"It is time that more of us spoke out."
May 9, 2015 2:13 PM   Subscribe

What Caused the Crime Decline? [Brennan Center for Justice]
"What Caused the Crime Decline? examines one of the nation’s least understood recent phenomena – the dramatic decline in crime nationwide over the past two decades – and analyzes various theories for why it occurred, by reviewing more than 40 years of data from all 50 states and the 50 largest cities. It concludes that over-harsh criminal justice policies, particularly increased incarceration, which rose even more dramatically over the same period, were not the main drivers of the crime decline. In fact, the report finds that increased incarceration has been declining in its effectiveness as a crime control tactic for more than 30 years. Its effect on crime rates since 1990 has been limited, and has been non-existent since 2000."
A report by Oliver Roeder, Lauren-Brooke Eisen, and Julia Bowling, with a foreword by Joseph E. Stiglitz and an executive summary by Inimai Chettiar Brennan Center for Justice, NYU Law School, 134 pp. [Foreword] [PDF] [Scribd]

Related:

Mass Incarceration: The Silence of the Judges, a critical essay by Jed S. Rakoff. [New York Review of Books]
posted by Fizz (92 comments total) 44 users marked this as a favorite
 
The top line summary and the state-by-state summaries on the right hand side of the linked page are all pretty much the same in that they mostly say what didn't cause it (world-beating rates of incarceration), and that a lot of other things did.

I wonder if they give any house room to the much-publicized Freakonomics correlations between legal abortion and a subsequent generational decline in crime rates.
posted by George_Spiggott at 2:24 PM on May 9, 2015


Video games and HBO.
posted by yeolcoatl at 2:25 PM on May 9, 2015 [6 favorites]


Looks like they dismiss the "lead in gasoline" theory, in part because they don't have state-by-state data, without addressing the lead that used to be found in paint.
posted by Halloween Jack at 2:26 PM on May 9, 2015 [16 favorites]


From the NYRB essay:
The basic facts are not in dispute. More than 2.2 million people are currently incarcerated in US jails and prisons, a 500 percent increase over the past forty years. Although the United States accounts for about 5 percent of the world’s population, it houses nearly 25 percent of the world’s prison population. The per capita incarceration rate in the US is about one and a half times that of second-place Rwanda and third-place Russia, and more than six times the rate of neighboring Canada. Another 4.75 million Americans are subject to the state supervision imposed by probation or parole.
!
posted by Fizz at 2:30 PM on May 9, 2015 [2 favorites]


In the NYRB article Rakoff argues that the Brennan study -and really, all of the studies on this topic- are woefully incomplete. It's a pretty important footnote.
posted by Going To Maine at 2:30 PM on May 9, 2015 [1 favorite]


I wonder if they give any house room to the much-publicized Freakonomics correlations between legal abortion and a subsequent generational decline in crime rates.

Legalized Abortion & Crime:
Te authors do not draw a conclusion on this theory because they could not secure complete state-level data on this variable for all the years examined. Based on past research, it is possible that legalized abortion could have affected the crime decline in the 1990s. However, even if there was any such effect, it likely waned in the 2000s. Te first cohort that would have been theoretically affected by abortion, 10 years after the 1990s, would be well beyond the most common crime committing age in the 2000


They seem to say similar stuff for most of the explanations. Is there any summary where they get down to the most likely cause?

(I still say lead.)
posted by Drinky Die at 2:35 PM on May 9, 2015 [5 favorites]


My current fave explanation is banning lead from petrol helped would-be criminals to make better decisions.
posted by devious truculent and unreliable at 2:36 PM on May 9, 2015 [6 favorites]


Decline? To hear my local news tell it, we're a liquor store holdup away from martial law being declared.
posted by Thorzdad at 2:37 PM on May 9, 2015 [29 favorites]


we're a liquor store holdup away from martial law being declared

Hey, wait your turn. Texas has dibs on first.
posted by yoink at 3:03 PM on May 9, 2015 [12 favorites]


Looks like they dismiss the "lead in gasoline" theory, in part because they don't have state-by-state data, without addressing the lead that used to be found in paint.

Via: [Mother Jones] : America's Real Criminal Element: Lead
THERE ARE, IT TURNS OUT, plenty of theories. When I started research for this story, I worked my way through a pair of thick criminology tomes. One chapter regaled me with the "exciting possibility" that it's mostly a matter of economics: Crime goes down when the economy is booming and goes up when it's in a slump. Unfortunately, the theory doesn't seem to hold water—for example, crime rates have continued to drop recently despite our prolonged downturn.
...
But there's a problem common to all of these theories: It's hard to tease out actual proof. Maybe the end of the crack epidemic contributed to a decline in inner-city crime, but then again, maybe it was really the effect of increased incarceration, more cops on the beat, broken-windows policing, and a rise in abortion rates 20 years earlier. After all, they all happened at the same time.
posted by Fizz at 3:06 PM on May 9, 2015 [3 favorites]


It didn't get a lot of publicity (since "legalized abortion reduces crime!" makes for better headlines than "reduction in crime rates have complicated cause!") but the major proponents of the legal abortion theory muffed their data. Once their calculation errors are fixed their supposed causation disappears pretty much entirely.
posted by Justinian at 3:15 PM on May 9, 2015 [3 favorites]


The next trick will be figuring out how to convince most people that crime has actually declined.
posted by octothorpe at 3:16 PM on May 9, 2015 [27 favorites]


Maybe a large number of the potentially criminal just got into "white collar crime" which is much more likely to not only go unpunished but never reported as "crime".
posted by oneswellfoop at 3:19 PM on May 9, 2015 [14 favorites]


A few semesters back I took a criminal justice class. One of the factors for the drop in crime mentioned was the legalization of abortion. The idea was that the drop started right around 16 years after abortion was legalized, the time a whole lot of unwanted children would be coming of age.

There were other factors discussed though; new surveillance and policing technologies, changes in drug culture, and also the displacement of the violence and crime into prisons, where it doesn't "count".
posted by quirkyturky at 3:27 PM on May 9, 2015


I guess I don't understand the logic behind the follow-on assumptions being made. It seems that, crime went down, and there are any number of factors that could have contributed to that decline. And while you can say that mass incarceration did not play as significant role as other factors, it is not logical to assume that any of those factors didn't act as a force multiplier for the others. You would like to say that this is a clear repudiation of mass incarceration, but it seems impossible to disentangle it from the result.
posted by Cool Papa Bell at 3:27 PM on May 9, 2015 [1 favorite]


Its not impossible. You could easily test that using an interaction effect.

for Justinian: can you point me to where Leavitt's data was muffed? Thanks,
posted by MisantropicPainforest at 3:32 PM on May 9, 2015 [4 favorites]


How about: The increase in police procedural dramas and "true grit" methodology-based crime specials - Combined with the reduction of atmospheric lead - have produced a better breed of criminal: There's just as many murders and robberies going on, but the criminals are too smart to catch, or in some cases, even detect.
posted by Orb2069 at 3:32 PM on May 9, 2015 [3 favorites]


People never give enough credit to mean reversion.

