It is a great article. It is convincingly informative and interesting while telling a relatively simple story (with important complexity in the details).

I didn't find it terribly convincing. I mean, it was convincing that lead plays some role in crime rates. But I don't think his argument for the primacy of lead as a causal factor was that well made.

You're still rooting for Roe v. Wade, aren't you.

Fight!

I mean, it was an integration of various fundamentally correlational pieces of research. It seemed to draw on a couple of fairly significant bodies of research, which, good, but making a causal chain out of bodies of correlational research across various fields seems tricky to do, especially in a narrative rather than an actual scientific context.

3: it's not like he convinced me that he's wrong, either. I'm just not sold.

it's not like he convinced me that he's wrong, either. I'm just not sold.

It's true, a big part of his argument is just, "it's hard to think of any other affects which would be as uniform across different cities." I found that convincing but it leaves a lot of room for somebody to propose an alternate theory.

No, not you two. Oudemia was instructing Roe and Wade.

What about the fluoride, huh? Won't someone please think of the children?!

I'm going with rock 'n' roll and video games.

Shamefacedly, I admit that my primary two reactions were:

1. Doesn't everybody know this already? Seriously, I remember thinking it was old news when I read a post on Obsidian Wings about it at least five or six years ago.

2. My Malcolm Gladwell "just-so story" b.s. detector is going off. This is just too cleverly written for my comfort, even if the underlying research is solid as all get-out.*

*Which, fwiw, I believe it is.

Don't listen to me, though. I'm a grump who is definitely not the primary audience for this article, and also the person who this morning found out that her city gets 150,000 domestic violence reports a year. -- i.e., 1 for every 10 residents.**

**Yes, yes, that's not an unduplicated number. It's not actually 150,000 people. Still.

OT: My drink just finished a bottle of Drambuie. I've never seen an empty one before.

That's odd, because there were 150,000 empty Drambuie bottles in your city last year, all of them from your house.

12: If it's causing you to go blind, it may not actually be Drambuie.

I'm going with rock 'n' roll and video games.

Rock & Roll, Video Games, and lawyers, of course.

Doesn't everybody know this already?

I've certainly seen discussion about lead before, but this had more numbers than what I had remembered (though, of course, that only adds value if the numbers are correct).

I think wood alcohol humor is funnier than domestic violence humor.

But I don't think his argument for the primacy of lead as a causal factor was that well made.

I didn't think he was making an argument, I thought he was giving a brief digest of the results of research that actually does make an argument.

Maybe that means I'm agreeing with you. Anyway, it sounds like there might be sufficient statistical evidence to be convincing, but it's hard to know without going to the primary literature. I'm also a little wary about the "Why has the lead/crime connection been almost completely ignored in the criminology community?" bit, because cases I know in which people talk to reporters about how they've done amazing research and the stodgy old academic establishment won't listen to them are usually just crackpots. But due to the interdisciplinary nature of this subject I guess I could imagine it's actually being ignored even if it's really solid.

Rock & Roll, Video Games and Tetraethyl Lead ought to be a convenience store.

I would like to say that, on my slightly rushed reading, this seems like a lot more solid results (in terms of correlations and quality of underlying studies) than the vast majority of research into societal issues typically yields. The multiplicity of research methods finding the same results is also good, since results so often evaporate on further examination.

I'm marking off the years until I can become a crackpot.

20: Why wait?

That will be 2036 for those keeping track. Or as soon as I get Fuck Yinz money.

21: Family obligations.

I guess famous surfer dude crackpot doesn't have any kids.

I only have one. Maybe I'll move the schedule up to 2030 or so.

The factoid about crime rates in cities of different sizes converging was very interesting.

17.2: yup. All that.

And a bag of chips.

Chips might also play a causal role. There's like 43 flavors of Doritos now.

I only eat Cool Ranch, because the dust has blue and red bits.

And I'll cut anyone who says multi-colored dust doesn't have more vitamins than orange dust.

The criminology community ignoring this says more to me about them than the theory, but I lean crackpot. If they looked into this and had solid reasons for dismissing it then lets hear them.

the person who this morning found out that her city gets 150,000 domestic violence reports a year. -- i.e., 1 for every 10 residents.

At least if this book can be trusted, despite the fact that Ciudad Juárez is known worldwide for the murders of women, your city is actually more dangerous for them.

(That book is outstanding, even if you're not interested in soccer at all.)

I'm going to lick the innards of a car battery, just so people know I'm serious about the cutting.

Those blue and red bits are actually flecks of lead paint. A rather questionable choice by the Frito-Lay company, I think, but you can't argue with deliciousness.

31: Theory not valid for paint chips.

17

Maybe that means I'm agreeing with you. Anyway, it sounds like there might be sufficient statistical evidence to be convincing, but it's hard to know without going to the primary literature. ...

He has more than a statistical association with crime rates, he has a plausible proposed mechanism (lead causes brain damage causes crime) although I wonder how solid the brain damage estimates are, particularly for the lower exposure levels.

Just to be clear, I'm only going to cut people who aren't in my household.

I only eat Cool Ranch

Is it no longer called "Cooler Ranch"? And if not, why didn't they upgrade it to "Coolest Ranch"?

Sonofabitch. Spike probably ate fewer chips than me.

God, you guys are a bunch of buzzkills. I thought the article was great.

The Ethyl Corporation. Kind of like RCA. Still in business too. They did lots of advertising in popular magazines.

General Motors had the "use patent" for TEL as an antiknock, based on the work of Thomas Midgley, Jr., and Esso had the patent for the manufacture of TEL.

I'll cut the marketing department at Frito Lay if the name doesn't go back.

26: Yeah. A few other bits impressed me, like the claim that murder rates rose and fell with lead paint use in the early 20th century. Being correlated over one rising-and-falling cycle is interesting, but over two such cycles seems much more compelling.

God, you guys are a bunch of buzzkills.

You just noticed?

This talk of cycles reminds me that my first conference presentation ever was on Kondratiev waves. Fucking Russians.

I'm somewhat where Tweety and essear are--and wondering about potential confounding variable(s). Pretty interesting, however, and more than I had seen pulled together before.

But who cares what I think, I was a young child during the mid-late 50s local maxima for lead. (But really watch out for the ~40-45 year olds.)

Anyway, it's interesting to see the skepticism here. I suspect reactions to this vary considerably by academic and disciplinary background.

They didn't even learn to make chips until 1993 but they kept making economists.

WAIT A MINUTE! TEO IS CALLING US BUZZKILLS?

47.2: See 31 and following.

51: Ah, I thought that was all just some Nebraska thing.

50: Heh.

Aside from the underlying research and the question of whether or not it's convincing, I think this article is a great example of good science journalism, in contrast to the usual terrible kind that we were talking about the other day. He explains the issues at hand clearly and refers to lots of different studies instead of just hyping the newest one, among other things. Of course, this makes the article way longer than a typical science journalism piece. It's noteworthy that Drum's background is in blogging rather than journalism.

To be clear, I think the issue is important and the evidence is compelling.

44: There are also studies on the effects of lead levels in different countries that phased lead out at different times.

There was also one study that found a correlation between childhood lead levels and arrest rates 30 years later among the very same people.

52: Just before that, somebody was laughing at me because I had to do math to figure out how old I was. I was very conscious of my age at the moment.

General Motors had the "use patent" for TEL as an antiknock

I'm still convinced the anti-TEL people are in the pocket of Big Knockers.

It's really more of a cleft.

My exposure to lead as a child must have been off the charts. In addition to having leaded gasoline and lead paint everywhere, we had actual pieces of lead lying around the house. My brother was a muzzle-loading enthusiast, and used to cast bullets in the kitchen, melting lead ingots on the stove and pouring the molten metal into molds. I would play with these amusing balls of soft metal, fascinated at how I could crush them with my teeth (!) or pound them flat and use them to write with. I had zero consciousness that the stuff was hazardous. Sometimes I wonder whether I wouldn't have amounted to more if I hadn't poisoned my brain all those years.

I agree with 54.

