To disprove “scare headlines” about a rise in crime following the Ferguson agitation (which started 8/9/2014), the good folks at ESPN’s FiveThirtyEight have assembled homicide data for 59 of the 60 biggest cities in the country for 1/1/2014 through 8/8/2014 and for 1/1/2015 through 8/8/2015:
Scare Headlines Exaggerated The U.S. Crime Wave
A full list of the top 60 cities gives a more nuanced picture.
… The wave of crime-wave reporting began this spring with NPR, CNN, the BBC and USA Today, based on just a handful of cities. In August, after the Major Cities Chiefs Association (MCCA) announced results of a survey of a few dozen of its members about crime in their cities, many other outlets — including Reuters, Voice of America and Time — added to the chorus. And last week, The New York Times put the crime wave on its front page, saying that “cities across the nation are seeing a startling rise in murders.” The Times article was accompanied by a chart showing crime trends in 10 cities, which in turn provoked refutations from The Washington Post and the Marshall Project, based on fewer than 20 cities each.
I’ve taken FiveThirtyEight’s table of data and created the graph above, with cities that have seen an increase in homicide numbers in red and cities that have seen a decrease in green.
It looks like the “scare headlines” were right.
There’s always a lot of randomness in homicide stats, but consider this: the biggest decline in absolute number of homicides was in Boston, with 15 fewer dead bodies so far in 2015 than in the same time period in 2014. In contrast, among the cities with increases in homicide, a dozen cities have gone up more dead bodies than Boston went down (with Baltimore leading the way with 77 more homicides).
How about cities with changes, up or down, of at least 5 homicides? Three cities had declines of that magnitude or greater, 30 had increases.
Summing across all 59 cities, homicides were up 16%. That’s a scary 482 more dead bodies so far in 2015 v. 2014.
And as this graph shows, the trend is pretty widespread, especially in the blacker cities.
You could argue that the trend wouldn’t be quite so obvious if you left out the top two increases in homicides: Baltimore (Freddy Gray) and St. Louis (Michael Brown). But, of course, that just increases the evidence that this homicide wave is related to the campaign by Eric Holder and friends against the police.
#BlackLivesMatter = #BlackDeathsDon’tMatter
So how big is this change in homicides? The FiveThirtyEight guys go on at some length about statistical significance in each city, but here’s a neat example over almost exactly the same time period that helps you get a better feel for how confident you can be: according to the National Golf Foundation, the number of rounds of golf played in the first 7 months of 2015 is up 0.4% versus the first 7 months of 2014.
Does that mean the Golf Recession is finally over after a decade and a half?
Eh, hard to say. If you look at individual markets, you can see that many are up and many are down. A lot of golf rounds played statistics depend upon the weather, which the NGF report includes to help you interpret whether trends are climatic flukes or not. So far this year, the best market for golf has been the Twin Cities in Minnesota, which are up 14%, while the worst is Houston, down 13% (maybe due to oil prices being down?).
Milwaukee, which has had a 76% increase in homicides, is also up 10% in terms of rounds of golf. Perhaps nice weather is to blame for both?
My guess would be that homicides are more statistically volatile than rounds of golf because they are so few in number in many places. If Arlington, Texas is down from 8 murders in the first 7.25 months of 2014 to 4 in the same time period in 2015, well, randomness is probably playing a big role.
But, still … this graph suggests that our society insisting that blacks should be angry at law & order has led to more blacks killing each other in substantially black cities. No doubt that’s not the only reason for what we see on this graph, but it sure looks like the most obvious reason.
Hopefully, this trend won’t continue. People sometimes learn from past mistakes. In the mid-1990s, for example, a lot of people figured out that the trigger-happy crack-dealer lifestyle that seemed so glamorous on their gangsta rap albums wasn’t a good idea at all.
P.S. More on this topic here.
Save ListCancel
Save ListCancel
If someone went through the list and noted percentage of black citizens per city, it would also show that black citizens matter per increase or decrease of city homicide rate.
http://www.unz.com/isteve/homicides-2015-v-2014-there-seems-to-be-a-trend/#comment-1130394
#BlackLivesMatter = #BlackDeathsDon’tMatter
Since I rather scorn the closet, rather impetuously escape the closet for catharsis sake, and since its easy enough to google—when I was sixteen I got indicted for murder and plead guilty to manslaughter after a fist fight turned into a freak accident, an awful tragedy, and a five year sentence, plus a decade of probation. Lots of relevant things to say about my case (not the least of which is the unfair ease with which I was indicted for murder and by that certified as an adult, yet Michael Brown’s cop wasn’t indicted, not even indicted) but ultimately I wasn’t railroaded, though there was a purposively anti-lynching statue on the books under which I could have been, because everyone intuitively knew at the end of the day that the tragedy was not about malice, but more about boys being boys. And one expert said exactly that on the stand. My point is, there are good people who should be more generous about the optically jarring disconnect between blacks protesting cops and not protesting gangsters. We should understand these people by now. This is black boys being black boys, and most street murders are simply not about malice, they’re about status. And I just mentioned along these lines elsewhere, but black people have a very keen sense of tragedy, and not unoften welcome it, the noble intellection that there is a beauty to sad stories, and protesting fallen soldiers like the protests in Ferguson would ruin that solace for many of them. Anyway there’s not much to protest when its actually just black boys being black boys, because the killers are people they know, people they don’t consider evil, and they will never think of them like they think of white cops. And that’s not stupid of them. Nor does flaunting the hypocrisy of black lives matter illuminate something that people don’t already see.
No of course not. Cops are generally far safer to them than the people they know and treat them with more respect. Of course they should fear their own people more. They would be stupid not to.
It may be even worse than your chart suggests.
Boston (-15 homicides for 2015) may be the outlier for one simple reason: snow.
Boston winter ( which includes the first three months of 2015) set a record for snowfall.
It is hard for thugs to be out shooting each other when there’s a foot of snow on the ground.
It would be interesting to stratify Boston homicides by month and see if the (-15) can be attributed to deviation from monthly norms in Jan, Feb and March.
I wouldn’t be surprised if those months accounted for most of the deviation.
http://www.weather.com/news/news/new-england-boston-record-snow-tracker
http://www.weather.com/news/news/new-england-boston-record-snow-tracker
Milwaukee has had unusually nice weather, I believe.
http://www.aos.wisc.edu/~sco/clim-history/stations/mke/mke-rts-2015.gif
The UC Berkeley student body is only 4 percent Black and this qualifies as the biggest crisis facing The United States in 2015 according to the racial grievance left wing megaphone.
http://news.yahoo.com/uc-berkeley-plan-boost-black-student-numbers-affirmative-230143234.html
The only explanation for the low single digit percentage of Blacks at UC Berkeley is obvious, White supremacists run the city of Berkeley with an iron fist.
As TNC would likely say, 96 percent of The UC Berkeley student body consists of people who think they are either White or some other type of Nonblack.
Turns out your belief is correct:
http://www.aos.wisc.edu/~sco/clim-history/stations/mke/mke-rts-2015.gif
This increase in homicide is 10x bigger than the increase in rounds of golf played.
City / Black %
Baltimore 63
St Louis 49
Chicago 33
Milwaukee 40
Houston 24
Washington 50
New Orleans 60
Louisville 23
USA 13
Fresno 8
Miami 19
Arlington TX 19
Riverside 7
Las Vegas 11
Memphis 63
Boston 24
About those 2 Black women in Seattle who verbally harassed poor weak frail grandpa Bernie Sanders, it would be some major irony if in the future one of their Black male family members ends up being murdered by another Black male. Knowing them they would somehow find a way to still blame it on White people. Big Worm would not have murdered my cousin T-Dawg if the White man would have just hooked Big Worm up with a job. It’s institutional racism by the White man to keep Black male bodies unemployed. White employers throwing Black people’s resumes in the trash can is the cause for Black on Black murder.
Irony? It's a near certainty if you include step-siblings, cousins to the second or third degree and relations by marriage. Somebody they know will be killed by another young black male. And they'll either blame the white man or the gun or drugs or a youth that went wrong. Never will they look at the real reason.
http://www.aos.wisc.edu/~sco/clim-history/stations/mke/mke-rts-2015.gif
Total rounds of golf in the U.S. were up 1.6% in the first half of 2015 over the first half of 2014, which is pretty rare in this century. But in local markets, number of rounds of golf played goes up and down a lot due to the weather. The North Central has had a big year for golf, but that’s probably due to weather.
This increase in homicide is 10x bigger than the increase in rounds of golf played.
Your next task is to look at the correlation between City deaths and GDP per capita over the past decade. It may well be positive, that of golf games is likely negative.
If bone chilling icebox weather prevents most Black thugs from their daily mugging routine, than the safest time for White people to go to Detroit for example is when it’s like 20F degrees outside. Most Black thugs will prefer to stay inside their warm housing project apartments and catch up with some episodes of Empire.
I’m doing one of those off-topic comments where that people seem to be doing here, bringing up some article or text from the left and hoping that Steve will take it on.
http://www.theguardian.com/commentisfree/2015/sep/10/end-online-comments
This is just the latest volley in the left’s icreasingly intense war on comment sections. Moral posturing about what a great person you are for trying to shut down your opponents instead of actually dealing with their arguments. I’m Swedish so I’m well aware of this phhnomena. Today it’s nigh impossible to find any article on immigration written in any Swedish establishment media that you can comment on. The Guardian still has comment sections for articles on immigration, but not for much longer I would expect.
Update from Sweden for those interested. Three different polling organizations have now confirmed the Swedish Democrats as the biggest party. Latest poll released after a massive propaganda blitz coinciding with the drowned boy in Turkey.
I am often surprised to see comments sections so diametrically opposed to the articles they are commenting on when I mistakenly assume the commenters are the customers. The media knows they aren't the customers I guess.
By the way, I loved this recent Swedish movie:
http://youtu.be/bS5P_LAqiVg
But who is the director/main character?
He doesn't look Swedish.
This increase in homicide is 10x bigger than the increase in rounds of golf played.
“This increase in homicide is 10x bigger than the increase in rounds of golf played”: a Sailerism for the centuries.
I found Sailer's comparison between the increase in golf rounds and the increase in homocides very interesting but shallow. I wonder why he didn't more deeply explore the golf/homicide connection by telling us whether the number of holes-in-one has changed substantially over the period. Maybe that's where the connection lies. Or bogies. (This is just a WAG, but maybe it's all those frustrated golfers who go out and murder people. I have seen numerous instances of such behavior on televised golf tournaments where a pro golfer gets so frustrated by a bad shot that he breaks a perfectly good gulf club. Who knows what such individual is capable of once the round is over and the TV cameras are off? ) Or whether handicaps have gone up or down. It looks like he might be working on a proposal for a federal grant to study the issue, in which case I would suggest broadening the study to include a comparison of birdies and eagles. Once those federal dudes see that wildlife is involved, I think the proposal will win speedy approval.
