Go to qualitypolicing.com

Tag: stats

  • Disparities in Police-Involved Shootings by City and County

    Disparities in Police-Involved Shootings by City and County

    So I’ve done a little work using the data from FatalEncounters.org on people shot and killed by police. Fatal Encounters is like the Washington Post database, but for adults. I combined/merged this with a city or police department’s population, number of cops, average number of murders in the jurisdiction (over 4 or 3 years), median household income, percentage Black, and percentage Latino/Hispanic. The dataset includes every city/town where cops killed somebody between 2015-2019 and also every city above 100,000 population. I end up with 2,872 cases.

    I also looked at counties, which nobody has ever seem to have done before. If you live in a state like Maryland, Texas, California or Arizona, you probably know that county police of sheriff can be the major police department. Some of the counties are huge, and their very existence is seemingly noticed by research despite the fact that there are 88 county police departments that have jurisdictions of more than 100,000 people. The police departments of 20 counties police more than 500,000 people. County data is tricky. So take this with grain of salt. Population (the denominator is the rate) is based not on the entire county but on the population policed by the department. It could be wrong (corrections welcome). And I tried to exclude jail operations from cop population (by taking only sworn officers).

    LA County Sheriff’s Department kills an average of 12 people a year (2015-2019). That’s a lot. Their rate is 11 per million population (if my population figures are correct, which is tricky for county police and overlapping jurisdictions). The rate for Los Angeles City Police Department is 4.2. The national average is about 3. Riverside County and San Bernardino Counties also have very high rates. Riverside County is 32 per million, the highest in the nation. But that is only if Riverside County Sheriff’s Department polices but 180,000 people (which is the population of Riverside County minus the cities that have their own police department… but maybe that’s not a good way to figure it out; the population of Riverside County is 2.4 million). Either way, 1,795 cops killing 5.8 people a year over 5 years is a lot. That’s 1 killing for every 310 cops. In NYC, the comparable figure is 1 killing for every 4,605 cops.

    The Bernalillo County Sheriff’s department (Albuquerque) has a rate of nearly 20 (per million). Three-hundred Sheriff Deputies killed 10 people over 5 years. That’s a lot. Could be bad luck. Could be unfortunate but necessary shootings in cases for which there was no less-lethal alternative. But if the NYPD killed two people for every 300 cops, it would be over 200 police-involved shooting deaths a year in NYC. Last year in NYC police shot 15 people and killed 5.

    Other county sheriff departments in which there aren’t that many cops and kind of a lot of people killed are Spokane WA, Pierce WA, Clark WA, Volusia FL, and Lexington SC, King WA, and Greenville SC

    Riverside County CA and Bernalillo County NM are interesting because the largest city police departments in their county (Riverside City and Albuquerque, respectively) also shoot a lot of people (but not nearly at such a high rate). Here are the cities of over 100,000 population with the highest rate of people shot and killed by police.

    Every single city on this list is west of the Mississippi (or in Florida). Every single one. The mean rate for cities in eastern states is 3.8. If you take Florida out of the east, the mean goes down to 3.5. For cities in western states, the mean rate is 5.4. That’s a big difference. (The median is 3.2 and and 4.2.) And whatever real differences account for the arbitrary geographic difference, there are many department in cities over 100,000 that shot and killed few few people from 2015-2019, or at a rate less than the national average: Plano TX, Irvine CA, Fairfield CA, Grand Prarie TX, Pasadnia CA, Mesquite TX. Were they just lucky? Or were they doing something right. Or maybe both.

    Maybe population greater than 100,000 isn’t the right cut off. The top cities just make the greater than 100,000 list. The total n (for 5 years) is between 8 and 35. So a little good or bad luck can affect the rate a lot. But still, a lot of shooting goes on in cities of this size. Also, the murder rate is high in a lot of these cities… but not all of them. And the murder rate is also high in Birmingham, Baltimore, New Orleans, Jackson, and Detroit, and they’re not on the list. And a lot of cities that are on this list have very few black people (Las Cruces, Pueblo, Westminster, Billings, Albuquerque, Tucson, Spokane, Salt Lake City).

    Once you start getting into larger cities, I should look not only at places where cops shoot a lot, but also at places where cops shoot very little. Sure, since shootings are rare, at might just be luck. But it might be police departments are doing something right.