I've always found the lead theory compelling (which says just as much about my tastes as it does about any evidence), but it's always bothered me that it doesn't seem to address the crack epidemic. Is it supposed to be a coincidence that the crack epidemic wound down at the same time or did decreasing lead exposure play a role in ending it?

61: You'd have probably just had a brain good enough to do well on the GRE but not do anything useful. So not worth it. Trust me on that one.

Is it supposed to be a coincidence that the crack epidemic wound down at the same time or did decreasing lead exposure play a role in ending it?

I don't know if any studies have addressed this, but one plausible mechanism for the latter would be that kids with less lead-addled brains were less interested in smoking crack.

Teo is too young to have enough lead in his clean, clean brain.

So we just need to do the marshmallow test but with crack.

So we just need to do the marshmallow test but with crack.

Right. Let's start with Knecht.

Yes, unleaded brains knew that crack was whack. (Wack?)

Because my wife got all fancy, we have some organic marshmallows. They spoil much faster than the corn syrup marshmallows that nature intended.

Because my wife got all fancy, we have some organic marshmallows. They spoil much faster than the corn syrup marshmallows that nature intended.

Next she'll be insisting on artisanal crack.

The crack epidemic wouldn't have been so bad if people hadn't been smoking it out of lead pipes. Like the Romans.

Sometimes I wonder whether I wouldn't have amounted to more if I hadn't poisoned my brain all those years.

Probably you'd be doing something other than commenting on Unfogged. So we're all benefiting from your lead exposure.

What are the other theories for the rise and fall of the crack epidemic?

She won't let me have any crack.

She won't let me have any crack.

So selfish.

74 to 70.

Teo is too young to have enough lead in his clean, clean brain.

In grad school I moved into a really old building that probably had lead paint and had those annoying painted-over old windows that are hard to open and that send paint flecks into the air whenever you open or close them. Maybe that explains why I feel stupider than I used to be.

Why don't we all post our blood lead levels?

I took (twice!) a class on rebuilding stained glass windows with lead cane. It was fun.

I had to constantly remind myself not to lick my fingers.

The cross-national evidence sounds pretty compelling. I was a bit worried before reading that the article would only discuss the US.

||

I pass this along for your amusement.

|>

82: The full list is impressively insane. Seamstress? Hair stylist? Drill press operator? On the other hand, they seem to be doing a bang-up job of repackaging public-domain information collected by the government, sticking it on a website, and selling ad space, which I presume is their actual goal.

"They" in 83 being CareerCast. Forbes seems to just be repackaging CareerCast's information and selling yet more ad space.

Yeah, I guess it's turtles all the way down.

Based on the CareerCast article, I should be asking for a huge raise.

I guess they confused "full" and "full-time".

"Minimum travel demands" amused me, but I guess it's consistent with heebie's experience.

Guess, guess, guess I should get some sleep.

Is an implication that there will be a crime wave in 20 years from the children of Brooklyn brownstone restorers?

Yes. Empirically testable predictions! Woo!

the explanatory mechanism for the end of the crack years is just as suggested: having been born 18 or so years after big lead stopped leadin' up the joint, the youth agreed that smoking blunts and drinking 40s at home on their mom's couch, listening to nas's "illmatic" was the dopeness, while crack was, as had been suggested, wack. I'm amused and impressed by knecht's "lead: delicious but deadly" exposure. my concern is that I might as well have eaten 10 pounds of aluminum, given all the burnt-up tin foil I inhaled while chasing the dragon. and, I guess, smoking stuff out of a macguyvered coke-can pipe every once in a way. if that stuff gives you alzheimer's I better get the shotgun while I can still--squirrel!!

watch out for the prions in the squirrel brains.

I haven't read it yet, but wouldn't the crack issue be explained if (as I've read elsewhere) the "crack baby" thing was actually a myth, and much of the epidemic was actually an epidemic of shouting about black people? Casual empiricism suggests such events are not unknown.

This research might explain why both my grandfather and I are much smarter than my dumb wingnut dad.

On the question of how convincing it is, I thought the time series showing lead levels tracking crime in the same way in different locations at different times looked impressive.

Or the crack epidemic could have been a real drug fad that happened at the same time as a lead-caused violence epidemic. Possibly an unleaded generation of crackheads wouldn't cause nearly as much trouble. I mean, case for case, meth makes you as messed up as crack, but current meth problems aren't driving high crime.

91: I am one who'd say the "crack baby epidemic" was a myth, and yet (to tie threads together) there are a lot of preemies among the babies with prenatal cocaine exposure and that has its potential negative impacts, as do inadequate prenatal care and nutrition, potentially being born into in a lead-heavy inner city area, and having an addicted mother who may not be prepared to parent appropriately. But there was (AFAIK) nothing special about crack in that respect.

My response to the article was pretty much word-for-word Witt's 11.1 and 11.2, but it's been nice to see friends from here passionately sharing the article elsewhere.

I had to constantly remind myself not to lick my fingers.

Same thing while setting old type for letterpress printing.

I didn't know old type had any interest licking my fingers.

oh, I wasn't referring to the 'crack baby' thing which has been, I thought, pretty seriously debunked, but rather the question, 'how does this explain the end of the crack era,' and I thought the drum article was persuasive on that point, namely, that was the trailing end of the kids exposed to leaded gasoline in concentrated doses. I have lived in two pretty old (by narnian standards) houses for my last 2; edwardian, let's say, and we've always drunk water from one of those big things you'd have at the office, because I was concerned the old piping was soldered together with lead. doesn't this mean community gardens in urban areas are a bad idea though?

I mean, case for case, meth makes you as messed up as crack, but current meth problems aren't driving high crime.

Aren't they? I thought there was an awful lot of meth-related crime. It may all be happening in rural areas, which don't officially have high crime rates because Real America.

There's meth related crime, but not enough to make overall crime levels high by standards of prior decades.

98.last: I think urban gardens often use raised beds with imported soil for that reason.

I thought the time series showing lead levels tracking crime in the same way in different locations at different times looked impressive.

The time series for different cities in the US looked impressive, but the ones for different countries look like a mess to me. I haven't read the full paper, but just going through it and looking at the figures, it doesn't look like the curves track each other as closely as the ones in the Mother Jones article....

I just looked at the "other countries" article. At first glance it looks like he used a best-fit lag for various different crimes to get the curves to match up? I imagine that could be justified but it seems like it would risk overfitting to me. This may definitely be a case where I don't understand how a given scientific field does things.

The voxel-based morphometry study looking at gray matter volume also bothers me a little bit. One, those studies might have problems generally, but two I don't understand what the mechanism would be, or if I come up with one it doesn't seem very related to the myelination thing, at first glance. Also, the link he gives for the myelination thing doesn't mention myelination (actually, that makes me feel better about the actual research, if not Drum's article. Problems with myelin formation tends to lead to things like motor deficits and neuropathy, not aggressiveness, I'd think) at all.

What that article does say:

A number of mechanisms may be at work. Lead interferes with synapse formation, disrupts dopamine systems, and lowers serotonin levels. Lead exposure has been shown to reduce MAO A (monoamine oxidase A) activity, and low MAO A activity has been associated with violent and criminal behaviors

So, okay, some of those could certainly lead to mood dysregulation disorders. But at a first pass it seems like the neurological mechanism is not terribly clear (which makes sense, of course. Nothing in neuroscience is terribly clear).

None of which, again, is to say that the scientific case isn't actually quite strong. It's just not totally clear to me from Drum's article whether or not that's the case, and looking at the literature (and thinking about the kinds of information you'd need to make a solid case) I could imagine somebody with a different theory making a strong case for the primacy of some other factor that would be just as believable to me as a non-expert. (Now, maybe nobody has done that, and this is basically the scientific consensus. That's not the sense I got from Drum's article, though.)

There are a bunch of mentions of myelin in this review article, though.

Ah, indeed there are.

The "replaces calcium ions in calcium channels" thing certainy sounds icky.

