Lead and crime Arguments that lead has a lot to do with crime levels, and discussion of why this has gotten so little attention.
Just to indulge in a little evolutionary psychology..… Punishing people and helping people are both strong drives, but spending a lot of money on lead abatement (the lead from gasoline is still in the soil, and it keeps coming back—lead paint is still a problem, too) is pretty boring.
ETA: And worse, progress with lead abatement is literally invisible (you don’t have a dam or a highway so it looks like you’re doing something) and the good effects take some 15 or 20 years to be obvious.
The basic point is reasonable, but there are so many things that bother me about that article.
Drum’s credulity varies a lot in this article. His lowest level is about where I stand. I have to wonder if that actually reflects his beliefs and the rest of it is forcing enthusiasm on himself because to reflect value rather than truth; that is, he is doing an expected value calculation. Certainly, he should be applauded for scope sensitivity.
Perhaps the biggest thing that bothers me is that Drum tries to have it both ways: small amounts of lead matter and big amounts of lead matter. It seems rather unlikely that this is true. Maybe 10μg/dL has a huge effect, but if so, I doubt that 20 has double that effect, and this ruins all the analysis of the first half of the article. This is important because there is a logical trade-off between saying that past lead reduction was useful and saying future lead reduction will be useful. In particular, Drum says that Kleiman says that if the US were to eliminate lead, it would reduce crime by 10%. Did he just make up this number, or does it come out of a model? I’d like to see the model because even if he pulled the model out of thin air, it forces him to deal with the logical trade-off.
In Kleiman’s book, he says that eliminating lead paint would reduce crime by 5% and attributes it to Nevin 2000. On the same page, he misquotes Nevin in a way that makes me not trust Kleiman with models. But that’s OK because he has a citation, not model. I cannot find the claim in Nevin’s paper. There is a model on p19 that says that 6 points of IQ, applied to the lowest 30% of the population could explain the past decline. And that’s at a rate of 2 points of IQ for 10μg/dL, a small enough rate I’m willing to extrapolate linearly. If you assume crime in linear in lead, the 5% number is reasonable, except for the assumption that lead explains all of the past decline. (I’m not sure Nevin actually makes this assumption because I don’t think he makes a prediction about eliminating lead; in this section, I think he’s just doing a reality check that the known IQ effect of lead plus the known correlation of IQ and crime is big enough to explain the whole drop in crime.)
So I am bothered by Drum’s language about the effects of low levels of lead, even though the suggestion of a 10% drop in crime maybe survives the trade-off between past and future. (And how does Kleiman’s 5% turn into “Kleinman’s” 10%? windows vs windows+soil?)
From the first half of the article:
the field of econometrics gives researchers an enormous toolbox of sophisticated statistical techniques
Econometrics gives people enough rope to publish themselves. Plus they implement these algorithms in spreadsheets, to hide the bugs from themselves.
murder rates have always been higher in big cities than in towns and small cities
If lead explains everything, this should not always have been true. In fact, I think it was not true in 1960. The graph Drum cites starts in 1975, after most of the increase in national murder rates has already happened, but there is very little dependence on city size until later. The graph seems to me evidence against the claim that lead explains this detail. Anyhow, such bucketed graphs are a bad way to test this hypothesis. In particular, there are only 9 “big cities” and NYC has 1⁄3 of this population. The convergence today is probably driven just by NYC now having a lower murder rate than small cities.
Drum says that Newarks’s crime rate dropped 75%. That is true and but it is also true that Newark’s murder rate has rebounded to its peak. I don’t know how to resolve this. I usually prefer murder rates because they are harder to fake, but there are only about 80 murders in the worst years, making the data quite noisy.
That the graphs of leaded gasoline and crime match perfectly, up until year that Nevin’s first paper was published screams publication bias.
Crack:
Trying to explain the crack epidemic in terms of childhood seems like a serious error to me. It seems very clear to me that it was contagious. How it spread and why it burnt itself out, I do not know. Regardless, one can disprove Nevin’s model’s claim to explain the crack epidemic, like Levitt’s spreadsheet fraud before it, because it assumes that the age of criminals is constant in time. In fact, the crack epidemic involved young murderers, born after lead levels had started to decline. I think Nevin worries about this in later papers, but I don’t know what he does.
Here is a suggestion for a better model for testing Nevin’s hypothesis than he used in 2000: instead of lagging on some constant, create a new time series of murder by age of birth. This also corrects for the demographic problems such as the baby boom. The disadvantage is that this loses exogenous effects, such as the crack epidemic, which hit multiple ages simultaneously. Yet another time series, to avoid the problem of missing data, uses the age of the victim rather than of the perp.
So Nevin fails to explain the crack epidemic, but if he just explains the big rise and the big fall, that’s a big deal. Unfortunately, the presence of the crack epidemic masks the big fall. In the absence of crack, when would crime have started falling? Perhaps it would have started falling earlier, but was elevated by crack. Or perhaps all those dead or jailed young teens would have become 25 year old criminals and so the effect of crack was to speed things up, including the falling crime rate.
There’s a lot you can do to remediate lead and the bioavailable forms of it, fortunately (been working on a garden in an urban area, and bioremediation is a chief concern) -- it doesn’t just have to involve removing it. Unfortunately, it’s still likely to be rather expensive and unglamorous, so it’ll be a tough sell as a point of policy.
The sexy project would be to figure out how to undo the effects of lead on people years after they’d been exposed as children. I think succeeding at this would wonderful, but I wouldn’t put off cleaning up lead in the environment in the meanwhile.
That’d be beyond “sexy”; the effects of lead poisoning on the central nervous system are generally considered irreversible. I daresay anything that could repair that sort of brain damage would have a whole host of other applications...