    Thirty-one cities have rates under 1 per million. All but 4 have fewer than 200,000 people. So maybe they’re lucky. Irvine California is on the list. But hey, Irvine is rich. But what about Hialeah FL? Or Lexington KY? Or Lubbock TX? Zero fatalities all. What about New York City? 8.5 million people. And a rate of 0.89, less than a third the national average? That’s not an accident. That’s policy, training, and leadership. Why not learn from the cities doing it right?

    Βetter cities (rate < 1.5 / million, half the national average) in the 200,000 to 300,000 range (n = 52), include Lubbock, Hialeah, and Greensboro. They aren’t rich. (Irvine, Oxnard, Glendale, Plano, and Jersey City are also on the good list.) In the most-shooting category (rate > 10 / million, 3 times the national average) are Orlando, Baton Rouge, Tacoma, Spokane, Salt Lake City, Birmingham AL, Richmond VA, and Modesto CA. These are mostly middle income places with a wide variety of racial demographics.

    In the 300,000 to 500,000 category (n=29), only Lexington KY and Raleigh NC stand out as better than average (rate < 2). Though Virginia Beach, Minneapolis, Pittsburgh have rates < 4. On the high end (rate > 10) are Miami, Bakersfield, Tulsa, and St. Louis. St. Louis tops the chart at a whopping rate of 22.2 / million. Though St. Louis has a terribly high murder rate of 60 (per 100,000). Though New Orleans has a high murder rate of 39,000 and a cop-involved killing rate of (just?) 4.5 per million. (The US murder rate is about 5 per 100,000.)

    Above half a million population, the range in rates of killed by police goes from above 8 in Albuquerque, Tucson, Denver, Mesa, Oklahoma City down to New York City with a rate of 0.89. Nothing comes close. Nashville, Philadelphia, Boston and San Diego have annual rates between 2 and 3 per million.

    (Note I’ve changed the scale from the above charts. The x axis went to 30. Now it’s 14.)

    Keep in mind there are hundreds of smaller cities and counties between Albuquerque and New York City. But the disparity between cities at the top and bottom of the list! It’s immense. And nobody sees to be able to look up from the latest outrage and ask, why?

    So let’s give credit where it is due. By my figuring these departments all have killing rates under 1 per million (and serve populations over 180,000. If my data is correct, which it may not be). Their success should be applauded and emulated:

    Travis County Sheriff’s Office
    Montgomery County Department of Police
    New Castle County Police Department
    Gwinnett County Police Department
    Loudoun County Sheriff’s Office
    Chesterfield County Police Department
    Prince William County Police Department
    Santa Clara County Sheriff’s Office
    Fairfax County Police Department
    Monroe County Sheriff’s Office
    Arlington County Police Department
    Macomb County Sheriff’s Office
    Oxnard Police Department
    New York City Police Department
    Lubbock Police Department
    Lexington Police Department

    For those who understand such things, I also ran this regression for cities > 100,000. Dependent variable being the rate of police killings and independent variables being median household income, percentage black, murder rate, cops per capita and Hispanic/Latino percentage. Income matters (not a surprise). So does murder rate (obviously). But the negative correlation with Black percentage is of note. I was not expecting the lack of correlation with Hispanic/Latino percentage. My knowledge of advanced statistics doesn’t get much advanced that this, alas.

    And this is all subject to errors and corrections. This a blog. Not a peer-review article. Leave a comment or better yet email me. Or twitter @petermoskos

    Methods and sources:

    Fatal Encounters. https://fatalencounters.org/
    Population and police numbers mostly from here: https://ucr.fbi.gov/crime-in-the-u.s/2018/crime-in-the-u.s.-2018/tables/table-78/table-78.xls/view.
    City murder number I mostly keep track of. But through 2018 from this kind of source: https://ucr.fbi.gov/crime-in-the-u.s/2016/crime-in-the-u.s.-2016/tables/table-6/table-6.xls/view
    Other number from wikipedia and police department websites.
    And here: https://www.census.gov/quickfacts/fact/table/US/

    Killed by police data is from https://fatalencounters.org/. I gave $100; you should give few bucks, too. This is really important data, and it’s all the work of one guy. Plus he puts the format of the Washington Post’s gathering of similar data to shame.