This is interesting:

Lead as a neurotoxin can carry a lethal legacy. Young women who live in lead‐contaminated housing or who were lead‐poisoned themselves as youngsters can pass lead on to their unborn fetuses. There is a strong correlation between maternal and umbilical cord blood lead levels, indicating the transfer of lead from mother to fetus

The New Zealand plots in the "other countries" article look particularly bad for the hypothesis: similar crime curve to other countries, but lower lead growth over time.

And yet, the correlation coefficient for New Zealand is very high-- the trouble is that modeling the relationship would give very different parameters than it would in other countries, I think. Perhaps we are to believe that New Zealanders are especially sensitive to lead compared to other countries.

In general, talking exclusively about R² seems kind of nutty to me. I'd rather see something like a model for crime rates versus lead cooked up on the basis of, say, US data, and then ask how much of the variation in other countries is explained by the model.

109.2: yup. It is all purely correlational (with those best-fit time lags!) in a way that instinctively worries me.

I mean, I imagine doing it that way would be insanely hard and/or, because of the stuff that he mentions about different reporting rates from different countries and so on. He also doesn't appear to be controlling for very much which, on the one hand, certainly explains the variability across countries, but on the other hand makes me nervous. Dude needs to get Andrew Gelman involved.

110: The best fit lags for the age-specific effects actually make me more comfortable with the approach. Pregnancy at 15 had a best fit lag of 15 yrs. Incarceration at 19 or whatever had a best fit lag of 19 yrs.

The inter-US data also seemed quite strong, and attempted to control for several factors that made sense.

||
The nice guys of OKCupid are cringe-worthy and sometimes very funny in their obliviousness.
|>

You know the Romans, with their lead water pipes, were some violent motherfuckers. Presumably there's a study to be made out there correlating plumbing vs. violence for various cities in the ancient world.

114 is an interesting thought, but it seems like it would be hard to be quantitative about it.

Archaic-style violence.

116. A sword is an armament in the plain meaning of the term (see other thread). How is PA on concealed carry and stand your ground?

I've never heard of any laws against concealed carry of a blade in a general public setting (i.e. not the airport or something). For guns, it's fairly easy to get a concealed carry permit.

I'd have to look up what PA's laws are to know if we have a stand your ground provision, but if the guy actually needed the cane, he's got a pretty good argument about why he couldn't flee from a guy 20 years younger.

He wasn't concealing it, though, it was right there in his hand.

re: 118

Really? Pretty much any weapon is illegal in the UK, but I'd be surprised if a _sword_ wasn't illegal for carry in the US. But what do I know?

According to the sword cane store (I have it bookmarked), they won't ship to California, Massachusetts, or New York. That doesn't appear to apply to just plain swords.

120. So's a gun when you're firing it. The point about open carry is that everybody can see that you're armed before you do so. I'd have thought a sword stick fell pretty neatly into the category of carrying a concealed weapon.

Then again, in the UK, a stick could quite easily count as an illegal weapon.

Massachusetts law on the subject:

(b) Whoever, except as provided by law, carries on his person, or carries on his person or under his control in a vehicle, any stiletto, dagger or a device or case which enables a knife with a locking blade to be drawn at a locked position, any ballistic knife, or any knife with a detachable blade capable of being propelled by any mechanism, dirk knife, any knife having a double-edged blade, or a switch knife, or any knife having an automatic spring release device by which the blade is released from the handle, having a blade of over one and one-half inches, or a slung shot, blowgun, blackjack, metallic knuckles or knuckles of any substance which could be put to the same use with the same or similar effect as metallic knuckles, nunchaku, zoobow, also known as klackers or kung fu sticks, or any similar weapon consisting of two sticks of wood, plastic or metal connected at one end by a length of rope, chain, wire or leather, a shuriken or any similar pointed starlike object intended to injure a person when thrown, or any armband, made with leather which has metallic spikes, points or studs or any similar device made from any other substance or a cestus or similar material weighted with metal or other substance and worn on the hand, or a manrikigusari or similar length of chain having weighted ends; or whoever, when arrested upon a warrant for an alleged crime, or when arrested while committing a breach or disturbance of the public peace, is armed with or has on his person, or has on his person or under his control in a vehicle, a billy or other dangerous weapon other than those herein mentioned and those mentioned in paragraph (a), shall be punished by imprisonment for not less than two and one-half years nor more than five years in the state prison, or for not less than six months nor more than two and one-half years in a jail or house of correction, except that, if the court finds that the defendant has not been previously convicted of a felony, he may be punished by a fine of not more than fifty dollars or by imprisonment for not more than two and one-half years in a jail or house of correction.

Massachusetts really needs an editor.

I'm confused about whether or not I can carry a regular sword in MA or not.

116: An epidemic! There was a similar sword-cane incident in Beaver County in 2009. And commented on here at the time by Moby Hick.

I'd be behind arming the police with swords. Most of the weapons used by criminals are knives, and a sword beats a knife. Plus, very difficult to stab the wrong person - much more so than shooting the wrong person.

127: single-edged, and not in most cities.

re: 129

Plus a whacking great sword -- katana, cavalry sabre, scimitar or whatever -- has a certain intimidation value.

'Move along now.'
'Fuck off'
[ssssssshhhhwwwwwwick]
'Move along now.'
'Yes, officer.'

There's meth related crime, but not enough to make overall crime levels high by standards of prior decades

Is that about industry structure? You can prepare meth from commonly available precursors, but if it's crack you want, you've got to buy cocaine from the cocaine gang. As a result, you'd expect lots of smallish meth producers rather than a few big cartels (which I think is what happened), and perhaps lower prices.

You can make a case that the cartel would tend towards stability and less fighting (oligopolistic stability) and the competitive scenario would be more, uh, competitive. You could also, however, taking an international relations/strategic studies view rather than an economics one, argue that more than a very few large actors is an unstable balance of power, like Europe in 1914 rather than in 1978.

Except that has the precursors for meth have gotten harder and harder to obtain (I can buy a sword without ID but not cold medicine), violence hasn't increased to my knowledge.

132 was my thought, but also I think a lot more US drug production and distribution has been outsourced to the outrageously violent Mexican drug cartels rather than on-the-ground US gangs, though what the hell do I know?

132: Relevant in some alternate universe.

city gets 150,000 domestic violence reports a year. -- i.e., 1 for every 10 residents.**

If it helps at all a huge number of these are just dysfuntional people being shitty to each other rather than someone being victimized.

Crack is still common but has to compete with meth for the smokable stimulant market. Mexican labs churning out a cheap quality product plus domestic restrictions on precursors really killed off domestic meth production. It's not totally gone but pretty rare. Labs used be a regular occurrence in this jurisdiction but I'm in my fifth year on the job and have yet to see one.

If it helps at all a huge number of these are just dysfunctional people being shitty to each other rather than someone being victimized.

This distinction leaves a lot of gray areas.

Only semi OT:

I've been kind of intrigued by the research showing that areas with relatively high levels of lithium in the water seem to have lower suicide rates. I wouldn't put it in the water because of the risks of renal toxicity, but I wonder whether it wouldn't be worth prescribing it in minute doses to children who are at very high risk of developing mood disorders.

I wonder if the leaded-gas thing also holds true for Mexico. It seems so absurd but I guess given delays in imposing catalytic converters it might.

Whatever the deal is with lead the crack epidemic was real and really led on its own to extreme violence, in large part because ever after Freeway Rick the main distribution method in the US was preexisting street gangs who were otherwise inclined to violence. Meth has produced waves of crime and violence where it's existed, but this tends to mostly be outside of big cities and thus somewhat more invisible, and the distribution channels are different.

I guess what bugged me about the Drum article is that there seemed to be an implication that there was a single cause of the drop in crime, lead abatement. Even if its a significant factor I have a hard time believing that the other factors weren't also significant.

139: see
http://www.irishtimes.com/newspaper/health/2012/1023/1224325572806.html
(The Irish Times have confirmed that they don't really stand over the ridiculous proposition that they can charge for links to their site.)

there seemed to be an implication that there was a single cause of the drop in crime, lead abatement.

Implication? Saying that a factor accounts for 90% of the variance isn't exactly an "implication" so much as an explicit claim.