We don't have abnormally low crime rates now, we had abnormally high rates in the 1970s and 1980s.

There was truly a perfect storm in the late 60s and early 70s: a dramatic weakening in social and family strictures, the collapse of the post war economic boom into stagflation, and a brief but immediately disastrous experiment with leniency on criminals. And the biggest reactions: white flight to the suburbs, and the War on Drugs, aggravated the situation.

In the face of that much bad luck, anything was likely to have facilitated the reduction of crime, and most likely everything that was tried had some benefit.
posted by MattD at 3:35 PM on May 9, 2015 [22 favorites]


Except they still have declined not relative to those high rates, but relative to more recent rates.
posted by MisantropicPainforest at 3:38 PM on May 9, 2015 [3 favorites]




Video games and HBO.

I think there might be something to this. Young men (and some women) who, in the past, might have committed a first crime out of boredom/for jollies, got arrested, and launched on the career-criminal path - instead sit at home, play games, watch TV (often on the Internet) and surf the web. They can get vicarious jollies without ever leaving their rooms. And they're not gathering in groups in person where one or two bold individuals can egg everyone else on to do something illegal - they're socializing on Facebook or via text.

I surmise better psychiatric diagnosis and treatment might play a part, as well. Yesterday's "bad kid" today gets diagnosed with ADHD or a learning disability, gets meds and/or therapy, and thus a much better likelihood of a happy, law-abiding life. Underdiagnosis in poor communities of color is still an issue (black boys, in particular, get written off as troublemakers) but I think far more kids get the help they need.
posted by Rosie M. Banks at 3:44 PM on May 9, 2015 [4 favorites]


I happen to have read this for a paper I was doing until just yesterday.

Page 8 of the pdf has the juicy summaries, though it's a bit underwhelming. It's all couched in "0-10%" for the various effects, except for the major losers (death penalty and liberalized conceal-carry laws, both of which were excluded as possible effects). Levitt's abortion theory has been roundly criticized but still may account for some of the effect during the 1990's. Lead is still a pretty good potential, but only for some of the drop, it's cited here as being "possibly some effect". I looked around and there is no serious refutation of Reyes article which popularized the lead-crime hypothesis, but there is a lot of criticism that her statistical analysis wasn't rigorous enough to draw the conclusion she did (56% of the 20-year decline due to lead 25ish% due to abortion). However, it felt like to me (a non-statistician, non-criminologist) that Reyes work was much better backed by data than Levitt. She was able to look at sales of various levels of leaded gas use state by state for two decades leading up to and during phase-out of leaded gas.

Levitt had a number of problems, the biggest off the top of my head was that his original paper counted abortions in the state they were performed rather than in state of residence. This even though data was available for the latter. Counting abortions where they occurred hugely inflated the numbers for states like New York and hugely deflated them for New Jersey, when in fact it seems that the number of abortions per resident for the two states was close to the same. This pretty much ruins a good chunk of the correlation Levitt based his conclusion on, which was that the states (New York, California, Hawaii, Alaska, Washington, plus D.C.) that had legal abortion before 1973 also had reduced crime rates that began sooner than non-abortion states.

There were also about 20 articles regarding Levitt's other statisical problems, but honestly, most of them went over my head.
posted by skewed at 3:44 PM on May 9, 2015 [10 favorites]


Here is the original paper by Jessica Reyes that popularized the lead-crime hypothesis, the first 10-15 pages are pretty easy reading and give a good overview of the basis of the claim.

Some shocking stuff about environmental lead levels from before the Clean Air Act: average lead-blood levels of 17 micrograms per dL in 1975 before the phase out, compared with 3 micrograms today. Current guidelines call for public health action for communities with levels higher than 5 micrograms, and actual treatment for any children with levels above 10 micrograms. Think about that, we prescribe treatment for lead toxicity for children who have lead levels substantially below what was average in 1975. Think of that next time someone grumbles about government regulation.

And we knew that even small doses of lead were potentially harmful as much as a century ago, there's a quote from the surgeon general in 1922 questioning the introduction of lead into gasoline, just two years after the process was invented. The whole thing just makes me sick to think about.
posted by skewed at 3:56 PM on May 9, 2015 [53 favorites]


I guess I don't understand the logic behind the follow-on assumptions being made. It seems that, crime went down, and there are any number of factors that could have contributed to that decline.

That's right. The crime rate decreased significantly, and there are multiple plausible hypotheses about the cause, and possibly there are multiple relevant causes.

And while you can say that mass incarceration did not play as significant role as other factors, it is not logical to assume that any of those factors didn't act as a force multiplier for the others.

Surely, what's illogical is making untested assumptions about causal relationships, period? I mean, think for a moment about how complex this could easily get: if variable A causes an increase in the outcome variable, except when variable C is present regardless of its value, and variable B decreases variable A's effect unless variable B is over a certain threshold at which point its effect on A is reversed, etc etc.

You would like to say that this is a clear repudiation of mass incarceration, but it seems impossible to disentangle it from the result.

Like MisantropicPainForest said, this is not only not impossible, but instead it would be impossible for statistical analysis to be useful if we couldn't disentangle discrete factors such as the rate of incarceration from the effect (i.e., the crime rate decline) that we're trying to explain. This is what people mean when they say that they're "controlling for" a variable.
posted by clockzero at 4:13 PM on May 9, 2015 [1 favorite]


for Justinian: can you point me to where Leavitt's data was muffed? Thanks

Wikipedia has footnotes pointing to more than you probably want to know about the subject in the article Legalized Abortion and Crime Effect which is mostly about Levitt's work.
posted by Justinian at 4:13 PM on May 9, 2015


But for those people who are TL;DR -
Later in 2005, Christopher Foote and Christopher Goetz claimed that a computer error in Levitt and Donahue's statistical analysis led to an artificially inflated relationship between legalized abortion and crime reduction. Once other crime-associated factors were properly controlled for, they claimed that the effect of abortion on arrests was reduced by about half. Foote and Goetz also criticize Levitt and Donahue's use of arrest totals rather than arrests per capita, which takes population size into account. Using Census Bureau population estimates, Foote and Goetz repeated the analysis using arrest rates in place of simple arrest totals, and found that the effect of abortion disappeared entirely.[7]

Donohue and Levitt subsequently published a response to the Foote and Goetz paper.[8] The response acknowledged the mistake, but showed that with different methodology, the effect of legalized abortion on crime rates still existed. Foote and Goetz, however, soon produced a rebuttal of their own and showed that even after analyzing the data using the methods that Levitt and Donohue recommend, the data does not show a positive correlation between abortion rates and crime rates.
posted by Justinian at 4:14 PM on May 9, 2015 [7 favorites]


it would be impossible for statistical analysis to be useful if we couldn't disentangle discrete factors such as the rate of incarceration from the effect (i.e., the crime rate decline) that we're trying to explain. This is what people mean when they say that they're "controlling for" a variable.