I did that analysis. There is little relation between increases in homicide and the share of blacks in a city’s population. The correlation is 0.18 (p=0.163, N=59). Controlling for population size or absolute number of homicides doesn’t change the results.
The correlation between % black and number of homicides in in these cities in 2015 is 0.56. Controlling for population size, the correlation is 0.69.
The data, with population figures from Wikipedia, are here. Copy and paste them to a text file and it should open in Excel.
If you want to know the crime level of a city, you find out what percentage of it is black. Fits the data like a glove.
Maybe you should review their studies and see why they differ?
Additionally why don't you tell us what methodology you used? That's a lot more useful than a link to a spreadsheet. I suspect you have some holes in your model.
I calculated homicide rates and the change from 2014 to 2015. The change in homicide rates correlated 0.45 with share of blacks in a city’s population. I think looking at the change in rates rather than change in the absolute number of homicides is more appropriate.
I think this contradicts your assertion "There is little relation between increases in homicide and the share of blacks in a city’s population." For example, see http://sites.stat.psu.edu/~jls/stat100/lectures/lec16.pdf which states:For more fun I fit some linear regression models. Regressing the rate change on the share of blacks the R^2 was 0.2 (=0.45^2 which makes sense).
I tried adding a flag for cities which had incidents in 2015 (i.e. Baltimore and St. Louis) and was shocked to see the R^2 increase to 0.68 which is amazingly high for such a simple model in the social sciences. Here is the R output (a p value for the model of 1.2e-14 is rather convincing IMHO):What do you think? Again, thanks for the data!
P.S. Steve, if you would be interested in following up on this I can email you the R markdown code and/or the generated HTML or PDF.
P.P.S. There was a time when I looked to 538 for interesting analytical work. It seems they have given up on that and are now fully committed to narrative damage control.
If you did, it would be a way to test if the increase in homicides was due to lower policing intensity as opposed to other factors.
If arrest rates went down, while homicide rates went up, it would be strong evidence that a decline in policing intensity caused the increase in homicides. It would contradict the hypothesis that the homicide increase is due to other causes, because if the increase were due to other causes, we would expect an INCREASE in arrests.
This is a pretty standard type of media story–there’s an apparent statstical blip that could be spun to support a story you’d like to tell, so you jump in and tell it. It’s broadly possible that the rise in homicide in some big cities is due to the backlash against police shootings. It’s also possible it’s a completely unrelated blip, caused by some other thing–a change in the drug market, an unusually hot summer, whatever. (My guess is there’s a real effect in Baltimore and St Louis, and probably the rest is random fluctuation in the numbers.)
In general, crime rates go their own way, paying little attention to most policy changes intended to lower them–I think to reliably have an impact, you have to do something very big like doubling the number of people you put in prison.
I’ve decided that the earnest white progressives (if there are any) who see these charts have to latch onto nebulous ideas like “structural” or “systemic” racism or else they would have no other choice than to stop being progressives. Anytime real world data blows up some abstract progressive concept, they simply double down on the latest unverifiable nonsense to come out of academia or the media. That’s why the data wonks like 538 are so fascinating, because their editorially meetings must be a total minefield of progressive thoughtcrime to navigate.
Funny La Griffe du Lion and the host of this site (Ron Unz) did similar analyses, and came to exactly the opposite conclusion. In general crime in America is linked to black people.
If you want to know the crime level of a city, you find out what percentage of it is black. Fits the data like a glove.
Maybe you should review their studies and see why they differ?
Additionally why don’t you tell us what methodology you used? That’s a lot more useful than a link to a spreadsheet. I suspect you have some holes in your model.
http://news.bbc.co.uk/1/hi/uk/607623.stm
After the trial of whites (known thugs) for a street killing of a black teenager collapsed, the British home secretary Jack Straw forced (threat of sacking) the London police commissioner to publically confess the police were institutionally racist. There was a substantial increase in crime. But it will be OK to set up a special squad to tackle black on black crime, London had one called Operation Trident. Black on white crime isn’t so serious, obviously.
I wonder why they didn’t compare 8/9/13 – 8/8/14 to 8/9/14 – 8/8/15?
Why leave out the almost 4 full months 8/9/14 – 12/31/14?
Why is the weather always dragged into all this? I’ve experienced many very hot summer days yet never once thought that shooting at someone would be a good way to express myself. Nicholas Wade and others have shown that primitive tribals engage in a lot of warfare and have high death rates as a result. This is what we have, tribals living in our midst doing what they do. We’re there to pay the medical bills and put it on our nightly news.
These developments involve narcotics business.
In particular, Michael Brown in Ferguson and Freddie Gray in Baltimore were acting in narcotics business when their fatal incidents occurred.
The neighborhoods where the incidents occurred are saturated with residents who are involved in narcotics business. Those particular residents are highly motivated to intimidate, discredit and attack police officers. Those particular residents have the free time and the angry energy to protest hysterically and endlessly after an incident. Many of them escalate their protests by terrorizing their law-abiding neighbors and by looting and burning local businesses and property.
Feeding on this situation is a parasitical group of local lawyers who make their livings as public defenders of drug traffickers. These low-class lawyers aspire to to strike it rich by winning huge lawsuits against the local law-enforcement establishments. These lawyers use the mass media to agitate liberal citizens and politicians.
These various actors work together to make narcotics laws impossible to enforce.
The major consequence is that the worst neighborhoods become even more saturated with narcotics and intimidated by narcotics traffickers. All the following phenomena increase:
* Shop-lifting
* Theft
* Robbery
* Violence
* Murder
* Incarceration
* Single motherhood
* Poverty
* Failure of legitimate business
* Unemployment
* Government bankruptcy
Groan, what an exercise in statistical sophistry. Yes, at the city-by-city level, many of the homicide increases are indeed not statistically significant. The overall increase, however, is extremely, stupendously, screamingly statistically significant. By my calculation, 2015′s total is almost nine standard deviations (!!) above 2014′s.
Trigger warning: don’t read any further if your last statistics course gave you PTSD.
The statistical model that applies to rare events like homicides is called the Poisson distribution. (I’m certain it’s what the author’s stats geek Bruce Frederic (who “outlined his methodology in an email”) used.) For the Poisson distribution, one standard deviation is the square root of the number of incidents. Thus Atlanta’s rise from 45 in 2014 to 52 in 2015 represents a (52-45)/sqrt(45)=1.04 standard deviation increase. That sort of thing will happen by chance about 30% of the time, which is not considered statistically significant (in the article, they took “significant” to mean, “won’t happen more than 5% of the time”).
But now let’s look at the bottom line. Adding up the city-by-city numbers I see there were 3437 murders in 2015, as against 2955 in 2014. Number of standard deviations difference: (3437-2955)/sqrt(2955) = 8.9. So there you go. Statistically significant at the, who knows, one-googleth percent level.
In the service of his plainly ideological motive, the author could have (albeit after more effort than he was willing to devote to this thing) presented the numbers broken out not by city but by zip code. Then, because individual zip codes will have usually just one or two murders, he could have reported that a mere 1% (or something) of his zip code-level murder increases were statistically significant.
One more technical note. Some of you will have observed that the formula I gave above for standard deviations of the Poisson distribution makes no sense when you get to zero; in a town with 4 murders per year, the standard deviation is 2, which sounds like I’m saying there’s about a 2.3% chance to observe a negative number of murders next year! OK, the deal is this: the Poisson distribution is not the Normal (aka Gaussian aka Bell-curve) distribution. So the percentiles that correspond to 1, 2, 3 etc standard deviations away from the mean are not the same. However, they are very close when the mean is moderate or large — roughly, twenty or more. And when the number is in the thousands (e.g. the 2955 murders in January-August 2014), the two distributions are for all practical purposes identical.
Trigger warning: don't read any further if your last statistics course gave you PTSD.
The statistical model that applies to rare events like homicides is called the Poisson distribution. (I'm certain it's what the author's stats geek Bruce Frederic (who "outlined his methodology in an email") used.) For the Poisson distribution, one standard deviation is the square root of the number of incidents. Thus Atlanta's rise from 45 in 2014 to 52 in 2015 represents a (52-45)/sqrt(45)=1.04 standard deviation increase. That sort of thing will happen by chance about 30% of the time, which is not considered statistically significant (in the article, they took "significant" to mean, "won't happen more than 5% of the time").
But now let's look at the bottom line. Adding up the city-by-city numbers I see there were 3437 murders in 2015, as against 2955 in 2014. Number of standard deviations difference: (3437-2955)/sqrt(2955) = 8.9. So there you go. Statistically significant at the, who knows, one-googleth percent level.
In the service of his plainly ideological motive, the author could have (albeit after more effort than he was willing to devote to this thing) presented the numbers broken out not by city but by zip code. Then, because individual zip codes will have usually just one or two murders, he could have reported that a mere 1% (or something) of his zip code-level murder increases were statistically significant.
One more technical note. Some of you will have observed that the formula I gave above for standard deviations of the Poisson distribution makes no sense when you get to zero; in a town with 4 murders per year, the standard deviation is 2, which sounds like I'm saying there's about a 2.3% chance to observe a negative number of murders next year! OK, the deal is this: the Poisson distribution is not the Normal (aka Gaussian aka Bell-curve) distribution. So the percentiles that correspond to 1, 2, 3 etc standard deviations away from the mean are not the same. However, they are very close when the mean is moderate or large -- roughly, twenty or more. And when the number is in the thousands (e.g. the 2955 murders in January-August 2014), the two distributions are for all practical purposes identical.
To double-check that the author relied on the Poisson model, the way I described it, I counted how many of his 59 cities saw an increase of 1.645 standard deviations (1.645 being the 95th percentile of a one-sided Normal distribution) and I got the same number he reported — 16.
Not that there was much doubt in the first place about what he did. As he says,
Of course this isn’t the only way to look at murder numbers. He could, instead, have looked at the standard deviation of the year-over-year changes going back, say 20 or 30 years. I imagine he didn’t do that, at least in part, because, as he says, it was hard enough to gather just one year of January-through-August data from each city. Another reason not to do it that way is that you’d be assuming the underlying “process” (statistics speak) driving the data is the same in every year — which clearly is not the case, what with the rise and fall of the crack epidemic, changes in the willingness to incarcerate, etc.
When I saw Boston with the biggest homicide drop, I was thinking of what might be the reason. Then I remembered – I live in New York, which had a relatively snowy and very cold winter. But every time I watched the news, somebody was up in Boston talking about an additional two feet of snow. The winter was so bad up there that some snowbanks didn’t melt until the summer. Snow cover was in the 3-4 feet range for much of January and February, which probably makes it harder to run from the cops or get away after a violent crime.
I don’t know how the winter was in the Midwest, but since I didn’t hear about it, I figure it was probably pretty ordinary, which means cold and snowy but not what Boston had.
Baltimore and St. Louis, the two cities most identified with Obama’s riots and “Black Lives Matter”, had the biggest increases in homicides. It’s no coincidence, and GOP presidential candidates need to start talking about it.