Wow, 125 is hard to parse. I can't tell if the four-inch blade on my Leatherman is OK or not. It's more than one and a half inches, so it's bad? But it doesn't lock, so that's OK?

Well, right. That just seems implausible on its face, particularly given the weakness of the evidence for the hypothesized causal mechanism.

142,144. What? no, it's a correlation. The canonical example is the allele responsible for an epicanthial fold and chopstick use.

The cross-country datasets are (usefully, competently as far as I can tell based on a quick read) chosen to eliminate a bunch of possible alternate explanations. The lack of a lower limit on safe exposure is also strongly suggestive, all of this should be a spur to use public money, tax breaks, zoning, whatever, to clean up existing environments ASAP rather than dink around with fiurther study.

But if violence is taken as a phenotype, this study isn't enough to draw a conclusion, though it's well done and suggests a bunch of easily-performed followups (comparing lead levels in offenders and a suitably matched cohort of non-offenders).

WeSome people I know only in passing had problems with this in SCA, which requires you to carry a fairly large knife holstered in a sheath if you want to eat at any of their banquets.

re 125,
The New Jersey weapon statute:

2C:39-5. Unlawful possession of weapons

Other weapons. Any person who knowingly has in his possession any other weapon under circumstances not manifestly appropriate for such lawful uses as it may have is guilty of a crime of the fourth degree.

I was recently on a grand jury. This was used to add a charge to anyone who was frisked by police on some other charge, and found to have any kind of knife in their pocket. One guy was a misdemeanor shoplifter. Inother guy was arrested for drunk driving, and was charged on this statute for having a switchblade in his pocket while inside his own car.

Massachusetts statutes are written in unusually complicated language, IME. But a Massachusetts lawyer took offense when I mentioned this casually during a negotiation. Maybe the laws are like the roads and you have to be from here to criticize them.

146: Wait, which is the requirement: that you carry a knife, or that if you're carrying a knife it be in a sheath?

Maybe we should move the knife-fight portion of Saturday's meetup to New Hampshire.

To be acceptably period, generally had to be double edged, which I think it what made it illegal. And you had to carry it around so you could bring it to meals so ideally you would have a sheath since you wouldn't want to carry an open double edged blade around all the time.

Back to the lead topic a bit, my favorite aspect of the tetraethyl lead story is that the person who discovered their use in auto engines, Thomas Midgley, Jr., also invented chlorofluorocarbons.

When the Pater last lived in Massachusetts, somebody was running a (perfectly legitimate, though mind-boggling in its offerings' variety) knife e-commerce site out of their home a few doors down the street.

As of 2011, Pennsylvania has a wide-reaching Castle Doctrine covering assault by people who feel threatened.

In this case, a guy who does sound like he was pretty clearly defending himself killed someone with a bow and arrow and was not charged.

145: I know what it is, but the way the correlative evidence is presented in more than an implication.

155: Where, I ask, are the folk musicians? That's a ballad right there.

I'm baffled by the description of this evidence as weak.

The objection seems to be that the evidence is too strong.

Without a detailed mechanism, there is no demonstrated causation.

Violence is a complicated phenotype. With a detailed mechanism, you could hope for a way to reverse some effect of exposure for instance, or investigate the effect on lead on digestion as well as behavior. Dosing rats would be a way to get at this.

Again, for social policy, this is great work, but it's not biology yet.

159: Paint changed lead loads in the US in the early 20th. The expected lagged changes showed up. That's not reversing effect, but significant support to dose effect.

Sure there are details we don't yet understand, but you can't dismiss the statistical evidence and physiologic clues because we don't yet have a complete description of the chemical causes of criminal tendencies. Also, a detailed mechanism won't necessarily help in individual cases. Getting heavy metals out of the body is very difficult.

152: Yes, and celebrated in the archives accordingly. My favorite quote on Midgely. [Midgely] had more impact on the atmosphere than any other single organism in Earth's history.

Also from that comment:

In 1940, at the age of 51, Midgley contracted poliomyelitis, which left him severely disabled. This led him to devise an elaborate system of strings and pulleys to help others lift him from bed. This system was the eventual cause of his death when he was accidentally entangled in the ropes of this device and died of strangulation at the age of 55.

re: 147

The UK law isn't unlike that. You can carry things that are weapon-y as hell, if you have a good reason to be doing so. Maybe you are an iado instructor on the way to class, or a gamekeeper travelling home from a shoot, or someone who repairs and restores antique weapons. But if you don't have a manifestly good reason for having it, you could be done for almost anything.

151: The Massachusetts police don't recognize peace-tying? Awwww, mannnnnn.

163 How antique does it have to be? Could you get away with hauling a Lewis gun around?
"Well you see officer, I needed a new main spring collet as the old one had wore out and...what's that you say? How did the old one wore out? Well funny story that..."

had more impact on the atmosphere than any other single organism in Earth's history.

Obviously the person who wrote this never spent any time in an enclosed space with my college roommate.

Spike's college roommate was Storm from the X-Men.

Of course they had co-ed dorms. Spike and Storm went to Harrad.

Pre-internet porn had such involved plots.

Storm ate a lot of beans.

had more impact on the atmosphere than any other single organism in Earth's history

This is incredibly unfair to the bacterium that first evolved oxygenic photosynthesis.

145: I know what it is, but the way the correlative evidence is presented in more than an implication.

155: Where, I ask, are the folk musicians? That's a ballad right there.

I feel like I should be getting that joke, but it's sailing right past me. Is there an explanation on Standpipe's blog?

156 was actually to 154, which linked to a story about a man killing his wife's lover with a bow and arrow.

Not that this should stop any lurking folk musicians from composing ballads about statistical correlation.

170- Single organism, I don't think any one bacterium suddenly had the whole fully optimized pathway.

The number of organisms that have lived is a finite integer, presumably, so there had to have been some organism that first had it. Of course it was standing on the shoulders of giants previous bacteria.

174: For something that reproduces by mitosis, I think maybe you can.

175: so really we should be looking for the first bacterium with shoulders.

I'm guessing some organism had one piece of it that gave a selective advantage but didn't actually make much O2, then another organism optimized that to be more efficient, etc., on down through the stack of turtles.

I don't see why people are getting so pious about the correlation vs. causation thing. In all the social and human sciences and epidemiology correlation is pretty much the name of the game, and where there are true experiments they often lack external validity. The search for some gold standard causal evidence in these areas often just leads you down a statistical rabbit hole or to overgeneralizing from experiments that were done in a different context. I would much rather have someone just present their correlations honestly and their explanatory story for those correlations, do lots of robustness checks to test other alternative explanations tested, and see what the accumulation of circumstantial evidence seems to say. Then everything is on the table.

With that said, you would clearly have to go to the original studies behind this article in order to fully judge how good they are, and look at evidence for other causes too. And no reporter should ever present a regression r-squared as though it was the proportion of the 'true cause' someone had found.

177: Or maybe the first one that could stand.

178: The two photosystems evolved independently, at least. Maybe we should give the prize to the bacterium that first evolved the water-splitting complex.

(I mentioned before that I think this book is awesome. If I wasn't in the middle of reading it right now I wouldn't be able to fake this conversation nearly as well.)

Storm ate a lot of beans.

probably not, given her bio: Storm almost certainly kept paleo.

Storm ate a lot of beans.

probably not, given her bio: Storm almost certainly kept paleo.

Assuming Drum is accurately describing the studies, this seems like as strong evidence as you're going to get from an epidemiology study. This is "smoking causes lung cancer" level evidence.

So not enough evidence to keep LB from disputing it.

In a strictly professionally ethical manner, of course. Everyone deserves a lawyer.
(previous was me.)

Oh, the tobacco companies aren't contesting that it causes lung cancer any more, at least not in any case I worked on. The defense these days is some variant of "Come on, you all knew. Everyone knew."

Pot legalization may reduce drunk driving. Not exactly relevant, but close enough that I'm not going to do the pause, play thing.

Probably increases stoned driving, though.

