Lead and crime: a hole in the theory

Kevin Drum, whose “Criminal Element” essay linking violence rates to lead exposure has been (as it deserves to be) extraordinarily influential, links to a BBC story in which a more-than-usually-dim criminologist explains that he disbelieves in the lead-crime link because all biological explanations of crime are outside “the mainstream” (presumably meaning “the mainstream of criminological research”). Kevin is right to say that the multiple levels of evidence for the lead-crime link (brain-imaging work, case-control studies at the individual level, studies of crime changes in neighborhoods that suffered greater and lesser increases in lead exposure after WWII and decreases in lead exposure after the mid-1970s, and cross-national studies) deserve more than a casual blow-off. It’s not, after all, as if “mainstream” criminology had a bunch of well-worked out causal theories with proven predictive power.

Like Kevin, I’ve been frustated that the evidence about lead hasn’t led to more action, if only some serious experimental work removing lead from randomly selected buildings or neighborhoods and then doing developmental measurements on the childen in the chosen areas compared to children in control areas. And I was puzzled when I heard that the hyper-sensible Phil Cook of Duke was among the skeptics.

Alas, Phil seems to have good reason to raise questions, especially about the ultra-strong claims sometimes heard about the percentage of the crime drop since 1994 attributable to changes in lead exposure. (Note that in a system with multiple positive feedbacks, the change-attribution factors needn’t sum to unity, so the claim that “the change in X accounts for 90% of the change in Y” is dubious on its face.)

Cook writes:

My skepticism about the “lead removal” explanation for the crime drop stems from the same source as my skepticism about the Donohue-Levitt explanation in terms of abortion legalization. They have an exact parallel to the previous explanations for the surge in violence of the late ‘80s by John DiIulio and others. The focus in both cases is the character of the kids – the belief that during the 80s the teens were getting “worse,” and during the 90s they were getting “better.” Even a fairly casual glance at the data demonstrates that whatever the cause of the crime surge, and then the crime drop, it was not associated with particular cohorts. It was an environmental effect – 10 cohorts were swept up in the crime surge simultaneously, and the drop has the same correlated pattern.

There is a natural inclination to assume that the reason the murder rate is increasing is because there are more murderers, and the reason we have fewer is that there are fewer murderers. It’s not that I rule out such explanations – I’m open to the idea of lead removal and abortion legalization – it’s just that I don’t think it explains the actual pattern of the youth violence epidemic, either up or down. More generally, my instinct is to look to the social and economic environment to explain large shifts in population outcomes.

Here’s Cook’s paper with John Laub.

Now, since youth violence surely has impacts on adult violence, finding that all the cohorts decreased their violence at the same time doesn’t quite rule out the idea that the younger cohorts were largely driving the train. And those brain-imaging, case-control, neighborhood, and cross-national results are still out there. But Cook’s objection remains a powerful one: the theory predicts that we should have seen the crime declines specifically in the birth cohorts exposed to less lead, and that’s not what we saw in fact.

I’d still like to see some experimental work to figure out how much additional lead removal would turn out to be cost-justified (with crime-reduction benefits a significant, but but not the dominant, part of the benefit story), but the simple story “Crime declined because lead declined” now has a major hole in it. That’s too bad, because attempts to find the relevant social and economic causes haven’t gotten very far. To my mind, the crime boom that started in the late 1950s or early 1960s and the crime bust that started in the mid- 1990s (and other large-scale, decades-long swings up and down in crime rates elsewhere) remain largely unexplained in retrospect, as they were entirely unpredicted in prospect.

Footnote And that, of course, is the difference between science and policy analysis on the one hand and mere advocacy on the other. Of course scientists and analysts are influenced by their values and prejudices. I don’t pretend that Jessica Reyes’s work didn’t make me happy or that Cook’s response doesn’t make me sad. The lead-crime story fits all my political prejudices, as well as my taste for simple and surprising explanations with clear policy implications. And I’ve been a loud advocate for it. But I don’t want to believe it if it isn’t true.

Comments

  1. RhodesKen says

    My recollection is that in his first post on the subject, Kevin showed an interesting time-offset correlation in different countries that had different schedules of lead reduction. This is about as good as you can do for accidentally creating a huge "experiment" to test the hypothesis. It's late now on the East Coast but tomorrow I will try to go back and find those data and post them here.

  2. JamesWimberley says

    Lead should go simply because of the known neurological damage and the IQ effect. The argument is fighting over the past. In a sense, the lead theory is pessimistic, since now lead has gone from motor fuel, there's little more to be gained that route. The more crime is explained by current environmental influences, the more hope there is to reduce it further. One strong argument for decriminalizing lesser drugs, and emphasising non-penal sanctions for the drug offences that remain, is that prisons are academies of crime.

  3. NedW says

    The hypothesis as I understand it is that lead exposure explains much (most) of the dramatic increase and decrease in violent crime over the past century (and possibly past century, in the case of lead paint) but that other factors explain the baseline rate of violence upon which that rise-and-fall was superimposed.

    So it will still take a lot of work to identify and address those *other* factors if we want to reduce crime much further (lead abatement efforts may still produce some improvement but even eliminating 100% of lead exposure wouldn't completely stop violent crime).

    If people misinterpret the past few decades' rise-and-fall of crime as being due to their preferred factor X when it was actually due to lead exposure, then that will distort future efforts to reduce violence. We need to understand what happened in the past to have a realistic sense of what can be achieved in the future. This is not exactly "pessimistic" since there is still hope to reduce violence by other means. But it is "realistic" because it doesn't attribute to those other means the brilliant success of the 1990s-2000s crime decrease.

