The Plume and the Alibi

An answer to Caroline Fraser, whose book Murderland argues that lead and arsenic from the smelters of the Pacific Northwest helped produce the serial killers of the 1970s and 1980s. A reader sent me a fragment of an interview with you. You are asking why the Pacific Northwest produced so many serial killers, and you have arrived at smelters, at lead, and at a pair of graphs that lie on top of each other. You won a Pulitzer for Prairie Fires and you do not deserve a lazy reply, so here is the long one. One half of what you are saying is sound. The other half is rotten, and the rot is spreading into the half that deserves to survive.

Start with what you have right, because it is a great deal. Lead is a neurotoxin. There is no known safe level of exposure to it in a child. It crosses into the developing brain because the body mistakes it for calcium, and once inside it damages myelination, impairs nerve conduction, and reduces brain volume in ways that show up on imaging decades later. The smelters were real, and so were the stacks. That Tacoma plume settled across roughly a thousand square miles of Puget Sound carrying lead and arsenic, and the Washington Department of Ecology is still mapping the soil it left behind. You are right that the smelters ran for about a century and that almost all of them are gone now. None of that is in dispute here, and none of it needs a serial killer to make it terrible.

The scale of the American exposure is worse than most people understand. Between 1976 and 1980, the National Health and Nutrition Examination Survey found 88.2 percent of American children aged one to five carrying blood lead at or above 10 micrograms per deciliter, with a geometric mean of 15.2, against a CDC reference value that today sits at 3.5. A 2022 study in the Proceedings of the National Academy of Sciences, which won that journal’s Cozzarelli Prize, estimated that more than 170 million living Americans were exposed above 5 micrograms per deciliter in early childhood, that 54 million were above 15, and that four and a half million were above 30. The authors put the cumulative cognitive cost at 824 million IQ points. Children born between 1966 and 1970 lost an average of 5.9 points each. Concede all of it. I do.

Now the sentence where your argument goes wrong. “You take these and you put them over each other, and they are the same.”

That is the sound of two smooth curves being mistaken for a mechanism. A curve that rises through the sixties and falls through the nineties will lie neatly on top of every other curve with that shape, and mid-century America produced dozens of them. Legalized abortion tracks it, which is how Donohue and Levitt built their own explanation of the crime drop. Crack cocaine tracks it. Incarceration tracks it upside down. So does the share of the population made up of young men, and so does police staffing. Steven Levitt’s 2004 survey named four factors he thought explained the nineties decline and six he thought did not, and lead was not among the four. When a dozen candidate causes all fit the same silhouette, fitting the silhouette proves nothing about any of them. You need identification, meaning some source of variation in lead exposure that is not tangled up with poverty, old housing, highway proximity, and everything else that predicts crime on its own.

The field has done that work, and the results are not what your transcript suggests. In 2022, Anthony Higney, Nick Hanley, and Mirko Moro published the first meta-analysis of the lead-crime literature in Regional Science and Urban Economics. They pooled 542 estimates from 24 studies. They ran seven separate tests for publication bias, and every one of them came back positive, which means the literature’s effect sizes are inflated by the fact that null results do not get written up. Across the full sample they found a partial correlation of 0.16, a small to moderate effect. Then they split out the studies that address endogeneity with quasi-experimental designs, the seven papers using instrumental variables and difference-in-differences, the ones actually built to identify a causal effect. The mean partial correlation in that subsample was 0.01, within one standard deviation of zero.

Sit with that number. The better the study design, the smaller the effect. That pattern repeats when you sort by unit of analysis: research on crime rates in geographic areas produces a mean of 0.25, and research following individual human beings produces 0.03. Effects shrink as the evidence gets closer to actual people. Rick Nevin’s two famous time-series papers, the ones that generate the overlay graphs everyone shares, come in at partial correlations of 0.91 and 0.81. Nothing in social science correlates at 0.91. When a result looks like a law of physics, the first thing to suspect is the method.