You sure? I ran your numbers and got a correlation of 0.42. (I.e. that’s the correlation between your column 5 and the difference between your columns 3 and 2.) The T statistic on that 0.42 correlation is 3.47, i.e. significant at the 99.99% level.
Just on common sense grounds, it would be surprising if blacks, who commit the lion’s share of the murders in our cities, didn’t also commit the lion’s share of an increase in those murders. For that to happen, whites and others would have to step in and up their game just in time for the crime increase.
In cities like Detroit and Atlanta, Black males pretty much have a monopoly on being perpetrators of murder. So it is going to be extremely hard for White thugs there to step up their murder game.
Blacks have a monopoly on committing murder in the ATL and the Motor City like Asians have a monopoly on laundromat ownership in San Francisco.
In general, crime rates go their own way, paying little attention to most policy changes intended to lower them--I think to reliably have an impact, you have to do something very big like doubling the number of people you put in prison.
It’s more than a blip though. I thought it was a blip too, and I went scouring through uniform crime reports data to find evidence that calling it a “crime wave” was too sensationalist. But the fact is a 16% increase is significantly greater than most upticks in the midst of downward trends – not just in the most recent long-term downtrend but shorter downtrends that have occurred since 1960 (in state-level data). Hopefully there’s a sizable element of randomness in play here that peters out over the remainder of this year, but if the jump is more directly related to the climate of hysteria generated by leftist anti-white and anti-cop paranoia, and if this continues, then in a year or two you could really see some fireworks.
““This increase in homicide is 10x bigger than the increase in rounds of golf played”: a Sailerism for the centuries.”
I found Sailer’s comparison between the increase in golf rounds and the increase in homocides very interesting but shallow. I wonder why he didn’t more deeply explore the golf/homicide connection by telling us whether the number of holes-in-one has changed substantially over the period. Maybe that’s where the connection lies. Or bogies. (This is just a WAG, but maybe it’s all those frustrated golfers who go out and murder people. I have seen numerous instances of such behavior on televised golf tournaments where a pro golfer gets so frustrated by a bad shot that he breaks a perfectly good gulf club. Who knows what such individual is capable of once the round is over and the TV cameras are off? ) Or whether handicaps have gone up or down. It looks like he might be working on a proposal for a federal grant to study the issue, in which case I would suggest broadening the study to include a comparison of birdies and eagles. Once those federal dudes see that wildlife is involved, I think the proposal will win speedy approval.
I am now firmly convinced that Steve Sailer is truly a prophet, right up there with all those prophets in the Old Testament.
Thanks for the data!
I calculated homicide rates and the change from 2014 to 2015. The change in homicide rates correlated 0.45 with share of blacks in a city’s population. I think looking at the change in rates rather than change in the absolute number of homicides is more appropriate.
I think this contradicts your assertion “There is little relation between increases in homicide and the share of blacks in a city’s population.” For example, see http://sites.stat.psu.edu/~jls/stat100/lectures/lec16.pdf which states:
For more fun I fit some linear regression models. Regressing the rate change on the share of blacks the R^2 was 0.2 (=0.45^2 which makes sense).
I tried adding a flag for cities which had incidents in 2015 (i.e. Baltimore and St. Louis) and was shocked to see the R^2 increase to 0.68 which is amazingly high for such a simple model in the social sciences. Here is the R output (a p value for the model of 1.2e-14 is rather convincing IMHO):
What do you think? Again, thanks for the data!
P.S. Steve, if you would be interested in following up on this I can email you the R markdown code and/or the generated HTML or PDF.
P.P.S. There was a time when I looked to 538 for interesting analytical work. It seems they have given up on that and are now fully committed to narrative damage control.
P originally looked at the correlation between percentage black and percentage change in homicides.
But you look at the correlation between percentage black and the ABSOLUTE change in homicide rates rather than the PERCENTAGE change in homicide rates.
I am concerned that your results are simply due to the fact that homicide rates are correlated with percentage black.
Take an extreme example, two cities, each with a population of 1,000, one all-black with 100 homicides per year and one all-non-black with 10 homicides per year. Homicide rates are 10% and 1% respectively. Homicides in both double (to 200 and 2) for reasons unrelated to the riots (such as better weather). Therefore, homicide rates in both double (to 20% and 2%), and the PERCENTAGE change in crime rates is the same (both increase 100%). But the ABSOLUTE change is much bigger in the black city (an increase of 10% versus 1%).
Therefore, finding that the ABSOLUTE change in crime rates is higher in blacker cities does NOT prove that the increase in crime rates was not due to the weather.
If, however, the changes are due to less intensive policing of black cities for fear of causing riots, then we would expect to see a difference in the PERCENTAGE change in crime rates.
Hence, we want to look at the correlation of the percentage change in crime rates, not the absolute change in crime rates.
Now maybe, police are policing everyone less intensively and so crime is going up everywhere. This is possible, but we can't test that using this data. We need to look at some measure of policing intensity.
In general, crime rates go their own way, paying little attention to most policy changes intended to lower them--I think to reliably have an impact, you have to do something very big like doubling the number of people you put in prison.
See my analysis in another comment. The Baltimore/St. Louis effect is larger, but there looks to also be another real effect which can be modeled using share of black population as a predictive variable.
The Pearson correlation of 0.18 I reported is between the CHANGE column and the BLACK_POP column. The Spearman correlation between the two is 0.22 (p=0.09). The CHANGE column indicates the percent change in homicide between 2014 and 2015.
It’s true that the correlation between the differences in absolute numbers of homicides and BLACK_POP is 0.42 (p=0.01). Controlling for total population the correlation is 0.43.
I calculated homicide rates and the change from 2014 to 2015. The change in homicide rates correlated 0.45 with share of blacks in a city’s population. I think looking at the change in rates rather than change in the absolute number of homicides is more appropriate.
I think this contradicts your assertion "There is little relation between increases in homicide and the share of blacks in a city’s population." For example, see http://sites.stat.psu.edu/~jls/stat100/lectures/lec16.pdf which states:For more fun I fit some linear regression models. Regressing the rate change on the share of blacks the R^2 was 0.2 (=0.45^2 which makes sense).
I tried adding a flag for cities which had incidents in 2015 (i.e. Baltimore and St. Louis) and was shocked to see the R^2 increase to 0.68 which is amazingly high for such a simple model in the social sciences. Here is the R output (a p value for the model of 1.2e-14 is rather convincing IMHO):What do you think? Again, thanks for the data!
P.S. Steve, if you would be interested in following up on this I can email you the R markdown code and/or the generated HTML or PDF.
P.P.S. There was a time when I looked to 538 for interesting analytical work. It seems they have given up on that and are now fully committed to narrative damage control.
Yeah, the change in homicide rates is probably a better measure than the proportional change in absolute homicide numbers. Using it, I get r=0.45, too. So, I withdraw the claim that there’s little relation between % black and increases in homicide.
I reran your analysis using change in homicide rates instead of change in absolute number of homicides and got the same correlation (.18). Moreover, this is what you SHOULD get because you are just dividing the 2014 and the 2015 homicide rates by the same number (population of the city).
So, I don't understand where a correlation of .45 comes from.
But, I don't believe it is the proper correlation to look at, which I will explain in a post responding to @res.
If you want to know the crime level of a city, you find out what percentage of it is black. Fits the data like a glove.
Maybe you should review their studies and see why they differ?
Additionally why don't you tell us what methodology you used? That's a lot more useful than a link to a spreadsheet. I suspect you have some holes in your model.
If you read my comment again, you’ll see that I noted that there’s a correlation of 0.56 or 0.69 between the number of homicides and the percentage of black residents. So my analysis doesn’t contradict Griffe’s or Unz’s.
The particular claim I made was that the recent increase in homicides wasn’t related to % black residents. That’s a separate issue. However, the variable I used to examine this may have been faulty, see res’s comment.
As for the broader race/crime correlation, I covered the topic exhaustively in a long paper a couple of years ago, which you might want to examine. Here's the link to the section focusing on the correlations:
http://www.unz.com/article/race-and-crime-in-america/#the-pattern-of-urban-crime-in-america
And here's one of my numerous charts, showing the homcide correlations for the last quarter-century:
Incidentally, note that the results are aggregated by entire city rather than by zip code, which would be far more accurate but extremely difficult to calculate. Therefore, my results must enormously understate the true race/crime relationship, lumping together the Upper East Side and the South Bronx. My guess would be that a zip-code based race/crime correlation would be somewhere between 0.90 and 0.95, certainly the highest ever found anywhere in the social sciences...
“Just on common sense grounds, it would be surprising if blacks, who commit the lion’s share of the murders in our cities, didn’t also commit the lion’s share of an increase in those murders. For that to happen, whites and others would have to step in and up their game just in time for the crime increase.”
In cities like Detroit and Atlanta, Black males pretty much have a monopoly on being perpetrators of murder. So it is going to be extremely hard for White thugs there to step up their murder game.
Blacks have a monopoly on committing murder in the ATL and the Motor City like Asians have a monopoly on laundromat ownership in San Francisco.
For the Michael Brown and Freddie Gray affairs to be the cause of increasing crime rates, police departments must have become less aggressive in their policing. I do not know if there is any evidence to that effect. Or if it’s even something that can be quantified.
Peter
No, it would only require that the hubbub surrounding them made blacks more angry and murder-prone. Also, the robbery of 300,000 doses of opiate painkillers under cover of the Baltimore riots no doubt contributed to fights between gangs.
If arrest rates have gone down, while homicide rates went up, that would be strong evidence that it was a decline in policing intensity caused the increase in homicides. It would contradict Harry Baldwin's hypothesis that the homicide increase is due to other causes, because if the increase were due to other causes, we would expect an INCREASE in arrests.
Peter
For the Michael Brown and Freddie Gray affairs to be the cause of increasing crime rates, police departments must have become less aggressive in their policing.
No, it would only require that the hubbub surrounding them made blacks more angry and murder-prone. Also, the robbery of 300,000 doses of opiate painkillers under cover of the Baltimore riots no doubt contributed to fights between gangs.
This increase in homicide is 10x bigger than the increase in rounds of golf played.
It’s cheaper to kill someone than pay green fees.
Your next task is to look at the correlation between City deaths and GDP per capita over the past decade. It may well be positive, that of golf games is likely negative.
The kids couldn’t hurt Crack, they tried and tried and tried
They dropped things on his back, and lied and lied and lied and lied and lied…
I calculated homicide rates and the change from 2014 to 2015. The change in homicide rates correlated 0.45 with share of blacks in a city’s population. I think looking at the change in rates rather than change in the absolute number of homicides is more appropriate.
I think this contradicts your assertion "There is little relation between increases in homicide and the share of blacks in a city’s population." For example, see http://sites.stat.psu.edu/~jls/stat100/lectures/lec16.pdf which states:For more fun I fit some linear regression models. Regressing the rate change on the share of blacks the R^2 was 0.2 (=0.45^2 which makes sense).