To be sure though, if there are highly collinear covariates, what happens in practice a lot of times is that a marginal effect becomes statistically significant only because of a few unrepresentative observations, rather than any general trend.
posted by MisantropicPainforest at 4:20 PM on May 9, 2015 [2 favorites]


(Glad to see so many people in here pointing out that access to abortion reduces crime rates. I am never able to find much on the internet--just the Foote/Goetz paper and a lot of Catholic propaganda--but it's pretty generally recognized in criminal justice policy circles that the theory that ready access to abortions causes lower crime rates 15-20 years out is not robust. It's a very complicated thing--crime rates and social factors)
posted by crush-onastick at 5:20 PM on May 9, 2015


To be sure though, if there are highly collinear covariates, what happens in practice a lot of times is that a marginal effect becomes statistically significant only because of a few unrepresentative observations, rather than any general trend.

Sure. I was pointing out that the tools exist to distinguish between coincidental phenomena and causally-related ones, even though (as you say) in practice, the extent to which it's possible to incontrovertibly demonstrate causal relation depends on the actual data.
posted by clockzero at 5:29 PM on May 9, 2015 [1 favorite]


Another issue with the Levitt/Donohue "more abortion, less crime" hypothesis is that it probably doesn't hold up in cross-national analyses. Abortion in France was legalized in 1975, just two years after Roe v. Wade in the United States, but France experienced increases in crime in the 1990s, whereas the U.S. experienced decreases. Another problematic implication of the Levitt/Donohue hypothesis is that the increase of state-level abortion restrictions in the 1980s and 1990s should have led to a reversal of the 1990s crime drop in the 2000s, but the crime drop continued into the 2000s.
posted by jonp72 at 5:40 PM on May 9, 2015 [2 favorites]


go read David Simon on what Mayor O'Malley did in Baltimore:
The second thing Marty did, in order to be governor, involves the stats themselves. In the beginning, under Norris, he did get a better brand of police work and we can credit a legitimate 12 to 15 percent decline in homicides. Again, that was a restoration of an investigative deterrent in the early years of that administration. But it wasn’t enough to declare a Baltimore Miracle, by any means.

What can you do? You can’t artificially lower the murder rate – how do you hide the bodies when it’s the state health department that controls the medical examiner’s office? But the other felony categories? Robbery, aggravated assault, rape? Christ, what they did with that stuff was jaw-dropping.

So they cooked the books.

Oh yeah. If you hit somebody with a bullet, that had to count. If they went to the hospital with a bullet in them, it probably had to count as an aggravated assault. But if someone just took a gun out and emptied the clip and didn't hit anything or they didn't know if you hit anything, suddenly that was a common assault or even an unfounded report. Armed robberies became larcenies if you only had a victim’s description of a gun, but not a recovered weapon. And it only gets worse as some district commanders began to curry favor with the mayoral aides who were sitting on the Comstat data. In the Southwest District, a victim would try to make an armed robbery complaint, saying , ‘I just got robbed, somebody pointed a gun at me,’ and what they would do is tell him, well, okay, we can take the report but the first thing we have to do is run you through the computer to see if there's any paper on you. Wait, you're doing a warrant check on me before I can report a robbery? Oh yeah, we gotta know who you are before we take a complaint. You and everyone you’re living with? What’s your address again? You still want to report that robbery?

They cooked their own books in remarkable ways. Guns disappeared from reports and armed robberies became larcenies. Deadly weapons were omitted from reports and aggravated assaults became common assaults. The Baltimore Sun did a fine job looking into the dramatic drop in rapes in the city. Turned out that regardless of how insistent the victims were that they had been raped, the incidents were being quietly unfounded. That tip of the iceberg was reported, but the rest of it, no. And yet there were many veteran commanders and supervisors who were disgusted, who would privately complain about what was happening. If you weren’t a journalist obliged to quote sources and instead, say, someone writing a fictional television drama, they’d share a beer and let you fill cocktail napkins with all the ways in which felonies disappeared in those years.
and tell me what the actual fundamental sources of error in the data are and consequently what the error bars actually are.
posted by ennui.bz at 5:44 PM on May 9, 2015 [13 favorites]


Another issue with the Levitt/Donohue "more abortion, less crime" hypothesis is that it probably doesn't hold up in cross-national analyses. Abortion in France was legalized in 1975, just two years after Roe v. Wade in the United States, but France experienced increases in crime in the 1990s, whereas the U.S. experienced decreases. Another problematic implication of the Levitt/Donohue hypothesis is that the increase of state-level abortion restrictions in the 1980s and 1990s should have led to a reversal of the 1990s crime drop in the 2000s, but the crime drop continued into the 2000s.

That's not a problem with the hypothesis, that's an issue of scope conditions. A hypothesis can hold without being universal. Moreover, the Levitt argument is that other things being equal, abortion leads to less crime. You can't say that after abortion became legalized in some place, then the change in crime rate following the legalization demonstrates whether or not abortion reduces crime rates.
posted by MisantropicPainforest at 6:00 PM on May 9, 2015 [2 favorites]


Has no one taken into account the vast proliferation (and subsequent devaluation) of consumer goods in the same period? Like, maybe it was worth the risk to steal a computer or a big screen tv or a car stereo when they were $6000, $10,000, and $750...not so much when they are suddenly $1000, $500, and $50
posted by sexyrobot at 6:05 PM on May 9, 2015 [11 favorites]


and tell me what the actual fundamental sources of error in the data are and consequently what the error bars actually are.

Measurement error can be modeled and widespread measurement error only introduces uncertainty, not bias. Unless you have reason to believe that fudging the numbers are endogenous with some other cause of crime, your criticism is baseless. Dismissing a mountain of quantitative research that spans decades because of an anecdote is silly.
posted by MisantropicPainforest at 6:36 PM on May 9, 2015 [4 favorites]


Abortion in France was legalized in 1975, just two years after Roe v. Wade in the United States, but France experienced increases in crime in the 1990s, whereas the U.S. experienced decreases.

I wanted to say that France is a troublesome comparator here, because I've read that the political and administrative system has a history of disincentivizing regional governments and police forces from accurately reporting crime. But I'm not finding a good link for that.
posted by George_Spiggott at 6:39 PM on May 9, 2015 [1 favorite]


The answer is, probably, the removal of lead from paint and gasoline; the rise in crime rates in the mid-20th century tracks pretty well with the widespead use of motor vehicles in every industrial country (not merely in the USA); the subsequent decline in crime rates from their earlier highs also tracks very well with the removal of lead from paint, motor fuel, and other consumer products. Lead is a known neurotoxin that has frontal lobe effects (and lead exposure can lead to greater impulsivity and aggression). Once you take into account the similar decline in crime rates OUTSIDE the US as well as within it, and the correlation of that decline in crime rates with elimination of lead...it's not such an easy thing to dismiss.
posted by Pseudonymous Cognomen at 6:39 PM on May 9, 2015 [3 favorites]


Is it possible that the rise in obesity is a contributing factor in the drop in crime? It a lot harder to run to escape when one is overweight. Likewise breaking into residences and other crimes requiring maneuverability or physical exertion would be more difficult. What are the physical measurements/physical fitness of suspects arrested for such crimes?
posted by haiku warrior at 6:48 PM on May 9, 2015 [1 favorite]


Abortion in France was legalized in 1975, just two years after Roe v. Wade in the United States, but France experienced increases in crime in the 1990s, whereas the U.S. experienced decreases.