  4. FCB says

    See comments to Kevin’s reply. The cohorts do not move together. Youth declines were first and increases in violent crime in older age categories may still be increasing. Some other stuff going on with the data, but a cursory glance does not suggest crime went up and down in all cohorts simultaneously.

  5. tgranulosa says

    I'm a bit confused by the argument. Nevin and others have utilized cohort data as evidence for the lead hypothesis.

    Basically juvenile violent crime peaked first while older age groups were still seeing increases in violent crimes. If you look at the shape of the violent crime demographic it's been skewing older and older since the early 1990's. I believe we're seeing higher rates of violent crime in the 50-64 year old demographic than we ever have before.

    Just a couple of links from Kevin's comment thread:

    See the final figure and caption here: https://www.ncjrs.gov/html/ojjdp/jjbul2000_12_3/p

    And newer data on incarceration rates here: http://www.ricknevin.com/uploads/It_Will_Not_Take

  6. erasmuse says

    "the crime boom that started in the late 1950s or early 1960s and the crime bust that started in the mid- 1990s (and other large-scale, decades-long swings up and down in crime rates elsewhere) remain largely unexplained in retrospect, as they were entirely unpredicted in prospect."

    How is that? Plenty of people predicted in the 60s that Warren Court procedural innovations and the soft-on-crime attitude of courts would increase crime— and it did. Plenty of people predicted in the 1970s and 1980s that increasing prison terms would reduce crime—and it did. They pointed to left field first, and then hit the ball there.

    And of course that is the obvious theory— punish crime more and it will decline. Only criminologists seem to be skeptical of the theory and fooled in their predictions.

  7. erasmuse says

    I haven't read the papers, but since lead paint has been around forever and lead pesticides started being phased out around 1945, I think, is the lead theory really a theory that driving increased a lot after WW II and the lead in gasoline pushed lead levels up enough to increase crime? Since we've gotten rid of leaded gasoline, that probably would seem to be largely solved. I guess another implication is that in the 60s crime woudl have gone up a lot in cities but not changed in areas with low car air pollution.

    • tgranulosa says

      The basic idea is based on the effect of lead on brain development. Low levels of lead exposure are demonstrated to have a significant effect at the molecular scale in how growing neurons interact with other growing neurons and how many connections they develop. This shows up visually in the MRIs of children with higher early life lead exposure and it shows up statistically in things like IQ tests (not surprisingly, negatively). A bit unexplained is the fact that the MRIs show the largest changes in parts of the brain that may be associated with impulse control — the ability to interfere with and control anger, arousal, etc.

      The expected response is supposed to be related to people that were exposed to high levels of lead (urban areas between 1960 and 1980 having the highest exposure). The relationship with an increase in violent crime (not the existence of violent crime or all violent crime) presumably shows up with a 15-20 year offset because infants and toddlers are not prosecuted for block throwing and the natural peak age for violent perpetrators is in the teens and 20's.

      This shows up in national statistics with an offset increase, peak, and decrease in violent crime (murder, rape, assault, etc.). It also shows up in other countries with different curves depending on when leaded gasoline was used and phased out. In Central and South America late phase outs are associated with violent crime numbers which are still be increasing. It's not easy to study in smaller populations as people move and violent crime is rare (small population equals bad statistics) but there is some indications that lead smelting towns have higher violent than national averages. Evidence also shows up in large long-term studies of lead poisoned children and in the lead analysis of teeth of inmates being significantly higher than the general population. Kevin Drum's longer article on the subject is here: http://www.motherjones.com/environment/2013/01/le

      The current argument is whether violent crime also is associated with the appropriate lead poisoned cohort when it peaks. The 1994 data point in particular saw higher violent crime levels in both youth (as expected) and in older populations who should not have been as poisoned. I'd basically argue that you need to look at more than the 1994 data point and try to filter through the noise. Large scale trends are that youth violence did in fact start declining first and that older age groups either peaked later or are still increasing and becoming a greater proportion of the prison population. These older demographics aren't generally associated with violent crime but relative to the same age people in 1980 they are being significantly more violent today. In the late 80's it was the youth crime explosion. 15 years ago it was the 30-45 year olds who where being statistically aberrant (see my link in earlier on 1999 data) . Today it's retired people in Japan who are committing assaults and murder (at a greater rate than ever in the past).

      It's still a hypothesis but I personally don't think the "hole" being referenced here isn't really a hole. The linked paper made the argument back in 2002 (against the abortion hypothesis) but they're data is actually fairly weak and contradicted by more recent reports. I'd like to see age cohort graphs for all violent felons from 1980 to the present with all the FBI data. Graphs referencing victim age are indirect at best and the focus on 1994 raises the potential for highlighting noise over long-term trends.

  8. SteveSailer says

    As I pointed out in Slate to Steven Levitt in 1999 in arguing about his abortion-cut-crime theory, the homicide rate for teens born in the late 1970s was about triple what it was for teens born in the late 1960s:
    http://isteve.blogspot.com/2014/04/does-abortion-

    That's inconvenient both for his abortion theory and for the lead theory. Of course, that was the Crack Era that caused the spike in youth homicides around 1993, but that shows that hypothesized driving forces like abortion or lead aren't very important relative to changes in the drug business.

    • Keith_Humphreys says

      Terrific debate! I did not see that at the time but greatly learned from reading it just now. Thanks for posting it.

  9. paul says

    Seems to me that if you accept even a little bit of a multifactorial explanation for crime, you’re going to expect different cohorts to be at least somewhat correlated.

Trackbacks