I tried adding a flag for cities which had incidents in 2015 (i.e. Baltimore and St. Louis) and was shocked to see the R^2 increase to 0.68 which is amazingly high for such a simple model in the social sciences. Here is the R output (a p value for the model of 1.2e-14 is rather convincing IMHO):What do you think? Again, thanks for the data!
P.S. Steve, if you would be interested in following up on this I can email you the R markdown code and/or the generated HTML or PDF.
P.P.S. There was a time when I looked to 538 for interesting analytical work. It seems they have given up on that and are now fully committed to narrative damage control.
I’m nobody’s statistician, but I imagine what makes it tricky is that even within high homicide cities the bulk of the killing is confined to a few distinct neighborhoods. Or at least that’s my pretty strong impression: Cabrini Green in Chicago, the Magnolia Street projects in New Orleans, etc. I’d bet once you control for not just city, but neighborhood the correlation would be plain as day.
Presidential rounds of golf have shown a steady increase over the past several years.
http://obamagolfcounter.com/
Perhaps that is a leading indicator of national homicide rate.
Baltimore 63
St Louis 49
Chicago 33
Milwaukee 40
Houston 24
Washington 50
New Orleans 60
Louisville 23
USA 13
Fresno 8
Miami 19
Arlington TX 19
Riverside 7
Las Vegas 11
Memphis 63
Boston 24
Nevermind Las Vegas, What the Heck is up with Memphis. And how to replicate the heck out of that?
Your next task is to look at the correlation between City deaths and GDP per capita over the past decade. It may well be positive, that of golf games is likely negative.
But it’s also cheaper to kill him with a golf club than with a .45 iron.
Come to think of it, baseball bats are illegal in the UK, aren’t they? But not golf clubs. Hmm…
Another factor to conflate with rising crime: bleeding hearts.
http://www.judicialwatch.org/press-room/press-releases/judicial-watch-san-francisco-crime-surge-homicide-arrests-up-55-rapes-arrests-up-370/
I don’t understand this.
I reran your analysis using change in homicide rates instead of change in absolute number of homicides and got the same correlation (.18). Moreover, this is what you SHOULD get because you are just dividing the 2014 and the 2015 homicide rates by the same number (population of the city).
So, I don’t understand where a correlation of .45 comes from.
Peter
A way to test this hypothesis is to look at changes in arrests.
If arrest rates have gone down, while homicide rates went up, that would be strong evidence that it was a decline in policing intensity caused the increase in homicides. It would contradict Harry Baldwin’s hypothesis that the homicide increase is due to other causes, because if the increase were due to other causes, we would expect an INCREASE in arrests.
Do you have any data on arrest rates (or some other measure of policing intensity)?
If you did, it would be a way to test if the increase in homicides was due to lower policing intensity as opposed to other factors.
If arrest rates went down, while homicide rates went up, it would be strong evidence that a decline in policing intensity caused the increase in homicides. It would contradict the hypothesis that the homicide increase is due to other causes, because if the increase were due to other causes, we would expect an INCREASE in arrests.
http://www.theguardian.com/commentisfree/2015/sep/10/end-online-comments
This is just the latest volley in the left's icreasingly intense war on comment sections. Moral posturing about what a great person you are for trying to shut down your opponents instead of actually dealing with their arguments. I'm Swedish so I'm well aware of this phhnomena. Today it's nigh impossible to find any article on immigration written in any Swedish establishment media that you can comment on. The Guardian still has comment sections for articles on immigration, but not for much longer I would expect.
Update from Sweden for those interested. Three different polling organizations have now confirmed the Swedish Democrats as the biggest party. Latest poll released after a massive propaganda blitz coinciding with the drowned boy in Turkey.
I think the Sweden Democrat’s high polling is one of the things that made the EU decide they needed a mass invasion before the anti-EU parties got too strong.
That is interesting – a mystery.
I wonder what the age distribution is in those towns with a large black percentage but a low increase.
If black flight from hispanic immigration works the same way as white flight then it’s the parental age people who go leaving the older people behind.
(Hence less young men.)
In general, crime rates go their own way, paying little attention to most policy changes intended to lower them--I think to reliably have an impact, you have to do something very big like doubling the number of people you put in prison.
Or that the rise in homicides is the result of a rise in background anger caused by the media’s incitement.
I was confused in my earlier post questioning the .45 correlation. Please ignore it. I now understand where it comes from.
But, I don’t believe it is the proper correlation to look at, which I will explain in a post responding to .
should have read whole thread before commenting – apparently there’s no mystery after all
I calculated homicide rates and the change from 2014 to 2015. The change in homicide rates correlated 0.45 with share of blacks in a city’s population. I think looking at the change in rates rather than change in the absolute number of homicides is more appropriate.
I think this contradicts your assertion "There is little relation between increases in homicide and the share of blacks in a city’s population." For example, see http://sites.stat.psu.edu/~jls/stat100/lectures/lec16.pdf which states:For more fun I fit some linear regression models. Regressing the rate change on the share of blacks the R^2 was 0.2 (=0.45^2 which makes sense).
I tried adding a flag for cities which had incidents in 2015 (i.e. Baltimore and St. Louis) and was shocked to see the R^2 increase to 0.68 which is amazingly high for such a simple model in the social sciences. Here is the R output (a p value for the model of 1.2e-14 is rather convincing IMHO):What do you think? Again, thanks for the data!
P.S. Steve, if you would be interested in following up on this I can email you the R markdown code and/or the generated HTML or PDF.
P.P.S. There was a time when I looked to 538 for interesting analytical work. It seems they have given up on that and are now fully committed to narrative damage control.
I disagree. I don’t think you are using the right variables.
P originally looked at the correlation between percentage black and percentage change in homicides.
But you look at the correlation between percentage black and the ABSOLUTE change in homicide rates rather than the PERCENTAGE change in homicide rates.
I am concerned that your results are simply due to the fact that homicide rates are correlated with percentage black.
Take an extreme example, two cities, each with a population of 1,000, one all-black with 100 homicides per year and one all-non-black with 10 homicides per year. Homicide rates are 10% and 1% respectively. Homicides in both double (to 200 and 2) for reasons unrelated to the riots (such as better weather). Therefore, homicide rates in both double (to 20% and 2%), and the PERCENTAGE change in crime rates is the same (both increase 100%). But the ABSOLUTE change is much bigger in the black city (an increase of 10% versus 1%).
Therefore, finding that the ABSOLUTE change in crime rates is higher in blacker cities does NOT prove that the increase in crime rates was not due to the weather.
If, however, the changes are due to less intensive policing of black cities for fear of causing riots, then we would expect to see a difference in the PERCENTAGE change in crime rates.
Hence, we want to look at the correlation of the percentage change in crime rates, not the absolute change in crime rates.
Now maybe, police are policing everyone less intensively and so crime is going up everywhere. This is possible, but we can’t test that using this data. We need to look at some measure of policing intensity.
IMHO it is silly to try to decide on a single underlying reason. It is likely both decreased police presence and increased violence (and weather etc.) are relevant to a full explanation of the change. Of course, the key point is there IS a statistically significant change despite 538's analysis. I think the easiest way to show this is a paired t-test on the homicide rates for 2014 and 2015, but I would be interested in hearing other ideas.
When I do a paired t-test on the rates I get:
t = -3.117, df = 58, p-value = 0.002841
which is actually less significant than for the absolute numbers:
t = -3.7807, df = 58, p-value = 0.0003717
I was surprised to find that performing the paired t-test on rates excluding Baltimore and St.Louis gave a more significant result:
t = -3.2617, df = 56, p-value = 0.001888
It looks to me like removing the outliers decreased the SD by enough to overcompensate for the decrease in mean change.You raise a good point about the importance of this correlation. It turns out replacing percentage black with the 2014 homicide rate actually increases the R^2 of the model to 0.72
(this would be the narrative approved version of the model ;-)
This makes sense given that the 2014 homicide rate is probably a better measure of underlying violence than percentage black.
This of course raises the question of why the correlation exists (perhaps Ta-Nehisi Coates will address this in “The Enduring Myth of Black Criminality”). I suppose blacks might be preferentially attracted to areas with high criminality? (snark aside I suspect this is part of the picture, but unlikely to be the dominant cause)
The funny thing about this is the correlations we see here (e.g. see Ron's plot above) would unleash a flood of publications if they were about a PC topic given how high these correlations are compared to what is typically seen in social science.
P.S. @P thanks for both supplying the data and engaging in a civil disagreement. I would be interested in hearing your further thoughts on all of this.
http://www.theguardian.com/commentisfree/2015/sep/10/end-online-comments
This is just the latest volley in the left's icreasingly intense war on comment sections. Moral posturing about what a great person you are for trying to shut down your opponents instead of actually dealing with their arguments. I'm Swedish so I'm well aware of this phhnomena. Today it's nigh impossible to find any article on immigration written in any Swedish establishment media that you can comment on. The Guardian still has comment sections for articles on immigration, but not for much longer I would expect.
Update from Sweden for those interested. Three different polling organizations have now confirmed the Swedish Democrats as the biggest party. Latest poll released after a massive propaganda blitz coinciding with the drowned boy in Turkey.
Ive seen Trump referred to as the “comment section” running for president, and it made sense to me.
I am often surprised to see comments sections so diametrically opposed to the articles they are commenting on when I mistakenly assume the commenters are the customers. The media knows they aren’t the customers I guess.
By the way, I loved this recent Swedish movie:
But who is the director/main character?
He doesn’t look Swedish.
Well, I haven’t looked at the numbers for this particular matter, but just eye-balling the chart at the top, most of the cities showing decreased homicide rates (including my own San Jose) have few blacks, while most of the ones showing big increases are heavily black.
As for the broader race/crime correlation, I covered the topic exhaustively in a long paper a couple of years ago, which you might want to examine. Here’s the link to the section focusing on the correlations:
http://www.unz.com/article/race-and-crime-in-america/#the-pattern-of-urban-crime-in-america
And here’s one of my numerous charts, showing the homcide correlations for the last quarter-century:
Incidentally, note that the results are aggregated by entire city rather than by zip code, which would be far more accurate but extremely difficult to calculate. Therefore, my results must enormously understate the true race/crime relationship, lumping together the Upper East Side and the South Bronx. My guess would be that a zip-code based race/crime correlation would be somewhere between 0.90 and 0.95, certainly the highest ever found anywhere in the social sciences…
because the killers are people they know, people they don’t consider evil, and they will never think of them like they think of white cops. And that’s not stupid of them.
No of course not. Cops are generally far safer to them than the people they know and treat them with more respect. Of course they should fear their own people more. They would be stupid not to.
About those 2 Black women in Seattle who verbally harassed poor weak frail grandpa Bernie Sanders, it would be some major irony if in the future one of their Black male family members ends up being murdered by another Black male
Irony? It’s a near certainty if you include step-siblings, cousins to the second or third degree and relations by marriage. Somebody they know will be killed by another young black male. And they’ll either blame the white man or the gun or drugs or a youth that went wrong. Never will they look at the real reason.