In my experience stoned drivers are often safer. That is from hanging out with some real potheads though. I could always tell when she wasn't stoned because that's the only time she would speed or cut people off. When she was stoned she'd just hang out in the right hand lane at exactly the speed limit, smiling happily.

The evidence seems very convincing that lead pollution helped to cause crime waves. But the big question is how much if an effect it had -- there are lots of other plausible factors, right? The smoking/lung cancer thing doesn't seem quite right -- we've got evidence that smoking causes lung cancer, but Drum seems to be saying something like that we know for sure (or almost for sure) that a specific patient died from lung cancer because he smoked, even though we know the same guy worked as a welder* in a smoggy area, and therefore we shouldn't worry much about people working in welding or smog. It's the latter conclusion that seems more problematic.

*not sure how important this is for lung cancer, but you get the point.

In law school, they call that the "you were so gullible you believed our BS" defense.

"You fucked up; you trusted us."

192: The other factors don't explain much of the variation in the US data, while the lagged lead levels do. An example is poverty. It's certainly plausible that poverty causes crime, right? But during the current recession poverty levels have risen, and crime levels have dropped.

192: It doesn't say anything about any specific criminal/smoker, but, barring some confounding variable, it does say lead is a lot more important than those other factors.

Late, but this caught my eye:

except that, if the court finds that the defendant has not been previously convicted of a felony, he may be punished by a fine of not more than fifty dollars or by imprisonment for not more than two and one-half years in a jail or house of correction.

So 50 dollars or TWO AND ONE-HALF YEARS in the house of correction!? Heck of a contrast there.

(Also, "house of correction"; how antiquated. I remember in a history course on the Crime & Punishment in Britain, the British prof was very amused when I noted that Mass still called jails "houses of correction". )

I think that quote about Midgley comes from this book, which I remember being pretty good.

I think that quote about Midgley comes from this book, which I remember being pretty good.

But maybe not double-posting good.

This book (by one of my grad school professors) uses the removal of lead from gasoline as one of its case studies in environmental policy. It's a very interesting story.

Saying that a factor accounts for 90% of the variance isn't exactly an "implication" so much as an explicit claim.

For the sake of completeness I'll note that Drum elaborated on that 90% number yesterday:

It's true that one researcher has suggested that lead can explain 90 percent of the rise and fall of crime, but that's very much the high end of the estimates in the field. I'm a lot more comfortable with an estimate of around 50 percent, something I should have made clearer in the text of my piece. In other words, lead probably explains a very big chunk of the rise and fall of postwar crime in America, but it doesn't trump everything else. Drugs, poverty, urban gang warfare, education, policing tactics, and other things also play a role.

...

POSTSCRIPT: Another thing to keep in mind: Even if the 90 percent number is correct, it doesn't imply that lead is responsible for 90 percent of all crime. It only implies that it's responsible for 90 percent of the postwar rise of crime above its natural level, which is determined by a variety of other factors. Later, the drop in lead emissions was responsible for 90 percent of the decline of crime back to its natural level.

The correlations are really excellent -- per state, per country, per city, per neighborhood, the lead levels correlate with crime 23 years later almost perfectly.

Sure it's not the cause of all violent crime. It's clearly the *dominating* cause, ovrewhelming all other causes.

Somebody has done a much better job than I did of discussing what's so tough about making the causal claim and the places where the article maybe overreaches.

Which is to say I don't care if the thread's dead! Everybody read 204! Everybody!

READ IT

I will, I will!

Hooray!

Reading! Not impressed so far.

Reading is overrated, but not so much as math is.

Give Eggplant time, Moby. He might develop a taste for it.

Would it be correct to sum it up as follows: the historical data looks good, but it's fundamentally unreliable because there's no way to be assured that the specific individuals with high lead levels were characterized by a high crime rate. And while there's one study that tracks high lead individuals, and it does show a correlation, it's too small to rule out the possibility that lead had no effect.

Fair? Because, okay, but doesn't seem devastating to me.

It's not devastating. As I said from the start, I would in no way claim that there's any evidence that Drum isn't correct that lead is a primary cause of increases in crime. On the other hand, the evidence that he is correct is not dispositive.

212: That's what I got out of it. "Who cares about cross-sectional data. What you need is a cohort study. Oh, you mention one cohort study? Let me ignore the rest of the literature and focus on one part of your cohort study that only has mild statistical significance and skip past the parts that do."

Maybe reading isn't your thing after all, Eggplant.

This type of design is called a cross-sectional study, and it's been documented that values for a population do not always match those of individuals when looking at cross-sectional data. This is the ecological fallacy, and it's a serious limitation in these types of studies.
But we know those convicted of crimes tend to have high lead levels.

I actually have an cohort study.

On the cross-sectional data, there's an issue that I find really convincing but I'm not sure of the vocabulary for quantifying it: the shape of the correlated trend. Most of the 'correlation is not causation' jokes rely on monotonically increasing trends: all sorts of things are growing over time, so you can pick any two and talk about how strongly correlated they are, but how funny it is to think of one as causing the other. Here, we've got a hump, which is a less trivial shape to find. It's not as good as it would be if we were matching up several peaks and valleys, but it seems much better (in terms of arguing that either one causes the other or there's a non-trivial shared cause) than a steady increase or decrease.

Is there a term for the concept I'm talking about, or does it not make sense for some reason that's easy to explain?

216: Those whose crime is stealing lead will skew the average.

Or those who have been shot.

Here, we've got a hump, which is a less trivial shape to find.

Why?

Just as an anecdote, I've only ever seen the billboards reminding people from the UC childhood lead exposure study to come in for an update (and get cash, though that may just have been implied) in poorer, higher-crime neighborhoods than the median.

215: Care to be more specific?
219: I assume the counter hypothesis is that low-lead criminals find it easier to take advantage of their leaded victims.

223.1: I didn't find your summary to be an accurate characterization of what the blog post did or was trying to do.

there's a non-trivial shared cause

This one is of course a compelling (and far from trivially different!) alternate hypothesis.

221: Takes two variables instead of one to describe.

I just read Jetpack's link, and I think it seems fair enough. Still -- and I don't think this contradicts Jetpack's link -- the Drum piece is about as good as it gets for science journalism, and the findings are, given that they're rooted in historical correlations, about as persuasive as they can be.

Which is to say, if you're persuadable by that kind of data, and I mostly am (though I wouldn't be surprised if the conclusions were wrong), you'll be persuaded. But if you're not, you won't be.

No. It doesn't. Never mind.

Because it's got more corners on it? Any monotonic upward line correlates equally well with any other monotonic upward line. But a hump has a turning point that occurs on a specific date -- you can identify any timelag between the curves, and see if it's causally plausible (which it seems to be). A multihumped curve that matched up with high and low points separated by the right time interval would be better still, but even one hump is better than a line.

225: yup.

But even without that a priori any time series is as likely as any other, right? You can find a good match for essentially any time series here, although that's obviously extreme.

There were, in fact, two humps.
224: What would your summary look like? Is there correlational evidence that you, or this author, would accept?

And I can't remember -- does the curve show crime flattish at first and then with a corner where it starts to rise? Because if it does, that's another point that can be used to match a change in the postulated cause with a change in the effect at the same time interval.

227.1: it is extremely good as activist science journalism, which is a bit different from pure science journalism and more different yet from science, would be my take.

225: Although a shared cause doesn't leap to mind. It's not anything about cars or driving generally, because it increases with car usage until lead is removed from gas, and then diverges. For something monotonically increasing, the shared cause is usually going to be a function of population. For lead and crime, though, nothing that causes both high lead and high crime is obvious to me. Doesn't mean it doesn't exist, but it's not easy.

218: I think the term is cross correlation of the time series.

Any monotonic upward line correlates equally well with any other monotonic upward line.

What?

I agree that it's an excellent piece of journalism, but I do find the linked piece by Tweety "devastating" -- not in the sense that it proves Drum wrong, but in that the point of Drum's article is that we now *know* what the cause if decreased crime was (lead) and wasn't (eg changes in policing or incarceration rate). That's the bottom line that will be taken away by most readers, and it seems like way to big of a conclusion to reach from the data that Drun had.