    Then I filtered for intentional gun killings for each city, county, and police agency. From this I created a data set (one row) for each city, county, and/or agency. County data is tricky. Best I could, I figured out the population policed by large police agencies. But it’s not an exact science. (Basically take a county and subtract the cities and towns that have their own police.) There’s a lot of overlapping jurisdiction. There’s also the issue that a lot of sheriff department are responsible for jails, and I tried to exclude correctional officers (by leaving out non-sworn employees). But then in the end it turns out the number of cops per capita seems to not be that revealing, other than being correlated with murders per capita (yes, cities with more murders have more cops, presumable in that direction of causality).

    It’s also likely that some of the counties shouldn’t be included because their work is limited to courts and jails. Some of the police in these counties probably aren’t doing active policing, and hence shoot nobody. Also, murder data is probably accurate, because it comes from county departments reporting. And departments don’t generally claim other people’s murders. And some county department just don’t report any data. So some of the rates may be wrong. Long way of saying take county data with a grain of salt. But it’s still worth looking at.

    [Update] Here are the rates for every city in America with more than 200,00 people. Because somebody asked requested. This is the annual rate of people shot and killed by cops (2015-2019) in this city. Rate per million.

    Here’s county data. (Sorted by state, then city). Here I am including more data because I’m not confident about these rates. What is correct is the number of people killed by the agency in 5 years (Avg1yrKillAgcy). I’m not certain about the rate (KillMilAgcy) because I’m not certain about the population policed (Or the number of cops). If you know better, let me know.

    2020 caveat.

    Here’s some fancier statistical regression courtesy of Professor Gabriel Rossman. This is a work in progress.

    I think we get a few things from the Poisson:

    1. The satisfaction that it’s done right, or at least that it’s less wrong.
    2. Cops/1000 population is now significant. Given that the specification is technically better, as in the data better fit the model’s assumptions, you can probably trust this, or at least trust it at least as much as you could the OLS of rates
    3. You no longer need to worry about small n and zeroes biasing the models which means that even with a rare event you can include small cases. You no longer need to drop Mayberry from the dataset though obviously data cleaning is a pain with a bunch of small towns.

    12/7/2020 KillMilCity and KillMilAgcy are deaths as police homicides per million population.