The individual-level studies tell the same story. David Fergusson’s longitudinal work concluded that lead explained about one percent of the variance in offending, and that what effect there was ran through poorer educational outcomes rather than through aggression. Robert Sampson and Alix Winter followed a Chicago birth cohort and found school-age lead levels were not associated with later arrests at all; their pooled estimate is negative. The Dunedin cohort in New Zealand, tracked from birth, produced a small positive association and a conclusion from the authors that other factors matter far more. The meta-analysis’s published bottom line is that falling lead accounts for somewhere between 7 and 28 percent of the American homicide decline, which leaves most of the decline unexplained and requires other causes.

I want to be scrupulous here: that is not nothing. A real effect on ordinary violence, spread across a whole population, adds up to a large number of dead people. Lead abatement saved lives. The meta-analysis says so explicitly and so do I. Higney and his coauthors are in no sense lead skeptics; their paper states that the evidence of biological harm is overwhelming, that no safe level of exposure exists, and that abatement improves children’s health whatever it does to crime statistics. They are correcting an inflated figure while defending the case against lead on its own ground.

But ordinary violence is not what you are trying to explain, and this is where the argument stops being merely overstated and becomes indefensible.

Serial murder is not a marginal case of assault. Roughly 770 serial killers were active in the United States during the 1980s, at the peak, in a country of about 240 million people. The Radford and Florida Gulf Coast database documents about 3,613 in the entire country since 1900. Set that against four and a half million American children who carried more than 30 micrograms per deciliter in their blood, ten times the level that now triggers intervention. If every serial killer in the recorded history of this country had been one of those poisoned children, they would amount to less than a tenth of one percent of them. An exposure that nearly every American child in the postwar era shared cannot explain an outcome that almost no American child produced. The arithmetic does not permit it. What it permits is a story.

The mechanism fails too, and it fails in a way that should have stopped the theory at the door. The documented behavioral signature of childhood lead exposure is lowered IQ, shortened attention, impulsivity, poor school performance, and weakened self-control. Those deficits predict a particular kind of offender: disorganized, opportunistic, impulsive, caught early, cycling through the system. Ted Bundy was the photographic negative of that profile. He took a psychology degree from the University of Washington, went to law school, and volunteered on state political campaigns. He worked across six states, kept his crimes separated by geography and time, escaped custody twice, and served as his own attorney at trial in Florida. Whatever ruined him, it was not a deficit in impulse control or cognitive function. You are invoking a mechanism that predicts one kind of criminal to explain a criminal of the opposite kind. You can answer that other men on your list do fit the disinhibited profile, and some of them plainly do. That answer costs more than it pays. Tens of millions of American men grew up impulsive, badly schooled, and lead-burdened, and they went to work, raised children, and killed nobody. A profile shared by an entire generation cannot pick out the few dozen who became predators.

Then there is the map, which does not match. If smelter plumes made serial killers, then the most heavily poisoned children in American history should be the test case. They are identifiable and they are close by. In September 1973 the baghouse at the Bunker Hill smelter in Kellogg, Idaho caught fire, and Gulf Resources chose to keep running the smelter without filtration for eleven months because lead prices were high. The company dumped an average of 73 tons of lead a month onto Smelterville and Kellogg. When the CDC and the state screened the children, 173 out of 175 living within a mile of the stack had blood lead at or above 40 micrograms per deciliter. Their average was 67.4. A one-year-old tested at 164, the highest level ever recorded in an American child. That is the worst childhood lead poisoning event in the history of this country, it happened in the Pacific Northwest, and it happened to the exact birth cohort your theory would put in the field twenty years later. The Silver Valley produced a Superfund site of 1,500 square miles. It did not produce a Bundy.

