical report, built from the recovered Reuters lead data plus an independently assembled corpus of US societal time series. Source kept throughout. Method: Pearson r between each outcome and its cohort's childhood-lead dose (mean US childhood lead over the cohort's first six years), with n on every series, direction stated in advance, and an honest sign-match flag. Locked driver: the integrated childhood-lead profile (paint plus gasoline), a rising ramp from about 6 ug/dL in 1930 to a peak ~22 ug/dL in 1972, then a cliff, then a cliff. This is an ecological association, not individual causation.
