A research collection
America had a 1990s violent-crime drop nobody can fully explain. Fraud arrests peaked thirty years later. Personal bankruptcy peaked forty years later. Every cohort that walked through its twenties between 1990 and 2016 broke something the previous cohort did not. The thing they all share: they were born when childhood lead exposure was at its highest in US history.
Childhood lead exposure rose with the US car fleet. About 6 ug/dL in 1930. About 10 ug/dL in 1950. About 22 ug/dL at the 1972 peak, the highest a generation of children has ever been measured at. Then the catalytic-converter mandate forced gasoline lead out and the curve fell off a cliff. American children born roughly 1955 to 1980 grew up at or near that peak. They became the adults who acted out, decades later, each in the year they reached the age the behavior happens. This collection lays the argument out honestly, shows the evidence three different ways, and lets you see where your own cohort sits.
Ecological association, not individual causation. Lead is treated as a contributing, necessary-not-sufficient factor. Series that do not fit the pattern are kept with a stated reason, never deleted. Each tool below names its Source.
Start here. The report lays out the claim, the honest limits, and the data behind it. The recovery story is how we got it.
The full claim, the method, the nine CORE fits, the sequestered non-fits with stated reasons, and the falsification section.
Read the report StoryReuters quietly deleted their landmark childhood-blood-lead small-area dataset. The tiles were still cached and we pulled all 30,187 areas back.
How we got the data backThree different ways the cohort pattern shows up. Each is the same claim, looked at from a different angle.
Pick any of 17 series. See how its birth-year curve tracks the integrated childhood-lead driver. Strong CORE fits up top, sequestered non-fits kept with a stated reason.
Open the explorer NewThe leaded cohort (born 1971-76) vs the pre-lead cohort (1900-1923) vs the post-phaseout cohort (1995-2000), at the same modal ages. The gap is the lead signal.
Compare cohorts NewJapan 1986, Germany 1996, US 1996, UK 1999, Italy 2002, plus Mexico, France, Australia, South Korea, Canada. Each country's outcome turns at its own cohort offset, not all at once.
See the natural experimentFind yourself in the data. Find your county on the map. Watch the cohort move through the country year by year.
Enter your birth year. See your childhood lead index, which behaviors peaked when you hit each life stage, and how your cohort compares to the peak-leaded and post-phaseout cohorts.
Find yourself in the data NewCounty-by-county time map. Press Play and watch the leaded cohort hit each behavior age in each part of the country. Switch metrics from the rail: BLL, firearm, overdose, bankruptcy, more.
Open the time map UpdatedAll 30,187 recovered Reuters areas plus EJScreen paint, mortality, ARCOS, and bankruptcy layers. New: type your ZIP or county to see your area's percentile and neighbors.
Find your areaSource. Recovered Reuters "Off the Charts" (Pell & Schneyer 2016-17), CDC NCHS, CDC CBLS, EPA EJScreen, FBI UCR and BJS, FTC, U.S. Courts, Federal Reserve, NHTSA, NCES, Census, NIAAA, UNODC, WHO Mortality Database, OurWorldInData. Childhood-lead driver: integrated paint plus gasoline (Annest 1983, NHANES II, Pirkle 1994, CDC NHANES). Cohort frame per Nevin and Reyes. Murderland point-sources from Fraser 2025.
Defensibility. Ecological (population) associations between an outcome and the integrated childhood blood-lead of the birth cohort that produces it at its modal age. Lead is a contributing, necessary-not-sufficient factor. The driver curve has real measured anchors after 1976 (NHANES II onward) and reconstructed anchors before 1976 (Nevin 2007 USA panel, McFarland-Hauer 2022 PNAS birth-cohort shares above 5 ug/dL, gas-lead tonnage from Nriagu 1990, paint-era reasoning from Mielke). Pre-1976 absolute level carries roughly +/- 3 ug/dL uncertainty; the shape (low to 1930, rising through the gas era, peak ~1972, cliff after) is well constrained. Series that do not fit the cohort pattern are sequestered with a stated reason, never deleted.