    cops <- read_csv(file = "moskos_copshootings.csv")
    ## Parsed with column specification:
## cols(
##   .default = col_double(),
##   citystate = col_character(),
##   statecity = col_character(),
##   statecounty = col_character(),
##   state = col_character(),
##   agcy = col_character()
## )
    ## See spec(...) for full column specifications.
    glimpse(cops)
    ## Rows: 166
## Columns: 30
## $ citystate         <chr> "Kansas City KS", "Escondido CA", "Pomona CA", "S...
## $ statecity         <chr> "KS Kansas City", "CA Escondido", "CA Pomona", "M...
## $ murder1Avg        <dbl> 6.50, 4.50, 14.25, 15.00, 6.66, 1.25, 14.25, 4.00...
## $ statecounty       <chr> "KS Wyandotte", "CA San Diego", "CA Los Angeles",...
## $ FlagCityCounty    <dbl> 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0...
## $ spendCapita       <dbl> NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, N...
## $ Population        <dbl> 152958, 153073, 153496, 155179, 155503, 155637, 1...
## $ cop1K             <dbl> 2.4516534, 1.0125888, 0.9511649, 3.1511996, 1.929...
## $ Mur100K           <dbl> 4.2495326, 2.9397738, 9.2836295, 9.6662564, 4.282...
## $ BlkPer            <dbl> 23.5, 2.4, 6.0, 20.9, 18.0, 1.7, 24.1, 1.4, 1.3, ...
## $ HisPer            <dbl> 29.9, 51.9, 71.5, 44.7, 37.5, 17.3, 11.5, 23.1, 7...
## $ IncMedHouse       <dbl> 43573, 62319, 55115, 36730, 51917, 131791, 53007,...
## $ KillMilCity       <dbl> 9.152839, 1.306566, 5.211862, 0.000000, 3.858446,...
## $ KillMilAgcy       <dbl> 7.845291, 1.306566, 2.605931, 0.000000, 2.572298,...
## $ state             <chr> "KS", "CA", "CA", "MA", "FL", "CA", "TN", "CO", "...
## $ EastWest          <dbl> 2, 2, 2, 1, 1, 2, 1, 2, 2, 1, 1, 2, 2, 2, 2, 2, 1...
## $ agcy              <chr> "Kansas City Police Department", "Escondido Polic...
## $ Cops              <dbl> 375, 155, 146, 489, 300, 217, 278, 285, 151, 340,...
## $ countCity         <dbl> 7, 1, 4, 0, 3, 4, 3, 8, 2, 4, 1, 3, 6, 9, 1, 2, 1...
## $ killedByAgency5Yr <dbl> 6, 1, 2, 0, 2, 4, 2, 7, 2, 4, 2, 3, 4, 9, 0, 1, 1...
## $ CopsKill1Yr       <dbl> 0.003200000, 0.001290323, 0.002739726, 0.00000000...
## $ CopsKill20Yr      <dbl> 0.06400000, 0.02580645, 0.05479452, 0.00000000, 0...
## $ Murder4yrTotal    <dbl> 26, 18, 57, 60, NA, 5, 57, 16, 113, 244, 21, 32, ...
## $ LEO               <dbl> NA, 209, 269, NA, 394, 282, 342, 409, 204, NA, 40...
## $ Civs              <dbl> NA, 54, 123, NA, 94, 65, 64, 124, 53, 250, 88, 11...
## $ unique            <dbl> 26448, 19403, 24380, NA, 26303, 350, 25627, 27185...
## $ zip               <dbl> 66111, 92027, 91768, NA, 33024, 94089, 37042, 802...
## $ lat               <dbl> 39.11662, 33.14459, 34.05056, NA, 26.02650, 37.39...
## $ long              <dbl> -94.81942, -117.03364, -117.82068, NA, -80.22943,...
## $ `filter_$`        <dbl> 1, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 1, 0, 0, 0...

    Replicate post

    Reasonably good match for Moskos’s 7/5/2020 blog post but numbers aren’t exact. Perhaps it’s minimum population of 100,000 (blog) vs 150,000 (this notebook). Alternately may be a counties issue.

    summary(lm(data=cops,KillMilCity~IncMedHouse+ BlkPer + Mur100K + cop1K + HisPer))
    ## 
## Call:
## lm(formula = KillMilCity ~ IncMedHouse + BlkPer + Mur100K + cop1K + 
##     HisPer, data = cops)
## 
## Residuals:
##     Min      1Q  Median      3Q     Max 
## -6.3049 -1.6853 -0.1688  1.5078  9.7141 
## 
## Coefficients:
##               Estimate Std. Error t value Pr(>|t|)    
## (Intercept)  8.333e+00  1.347e+00   6.185 5.05e-09 ***
## IncMedHouse -4.975e-05  1.434e-05  -3.469 0.000673 ***
## BlkPer      -1.288e-01  2.242e-02  -5.742 4.61e-08 ***
## Mur100K      2.752e-01  3.707e-02   7.423 6.53e-12 ***
## cop1K       -2.078e-02  3.636e-01  -0.057 0.954492    
## HisPer      -1.659e-02  1.220e-02  -1.360 0.175703    
## ---
## Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
## 
## Residual standard error: 2.782 on 159 degrees of freedom
##   (1 observation deleted due to missingness)
## Multiple R-squared:  0.3511, Adjusted R-squared:  0.3307 
## F-statistic: 17.21 on 5 and 159 DF,  p-value: 1.368e-13
    cops %>% ggplot(aes(x=KillMilAgcy)) + geom_histogram() + labs(x='Police Homicides Per Million Population', caption='Agency, not city')
    ## `stat_bin()` using `bins = 30`. Pick better value with `binwidth`.
    cops %>% ggplot(aes(x=killedByAgency5Yr)) + geom_histogram() + labs(x='Police Homicides Over 5 Years, Raw Count', caption='Agency, not city')
    ## `stat_bin()` using `bins = 30`. Pick better value with `binwidth`.
    cops %>% ggplot((aes(x=cop1K,y=killedByAgency5Yr,size=Population))) + 
  geom_point() +
  labs(x='Number of Cops / 1000 Population',y='Police Homicides Over 5 Years, Raw Count')
    cops %>% ggplot((aes(x=Population,y=cop1K))) + 
  geom_point() +
  labs(x='Population',y='Number of Cops / 1000 Population')

    Poisson

    Because police homicides are events, they can be modeled with a count model. Assuming the events are independent net of observables, a Poisson is appropriate. This seems consistent with the histogram. If the histogram were much more right-skewed or if there were strong theoretical reasons to think police homicides were not independent, then a negative binomial could be appropriate.