Lead and crime Arguments that lead has a lot to do with crime levels, and discussion of why this has gotten so little attention.
Just to indulge in a little evolutionary psychology..… Punishing people and helping people are both strong drives, but spending a lot of money on lead abatement (the lead from gasoline is still in the soil, and it keeps coming back—lead paint is still a problem, too) is pretty boring.
ETA: And worse, progress with lead abatement is literally invisible (you don’t have a dam or a highway so it looks like you’re doing something) and the good effects take some 15 or 20 years to be obvious.
The basic point is reasonable, but there are so many things that bother me about that article.
Drum’s credulity varies a lot in this article. His lowest level is about where I stand. I have to wonder if that actually reflects his beliefs and the rest of it is forcing enthusiasm on himself because to reflect value rather than truth; that is, he is doing an expected value calculation. Certainly, he should be applauded for scope sensitivity.
Perhaps the biggest thing that bothers me is that Drum tries to have it both ways: small amounts of lead matter and big amounts of lead matter. It seems rather unlikely that this is true. Maybe 10μg/dL has a huge effect, but if so, I doubt that 20 has double that effect, and this ruins all the analysis of the first half of the article. This is important because there is a logical trade-off between saying that past lead reduction was useful and saying future lead reduction will be useful. In particular, Drum says that Kleiman says that if the US were to eliminate lead, it would reduce crime by 10%. Did he just make up this number, or does it come out of a model? I’d like to see the model because even if he pulled the model out of thin air, it forces him to deal with the logical trade-off.
In Kleiman’s book, he says that eliminating lead paint would reduce crime by 5% and attributes it to Nevin 2000. On the same page, he misquotes Nevin in a way that makes me not trust Kleiman with models. But that’s OK because he has a citation, not model. I cannot find the claim in Nevin’s paper. There is a model on p19 that says that 6 points of IQ, applied to the lowest 30% of the population could explain the past decline. And that’s at a rate of 2 points of IQ for 10μg/dL, a small enough rate I’m willing to extrapolate linearly. If you assume crime in linear in lead, the 5% number is reasonable, except for the assumption that lead explains all of the past decline. (I’m not sure Nevin actually makes this assumption because I don’t think he makes a prediction about eliminating lead; in this section, I think he’s just doing a reality check that the known IQ effect of lead plus the known correlation of IQ and crime is big enough to explain the whole drop in crime.)
So I am bothered by Drum’s language about the effects of low levels of lead, even though the suggestion of a 10% drop in crime maybe survives the trade-off between past and future. (And how does Kleiman’s 5% turn into “Kleinman’s” 10%? windows vs windows+soil?)
From the first half of the article:
Econometrics gives people enough rope to publish themselves. Plus they implement these algorithms in spreadsheets, to hide the bugs from themselves.
If lead explains everything, this should not always have been true. In fact, I think it was not true in 1960. The graph Drum cites starts in 1975, after most of the increase in national murder rates has already happened, but there is very little dependence on city size until later. The graph seems to me evidence against the claim that lead explains this detail. Anyhow, such bucketed graphs are a bad way to test this hypothesis. In particular, there are only 9 “big cities” and NYC has 1⁄3 of this population. The convergence today is probably driven just by NYC now having a lower murder rate than small cities.
Drum says that Newarks’s crime rate dropped 75%. That is true and but it is also true that Newark’s murder rate has rebounded to its peak. I don’t know how to resolve this. I usually prefer murder rates because they are harder to fake, but there are only about 80 murders in the worst years, making the data quite noisy.
That the graphs of leaded gasoline and crime match perfectly, up until year that Nevin’s first paper was published screams publication bias.
Crack:
Trying to explain the crack epidemic in terms of childhood seems like a serious error to me. It seems very clear to me that it was contagious. How it spread and why it burnt itself out, I do not know. Regardless, one can disprove Nevin’s model’s claim to explain the crack epidemic, like Levitt’s spreadsheet fraud before it, because it assumes that the age of criminals is constant in time. In fact, the crack epidemic involved young murderers, born after lead levels had started to decline. I think Nevin worries about this in later papers, but I don’t know what he does.
Here is a suggestion for a better model for testing Nevin’s hypothesis than he used in 2000: instead of lagging on some constant, create a new time series of murder by age of birth. This also corrects for the demographic problems such as the baby boom. The disadvantage is that this loses exogenous effects, such as the crack epidemic, which hit multiple ages simultaneously. Yet another time series, to avoid the problem of missing data, uses the age of the victim rather than of the perp.
So Nevin fails to explain the crack epidemic, but if he just explains the big rise and the big fall, that’s a big deal. Unfortunately, the presence of the crack epidemic masks the big fall. In the absence of crack, when would crime have started falling? Perhaps it would have started falling earlier, but was elevated by crack. Or perhaps all those dead or jailed young teens would have become 25 year old criminals and so the effect of crack was to speed things up, including the falling crime rate.
There’s a lot you can do to remediate lead and the bioavailable forms of it, fortunately (been working on a garden in an urban area, and bioremediation is a chief concern) -- it doesn’t just have to involve removing it. Unfortunately, it’s still likely to be rather expensive and unglamorous, so it’ll be a tough sell as a point of policy.
The sexy project would be to figure out how to undo the effects of lead on people years after they’d been exposed as children. I think succeeding at this would wonderful, but I wouldn’t put off cleaning up lead in the environment in the meanwhile.
That’d be beyond “sexy”; the effects of lead poisoning on the central nervous system are generally considered irreversible. I daresay anything that could repair that sort of brain damage would have a whole host of other applications...
International data about a connection between lead and crime
pdf
less technical pdf