Because they kill each other more during hot weather. They stay up later on hot nights and are more irritable. Most people are but then most people have better impulse control.
As for the broader race/crime correlation, I covered the topic exhaustively in a long paper a couple of years ago, which you might want to examine. Here's the link to the section focusing on the correlations:
http://www.unz.com/article/race-and-crime-in-america/#the-pattern-of-urban-crime-in-america
And here's one of my numerous charts, showing the homcide correlations for the last quarter-century:
Incidentally, note that the results are aggregated by entire city rather than by zip code, which would be far more accurate but extremely difficult to calculate. Therefore, my results must enormously understate the true race/crime relationship, lumping together the Upper East Side and the South Bronx. My guess would be that a zip-code based race/crime correlation would be somewhere between 0.90 and 0.95, certainly the highest ever found anywhere in the social sciences...
I don’t know how much it would cost, but aren’t most police records online now?
If someone wanted to make the effort and spend … I don’t have the slightest idea. I wouldn’t think it would cost more than 20 million or something. You could collect and collate the data by individual case across the country.
I think this is eminently viable. For example this site: https://www.fbi.gov/news/stories/2014/november/crime-statistics-for-2013-released/crime-statistics-for-2013-released
has an estimate of 14,196 murders in 2013. Assuming you got a standardized form for cases in an individual state you could write a bot to extract the data of interest, like zip code of occurrence, address of suspect, etc.
I’m not sure how useful this is. But it could be done relatively cheaply (as these things go) by someone who was interested.
Heck you could just outsource things and have a team of people in India read each case entry and do data input.
Kind of stupid, and wouldn’t be necessary if the government didn’t deliberately try to obfuscate things like this.
Or it could be done piecemeal, by individual teams in each state figuring out what happened in their state, and submitting them to a central team.
Call it “ShadowCrimeStats” or something.
Oh yeah, and wanted to add that it seems like something useful could be extracted from the fact that most murders of blacks are by other blacks.
Trigger warning: don't read any further if your last statistics course gave you PTSD.
The statistical model that applies to rare events like homicides is called the Poisson distribution. (I'm certain it's what the author's stats geek Bruce Frederic (who "outlined his methodology in an email") used.) For the Poisson distribution, one standard deviation is the square root of the number of incidents. Thus Atlanta's rise from 45 in 2014 to 52 in 2015 represents a (52-45)/sqrt(45)=1.04 standard deviation increase. That sort of thing will happen by chance about 30% of the time, which is not considered statistically significant (in the article, they took "significant" to mean, "won't happen more than 5% of the time").
But now let's look at the bottom line. Adding up the city-by-city numbers I see there were 3437 murders in 2015, as against 2955 in 2014. Number of standard deviations difference: (3437-2955)/sqrt(2955) = 8.9. So there you go. Statistically significant at the, who knows, one-googleth percent level.
In the service of his plainly ideological motive, the author could have (albeit after more effort than he was willing to devote to this thing) presented the numbers broken out not by city but by zip code. Then, because individual zip codes will have usually just one or two murders, he could have reported that a mere 1% (or something) of his zip code-level murder increases were statistically significant.
One more technical note. Some of you will have observed that the formula I gave above for standard deviations of the Poisson distribution makes no sense when you get to zero; in a town with 4 murders per year, the standard deviation is 2, which sounds like I'm saying there's about a 2.3% chance to observe a negative number of murders next year! OK, the deal is this: the Poisson distribution is not the Normal (aka Gaussian aka Bell-curve) distribution. So the percentiles that correspond to 1, 2, 3 etc standard deviations away from the mean are not the same. However, they are very close when the mean is moderate or large -- roughly, twenty or more. And when the number is in the thousands (e.g. the 2955 murders in January-August 2014), the two distributions are for all practical purposes identical.
Great post! Yay statistics! The Poisson distribution always makes me tingle.
I am often surprised to see comments sections so diametrically opposed to the articles they are commenting on when I mistakenly assume the commenters are the customers. The media knows they aren't the customers I guess.
By the way, I loved this recent Swedish movie:
http://youtu.be/bS5P_LAqiVg
But who is the director/main character?
He doesn't look Swedish.
They aren’t in many cases. As I understand it, National Review hasn’t made bank from subscriptions and ads in years, assuming they ever did; it depends for its existence on a few wealthy donors.
I am often surprised to see comments sections so diametrically opposed to the articles they are commenting on when I mistakenly assume the commenters are the customers. The media knows they aren't the customers I guess.
By the way, I loved this recent Swedish movie:
http://youtu.be/bS5P_LAqiVg
But who is the director/main character?
He doesn't look Swedish.
David Sandberg, born in Sweden in 1981 according to his website. Could be Jewish but it’s also a typical Swedish surname. Even in Sweden there are plenty of people with brown eyes and brown hair.
I don't know how the winter was in the Midwest, but since I didn't hear about it, I figure it was probably pretty ordinary, which means cold and snowy but not what Boston had.
Baltimore and St. Louis, the two cities most identified with Obama's riots and "Black Lives Matter", had the biggest increases in homicides. It's no coincidence, and GOP presidential candidates need to start talking about it.
The Mayor of Boston, Marty Walsh, is an old-school Irish union guy, and he’s taken a hard line against BLM protestors as well as been a consistent supporter of the police. So that may have something to do with it, too.
http://www.judicialwatch.org/press-room/press-releases/judicial-watch-san-francisco-crime-surge-homicide-arrests-up-55-rapes-arrests-up-370/
Are you saying SF (increasing crime) is more bleeding heart than Boston (decreasing crime)?
P originally looked at the correlation between percentage black and percentage change in homicides.
But you look at the correlation between percentage black and the ABSOLUTE change in homicide rates rather than the PERCENTAGE change in homicide rates.
I am concerned that your results are simply due to the fact that homicide rates are correlated with percentage black.
Take an extreme example, two cities, each with a population of 1,000, one all-black with 100 homicides per year and one all-non-black with 10 homicides per year. Homicide rates are 10% and 1% respectively. Homicides in both double (to 200 and 2) for reasons unrelated to the riots (such as better weather). Therefore, homicide rates in both double (to 20% and 2%), and the PERCENTAGE change in crime rates is the same (both increase 100%). But the ABSOLUTE change is much bigger in the black city (an increase of 10% versus 1%).
Therefore, finding that the ABSOLUTE change in crime rates is higher in blacker cities does NOT prove that the increase in crime rates was not due to the weather.
If, however, the changes are due to less intensive policing of black cities for fear of causing riots, then we would expect to see a difference in the PERCENTAGE change in crime rates.
Hence, we want to look at the correlation of the percentage change in crime rates, not the absolute change in crime rates.
Now maybe, police are policing everyone less intensively and so crime is going up everywhere. This is possible, but we can't test that using this data. We need to look at some measure of policing intensity.
You have it backwards. Perhaps you should download the data and run the analysis yourself to confirm. (this is somewhat funny coming from someone with a username of FactsAreImportant)
Agree this would be interesting. Arrest rate (suggested by others above) would be a good proxy. Does anyone have appropriate data?
IMHO it is silly to try to decide on a single underlying reason. It is likely both decreased police presence and increased violence (and weather etc.) are relevant to a full explanation of the change. Of course, the key point is there IS a statistically significant change despite 538′s analysis. I think the easiest way to show this is a paired t-test on the homicide rates for 2014 and 2015, but I would be interested in hearing other ideas.
When I do a paired t-test on the rates I get:
t = -3.117, df = 58, p-value = 0.002841
which is actually less significant than for the absolute numbers:
t = -3.7807, df = 58, p-value = 0.0003717
I was surprised to find that performing the paired t-test on rates excluding Baltimore and St.Louis gave a more significant result:
t = -3.2617, df = 56, p-value = 0.001888
It looks to me like removing the outliers decreased the SD by enough to overcompensate for the decrease in mean change.
You raise a good point about the importance of this correlation. It turns out replacing percentage black with the 2014 homicide rate actually increases the R^2 of the model to 0.72
(this would be the narrative approved version of the model
This makes sense given that the 2014 homicide rate is probably a better measure of underlying violence than percentage black.
This of course raises the question of why the correlation exists (perhaps Ta-Nehisi Coates will address this in “The Enduring Myth of Black Criminality”). I suppose blacks might be preferentially attracted to areas with high criminality? (snark aside I suspect this is part of the picture, but unlikely to be the dominant cause)
The funny thing about this is the correlations we see here (e.g. see Ron’s plot above) would unleash a flood of publications if they were about a PC topic given how high these correlations are compared to what is typically seen in social science.
P.S. thanks for both supplying the data and engaging in a civil disagreement. I would be interested in hearing your further thoughts on all of this.
Is Boston a “sanctuary city?”
@Factsareimportant
In Baltimore arrests did massively decline as crime skyrocketed:
http://marginalrevolution.com/marginalrevolution/2015/06/in-baltimore-arrests-are-down-and-crime-is-way-up.html
And likewise in St. Louis/Ferguson, and New York.
“In November, Chief Sam Dotson of the St. Louis Police referred to the “Ferguson effect”: officers backing away from discretionary enforcement under charges of racism, thereby emboldening criminals. At that point, arrests in St. Louis city and county had dropped a third since the August shooting of Michael Brown in nearby Ferguson. Homicides in the city had surged 47 percent and robberies in the county were up 82 percent. In Baltimore, arrests dropped 56 percent this May since the protests and riots over the death of Freddie Gray, while shootings so far this year are up more than 60 percent compared to the same period last year.
In New York, arrests are down 17.4 percent through May 31 compared to the same period last year. Criminal summons, which encompass many discretionary, low-level offenses, are down 26 percent.
Pedestrian stops in New York are down 95 percent from the 2011 high. Guns are showing up daily in routine street encounters, N.Y.P.D. officials report.”
http://www.nytimes.com/roomfordebate/2015/06/04/have-fearful-police-brought-an-end-to-the-drop-in-crime/rise-in-crime-is-a-reason-to-fear-anti-police-rhetoric
The evidence is overwhelming.
If results in the "priming" studiess had been this strong, multiple Nobel prizes would have been awarded.
IMHO it is silly to try to decide on a single underlying reason. It is likely both decreased police presence and increased violence (and weather etc.) are relevant to a full explanation of the change. Of course, the key point is there IS a statistically significant change despite 538's analysis. I think the easiest way to show this is a paired t-test on the homicide rates for 2014 and 2015, but I would be interested in hearing other ideas.
When I do a paired t-test on the rates I get:
t = -3.117, df = 58, p-value = 0.002841
which is actually less significant than for the absolute numbers:
t = -3.7807, df = 58, p-value = 0.0003717
I was surprised to find that performing the paired t-test on rates excluding Baltimore and St.Louis gave a more significant result:
t = -3.2617, df = 56, p-value = 0.001888
It looks to me like removing the outliers decreased the SD by enough to overcompensate for the decrease in mean change.You raise a good point about the importance of this correlation. It turns out replacing percentage black with the 2014 homicide rate actually increases the R^2 of the model to 0.72
(this would be the narrative approved version of the model ;-)
This makes sense given that the 2014 homicide rate is probably a better measure of underlying violence than percentage black.