    Because cities/ agency jurisdictions vary wildly in size, it’s best to include population as an offset to model the different exposure. That is, more people means more people at risk of getting shot by cops and the model accounts for that.

    Compared to the OLS analysis of rates, the Poisson analysis of counts is similar but now everything is significant, including number of cops and percent Latino, both of which are negatively associated with the counts of police homicides.

    summary(glm(killedByAgency5Yr~IncMedHouse+ BlkPer + Mur100K + cop1K + HisPer + offset(log(Population)),
            data=cops,family="poisson"))
    ## 
## Call:
## glm(formula = killedByAgency5Yr ~ IncMedHouse + BlkPer + Mur100K + 
##     cop1K + HisPer + offset(log(Population)), family = "poisson", 
##     data = cops)
## 
## Deviance Residuals: 
##     Min       1Q   Median       3Q      Max  
## -3.9061  -1.2174  -0.1628   0.9152   3.3863  
## 
## Coefficients:
##               Estimate Std. Error z value Pr(>|z|)    
## (Intercept) -9.609e+00  1.748e-01 -54.973  < 2e-16 ***
## IncMedHouse -1.068e-05  2.140e-06  -4.993 5.95e-07 ***
## BlkPer      -2.789e-02  3.018e-03  -9.242  < 2e-16 ***
## Mur100K      5.070e-02  3.583e-03  14.149  < 2e-16 ***
## cop1K       -2.050e-01  3.459e-02  -5.926 3.10e-09 ***
## HisPer      -5.088e-03  1.536e-03  -3.312 0.000925 ***
## ---
## Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
## 
## (Dispersion parameter for poisson family taken to be 1)
## 
##     Null deviance: 654.06  on 164  degrees of freedom
## Residual deviance: 369.63  on 159  degrees of freedom
##   (1 observation deleted due to missingness)
## AIC: 960.63
## 
## Number of Fisher Scoring iterations: 4

    Percent Black vs Murder Rate

    There is a 0.772 correlation between % black and the murder rate, which suggests possible collinearity. As such,

    Note that the murder only version has a lower AIC so if forced to choose that’s the better model. Also note that when only one at a time is included, murder remains positive and black remains negative. Whatever is driving the murder and black effects, it is not collinearity.