Leaded gasoline was eliminated in the US, by law, in 1986; in Europe? not until 1998.
posted by Pseudonymous Cognomen at 6:51 PM on May 9, 2015 [2 favorites]


The lead hypothesis has some weird hold on the internet way out of proportion to its reputation in criminology, epidemiology, sociology, and other social sciences. Lead and crime went up together and down together in most US states and many other countries. However, a whole lot of things also went up and down together. Showing two curves that go together is not remotely sufficient to demonstrate causality. Reyes and others attempt to do a bit better by looking at the variation across states and countries, but their results are pretty weak. The main problem is that there are a million things that can both cause a state/country to start cleaning up their lead, crime, and a multitude of other social ills at the same time. Since lead was a recognized problem, its cleanup was not random, but subject to the same societal forces that affect all the other stuff, including crime. Similarly, the variation in the onset of lead, either in gas or paint, correlated with a lot of other developmental indicators. This might seem to set the bar on the theory very high -- you have to rule out dozens if not hundreds of other factors -- but that's the way science works. The null is no effect, and that null is especially strong in the social sciences, where there are thousands of colinear and causal pathways. If you have a new hypothesis that purports to explain a significant chunk of the causation, that's a high bar to pass. A couple middling-power econometric studies and a plausible medical story doesn't cut it. And a good thing too -- these sorts of correlations are easy to find, and people like Murray or Levitt are constantly claiming that they've controlled for everything when they've really just controlled for what they could get their hands on.* Big claims require big proof, and lead is not even close.

* Levitt of course claimed that he didn't need to control for all that other stuff because of his slick methodology that had state, time, and state*time fixed effects. Leaving aside the ridiculousness of trying to estimate a model using regression with effectively hundreds of (dummy) variables, I've seen with my own eyes their Stata code that famously "accidentally" lacks the all-important state*time fixed effects. And of course as Foote and Goetz point out, there are numerous other problems with Levitt, and Levitt's pathetic "collage of evidence" defense show just how poorly his abortion baloney has turned out. This stuff is really hard to prove, and Reyes, although she presumably doesn't make the same coding errors Levitt does, is subject to many of the same general criticisms. Criminologists are dubious of all these one-big-freaky-cause stories because there are a hell of a lot of variables bobbing along together in the social stew.
posted by chortly at 7:16 PM on May 9, 2015 [14 favorites]


What about the aging population? There are more old people as compared with young people and most crime is committed by young people.
posted by latkes at 7:27 PM on May 9, 2015


Why didn't anyone ever think of that?
posted by Justinian at 7:42 PM on May 9, 2015 [2 favorites]


Ugh, my snarkmeter broke today, sorry. Let me rephrase: That is the first thing they take into account.
posted by Justinian at 7:43 PM on May 9, 2015 [1 favorite]


If not lead, then what?

My own pet theory is Mr. Rogers Neigborhood and Sesame Street.

If you are of a certain age, and you know someone who had it really, really rough growing up, they all have firm and fierce opinions on The Land of Make Believe and their favorite muppets. The explosion of nationwide mass media, the establishment of Public Television, and the perfect people in the perfect place at the perfect time - Fred Rogers and Jim Henson - resulted in vicious cycles being broken, a generation of children born into imaginary families that they deserved and could retreat to when the real world got too much.

Mr. Rogers went nation-wide in 1968, Sesame Street in '69. Barack Obama would have been seven years old, then.

Hillary is still of the Boomers, but her cabinet will likely be comprised of people who can sing "Tomorrow" and "Ladybugs' Picnic" by heart.
posted by Slap*Happy at 7:44 PM on May 9, 2015 [13 favorites]


What about the aging population? There are more old people as compared with young people and most crime is committed by young people.

The overall aging of the population trend is very tiny compared to the decrease in crime. More likely is that the lead poisoned young people are aging out of crime and being replaced by non-lead poisoned young people, a much more rapid phenomenon.
posted by JackFlash at 7:46 PM on May 9, 2015


Honest question here: If it really was legalized abortion, shouldn't crime be going up (with 16 year or so lag time) as abortion becomes essentially unavailable in states with conservative governments? If so, the answer might become apparent in upcoming years.
posted by LastOfHisKind at 8:29 PM on May 9, 2015 [1 favorite]


Probably nobody has thought about this, but I think it might have been lead.

Seriously though... I think chortly is right here, the lead-hypothesis is more notable for how appealing it is, rather than its actual ability to explain the drop in crime. The tough thing about any account of the crime drop is that, when it even gets attention in the popular media, it's gotta be a single explanation or else no one's gonna write about it. The bottom line of the Brennan report that this post is supposedly about isn't even about lead or abortion, it's that there are many factors that might have contributed to the drop, and most notably that incarceration is at most a very small part of the equation. But you can't generate clicks with a headlines with that.

The reason I think the lead hypothesis is so appealing to communities like Metafilter is that when we say lead causes crime, we're basically saying that corporate polluters, ignoring available evidence from the scientific community, fought for unregulated rights to poison the population and then externalized the costs, leaving poor and minority populations to pay the price. I mean, that's not a narrative I can turn down, I'm not made of stone here.

In the back of my mind, I know that it's too simple to be more than a very partial account of the crime drop, but even if it's only 2-3%, fuck it, the rest of it's true--we know industry fought like hell against decades of mounting research showing that even small doses of lead can be harmful, especially to children. Maybe leaded gasoline didn't cause huge amounts of crime, but the sentiment holds nonetheless.
posted by skewed at 8:29 PM on May 9, 2015 [16 favorites]


I'm surprised no one here is looking at the obvious explanation :

Batman.
posted by webmutant at 9:47 PM on May 9, 2015 [4 favorites]


My own pet theory is Mr. Rogers Neigborhood and Sesame Street.