I will be fair about the limit of that argument. Kellogg was a small town, serial killers are rare, and a single town cannot disconfirm anything by itself. But that cuts in both directions, and it cuts harder against you. If Smelterville is too small a sample to refute the theory, Tacoma is too small a sample to establish it. You cannot claim one plume as evidence and dismiss another as noise. And when you widen the frame, the per-capita rankings of serial murder by state, whatever methodology you pick, tend to put Alaska, Montana, Louisiana, and the District of Columbia at the top. Robert Hansen hunted women in the Alaskan backcountry for more than a decade. Anchorage has no smelter, and neither does the District of Columbia.

The deepest problem is that your theory has no edges. Bundy spent his first three years in Philadelphia and moved to Tacoma at three or four, so Philadelphia is folded in as the City of Smelters. Richard Ramirez grew up in El Paso, so the El Paso plant is folded in. Dennis Rader grew up near the Kansas lead belt, so that is folded in. Jack the Ripper is folded in on the strength of Victorian London’s coal smoke and lead paint. Charles Manson is folded in because he was imprisoned on McNeil Island, downwind of the Tacoma stack. Manson was born in Cincinnati in 1934, raised between Ohio, West Virginia, and Indiana, and arrived at McNeil Island in 1961 as a twenty-six-year-old convicted check forger. His brain had finished developing more than a decade before he ever breathed Puget Sound air. If adult incarceration downwind of a smelter counts as childhood developmental exposure, the theory has stopped making claims and started collecting pins. You call the structure a crazy wall of connections yourself. The trouble with a crazy wall is that the string goes wherever the hand takes it, and a wall that can accommodate Cincinnati, Roxborough, El Paso, Wichita, Whitechapel, and a federal penitentiary is a wall with no wrong answers on it.

That is the fatal move, and it should be named plainly. Every city in mid-century America ran on leaded gasoline. Every child in the country was breathing it. A variable that is present everywhere cannot explain variation anywhere. The moment you accept Philadelphia and Tacoma and El Paso and Wichita and London and a federal penitentiary as evidence for the same proposition, you have built a theory that no observation can contradict, and a theory that nothing can contradict has told you nothing.

Your curve carries a second problem the transcript never reaches. What it plots is serial killers who were caught and identified, which is a different thing from serial killers who existed. Active serial killers in the United States went from 198 in 1987 to 12 in 2018. Michael Aamodt, who built the database that produces those numbers, has been candid that the recent figures run low because identification lags by years. He also offers the mundane explanations: 79 percent of American serial killers had been in jail or prison before their first murder, and since 1950 nearly 17 percent of them killed again after being released for a prior homicide, so longer sentences and tighter parole removed a meaningful share of them from circulation. Fewer people hitchhike. Fewer people walk to school. DNA typing entered criminal casework in the late 1980s and the national index followed in the nineties. The FBI’s violent criminal apprehension program started in 1985. The graph you overlaid on the lead curve is partly a record of forensic science and parole policy, and disentangling that from anything about brain chemistry is a serious research problem that nobody has solved by looking at two lines.

Finally, test the premise itself. Washington was not an unusually murderous state. The Washington State Institute for Public Policy, working from vital statistics running from 1908 through 1998, reported that Washington’s homicide rate has been below the national average for most of the twentieth century, and that its rate in 1998 was 44 percent below the national figure. The lowest homicide rates of the whole century in that series fall in the 1950s and early 1960s, the years the Tacoma stack was at full throat. British coverage of this thesis has repeated a claim that Washington’s murder rate in the mid-1970s was almost six times the national average. The national rate then ran near ten per hundred thousand, so that claim requires a Washington rate near sixty. The same vital statistics series shows the state sitting below the national line throughout, which is the opposite of the claim by a wide margin. Whatever made the Pacific Northwest distinctive in those decades, a high murder rate was not it.