236: Straight line. If you've got Thing A= M(time) + B, and Thing B = N(time) + C, where M, N, B and C are constants, aren't they perfectly correlated regardless of the value of M and N? (I'm arguing about statistics in baby-talk again, because it's all I know. If I'm deeply confused, correct me.)

"Thing B" and "B" weren't meant to both use the same letter.

Drum addresses the link.

I do not think the imaging studies make as much of a difference as he does, but anyhow.

238: no, unless M=N.

238: If you really had all your points fitting perfectly on lines like that, they would be perfectly predictable, and any point would tell you everything you needed to know about any other, but that would be equally true if all the points were lying perfectly on curves with lots of humps. The strength of a correlation lies in the size of the scatter of points around the line or curve you attempt to draw to predict them.

241: But M and N are in different units, so I don't understand how they could meaningfully be said to be equal or not. I mean, M is something like micrograms of lead in soil, and N is arrests per capita. I'm not saying you're wrong, because god knows I never knew much stats and I know less now, but how does that work?

If you want to dismiss these correlational studies, the ecological fallacy is not the way to go (for starters, it's not true in this case: we can measure lead levels in the criminal population and see that they are high). Multiple comparisons is a bit more promising, but this has explanatory power well beyond the initial studies (for instance, the convergence of crime rates) and large amounts of evidence in cohort studies.

Oh, I see what you're saying. Sure, I guess.

242 to 243. The variance about the line is important.

I guess that is to say that I disagree with Tweety. If you have a bunch of points, and they lie on two straight lines, then yes, the two lines are perfectly cross correlated, but I guess I should go look at the formulas to make sure.

Ugh, I wanted to be participating in this discussion, but I should be packing.

If your data were actually two different lines, you wouldn't mess with a correlation.

241 is incorrect. 242 is true, but neglects the lack of independence in the time series, and doesn't address LB's point that most sociological variables (I'm guessing here) are monotonic over the time period.

250 before 245 and 247, obviously.

but I should be packing.

Yeah, the way things are going, you'll need a gun if you want to win this argument.

233: fair enough, though you're making me think I probably don't read "pure science journalism". Regardless, I haven't read most of this thread, but your link reminded me of a loud discussion I had with a group of colleagues, none of whom are historians, the other day.

247 written before 245, 249 is definitely true, and to 250, I wasn't trying to make a point about the underlying issue, just about what correlation is.

249 update: That is, if your line isn't a regression line with an error term, you'd just have deterministic functions.

and doesn't address LB's point that most sociological variables (I'm guessing here) are monotonic over the time period

Well, crime isn't.

254: Fair enough. I was trying to get at why LB's intuition about humps was good.

257: I was trying to get at why LB's intuition about humps was good.
Her humps her humps her humps.
Her laggy leaded humps.

No time to read the thread, but LB's intuition about humps seems only good if you (a) find that the same time lag always fits or (b) have good a priori reasons to think the time lag is meaningful. In this case the time lag is roughly the time for babies to grow to typical crime-committing age, so that does support (b). Also, (a) has some support, but e.g. in the comparisons between countries it didn't seem to always be precisely true. Also, different time lags were found for different types of crimes, but maybe that could be (b) if you had independent evidence that different crimes are typically committed by people at different ages, for instance. Maybe this is in some of the papers; I didn't look closely enough to know.

Fetal lead exposure leads to increased aggression in male mice.

The biggest weakness with the cited epidemiological lead study IMO is the variability of the correlation between countries; very weak in NZ. I think that this is pretty clearly a dispute without much consequence, lead is awful.

I learned from Wikipedia's TEL page that until recently NASCAR still ran with leaded gas, and I was happy to see that it's no longer an additive in the Americas.

259: The different types of crimes didn't vary that much. I didn't see intervals for that, either.

Further to that, at least one found that things like pregnancy at 15 had a best-fit lag of 15 years, incarceration (or something) at X had a best-fit lag of X. It was really quite impressive.

242, 246: I got you -- while the best fit curve for two variables may be a line in both cases, the correlation between them depends on the variance of each around the line.

But holding variance constant, doesn't increased weirdness (as measured in number of humps and valleys) of the shape of the best-fit curve, if it matches for two variables, tend to persuade you that there's some real causal connection? The word for why it's persuasive probably isn't correlation, but there seems to be something worth paying attention to there.

I'm also a little bit concerned about the look-elsewhere effect / trials factor / multiple comparisons / whatever we're calling it today and in this context: how many different explanations for crime did people play this kind of game with? Do I know that I'm seeing all the cities and countries for which data was available and not just the ones that match the story? And so on.

There were two definitions of crime in the figures, one fits much better than the other. The lag of best fit is a free variable, reassuring that it comes out reasonably, but for instance if maternal age varies between countries and is available from census, that's not used. This is probably the best possible epidemiological data, very helpful for hypothesis generation, but with controlled experiments and more detailed knowledge it is possible to do much, much better.

Crime reports are a crappy and hugely variable phenotype for thinking about any of this.

I haven't read this stuff and I've not expressed an opinion about the connection.

265 to 262.

261. Disagree, the broad definition of crime showed a much weaker correlation.

260 The biggest weakness with the cited epidemiological lead study IMO is the variability of the correlation between countries

And not just in the correlation coefficient, but in the size of the effect even in the cases where it is highly correlated. E.g., you could in principle have perfect correlation in two different countries but in one such-and-such an amount of lead is leading to 100 murders per million people and in another to 1000; it's not clear to me how you should interpret that, because it seems like the same thing is not happening in the two cases even if the correlation is very strong in each case.

But I think I already said 268 somewhere upthread and now I need to figure out where I put my notebook.

265: I don't mean specifically with lead, but generally -- if you're trying to match up two variables that are both increasing over time with some variance, that's a correlation, but it's not all that exciting in the absence of a good story. If, on the other hand, you were trying to match up two variables that were both increasing over time for six years, then decreasing over time for two, then increasing for nine, then decreasing for five, that seems like much better evidence for a causal connection (either directly or through a non-trivial shared cause) to me.

On the lead thing specifically, it looks not absolutely certain, because nothing is, but certainly good enough to base policy on considering how low a standard that is generally.

262 doesn't increased weirdness (as measured in number of humps and valleys) of the shape of the best-fit curve, if it matches for two variables, tend to persuade you that there's some real causal connection?

Certainly.

263 is a concern. The criminology community doesn't seem to have played the multiple comparisons game very successfully, but still.

look-elsewhere effect / trials factor / multiple comparisons / whatever we're calling it today

Is this what I've seen called 'data mining'?

270: That sounds like you are just describing a stronger correlation. You'd plot X against Y and see a stronger correlation if they co-varied as you describe in the second hand. It is more complicated when you have two time series with various degrees of autocorrelation, but you can sort of think of it as plotting change in X against change in Y.

I don't think you can evaluate any of this stuff atheoretically.

273: Since I started, "data mining" as taken on a different connotation. I now see it mostly used to describe deliberately maximizing the variance explained (for example by regression trees) and not testing one hypothesis after another until one works and then just reporting the "success."

That sounds like you are just describing a stronger correlation.

She's describing a different kind of correlation, not just a stronger one. In other words, it's not just plotting X against Y and looking at a correlation coefficient for the scatter plot, which would lose the time information of which points happened when. Correlation of the time series is more interesting than just a scatter plot of the points. (I don't know the right jargon for this.)

If the size of the effect demonstrated in the correlation studies is giving people pause, perhaps someone could use the arrest rates from the cohort studies and the historical blood levels and see if the magnitudes correspond.

267: True, the expansive definition of "crime" had the worst performance. Also the worst category to consider.

But the range of best fit lags seems to be pretty relative to the distribution of offenders and other factors.

I'd also add back that the congruence between the best fit lags of different different social mal-behaviors and the observations (teen preggers at 15 lag by 15, unwed preggers 15-17 lag by 17, etc) also set my mind at ease.

276: I think it is called random walks, but I've never done a correlation like that. I'd just do a mixed model.