This of course raises the question of why the correlation exists (perhaps Ta-Nehisi Coates will address this in “The Enduring Myth of Black Criminality”). I suppose blacks might be preferentially attracted to areas with high criminality? (snark aside I suspect this is part of the picture, but unlikely to be the dominant cause)
The funny thing about this is the correlations we see here (e.g. see Ron's plot above) would unleash a flood of publications if they were about a PC topic given how high these correlations are compared to what is typically seen in social science.
P.S. @P thanks for both supplying the data and engaging in a civil disagreement. I would be interested in hearing your further thoughts on all of this.
Okay, so what we seem to be seeing are high 2014 homicide rate cities — e.g., St. Louis, Baltimore, Milwaukee — tending to get worse in 2015, while low homicide rate cities — e.g., Arlington (a nice suburb of DFW) — are behaving randomly.
That fits with the overall idea of a Ferguson Effect, although we can speculate about what precisely makes up the Ferguson Effect: angrier blacks? Cops retreating to the donut shop?
First, cities with major BLM issues (my INCIDENTS variable) show a large increase in homicide rates. Whether this is due to angrier blacks, retreating cops, weather, sunspots, or something else is hard to say conclusively.
However, the first link from @Correlation with reduced policing intensity above:
http://marginalrevolution.com/marginalrevolution/2015/06/in-baltimore-arrests-are-down-and-crime-is-way-up.html
is great and leads to the (current) Baltimore data we need (to look at arrests and homicides together) at:
https://data.baltimorecity.gov/Public-Safety/BPD-Arrests/3i3v-ibrt?
https://data.baltimorecity.gov/Public-Safety/BPD-Part-1-Victim-Based-Crime-Data/wsfq-mvij
The Baltimore arrest data includes race so it might be interesting to see if the racial balance of arrests changed after the BLM riots.
I have yet to work with this data, but the MR plots support the assertion that the police retreated. I intend to replicate the MR homicide and arrest plots and go from there. Also see:
https://github.com/khughitt/baltimore_ggplot2
for useful infrastructure. I just downloaded that code and once the correct libraries are installed it worked for me.
Second, is an overall effect for all cities. I think this is based on underlying levels of violence in each city (measured by either 2014 homicide rate or percentage black). Causality is even less well defined here, but I suspect the parsimonious explanation for this is angrier blacks (but do not claim to have proof) so chose percentage black in my model. It is worth reiterating that the 2014 homicide rate actually gives a slightly better fit.
The second effect is not as large and there is a great deal of noise. It might be worthwhile to look at some other cities closely. Two that come to mind are Philadelphia and Detroit--both of which had minimal increase in homicide (Detroit's 2014 rate was very high though).
It is worth remembering that if blacks commit murder at 6x the rate of other races they are committing the majority of murders even at 13% population prevalence (i.e. an angrier blacks explanation is tenable even at a surprisingly low proportion of the population).
Not yet (the political climate is much saner than SF, not that that’s saying much), but it’s moving in that direction.
My take is that Baltimore, St. Louis, and Milwaukee all had major BLM issues with police killings in 2014-2015 (I had missed the Milwaukee mention in my first reading of the NYT article and had not heard about it before). I think we are seeing two effects corresponding to the two variables in my R model.
First, cities with major BLM issues (my INCIDENTS variable) show a large increase in homicide rates. Whether this is due to angrier blacks, retreating cops, weather, sunspots, or something else is hard to say conclusively.
However, the first link from @Correlation with reduced policing intensity above:
http://marginalrevolution.com/marginalrevolution/2015/06/in-baltimore-arrests-are-down-and-crime-is-way-up.html
is great and leads to the (current) Baltimore data we need (to look at arrests and homicides together) at:
https://data.baltimorecity.gov/Public-Safety/BPD-Arrests/3i3v-ibrt?
https://data.baltimorecity.gov/Public-Safety/BPD-Part-1-Victim-Based-Crime-Data/wsfq-mvij
The Baltimore arrest data includes race so it might be interesting to see if the racial balance of arrests changed after the BLM riots.
I have yet to work with this data, but the MR plots support the assertion that the police retreated. I intend to replicate the MR homicide and arrest plots and go from there. Also see:
https://github.com/khughitt/baltimore_ggplot2
for useful infrastructure. I just downloaded that code and once the correct libraries are installed it worked for me.
Second, is an overall effect for all cities. I think this is based on underlying levels of violence in each city (measured by either 2014 homicide rate or percentage black). Causality is even less well defined here, but I suspect the parsimonious explanation for this is angrier blacks (but do not claim to have proof) so chose percentage black in my model. It is worth reiterating that the 2014 homicide rate actually gives a slightly better fit.
The second effect is not as large and there is a great deal of noise. It might be worthwhile to look at some other cities closely. Two that come to mind are Philadelphia and Detroit–both of which had minimal increase in homicide (Detroit’s 2014 rate was very high though).
It is worth remembering that if blacks commit murder at 6x the rate of other races they are committing the majority of murders even at 13% population prevalence (i.e. an angrier blacks explanation is tenable even at a surprisingly low proportion of the population).
Excellent links as well. The MR link is absolutely devastating.
I didn't know there was a BLM event in Milwaukee. Adding that to the mix, I think the evidence is overwhelming that BLM is responsible for a lot of dead bodies.
I also think the evidence is strong enough to attribute the entire 16% increase in homicides to BLM et al. We know from the INCIDENTS analysis that police respond to BLM and DoJ witch hunts. It is impossible to believe that police in other cities have not felt similar pressure. So, until someone comes up with an equally forceful explanation for the homicide increase, the increase should be attributed to BLM et al. The burden of proof has shifted.
Did something happen in Tulsa, OK over the last year? It had the largest percentage increase in homicides from 2014 to 2015 (from 21 to 38).
For what it’s worth, the INCIDENTS/BLACK_POP model was a terrible fit for the 2015/2014 homicide ratio (and the 2014 rate version was not any better).
But this does not seem like a complete explanation given that many of the additional homicides happened in January and March. See https://www.tulsapolice.org/content/crime-numbers-ucr.aspx
Looking more closely at those numbers it looks like part of the issue is that January-July 2014 had an unusually low number of homicides in Tulsa.
For what it's worth, the INCIDENTS/BLACK_POP model was a terrible fit for the 2015/2014 homicide ratio (and the 2014 rate version was not any better).
Eric Harris was killed by a Tulsa County deputy 4/2/15. See http://www.cnn.com/2015/04/13/us/tulsa-police-shooting-eric-harris-deputy-charged/ and http://bluenationreview.com/fck-breath-tulsa-officers-shoot-unarmed-man-refuse-render-immediate-aid/
But this does not seem like a complete explanation given that many of the additional homicides happened in January and March. See https://www.tulsapolice.org/content/crime-numbers-ucr.aspx
Looking more closely at those numbers it looks like part of the issue is that January-July 2014 had an unusually low number of homicides in Tulsa.
Well, I predict crime will go up then.
Boston isn’t, but it’s immediate area is, as Cambridge, Somerville and Chelsea are. Recovering alcoholic Walsh is still pondering, after curtailing the city’s participation in the Secure Communities program:
http://www.bostonglobe.com/2013/11/26/mayor-elect-walsh-says-wants-boston-pull-out-secure-communities-program/BikbYna0qIGwMWmM06uPrO/story.html?s_campaign=sm_tw
Boston is an outlier; it attracts smart blacks who attend the local universities, and a certain number remain, as the higher education biz is an affirmative action hotbed. Massachusetts keeps the dumb ones fat and happy with an average $50K per annum welfare package.
IMHO it is silly to try to decide on a single underlying reason. It is likely both decreased police presence and increased violence (and weather etc.) are relevant to a full explanation of the change. Of course, the key point is there IS a statistically significant change despite 538's analysis. I think the easiest way to show this is a paired t-test on the homicide rates for 2014 and 2015, but I would be interested in hearing other ideas.
When I do a paired t-test on the rates I get:
t = -3.117, df = 58, p-value = 0.002841
which is actually less significant than for the absolute numbers:
t = -3.7807, df = 58, p-value = 0.0003717
I was surprised to find that performing the paired t-test on rates excluding Baltimore and St.Louis gave a more significant result:
t = -3.2617, df = 56, p-value = 0.001888
It looks to me like removing the outliers decreased the SD by enough to overcompensate for the decrease in mean change.You raise a good point about the importance of this correlation. It turns out replacing percentage black with the 2014 homicide rate actually increases the R^2 of the model to 0.72
(this would be the narrative approved version of the model ;-)
This makes sense given that the 2014 homicide rate is probably a better measure of underlying violence than percentage black.
This of course raises the question of why the correlation exists (perhaps Ta-Nehisi Coates will address this in “The Enduring Myth of Black Criminality”). I suppose blacks might be preferentially attracted to areas with high criminality? (snark aside I suspect this is part of the picture, but unlikely to be the dominant cause)
The funny thing about this is the correlations we see here (e.g. see Ron's plot above) would unleash a flood of publications if they were about a PC topic given how high these correlations are compared to what is typically seen in social science.
P.S. @P thanks for both supplying the data and engaging in a civil disagreement. I would be interested in hearing your further thoughts on all of this.
Thanks for the reasoned response.
With all respect, I’m reasonably sure I’m right about this. Before commenting, I downloaded the data, ran the numbers, and verified both P’s .18 correlation and your .45 correlation.
To replicate P’s calculation, I calculated Corr(CHANGE, BLACK_POP) = 0.18, where CHANGE and BLACK_POP are from P’s dataset, and P calculated CHANGE = HOMICIDE_Y2015 / HOMICIDE_Y2014 – 1.
Note that P used the percentage change in homicides.
To replicate your calculation, I first calculated HOM_RATE_2014 = HOMICIDE_Y2014 / TOTAL_POP, and HOM_RATE_2015 = HOMICIDE_Y2015 / TOTAL_POP. Then I calculated CHANGE_HOM_RATE = HOM_RATE_2015 – HOM_RATE_2014. And then I replicated your result by calculating Corr(CHANGE_HOM_RATE, BLACK_POP) = .45.
Note that you used the absolute change in homicide rates, not the percentage change in homicide rates.
For New York, the relevant variables are: HOMICIDE_Y2014 = 190, HOMICIDE_Y2015 = 208, CHANGE = 0.09, BLACK_POP = .255, TOTAL_POP = 8500000, HOM_RATE_2014 = 2.235, HOM_RATE_2015 = 2.447, CHANGE_HOM_RATE = .095.
Maybe I missed something, but this sounds right to me. I also strongly expected a priori that your correlation would be much higher than P’s for the reasons I gave.
On closer reading you are right (and especially emphatic apologies for "correcting" you when you were right--I hate when people do that to me). The thing that keeps crossing me up reading/writing/thinking about this is that the percentage change in absolute # of homicides is the same as the percentage change in the homicide rates here (because we are assuming the population is unchanged, which over 1 year in the current conditions seems reasonable enough).