You hedge, sensibly, and you have hedged the same way in print: lead alone will not send a man on a killing spree, every individual is different, and it is one factor among many. In the transcript it comes out as a tipping point, “one more drop in the bucket is going to just spill it all over.” I understand the impulse behind that hedge, and it is the most dangerous sentence in the whole exchange. A tipping point that cannot be measured, predicted, or specified in advance is compatible with every possible outcome. If the poisoned child kills, the lead tipped him. If the poisoned child becomes an accountant, the bucket had room. The hedge is what keeps a hypothesis alive after the evidence has declined to support it, and a claim that cannot be wrong cannot be right either.

Which brings me to the line I have thought about most: “The amount of research that has been done to connect lead to crime is itself a powerful political argument.” That inverts the relationship between evidence and belief. The volume of a literature reflects what funders fund and what journals print, and this particular literature has been shown, by seven separate statistical tests, to be inflated by publication bias. Higney and his colleagues note the reason with some dryness: it is theoretically implausible that lead reduces crime, so researchers who find null results tend not to write them up. Volume is the symptom being diagnosed, not the diagnosis.

So let me lay out the available explanations for the Pacific Northwest cluster and tell you where I come down.

Environmental toxicology, your position, is not effective because the exposure was nearly universal while the outcome was nearly nonexistent, because the best-identified studies find effects statistically indistinguishable from zero, because the behavioral signature of lead predicts the wrong kind of offender, and because the theory absorbs every counterexample instead of being tested by any of them.

Geography and infrastructure is effective because it explains the things serial predation actually requires. The I-5 corridor moves an enormous transient population between Seattle, Tacoma, Portland, and the Canadian border. The Cascade foothills and the Green River valley put dense second-growth forest within thirty minutes of a freeway exit, which solves the disposal problem that stops most killers. Dozens of small municipal police departments with no shared database allowed a man to work across jurisdictional lines for years. Bundy, Ridgway, and Randall Woodfield all worked that corridor, and the corridor explains far more about their method than their childhood addresses do.

Investigative and cultural amplification is effective because it explains why these particular names recur. Washington built one of the earliest dedicated serial murder task forces, and detectives there, Robert Keppel foremost among them, went on to write the books and design the systems that taught the country the category. A region that investigates, names, and publishes its killers will appear to have more of them than a region that files the same murders as unsolved. Half the sensation of the Pacific Northwest cluster is the sensation of good police record-keeping.

Chance is effective because rare events cluster. Eight or ten offenders spread across three states and thirty years is precisely the kind of pattern randomness produces, and any explanation has to beat the null before it earns a hearing.

My recommendation is that the corridor and the amplification account for the cluster together, that chance has to be cleared first and probably absorbs a good portion of what remains, and that lead belongs in the story of American violence as a modest population-level contributor to ordinary homicide and nowhere in the story of serial murder at all. I hold the first three loosely. They are hypotheses, and I have no more proof of them than you have of yours. What separates them is that none requires a mechanism the evidence contradicts, and none has to explain why four and a half million poisoned children produced no killers.

Here is what worries me most about your argument, and it has nothing to do with Ted Bundy. You are spending the credibility of environmental science on a claim it cannot support. The case against lead is already overwhelming on its own ground: cognitive loss measured in the hundreds of millions of IQ points, cardiovascular and kidney damage, ruined school outcomes, a poisoned generation the country has never compensated. That case is built on evidence that survives every test anyone has run at it. Tie it to serial killers, and when the serial killer claim collapses, as it will, you hand every polluter’s attorney a demonstration that the people warning about lead overreach. The Guggenheims already answer for Smelterville. That charge sheet runs long enough to convict without Ted Bundy’s name anywhere on it.

And there is a moral problem underneath the empirical one. A theory that makes Bundy a casualty of ASARCO hands him the one thing he spent his life trying to manufacture, an explanation that lets him off. He lied about that with great skill while he was alive, and the theory now does the lying on his behalf. Meanwhile the real victims of the plume are the children of Ruston and Smelterville and Tacoma who lost five IQ points apiece and never hurt anyone, whose injuries are permanent and undramatic and will never carry a true crime book. Your argument takes their story and spends it on his alibi. They deserve better from you, and so does the evidence.