274: Is it a stronger correlation? I think LB is describing a correlation that wouldn't be picked up by a simple time trend. Two variables that went up an identical amount each year forever would show a perfect correlation or close to it but it would disappear or become indeterminate if you added a simple time trend.

Nope. Random walks have nothing to do with correlation. I'm over thinking. You'd just need the coefficient from a bivariate mixed model.

Covariance, summarizable as autocorrelation, is not a single number, but a function (of another two functions rather than of a scalar). Talking about these methods for this problem are like taking an electron microscope to measure how warped is this 2x4.

Actually, that would be more for when you had relatively few time points (which is what I'm used to working with) and not a time series.

283 before 282. I mostly deal with 2x4s.

267, 278: Is that weaker correlation explained by different lags for various types of crime (which I had been assuming was true and backed by independent research, but I of course don't know)?

Except today I've been avoiding it.

Or maybe I mostly work with electron microscopes. I'm confused by the metaphor.

Time series analysis is its own whole thing. The fundamental problem is autocorrelation. This is often handled by various means of detrending or differencing. Inclusion of a time trend is a form of detrending. There are lots of things that could appear correlated because each is on an upward trend.

Why are we talking about autocorrelation when we're wanting to correlate two different time series?

The only difficulty I see in thinking about the correlation of time series is how to normalize the data; if it's clearly fluctuating in a relatively Gaussian way around a mean value, things are simple, but for a case like crime statistics it's less clear that's the right basis for starting from, so I'm not sure if the usual definition of covariance is precisely what you want.

I would consider maybe taking the covariance of the time derivative of the log of the two functions, or something along those lines.

In other words: at time t, to what extent does a given fractional change in X(t) produce a proportional fractional change in Y(t)?

Anyway, off to swim the airport.

The conversation is now over my head, but 276 sounds like a more sophisticated way of saying what I was struggling towards.

Where's the Cosma signal. He's needed on two threads.

Why are we talking about autocorrelation when we're wanting to correlate two different time series?

Because autocorrelation screws up the fundamental assumptions of multiple regression -- error terms are not independent between observations. There's a time pattern stuck in the errors determining the independent variable. This ends up shrinking the standard errors and making things look more significant than they are.

I think that from an intuitive standpoint what is going on is that the regression thinks that each time period observation is a brand new draw so the stats are based on having a big sample, but in reality the observations are closely linked so just one or two coincidental trends would produce a high correlation. From this standpoint LB's hypothesizing about 'number of bumps' is very much on point.

Where's the Cosma signal.

Quick, someone ask him to apply for a grant!

293: No, it's not over your head, except to the extent that people are burying you and your perfectly simple and important point under a mountain of pointless jargon.

Oh, I didn't say I was wrong, just that people were using words I don't know.

I'm never going to put the work into getting sophisticated with stats unless something happens that makes it professionally necessary, but if I have an educational regret (and I have lots) it's that I didn't do more along those lines when I was in college.

if I have an educational regret (and I have lots) it's that I didn't do more along those lines when I was in college.

I took an upper division probability/stats course in college. It was really interesting, and I was glad I did, but I've forgotten most of it (For example, I remember that there are distributions other than the normal curve, but can't remember any of them or how they behave).

It did broaden my knowledge base and intuitions in useful ways but, like most things, if you don't use it regularly you will lose all of the details.

The problem with having a little stats knowledge is it tends to lead one to ask "which thing I know how to do can I do with the data?" instead of "what do I really want to know and how do I figure it out?"

Comparing one time series with a lagged other time series is cross-correlation, like Tia said upthread. If you're considering the two series jointly as a 2-dimensional vector compared to lags, then its vector autocorrelation, so in some sense the information contained in cross-correlation is contained within the information contained in the vector autocorrelation.

(Whoops -- I should have said dependent variable, not independent variable, in 295).

I do think that the connection to LB's basic insight can be made pretty intuitively, and in a way that connects you to the huge body of work on time series regression, which does get complicated and spends all its time on this stuff.

Intuitively, the autocorrelation problem is just this -- regressions assume each observation gives you brand new data. Think of taking a random sample of 100 people and asking them their gender and their income. Then correlate the two with regression. You could draw a conclusion with a pretty high degree of confidence about the relationship between gender and income, because you would have 100 separate independent observations -- a fair amount of data.

Now think of pulling two different variables for the last 100 years, like temperature and carbon emissions for example. You could run the regression and get the correlation, but if it is highly significant you may think that the result is 'just coincidence'. That's because you really don't have 100 different separate observations; the observations across years are clearly connected to each other even if you don't understand the underlying process. If the two time trends track each other you could just be seeing one or two coincidentally similar trends. If both trend upward smoothly, you could just be seeing one coincidence -- that both have an upward trend. With truly independent observations you really need quite a lot of coincidences (or a giant outlier, easy to spot) to get a strong result. With time series observations, not so much.

Mathematically, this is reflected in the autocorrelation problem. The independent error assumption is the assumption that each observation represents a new 'coincidence'. (An entirely new random draw). This assumption does not hold in autocorrelated time series.

People try to handle this problem by getting rid of the autocorrelation through some kind of detrending procedure. Intuitively, these procedures try to get rid of the 'trend' over time and just look at the 'wiggles' from period to period once the trend is gone. If the period-to-period 'wiggles' on two time series match up perfectly even ignoring the most obvious trends you may be more confident that there is a real causal relation.

The simplest detrending procedures may just get rid of one trend, implicitly assuming that the observations are independent once you've taken one broad trend away (e.g. both variables always going up through the entire time period). More complicated procedures can seek out and take away multiple trends, trends which can produce multiple 'bumps' in the time series or at least multiple changes in slope. It seems like LB is basically saying that something the more complicated a detrending procedure you would need to get rid of a time series correlation, the more convincing she finds that correlation. If a very simple detrending procedure would get rid of the correlation, she doesn't find it convincing. I think most people who work with time series data would agree with that.

The complexity comes in because in many cases we really don't know the underlying process creating these trends. Who knows how many 'bumps' a process could or should have for reasons besides the independent variable? So you end up with these very complicated debates about what kinds of mathematical transformations are appropriate, which I will confess to not understanding. Really, you need a natural experiment of some sort (or something similar) to be really convincing.

You could just put in time as a variable and do a multivariate regression. You don't want to overfit.

303: that was my point in 280. 'Adding a simple time trend' can be done by just adding year as a variable. That is one of the simple means of detrending. But there are all these canned detrending methods that time series analysts use, based on particular assumptions about the correlation structure.

302 is great, thank you.

I sent the Kevin Drum article to my dad, a quasi-libertarian who is also an industry expert on gasoline engineering.

This is what I got back:

I've heard this one before.

Too bad that so much time is wasted on bad statistics (it is not science). As you know you can correlate almost anything to anything. All you need to do is ignore reasonable mechanisms for why it should be so and any data that doesn't fit. It helps to be a little ignorant of what you are talking about as well. Aging demographics, the drugs& gangs epidemics, and improved policing are better explanations for the downturn in my opinion. In fact some big city mayors (Gulliani & Bloomburg) deserve some credit as well. Just look at Oakland: crime up with Dellums, down with Brown, and back up with Quon. The mayor's competence correlates well with the crime rate!

Another place to look is Latin America. Things are very bad and getting worse there. Guess what: lower age population, corrupt politicians, rampant drugs and gangs, but no tetraethyl lead. [JM: Drum says that leaded gasoline continued in Latin America until the mid-1990s.] People love to look elsewhere rather than in the mirror -- particularly if they can blame big oil and auto for all their problems. Did auto mechanics, service station attenders, parking lot cashiers, toll booth operators commit more crimes -- maybe?

Not that lead should be in gasoline -- we now know that it shouldn't. It is a poison and it does accumulate and can have unhealthy consequences akin to mercury poisoning. Besides who needs a 1955 Cadillac Seville with a 450 hp V8 requiring 98 octane at 6 mpg?