Don't forget the ABC After School Special as well. And a ton of what was on Saturday morning. It's astonishing to look back now and see the way for a few short years in the first half of the 1970s a certain folk ethos and definitely folk styles -- sometimes not even without notes of op-art and psychedelia -- went completely mainstream in children's programming, and it certainly was not limited to public broadcasting for that short while.
posted by George_Spiggott at 9:57 PM on May 9, 2015 [3 favorites]


Has no one taken into account the vast proliferation (and subsequent devaluation) of consumer goods in the same period? Like, maybe it was worth the risk to steal a computer or a big screen tv or a car stereo when they were $6000, $10,000, and $750...not so much when they are suddenly $1000, $500, and $50

That would only help explain changes in theft and robbery, and it requires an assumption of a rational economic action-orientation to the commission of crime which may or may not be valid.
posted by clockzero at 10:08 PM on May 9, 2015


Like, maybe it was worth the risk to steal a computer or a big screen tv or a car stereo when they were $6000, $10,000, and $750...not so much when they are suddenly $1000, $500, and $50

OR maybe the devaluation means you have to steal MORE to meet your financial need for that week.
posted by salvia at 10:43 PM on May 9, 2015


I guess different disciplines have different practices. I would never fit state*time as a fixed effect. Have these guys never heard of mixed models? [*]

[*]This is obviously inappropriate if you actually care about the estimates of the interactions. Typically, I want to account for variability due to time and location, but I'm not interested in the actual effects.
posted by wintermind at 11:42 PM on May 9, 2015


I'm a criminologist.

Chortly is right: I don't know a single criminologist who really gives the lead hypothesis much play. It's a pretty dangerous road that those kind of things lead you down, and it's troubling to see the hypothesis pursued so doggedly on the internet.
posted by still bill at 4:01 AM on May 10, 2015 [5 favorites]


Measurement error can be modeled and widespread measurement error only introduces uncertainty, not bias. Unless you have reason to believe that fudging the numbers are endogenous with some other cause of crime, your criticism is baseless. Dismissing a mountain of quantitative research that spans decades because of an anecdote is silly.

systematic (within baltimore crime stats) bias is exactly what Simon is describing, and endogenous to whichever crimes they wanted to show a decline in.

and of course if you read the forward they put "CompStat" as one of the three factors which led to a decrease in crime. it's telling that this report was written by two economists and a lawyer, no one here is actually qualified to do serious meta-analysis on crime stats.
posted by ennui.bz at 5:32 AM on May 10, 2015


I have a completely unsupportable theory: Children's books are responsible. Crime went up in the first place because children's authors like Dr. Seuss decided that a story with a moral was a boring story. Crime went down again because writers like the Berenstains went back to thwacking children over the head with morals and safety lessons.

Don't bother refuting this idea; it's obviously wrong. Instead, spread it as an example of an obviously wrong theory.
posted by clawsoon at 6:38 AM on May 10, 2015 [1 favorite]


I'm completely confident that econometric training is sufficient enough to analyze crime statistics. I do not see any reason not to be.

Anyway.

I guess different disciplines have different practices.

And I would never use Stata! Obvioulsy a hiearchical model is the way to go with these but a) Levitt's paper is about a decade old 2) frequentist hierchical models are kinda weird a decade ago even fewer people did bayesian statistics.
posted by MisantropicPainforest at 6:39 AM on May 10, 2015


Chortly is right: I don't know a single criminologist who really gives the lead hypothesis much play.

Okay, but -- and I don't mean this to be a dick -- how much of that is because it really is not very relevant and how much is because it's not the sort of theory/hypothesis that criminologists have been trained to find interesting and worthwhile?

I mean, similarly, theories of political behavior grounded in genetics and biology are unlikely to ever be an important part of political science, even if the few people who study this can demonstrate that biologically-generated predispositions explain a $BIG proportion of voting behavior or whatever. In part because these papers are, to us, less interesting -- they're not about individual decision-making or strategic action or the other things we've been trained to treat as important, so kinda who cares?
posted by ROU_Xenophobe at 6:52 AM on May 10, 2015 [2 favorites]


And I would never use Stata! Obvioulsy a hiearchical model is the way to go

You can do those with stata. It's had xtreg for a long while, and xtmixed for a good while (but maybe not when Leavitt did that).

Or if you want an experience that's marginally less painful than writing your own raw code and don't mind being virtually certain to end up with a model that is not what you thought it was, you could use gllamm.
posted by ROU_Xenophobe at 7:05 AM on May 10, 2015


I'm completely confident that econometric training is sufficient enough to analyze crime statistics. I do not see any reason not to be.

I'm sure they are confident of that too. But, the issue is whether it's simply a question of number crunching or whether the analysis of social data requires a theoretical context: this is exactly the question that trained economists are worst at, especially if (as for one of the author's) you did your PhD at U.Chicago.

There has been, since Nixon, a systematic devaluing and politicizing of social "science" research at the academic level and the idea that all you have to do is throw the right stats model at your data is part of that.

But, the Simon essay shows just how weedy that data really is. When you subtract the demographic effects of mass incarceration and the end of the 'baby boom', you start talking about changes on a much smaller scale than what is advertised
posted by ennui.bz at 7:19 AM on May 10, 2015


There has been, since Nixon, a systematic devaluing and politicizing of social "science" research at the academic level and the idea that all you have to do is throw the right stats model at your data is part of that.

Please do not confuse quantitative rigor for theoretical dillentantism.

So who, in your mind, is trained to analyze crime statistics? clearly not economists, especially if they went to Chicago. Maybe, say, a TV writer?
posted by MisantropicPainforest at 7:30 AM on May 10, 2015 [4 favorites]


Please do not confuse quantitative rigor for theoretical dillentantism.

So who, in your mind, is trained to analyze crime statistics? clearly not economists, especially if they went to Chicago. Maybe, say, a TV writer?


Yup. My point exactly. You really believe that the data is independent of a "theoretical" context. There are two questions: what do these numbers mean and what are their errors and biases. Those two questions are completely interconnected in the social "sciences." It's why they aren't sciences at all.

I would expect to see this kind of study coming from a federal bureau backed up by a academic research group... but that sort of thing is actually impossible to do honestly because of the years of politicization of crime and drug crime from the federal level on down.

But, I was making a specific point. Read what the TV writer who spent years studying the Baltimore police force says about the use of "CompStat" in Baltimore and tell me you believe the data that supposedly shows that nationally it led to a "10-15% decline in crime" per the linked study.
posted by ennui.bz at 8:18 AM on May 10, 2015 [2 favorites]


especially if (as for one of the author's) you did your PhD at U.Chicago.

None of the authors got a PhD from the University of Chicago, if that makes any difference to you.
posted by JackFlash at 8:19 AM on May 10, 2015 [1 favorite]


If you actually believe that social scientists do not place data in their context and do not judge the quality of data and believe that all data can be divorced from context and that there is no measurement error, then you are simply ignorant of contemporary social science research.
posted by MisantropicPainforest at 8:29 AM on May 10, 2015 [7 favorites]


None of the authors got a PhD from the University of Chicago, if that makes any difference to you.

oh, I misread the blurb, the first author was an undegraduate at Chicago. But it still remains that economists tend to be particularly bad about acknowledging "theoretical dilletantism" in the social sciences.
posted by ennui.bz at 8:31 AM on May 10, 2015


If you actually believe that social scientists do not place data in their context and do not judge the quality of data and believe that all data can be divorced from context and that there is no measurement error, then you are simply ignorant of contemporary social science research.

no true social scientist? come on. with the nationalization of "broken windows" policing i.e. mass arrests and CompStat, do you think that the problems Simon describes in Baltimore poisoned an isolated data set or reflect broader "problems" with the data. And how do you deal with poisoned data?
posted by ennui.bz at 8:45 AM on May 10, 2015


And how do you deal with poisoned data?