Rereading your earlier reply, your explanation for the increased correlation seems reasonable. I tend to prefer absolute change in the rate because percentage change tends to be dominated by statistical noise in the low crime cities (and looking at scatterplots there are many more low crime cities than high crime). Conversely, the absolute change in the rate tends to be dominated by high crime/black cities giving the correlation you noted. The two reasons I favor the absolute value of change in homicide rates are:
1. Less noise dominated as noted.
2. I believe it captures the real quantity of interest--additional deaths per capita. A change from 1 homicide per 100,000 people to 2 is less of an issue IMHO than a change from 10 to 15.
Regarding alternate explanations it's hard to say anything (beyond e.g. the Milwaukee anecdote) about the effect of weather without weather data for all cities. Even with that, saying anything useful would be hard because of difficulty modeling weather effects. For example, what's important: rain, average temp, max temp, temp over threshold, humidity, ...?
We talk about intensity of policing in other posts so I won't discuss here beyond saying I agree this is worth investigating.
One thing that frustrates me trying to analyze data like this is that in general everything negative correlates so highly with blackness (e.g. here HOM_RATE_2014 and POP_BLACK) that it becomes difficult to do reliable regression analysis with multiple explanatory variables. A classic example of this is disentangling the effects of blackness and poverty (IMHO one of the reasons the narrative maintains traction is this difficulty combined with strong priors for all parties). Some data/analysis is good enough to be compelling anyway (e.g. the race/income/SAT score analyses I have seen), but that seems to be the exception. If you have any suggestions in this area I would be interested. The approach which seems most fruitful to me is binning (as done by income in the race/income/SAT score analyses), but I don't see how to emulate that here.
First, cities with major BLM issues (my INCIDENTS variable) show a large increase in homicide rates. Whether this is due to angrier blacks, retreating cops, weather, sunspots, or something else is hard to say conclusively.
However, the first link from @Correlation with reduced policing intensity above:
http://marginalrevolution.com/marginalrevolution/2015/06/in-baltimore-arrests-are-down-and-crime-is-way-up.html
is great and leads to the (current) Baltimore data we need (to look at arrests and homicides together) at:
https://data.baltimorecity.gov/Public-Safety/BPD-Arrests/3i3v-ibrt?
https://data.baltimorecity.gov/Public-Safety/BPD-Part-1-Victim-Based-Crime-Data/wsfq-mvij
The Baltimore arrest data includes race so it might be interesting to see if the racial balance of arrests changed after the BLM riots.
I have yet to work with this data, but the MR plots support the assertion that the police retreated. I intend to replicate the MR homicide and arrest plots and go from there. Also see:
https://github.com/khughitt/baltimore_ggplot2
for useful infrastructure. I just downloaded that code and once the correct libraries are installed it worked for me.
Second, is an overall effect for all cities. I think this is based on underlying levels of violence in each city (measured by either 2014 homicide rate or percentage black). Causality is even less well defined here, but I suspect the parsimonious explanation for this is angrier blacks (but do not claim to have proof) so chose percentage black in my model. It is worth reiterating that the 2014 homicide rate actually gives a slightly better fit.
The second effect is not as large and there is a great deal of noise. It might be worthwhile to look at some other cities closely. Two that come to mind are Philadelphia and Detroit--both of which had minimal increase in homicide (Detroit's 2014 rate was very high though).
It is worth remembering that if blacks commit murder at 6x the rate of other races they are committing the majority of murders even at 13% population prevalence (i.e. an angrier blacks explanation is tenable even at a surprisingly low proportion of the population).
Excellent work, and very good ideas for future work.
Excellent links as well. The MR link is absolutely devastating.
I didn’t know there was a BLM event in Milwaukee. Adding that to the mix, I think the evidence is overwhelming that BLM is responsible for a lot of dead bodies.
I also think the evidence is strong enough to attribute the entire 16% increase in homicides to BLM et al. We know from the INCIDENTS analysis that police respond to BLM and DoJ witch hunts. It is impossible to believe that police in other cities have not felt similar pressure. So, until someone comes up with an equally forceful explanation for the homicide increase, the increase should be attributed to BLM et al. The burden of proof has shifted.
One thing I should be clearer about is so far I have been somewhat sloppy about defining INCIDENTS. Baltimore and St.Louis (Ferguson) seem obvious, but the other possible cases are more subtle (e.g. Milwaukee, Tulsa, others?). Milwaukee and Tulsa both had police killings of blacks, but I did not see whether or not they had BLM protests.
I'm not sure about how to handle this which is why I have not extended the INCIDENTS analysis beyond Baltimore and St. Louis except for noting that this issue is relevant (and running some tests with Milwaukee included). I think a proper analysis of the impact of police killings and BLM would include looking at the following:
- Cities where police killed black/white men (and perhaps the occasional woman) under questionable circumstances (including those where IMHO most of the questionableness is due to slanted media coverage, like Ferguson).
- Level of involvement of outside parties (e.g. BLM, DOJ, national media).
- Protests and/or riots with attention to their scope, level of violence, and involvement with BLM.
- Attempt to quantify any resultant change in intensity of policing.
- Timing of incidents (e.g. the Tulsa incident happened 4/2/15 in the middle of the January-July homicide data).
- Past history of killings and media coverage (for comparison and potential effects on 2014 data).
Doing a thorough version of this is beyond me, but seems like the kind of thing a DOJ interested in facts might do.
I'm thinking about handling INCIDENTS as a number between 0 and 1 where I attempt to account only for severity and timeline (e.g. Tulsa gets 4/7 for timeline multiplied by severity based on level of publicity/riots/etc.). I'm reluctant to put in the effort to do this though because of the difficulty of coming up with supportable numbers. Any thoughts?
I think the explanatory power of my basic model is sufficient to support (but not prove) a conclusion (similar but not identical to your last paragraph) that the bulk of the increase in homicides is attributable to the reaction to police killings last year (including BLM, DOJ, media, and individuals) in cities directly affected. I see less support (but still some) for the conclusion that this trend applies across all cities.
To editorialize OT a bit, I wish the conversation about excessive police militarization/violence and lack of responsibility was more separated from the BLM conversation. I think angry arguments (typically based on a minimum of facts) about the appropriateness of Michael Brown and Darren Wilson's actions are detracting from what seem like some common sense measures:
- Body and dash cameras for all police interacting with the public. Severe penalties if the camera is off for any reason during an incident.
- Real consequences for police guilty of bad behavior. I like the idea of docking pensions, but the unions would never approve. Different levels of consequences for different levels of offense is key.
- A recognition by the public that police have to make hard decisions under pressure. Not every decision will be textbook, but there are clearly levels of acceptability. This could also include a discussion of how police are trained to respond and whether it is appropriate. Also a further recognition that interacting with criminals can not always be done according to SWPL norms because it makes you look weak and/or exposes you to unnecessary danger.
- Redefining "good cop" to be not only avoiding bad behavior but also not participating in protecting "bad cops" from the consequences of their actions (or turning a blind eye). IMHO the population of "good cops" shrinks dramatically if this definition is used and that is a major part of the problem.
- A real conversation about the militarization of police. Do small cities (e.g. Davis, CA) truly need MRAPs?
I think the "law and order" contingent has an inadequate appreciation for how much footage of arrests would support "good cops" both in terms of consequences and public perception. I wish there had been video of the Ferguson encounter.
In Baltimore arrests did massively decline as crime skyrocketed:
http://marginalrevolution.com/marginalrevolution/2015/06/in-baltimore-arrests-are-down-and-crime-is-way-up.html
And likewise in St. Louis/Ferguson, and New York.
"In November, Chief Sam Dotson of the St. Louis Police referred to the “Ferguson effect”: officers backing away from discretionary enforcement under charges of racism, thereby emboldening criminals. At that point, arrests in St. Louis city and county had dropped a third since the August shooting of Michael Brown in nearby Ferguson. Homicides in the city had surged 47 percent and robberies in the county were up 82 percent. In Baltimore, arrests dropped 56 percent this May since the protests and riots over the death of Freddie Gray, while shootings so far this year are up more than 60 percent compared to the same period last year.
In New York, arrests are down 17.4 percent through May 31 compared to the same period last year. Criminal summons, which encompass many discretionary, low-level offenses, are down 26 percent.
Pedestrian stops in New York are down 95 percent from the 2011 high. Guns are showing up daily in routine street encounters, N.Y.P.D. officials report."
http://www.nytimes.com/roomfordebate/2015/06/04/have-fearful-police-brought-an-end-to-the-drop-in-crime/rise-in-crime-is-a-reason-to-fear-anti-police-rhetoric
Stunning links.
The evidence is overwhelming.
If results in the “priming” studiess had been this strong, multiple Nobel prizes would have been awarded.
The CW about chances of violence happening on or near any thoroughfare named after MLK is not nearly as wise as bad things happening at any sort of NAM “Festival”!
http://www.bostonherald.com/news_opinion/local_coverage/2015/09/shooting_at_cambridge_fest_turns_revelry_into_panic
Thanks for your reasoned responses as well. And sorry for the somewhat cranky part of my earlier reply.
On closer reading you are right (and especially emphatic apologies for “correcting” you when you were right–I hate when people do that to me). The thing that keeps crossing me up reading/writing/thinking about this is that the percentage change in absolute # of homicides is the same as the percentage change in the homicide rates here (because we are assuming the population is unchanged, which over 1 year in the current conditions seems reasonable enough).
Rereading your earlier reply, your explanation for the increased correlation seems reasonable. I tend to prefer absolute change in the rate because percentage change tends to be dominated by statistical noise in the low crime cities (and looking at scatterplots there are many more low crime cities than high crime). Conversely, the absolute change in the rate tends to be dominated by high crime/black cities giving the correlation you noted. The two reasons I favor the absolute value of change in homicide rates are:
1. Less noise dominated as noted.
2. I believe it captures the real quantity of interest–additional deaths per capita. A change from 1 homicide per 100,000 people to 2 is less of an issue IMHO than a change from 10 to 15.
Regarding alternate explanations it’s hard to say anything (beyond e.g. the Milwaukee anecdote) about the effect of weather without weather data for all cities. Even with that, saying anything useful would be hard because of difficulty modeling weather effects. For example, what’s important: rain, average temp, max temp, temp over threshold, humidity, …?
We talk about intensity of policing in other posts so I won’t discuss here beyond saying I agree this is worth investigating.
One thing that frustrates me trying to analyze data like this is that in general everything negative correlates so highly with blackness (e.g. here HOM_RATE_2014 and POP_BLACK) that it becomes difficult to do reliable regression analysis with multiple explanatory variables. A classic example of this is disentangling the effects of blackness and poverty (IMHO one of the reasons the narrative maintains traction is this difficulty combined with strong priors for all parties). Some data/analysis is good enough to be compelling anyway (e.g. the race/income/SAT score analyses I have seen), but that seems to be the exception. If you have any suggestions in this area I would be interested. The approach which seems most fruitful to me is binning (as done by income in the race/income/SAT score analyses), but I don’t see how to emulate that here.