In the 1970s most gasoline sold in the US was low lead (1 gm) versus high lead (2.5 gms) of the 1950s. Any inflection point on the correlation curves? I know it is cumulative so it could wash out, BUT still? Oil companies would like to make gasoline with out lead or ethanol. We have built refineries to do so with the needed octane supplied by alkylate and reformate (hydrocarbons derived from crude oil). This was a considerable expense but a sunk cost. The operating cost for producing higher octane with these facilities would still make lead unattractive except for small "fly-by-night" operations.

Ethanol is a whole other story. [JM: Long discursion on the chemistry of ethanol omitted] Today cars, pipelines, tanks, service stations are much tighter so vapors are not allowed to the environment and water is not allowed to get in as much. (at least from the terminal on to the service station - pipelines and tanks can still be a problem). That was NOT the case back in the 1950s. Ethanol would have been a disaster even if the suffurgettes [ JM: temperance advocates??!?] could have condoned it (not likely in the 1930s!)

Some thought from an former retired expert on gasoline. I'm now down from my soap box.

I replied blandly that the correlation gave me reason to hope that the downturn in crime wasn't an anomaly, that I was more intrigued by the data than he semed to be--although I granted his qualms about direct and immediate causation, and of course I pointed out that business from Drum's website about Latin America's only phasing out leaded gasoline in the mid 1990s.

I'm not sure he really looked at Drum's article or the science behind Drum's article, which makes me sad because I sure can't argue the chemistry.

I'm not sure he really looked at Drum's article or the science behind Drum's article

Indeed, the mention of Giuliani suggests strongly that he didn't.

The link in 204 is interesting, and definitely correct that a regulatory agency would need way more evidence than this to do something. Overall, though, I basically agree with this from VW:

if you're persuadable by that kind of data, and I mostly am (though I wouldn't be surprised if the conclusions were wrong), you'll be persuaded. But if you're not, you won't be.

This is more or less what I was getting at back in 48. For people from scientific disciplines, especially those with a strong experimental component, this sort of evidence isn't going to be remotely convincing. For people from quantitative social science disciplines, it's interesting but the statistical weaknesses are pretty apparent. For people from the non-quantitative social sciences and the humanities, it's pretty convincing but the weaknesses of this sort of quantitative approach are apparent. For people mostly familiar with the policy-oriented applied social science literature, this research, despite its flaws, is so much better than the incredibly shoddy and useless work that is all too common in those fields that it's astounding.

Andrew Gelman is not impressed with the link I posted.

Also the explication in 308.last is really interesting and I wish I'd asked for it sooner.

Right, 308 is a really good way of saying what I was thinking and sort of waving at from the end of 270.

Is reading Gelman's thing?

311: huh, I would never have gotten that from 270. I actually haven't been thinking about policy that much in any of my comments in this thread; lead remediation seems like an obvious good well past a Rachel Carson-ish standard, which the many Rachel Carson murals in my town have convinced me to adopt as my own standard. If anything, I would worry that Drum's article would take give ammunition to people who wanted to starve other worthwhile public policy (remediating the effects of poverty, for instance) efforts -- they could just claim (erroneously even per the terms of Drum's article, but anyhow) that lead explains 90% of the rise in crime so we don't need to do much of anything but gradual lead abatement.

312: in fact I think he may have misunderstood the point of that blog post in a similar way to the way you misunderstood it, but wasn't it nice of me to find and post a link that agrees with you and makes me look wrong?

It was nice of you. In what way did he (and I) misunderstand the post?
If anything, I would worry that Drum's article would take give ammunition to people who wanted to starve other worthwhile public policy (remediating the effects of poverty, for instance) efforts -- they could just claim (erroneously even per the terms of Drum's article, but anyhow) that lead explains 90% of the rise in crime so we don't need to do much of anything but gradual lead abatement.
I'm doubtful there are many such people. Conservatives and many centrists can hardly support programs aimed at the poor any less. Supporting lead abatement would be a giant step up.

Well, I don't think the post was meant as a quarrel with the literature, necessarily. I think he wanted to talk about the burden of proof that accrues in epidemiology, and he thought Drum's remediation calculations were handwavey and that some of his conclusions (the blindingly obvious bit) were oversold. In that context I thought showing that not all the evidence that could be there was there was a reasonable technique, even if from a statistical perspective it's not like he cast serious doubt on the broad conclusion. I might be reading the post too charitably, though, who knows.

I might be reading the post too charitably

That's what I think. Though I do agree that the remediation benefits calculations were handwavey.

For people mostly familiar with the policy-oriented applied social science literature, this research, despite its flaws, is so much better than the incredibly shoddy and useless work that is all too common in those fields that it's astounding.

I don't know about that -- in many areas the quantiative applied social science policy stuff is excellent. Look at evaluation organizations like the Manpower Demonstration Research Corporations or the MIT poverty lab. Policy experiments have included some of the most striking and remarkable efforts in applied social science ever -- look at e.g. the guaranteed income and health insurance experiments in the 1970s. Of course, it shares the shortcomings of all social sciences.

As I said somewhere far above, this article is fine but you can't fully evaluate it without going through the original articles carefully. For reasons that may be clear from the discussion above, I distrust time series evidence.

306: Agree or disagree with him, Jackmormon's Dad is certainly extremely articulate and smart, and seemingly able to have political disagreements without being obnoxious about it (rare!).

I think Drum agreed about the remedial calculations (which, to be honest, I didn't actually read through). I'd still like someone to do the calculation I mentioned in 277. Estimate the historical variation in blood levels, multiply by the criminal likelyhood factor from that cohort study, and then compare to the historical variation in arrest records. If they are of similar magnitudes that's strong evidence that lead is the culprit.

And to be clear on 317.1, I do disagree with Teo -- this work does not seem to me to be better than the state of the art in good policy oriented applied social science (I say 'good' because I'm not talking about some random crappy article somewhere, or some halfassed consultant's report). It's good stuff but not necessarily the best. The 'gold standard' in applied social policy work is large-scale randomized experiments, and this is not that and of course never can be because we're not going to poison people randomly. The real problem with randomized social experiments is external validity -- e.g. the results of the 1970s income maintenance experiments are totally fascinating and those experiments are spectacular achievements of social science in their way, but there is absolutely no reason to think the behavioral responses to those experiments would generalize to A) today, B) an broad-based and permanent income guarantee.

308.last is very well said.

To 306, 260.1 for some evidence in mice, there is a lot of animal model work.

So I guess maybe a $200 billion nationwide lead remediation effort would have been a better use of that money.

233: it is extremely good as activist science journalism, which is a bit different from pure science journalism and more different yet from science, would be my take

Sure, and I think it is interesting enough to be a starting point for some more focused actual science (or a continuation of the stuff he reports on).

313: lead remediation seems like an obvious good well past a Rachel Carson-ish standard

This. Despite the protestation of the fucking fucktards*, we're an extremely wealthy nation and the relative pittance to really knock the lead down is just that, a pittance--for a known mammoth increase in social and individual well-being, and a possible even mammother increase in same.

*Who almost to a man** are quite wealthy by any reasonable worldwide standard.

**In my dreams I'd love to cram a cubic fuckton of lead-impregnated dirt down George Will*** etc's throats.

***I'll see you in hell, motherfucker. I'll see you in hell.

Sure, and I think it is interesting enough to be a starting point for some more focused actual science (or a continuation of the stuff he reports on).

Oh yes!

317, 319: I do agree that there are exceptions to my generalization, and that economic policy is one of the biggest, due largely to its strong ties with academic economics. (Public health is probably another exception, with its similar ties to epidemiology.) I know nothing about criminology, but somehow I doubt it's in that category.

I think the best way to think of this research is that it's using epidemiological methods to address a question from criminology. As epidemiology it may well be suggestive but unconvincing, but as criminology it may be astoundingly convincing. At least to someone who's willing to accept the usefulness of these methods in criminology at all, which from Drum's telling most criminologists aren't. Note that Drum himself is interpreting it in the context of criminology; he starts the article by reading a whole bunch of theories proposed by criminologists to explain the crime surge, none of which he finds remotely convincing. Measured against that baseline the lead studies seem amazingly good.