Model it.
posted by MisantropicPainforest at 8:55 AM on May 10, 2015


'Okay, but -- and I don't mean this to be a dick -- how much of that is because it really is not very relevant and how much is because it's not the sort of theory/hypothesis that criminologists have been trained to find interesting and worthwhile?'

No, it's not because it's not interesting. Of course it's interesting. It's because it's far too focused on a single biological/environmental factor in explaining a hugely complex and intensely social thing.
posted by still bill at 9:02 AM on May 10, 2015


I blame the Beatles.
posted by jonmc at 9:09 AM on May 10, 2015 [6 favorites]


no true social scientist? come on. with the nationalization of "broken windows" policing i.e. mass arrests and CompStat, do you think that the problems Simon describes in Baltimore poisoned an isolated data set or reflect broader "problems" with the data. And how do you deal with poisoned data?

Just out of curiosity, what do you mean by "poisoned" data? I mean, there are statistical ways to control for bias.
posted by clockzero at 9:33 AM on May 10, 2015


The dismissals of the lead hypothesis in this thread seem a bit circular: "It's too focused on a biological factor, because crime is actually explained by social factors" - but that's the whole point, if it is true, the crime wave of the 70's and 80's was not "intensely social," it was mainly the result of widespread chronic lead poisoning.

It's not as if we're correlating pirates and global warming here. There's a correlation between lead and antisocial behavior, and an objective, biological explanatory mechanism. Even minute amounts of lead can cause long-term neurological, cognitive, and behavioral problems.

The young people with dentin lead levels greater than 20 ppm had a markedly higher risk of dropping out of high school (adjusted odds ratio, 7.4; 95 percent confidence interval, 1.4 to 40.7) and of having a reading disability (odds ratio, 5.8; 95 percent confidence interval, 1.7 to 19.7) as compared with those with dentin lead levels less than 10 ppm. Higher lead levels in childhood were also significantly associated with lower class standing in high school, increased absenteeism, lower vocabulary and grammatical-reasoning scores, poorer hand-eye coordination, longer reaction times, and slower finger tapping.

Very low lead exposures and children's neurodevelopment: Increased exposure is also associated with neuropsychiatric disorders such as attention deficit hyperactivity disorder and antisocial behavior. . . No level of lead exposure appears to be 'safe' and even the current 'low' levels of exposure in children are associated with neurodevelopmental deficits.

Freddie Gray grew up in a lead-painted house. "Before Freddie Gray was injured in police custody last month, before he died and this city was plunged into rioting, his life was defined by failures in the classroom, run-ins with the law and an inability to focus on anything for very long.


A 2013 article about lead poisoning in Baltimore:
[S]tudies have found that even infinitesimally low levels—down to one or two micrograms per deciliter—can reduce a child’s IQ and impair her self-control and ability to organize thoughts . . .

The most accurate national survey of lead poisoning was probably the 1976–1980 National Health and Nutrition Examination Survey, which found that 4 percent of all children under six—roughly 780,000—had blood lead concentrations exceeding thirty micrograms per deciliter[!!!], which was then thought to be the limit of safety.

Black children, the survey found, were six times more likely to have elevated lead than whites. The number of children with lead levels over five micrograms per deciliter—or for that matter over one or two—was obviously much higher
[likely, in the millions], but there’s no way of knowing how high it was. The 1985 leaded gasoline ban and the gradual renovation of slum housing have since reduced the number of poisoned children, so that today, the CDC estimates that some 500,000 children who are between one and five years old have lead levels over five micrograms per deciliter.

Also: Understanding international crime trends: the legacy of preschool lead exposure.This study shows a very strong association between preschool blood lead and subsequent crime rate trends over several decades in the USA, Britain, Canada, France, Australia, Finland, Italy, West Germany, and New Zealand. The relationship is characterized by best-fit lags (highest R2 and t-value for blood lead) consistent with neurobehavioral damage in the first year of life and the peak age of offending for index crime, burglary, and violent crime. Maybe someone can read this study and explain where Nevin went wrong?

How many criminologists have backgrounds in toxicology and neurology? I think people are so focused on the lead hypothesis because the link between lead and crime seems overwhelmingly obvious.
posted by mrbigmuscles at 10:04 AM on May 10, 2015 [14 favorites]


The dismissals of the lead hypothesis in this thread seem a bit circular:

No, the dismissals of the lead hypothesis are here because the evidence is weak.
posted by MisantropicPainforest at 10:54 AM on May 10, 2015


mrbigmuscles--gold star for digging up citations that I was too lazy to! I also find the lead-crime link blindingly obvious, but then I come from environmental science background. I mean, lead is a known neurotoxin with well-documented cognitive and behavioral impacts. That science is rock solid--it's part of the reason why we banned the stuff in the first place. And it's not like asbestos, where you just about have to live in an asbestos factory to be affected. I think it bears repeating that the current medical opinion is that there is no safe blood lead level for children. And yet we were going around spewing it out our tailpipes for decades. So after raising a whole generation of kids on lead-poisoned air (in urban areas in particular), why would we NOT expect the increased likelihood of these behavioral problems to result in increased criminal activity? I think it would be a miraculous finding if that generation had somehow not produced more criminals than any generation before or since!
posted by gueneverey at 10:54 AM on May 10, 2015 [7 favorites]


No, the dismissals of the lead hypothesis are here because the evidence is weak.

Your last few responses in this thread have been pretty weak, themselves. I don't know if you're just busy doing something else, but you're not contributing much right now that means anything to anyone who doesn't have exactly your body of knowledge, which I wager is not very many people at all.
posted by Steely-eyed Missile Man at 11:41 AM on May 10, 2015 [1 favorite]


The lead hypothesis is appealing to me in virtue of being poised to explain both the decrease in crime (from about 1990 through today) and also the increase in crime that preceded it (from about 1960 through 1990). The abortion hypothesis does not have the explanatory power of the lead hypothesis, since it cannot explain the rise in crime. So, I'm curious, are any of the explanations for the decline in crime considered in the Brennan report -- aging population, consumer confidence, decreased alcohol consumption, decreased unemployment, growth in income, increased incarceration, increased police numbers, or inflation -- poised to explain the previous increase in crime? Did the period from 1960 till 1990 see a rejuvenating (youthening?) population or decreasing consumer confidence or increased alcohol consumption or increased unemployment or fall in income or decreased incarceration or decreased police numbers or deflation? I haven't looked for myself yet, but I will be surprised if any of these has the sort of mirror image pattern that both lead and crime share in the 20th century.