Excellent links as well. The MR link is absolutely devastating.
I didn't know there was a BLM event in Milwaukee. Adding that to the mix, I think the evidence is overwhelming that BLM is responsible for a lot of dead bodies.
I also think the evidence is strong enough to attribute the entire 16% increase in homicides to BLM et al. We know from the INCIDENTS analysis that police respond to BLM and DoJ witch hunts. It is impossible to believe that police in other cities have not felt similar pressure. So, until someone comes up with an equally forceful explanation for the homicide increase, the increase should be attributed to BLM et al. The burden of proof has shifted.
Thanks for your kind words.
One thing I should be clearer about is so far I have been somewhat sloppy about defining INCIDENTS. Baltimore and St.Louis (Ferguson) seem obvious, but the other possible cases are more subtle (e.g. Milwaukee, Tulsa, others?). Milwaukee and Tulsa both had police killings of blacks, but I did not see whether or not they had BLM protests.
I’m not sure about how to handle this which is why I have not extended the INCIDENTS analysis beyond Baltimore and St. Louis except for noting that this issue is relevant (and running some tests with Milwaukee included). I think a proper analysis of the impact of police killings and BLM would include looking at the following:
- Cities where police killed black/white men (and perhaps the occasional woman) under questionable circumstances (including those where IMHO most of the questionableness is due to slanted media coverage, like Ferguson).
- Level of involvement of outside parties (e.g. BLM, DOJ, national media).
- Protests and/or riots with attention to their scope, level of violence, and involvement with BLM.
- Attempt to quantify any resultant change in intensity of policing.
- Timing of incidents (e.g. the Tulsa incident happened 4/2/15 in the middle of the January-July homicide data).
- Past history of killings and media coverage (for comparison and potential effects on 2014 data).
Doing a thorough version of this is beyond me, but seems like the kind of thing a DOJ interested in facts might do.
I’m thinking about handling INCIDENTS as a number between 0 and 1 where I attempt to account only for severity and timeline (e.g. Tulsa gets 4/7 for timeline multiplied by severity based on level of publicity/riots/etc.). I’m reluctant to put in the effort to do this though because of the difficulty of coming up with supportable numbers. Any thoughts?
I think the explanatory power of my basic model is sufficient to support (but not prove) a conclusion (similar but not identical to your last paragraph) that the bulk of the increase in homicides is attributable to the reaction to police killings last year (including BLM, DOJ, media, and individuals) in cities directly affected. I see less support (but still some) for the conclusion that this trend applies across all cities.
To editorialize OT a bit, I wish the conversation about excessive police militarization/violence and lack of responsibility was more separated from the BLM conversation. I think angry arguments (typically based on a minimum of facts) about the appropriateness of Michael Brown and Darren Wilson’s actions are detracting from what seem like some common sense measures:
- Body and dash cameras for all police interacting with the public. Severe penalties if the camera is off for any reason during an incident.
- Real consequences for police guilty of bad behavior. I like the idea of docking pensions, but the unions would never approve. Different levels of consequences for different levels of offense is key.
- A recognition by the public that police have to make hard decisions under pressure. Not every decision will be textbook, but there are clearly levels of acceptability. This could also include a discussion of how police are trained to respond and whether it is appropriate. Also a further recognition that interacting with criminals can not always be done according to SWPL norms because it makes you look weak and/or exposes you to unnecessary danger.
- Redefining “good cop” to be not only avoiding bad behavior but also not participating in protecting “bad cops” from the consequences of their actions (or turning a blind eye). IMHO the population of “good cops” shrinks dramatically if this definition is used and that is a major part of the problem.
- A real conversation about the militarization of police. Do small cities (e.g. Davis, CA) truly need MRAPs?
I think the “law and order” contingent has an inadequate appreciation for how much footage of arrests would support “good cops” both in terms of consequences and public perception. I wish there had been video of the Ferguson encounter.
On closer reading you are right (and especially emphatic apologies for "correcting" you when you were right--I hate when people do that to me). The thing that keeps crossing me up reading/writing/thinking about this is that the percentage change in absolute # of homicides is the same as the percentage change in the homicide rates here (because we are assuming the population is unchanged, which over 1 year in the current conditions seems reasonable enough).
Rereading your earlier reply, your explanation for the increased correlation seems reasonable. I tend to prefer absolute change in the rate because percentage change tends to be dominated by statistical noise in the low crime cities (and looking at scatterplots there are many more low crime cities than high crime). Conversely, the absolute change in the rate tends to be dominated by high crime/black cities giving the correlation you noted. The two reasons I favor the absolute value of change in homicide rates are:
1. Less noise dominated as noted.
2. I believe it captures the real quantity of interest--additional deaths per capita. A change from 1 homicide per 100,000 people to 2 is less of an issue IMHO than a change from 10 to 15.
Regarding alternate explanations it's hard to say anything (beyond e.g. the Milwaukee anecdote) about the effect of weather without weather data for all cities. Even with that, saying anything useful would be hard because of difficulty modeling weather effects. For example, what's important: rain, average temp, max temp, temp over threshold, humidity, ...?
We talk about intensity of policing in other posts so I won't discuss here beyond saying I agree this is worth investigating.
One thing that frustrates me trying to analyze data like this is that in general everything negative correlates so highly with blackness (e.g. here HOM_RATE_2014 and POP_BLACK) that it becomes difficult to do reliable regression analysis with multiple explanatory variables. A classic example of this is disentangling the effects of blackness and poverty (IMHO one of the reasons the narrative maintains traction is this difficulty combined with strong priors for all parties). Some data/analysis is good enough to be compelling anyway (e.g. the race/income/SAT score analyses I have seen), but that seems to be the exception. If you have any suggestions in this area I would be interested. The approach which seems most fruitful to me is binning (as done by income in the race/income/SAT score analyses), but I don't see how to emulate that here.
True, but you have to be careful about you are trying to accomplish. If you want to make statements about where the changes are most important, than absolute numbers is appropriate. But, if you are testing a specific hypothesis (such as black neighborhoods are more affected than non-black neigborhoods by something), then you have to think about what the model is and that will point to the correct variable. That is why I think in terms of simple thought experiments — it forces you to state your hypothesis clearly.
Some of this type of analysis is invaluable, but in many argumentative situations, people don’t understand enough statistics for this to work (see Ta’s latest article butchering crime rate data).
My preference is for simple logical tests. For instance, slavery, Jim Crow and other alleged discrimination is usually invoked to explain black crime levels. The best response is “OK let’s look at instances where there was no Jim Crow … how’s Haiti doing … how about Ethiopia … how about Detroit, which has been run by blacks for decades … how about Madison, which has been run by goodthinkers for more than half a century.
The DoJ’s research capabilities are hopelessly primitive. Always have been, and always will be. You need professionals to do this stuff right, and even the professionals often get it wrong (e.g., Raj Chetty).
WRT to analysis of INCIDENTS, I would recommend keeping it simple and breaking any analysis into little pieces. Document changes in policing intensity across cities and over time. First, just describe the data, then look for patterns. Group cities by percentage black and look for differences. Find some objective way to calculate INCIDENTS. For instance, number of hits from a news database for certain searches, such as [city name] + police + shooting + black + unarmed(?). Or, if possible, find someone else’s quantification of police misconduct or notoriety and use that. (Complaints against police per capita? Maybe Ta has spouted some indicator of police oppression.)
(While my tone may sound authoritative and confident here, I’m just winging it.)
Agree that all the racist lying obscures some real issues with policing. Unfortunately, nobody gets a cushy job as a university diversity czar by being careful and nuanced.
You don’t have to prove anything definitively (and you really can’t with statistics). But you can make powerful-sounding statements by saying things like “The evidence suggests that up to 2,200 people have died because of ….” Honest and forceful. Anyone repeating the result will state it much more definitively.
I think it is almost certain that there has been a decline in policing intensity and increase in crime nationwide. Put yourself in the shoes of a simple police officer observing what happened to Officer Wilson and the Baltimore 6. There is no way that didn’t have an effect on many police officers. Absent a compelling alternative explanation (and it is the burden of the BLM crowd to come up with one), it is completely reasonable to attribute the entire increase to BLM, Ta, and Obama. Indeed, it is almost certainly an underestimate, because the long-term trend in crime is downward. This analysis attributes only the INCREASE to Ta et al.
Good luck.
Thanks, guys, for all the work you’ve been doing.
I found Sailer's comparison between the increase in golf rounds and the increase in homocides very interesting but shallow. I wonder why he didn't more deeply explore the golf/homicide connection by telling us whether the number of holes-in-one has changed substantially over the period. Maybe that's where the connection lies. Or bogies. (This is just a WAG, but maybe it's all those frustrated golfers who go out and murder people. I have seen numerous instances of such behavior on televised golf tournaments where a pro golfer gets so frustrated by a bad shot that he breaks a perfectly good gulf club. Who knows what such individual is capable of once the round is over and the TV cameras are off? ) Or whether handicaps have gone up or down. It looks like he might be working on a proposal for a federal grant to study the issue, in which case I would suggest broadening the study to include a comparison of birdies and eagles. Once those federal dudes see that wildlife is involved, I think the proposal will win speedy approval.
Indeed, it turns out there may be a very real connection between golf and crime. http://www.nydailynews.com/news/national/seattle-arrested-black-man-carrying-golf-club-fired-article-1.2362131?utm_source=Sailthru&utm_medium=email&utm_campaign=Daily%20Newsletter%202015-09-16&utm_term=DailyNewsletter
I am now firmly convinced that Steve Sailer is truly a prophet, right up there with all those prophets in the Old Testament.
Thanks for the response. I want to echo this bit because I agree it is important.
If this is winging it I would like to hear you on a topic where you are more authoritative. In a similar vein, I should note that I am not a statistician (self trained, including substantial online course work and much reading and working through textbooks). Although I have a quantitative background any statistical pronouncements I make should be double checked by a real statistician.
I am a big fan of the technique of “find someone else’s quantification …” and using that in arguments. Works really well since so many people are innumerate and haven’t thought things through.
I agree with “I think it is almost certain that there has been a decline in policing intensity and increase in crime nationwide.” I think the latter has been shown across a range of cities. The former is more anecdotal and I have only seen the Baltimore data to check (as in MR).
My trouble now is coming up with a good follow on hypothesis to test. One that comes to mind is showing that BLM can be more conclusively tied to the increase in other cities (hence all my commentary on Milwaukee and Tulsa). You make some good suggestions regarding INCIDENTS which would help with this. I’m just not sure if doing this is worth the effort. I think I need to ponder about this for a while and see if I get any ideas. Another possibility is trying to separate the effects of “angry blacks” from “less intense policing.”
Thanks for the conversation and thanks to Steve for all of his excellent pieces on this and related topics.
Comments are closed.