I'm not a criminologist, and I don't pretend to know anything here. It just seems to me that if one wants to explain the decline of crime, one really needs to explain both the rise and the decline. And, all else equal, a single cause or a small collection of causes that were added and then removed is better than some factors called on to explain the rise and others called on to explain the decline. I would be happy to learn something here, though.
posted by Jonathan Livengood at 11:47 AM on May 10, 2015 [3 favorites]


No, the dismissals of the lead hypothesis are here because the evidence is weak.

Show me! Im perfectly willing to be shown it is wrong. I literally asked for it
posted by mrbigmuscles at 11:48 AM on May 10, 2015 [1 favorite]


Sure. For one, Levin doesn't even attempt to causally identify it in that paper. He just shows that the two time series of preschool levels of lead exposure and crime fit better after a lag than unemployment and crime. Note that the lag varies, depending on country, from 19 to 38 years. That's quite a big lag. And again, he's not attempting to causally identify the effects of lead on crime, and his paper shouldn't be read as such.

What needs to be done is either an insturmental variable or regression discontinuity approach. If you measure something that causes a decline in lead--but has no other effect on any other relevant variables--and use that as a proxy for a decrease in lead, and then measure crime, that would solve some of the massive endogeneity problems with these studies. I lack the contextual knowledge of this issue to think of ones, but those are tactics used whenever your goal is to causally identifying complex phenomena. You simply cannot control away the endogeneity--especially if the variables you are controlling for are affected by the rise in crime themselves. That will bias your results.

You simply can't say, "states that reduced their lead levels saw decreases in crime" and use that to infer causality. That's because states that reduce their lead levels also probably did other things that may reduce crime.

We have the causal mechanism, which you showed. Lead causes brains to be bad and that can cause crime. That's great, but it doesn't indicate population level-effects.

In order to causally identify effect of lead reduction on crime, someone should find out if there was any quasi-randomness in how states reduced their lead levels. Was there a private agency that randomly picked states to pressure them to reduce lead? That sort of thing.

On the theoretical level, I do find the argument somewhat unconvincing because, while lead makes brains bad, it doesn't just make brains do crime, it makes them do all sorts of bad things, which make crime more likely. Without attempting to causally identify the effect, does lead reduction also lead to a reduction in non-criminal behavior that a reduction in lead would effect? That would bolster the argument.
posted by MisantropicPainforest at 12:04 PM on May 10, 2015 [2 favorites]


All the environmental epidemiologists I know are pretty quick to say "we can't say that lead is responsible for the epidemic of certain types of crime in the 20th century," because it's pretty difficult to get proof according to the exacting standards of their profession.

However, they're also quick to point out that lead has a tremendous personal and social cost, lowers IQ, decreases educational outcomes, increases violence and impulsiveness, and is probably the most severe environmental justice issue of the 20th century through today.

It's a pretty powerful hypothesis. The fact that it's difficult to be conclusive about it is not particularly surprising.
posted by entropone at 1:55 PM on May 10, 2015 [3 favorites]


I wanted to say that France is a troublesome comparator here, because I've read that the political and administrative system has a history of disincentivizing regional governments and police forces from accurately reporting crime. But I'm not finding a good link for that.

This is an issue with any cross-national comparison of crime, because the reporting methods, reporting standards, and legal definitions for various crimes vary so much by country. However, the technique most criminologists use to get around that is to limit the analysis to homicides, because homicide is the crime least subject to being influenced by cultural differences (i.e., you either have a dead body or you don't).

I just brought up the U.S. vs. France comparison, because I had actually considered writing an article that did a cross-national analysis on the abortion-crime link (although my area of expertise was actually abortion not crime), but I ended up leaving academia before I ever started such an article.
posted by jonp72 at 2:07 PM on May 10, 2015


So, I'm curious, are any of the explanations for the decline in crime considered in the Brennan report -- aging population, consumer confidence, decreased alcohol consumption, decreased unemployment, growth in income, increased incarceration, increased police numbers, or inflation -- poised to explain the previous increase in crime? Did the period from 1960 till 1990 see a rejuvenating (youthening?) population or decreasing consumer confidence or increased alcohol consumption or increased unemployment or fall in income or decreased incarceration or decreased police numbers or deflation?

It's consistent with the inflation hypothesis in the sense that inflation was much higher in the 1960s and 1970s, then decreased from 1980s to 2010s. This pattern of increase and decrease also matches the general pattern of increase & decrease in crime on a decade by decade basis, but inflation isn't too useful in predicting regional variations in crime rates in the U.S.

The age cohort explanation could explain some of the increase in crime in the 1960s and 1970s, because you had a larger cohort of Baby Boomer males entering their prime "crime years" at the time. If crime is primarily driven by the number of young males in a society at a given time, then the increases and decreases in crime are going to be most dramatic as the Baby Boomer cohort "ages into" and "ages out of" the prime years for crime.

I'm less convinced of the consumer confidence explanation, because the 1970s was a decade that had both high crime rates and low consumer confidence.
posted by jonp72 at 2:14 PM on May 10, 2015 [1 favorite]


MisantropicPainforest: What needs to be done is either an insturmental variable or regression discontinuity approach. If you measure something that causes a decline in lead--but has no other effect on any other relevant variables--and use that as a proxy for a decrease in lead, and then measure crime, that would solve some of the massive endogeneity problems with these studies.

I seem to recall one study which set out to follow an apparent link between high-voltage power lines and various health effects, but instead found that the health effects followed freeways. (It was the fact that freeways and high-voltage power lines are often near each other which had led to the maybe-this-is-a-thing-but-it's-ambiguous results from earlier studies.) Would that sort of thing add enough orthogonal data to make conclusions easier to come to?
posted by clawsoon at 2:17 PM on May 10, 2015 [1 favorite]


Thanks for responding MP now we have something to sink our teeth into!
posted by mrbigmuscles at 2:33 PM on May 10, 2015 [1 favorite]


Think of that next time someone grumbles about government regulation.

I think of all sorts of things when someone grumbles about government regulation, but usually, those who grumble about government regulation are reality-challenged and presenting reality to them is entirely ineffective. These are the same people who sing to the high heavens about the benefits of deregulation when it's been shown, over and over again, that deregulation often only benefits a few people financially, and fucks over everyone else, including those who believe in deregulation.
posted by juiceCake at 2:46 PM on May 10, 2015 [2 favorites]


Would that sort of thing add enough orthogonal data to make conclusions easier to come to?

I'm not sure. The best way to think of how to do these things is: what would you do if you wanted to do this as an experiment? Ideally, we could have two very similar states, and treat one of them with lead removal, and use the other as a control. Then see how crime changes. Ok so we can't actually do that, but what approximates that?

The wikipedia entry for insturmental variables is overly technical but the examples may help.
posted by MisantropicPainforest at 6:50 PM on May 10, 2015


The fact that it's difficult to be conclusive about it is not particularly surprising.

I agree, I think the bigger question is though: why hasn't this been studied extensively? I mean, economists still write a butt ton of papers on the WTOs effect on trade, and still don't come to a consensus. It could be that its just at the margins of different disciplines.
posted by MisantropicPainforest at 6:53 PM on May 10, 2015


In order to causally identify effect of lead reduction on crime, someone should find out if there was any quasi-randomness in how states reduced their lead levels.

There were two studies linked in the Mother Jones piece that looked at state and international variations in lead exposure and crime, and the trend holds.

On the theoretical level, I do find the argument somewhat unconvincing because, while lead makes brains bad, it doesn't just make brains do crime, it makes them do all sorts of bad things, which make crime more likely. Without attempting to causally identify the effect, does lead reduction also lead to a reduction in non-criminal behavior that a reduction in lead would effect?

I can't find any studies offhand that look at non-criminal behaviors and lead on a population level (and I'm not even sure that's what you're looking for), but there are piles of studies that link behavioral problems, delinquency, etc to lead exposure. I honestly have a hard time wrapping my mind around what evidence you find lacking. I feel like your argument is akin to, "well I agree that smoking is a major risk factor for developing lung cancer, but aside from this major international trend of decreased deaths from lung cancer, do we have any evidence that non-lethal cases of lung cancer have gone down since smoking rates have decreased?"
posted by gueneverey at 7:18 PM on May 10, 2015 [2 favorites]


Or wait, sorry, you were talking about the Nevin study? Ignore me. I knew I shouldn't keep commenting into the night like this.
posted by gueneverey at 7:35 PM on May 10, 2015


I'm not sure.

Having better internal variation in the data seems to me to be a happy circumstance that can combat collinearity, but I don't offhand see how it would affect causal inferences.

The best way to think of how to do these things is: what would you do if you wanted to do this as an experiment?

That's part of it, anyway. The other part is to ask "What other relationships would we observe in the world if this theory were true?" Like when you noted that we should observe other population-scale cognitive effects of increasing and then reducing lead.
posted by ROU_Xenophobe at 9:46 PM on May 10, 2015


Everything is coming up roses as far as crime is concerned! First we were told that gun related homicides were way down over the past couple of decades. Then it's crime in general that has supposedly dropped like a rock -- which the Brennan Center authors are (incoherently) trying to explain. The common factor in all this research is the use of the FBI's Uniform Crime Reports (UCR) data. It's so flawed that I have my doubts if we know ANYTHING about crime.

What's wrong with the Uniform Crime Reports? The authors wait until p. 97 to mention it....somewhat. It's not mandatory for police departments to report their crime stats to the FBI for UCR purposes. You aren't getting data from the universe of U.S. police departments. You get data from a self-selected group of police departments. Also, there's a lack of uniformity in how police departments categorize certain crimes and so report the data to the FBI. Then there's an incentive to under-report crime. No P.D. wants to looks incompetent. You can call all of those things quality of data issues. As for the quantity of data collected, the FBI doesn't bother accumulating data related to pressing social issues -- like homicides committed on-the-job by cops, by what means (gun, choking etc.) and the races of the parties involved. You can guess why.

"Garbage in, garbage out" (i.e., GIGO) are words to live by when doing any kind of data analysis. If the data you're using to test your hypotheses is crap, all the statistical manipulation in the world isn't going to give results that you (or anybody else) can have faith in. I suspect we're up to our ears in garbage.
posted by bim at 11:56 PM on May 10, 2015


But the DoJ also just calls up 100,000 people every year and asks them if they've been the victim of any crimes, and has done since 197X, and the crime rate from the National Criminal Victimization Survey has also fallen like a rock.
posted by ROU_Xenophobe at 6:48 AM on May 11, 2015 [5 favorites]


I suspect we're up to our ears in garbage.

So, like ROU_Xenophone said, there isn't a single source this data is drawn on. Just one page after the Brennan report talks about the UCR they talk about the NCVS, which is a massive telephone survey done every year that has shown a similar drop in crime. The NCVS also allows for much broader array of criminal activity to be included, whether reported to police or not.

Also, the UCR has its faults but you make it out to be worse than it is. First, it doesn't cover every jurisdiction, which I think is unfortunate. But it covers jurisdictions that include 96ish percent of the population for every year that I've looked at.

Second, the fact that it doesn't have a special category for police brutality is . . . well, it's just a complete non-sequitur to the issue here.

It's not clear why you think the Brennan Report is incoherent, except maybe that it comes to a conclusion that you disagree with. Is there some evidence you're relying on which show that murder or overall crime rates are increasing, as you seem to suggest?
posted by skewed at 9:27 AM on May 11, 2015 [2 favorites]


Skewed & Rou_Xenophobe:

1) So, like ROU_Xenophone said, there isn't a single source this data is drawn on.

The dependent variable (i.e., the crime rate) in the Brennan Report comes from a single source -- the UCR (Brennan, p. 97). I am well aware of the NCVS. Since the Brennan Report doesn't use it (p. 97) nor does it explicitly discuss results from it (page number, please, if I'm wrong), it's irrelevant to the discussion at hand.

Are you also claiming that tossing more and more independent variables into a regression and using as many data sources as possible for these variables is a good thing? I sure hope not. That's wrong.

2) Regarding the UCR's data coverage, I'll concede that 96ish percent of the population isn't bad. That still leaves open the issues of misclassification of data and underreporting. If the data mixes apples with oranges, it negates even 100% coverage. In any case, the UCR is what it is. I realize that everybody uses it because there isn't much else.

3)The Brennan report is incoherent because it's trying to be all things to all people. For example, the graphs of each independent variable over time were a waste of space. They were entertainment for lawyers or social scientists who aren't much into statistics. What I would have preferred to see was a nice, neat table with the regression coefficients and t stats for every regression run by the authors. Then I could quickly ascertain what the authors were doing. Put it in an appendix if you're afraid of scaring away the average reader. BTW, laying out the regressions would also make it easier to ponder if the authors are using a bunch of endogenous variables as explanatory variables. That's yet another econometrics/statistics problem.

4) So...my complaints about the Brennan Center paper are mostly about about clarity. There was too much literature review and not enough clear presentation the authors' own empirical work. The paper should have been cut waaaay down from 142 pages. It was a lot of squeeze for not much lemon. For the record, I'm pro-gun control.
posted by bim at 2:20 PM on May 11, 2015


it's irrelevant to the discussion at hand.

Not really, because some people are saying "this data is shit, can't trust it" and others are saying, "well this data is corroborated from other sources too, so its not really shit, and its better than you think"
posted by MisantropicPainforest at 3:05 PM on May 11, 2015 [3 favorites]


I blame the Beatles.

My parameter estimate for Steely Dan is twice as large.
posted by lathrop at 7:32 PM on May 11, 2015 [1 favorite]


« Older Scroll Back: The Theory and Practice of Cameras in...   |   "people just aren't talking about [HIV] anymore." Newer »


This thread has been archived and is closed to new comments