    cops %>% ggplot((aes(x=BlkPer,y=Mur100K,size=Population))) + 
  geom_point() +
    labs(x='Percent Black',y='Murders per 100,000')
    ## Warning: Removed 1 rows containing missing values (geom_point).
    cops %>% ggplot((aes(x=Mur100K,y=killedByAgency5Yr,size=Population))) + 
  geom_point() +
  labs(x='Murder Rate',y='Police Homicides, Raw Count')
    ## Warning: Removed 1 rows containing missing values (geom_point).
    cops %>% ggplot((aes(x=BlkPer,y=killedByAgency5Yr,size=Population))) + 
  geom_point() +
  labs(x='% Black',y='Police Homicides, Raw Count')
    summary(glm(killedByAgency5Yr~IncMedHouse+ Mur100K + cop1K + HisPer + offset(log(Population)),
            data=cops,family="poisson"))
    ## 
## Call:
## glm(formula = killedByAgency5Yr ~ IncMedHouse + Mur100K + cop1K + 
##     HisPer + offset(log(Population)), family = "poisson", data = cops)
## 
## Deviance Residuals: 
##    Min      1Q  Median      3Q     Max  
## -4.068  -1.450  -0.369   1.020   4.553  
## 
## Coefficients:
##               Estimate Std. Error z value Pr(>|z|)    
## (Intercept) -1.015e+01  1.696e-01 -59.874   <2e-16 ***
## IncMedHouse -5.010e-06  2.025e-06  -2.474   0.0134 *  
## Mur100K      3.071e-02  3.195e-03   9.612   <2e-16 ***
## cop1K       -3.397e-01  3.132e-02 -10.846   <2e-16 ***
## HisPer       1.663e-03  1.378e-03   1.207   0.2274    
## ---
## Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
## 
## (Dispersion parameter for poisson family taken to be 1)
## 
##     Null deviance: 654.06  on 164  degrees of freedom
## Residual deviance: 458.59  on 160  degrees of freedom
##   (1 observation deleted due to missingness)
## AIC: 1047.6
## 
## Number of Fisher Scoring iterations: 5
    summary(glm(killedByAgency5Yr~IncMedHouse+ BlkPer + cop1K + HisPer + offset(log(Population)),
            data=cops,family="poisson"))
    ## 
## Call:
## glm(formula = killedByAgency5Yr ~ IncMedHouse + BlkPer + cop1K + 
##     HisPer + offset(log(Population)), family = "poisson", data = cops)
## 
## Deviance Residuals: 
##     Min       1Q   Median       3Q      Max  
## -7.4545  -1.3608  -0.2578   1.0028   7.3989  
## 
## Coefficients:
##               Estimate Std. Error z value Pr(>|z|)    
## (Intercept) -9.177e+00  1.750e-01 -52.428  < 2e-16 ***
## IncMedHouse -1.642e-05  2.183e-06  -7.524 5.33e-14 ***
## BlkPer      -6.959e-03  2.518e-03  -2.764  0.00572 ** 
## cop1K       -1.973e-01  3.238e-02  -6.094 1.10e-09 ***
## HisPer      -4.604e-03  1.548e-03  -2.973  0.00295 ** 
## ---
## Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
## 
## (Dispersion parameter for poisson family taken to be 1)
## 
##     Null deviance: 654.55  on 165  degrees of freedom
## Residual deviance: 533.19  on 161  degrees of freedom
## AIC: 1125.7
## 
## Number of Fisher Scoring iterations: 5

  • More on state differences in cops shooting people

    Inspired by some twitter threads — mostly this onewith Gary Cordner and this onewith Andrew Wheeler — I thought I’d look more at the cops getting killed as a factor in cops killing people.

    I like presenting this stage of research. In part because coming up with ideas and hypotheses and basic number crunching is what I like doing most. (I’ll leave the journal article submitted and advanced stats to others.) I’ll explain my steps partly to help others, but also to help me go through this on the old assumption that if you can’t explain it to others clearly, you don’t really understand it yourself. (I used Excel and PSPP.)

    I’m always partial to fewer better-data over more bad-data. So, as I often do, I’d like to stick with good old murder: officers shot and killed on duty (from the officer down memorial page, which over the years I’ve found close to faultless, which is more than one can say for the UCR or anything else.)

    The problem (from a statistical not a moral sense) is that there are many states in which very few officers are killed. So I went back to gather 20 years of data (for no particular reason, just a choice, it could have been 10 or 30) and got the number of officers killed between 1999 and 2018, by gunfire, for each state. 50 states. 990 total deaths. I dropped the states where n < 10. That leaves 33 states. Texas and California top the list, which isn’t surprising because they’re big states. But then come Georgia, Florida, and Louisiana. Interesting…

    But what’s the best denominator? I mean obviously one needs to look at population to get a rate. But which population? In order, I’m going to consider 1) number of cops, 2) population levels, and 3) violent crime levels, 4) population density, and 5) percent of population that is African American.

    1) Perhaps we should look at cops killed in terms of how many cops there are in any given state, so as to consider the chance of any given cop being killed on duty. Makes sense to me, the problem is that the official data even on how many cops there are looks dodgy. It seems unlikely to me, for instance, that Mississippi went from 5,222 cops in 2007 to 2,524 in 2014 (the two years anybody attempted to count, but reporting is voluntary). If I don’t trust the data, I don’t want to use it. But I still did the numbers, based on the average between 2007 sworn officers and 2014 sworn officers.

    For presentation purposes, let’s use the USA average (using all 50 states) as a baseline, set that to 0, and compare all the states:

    Cops are more likely to be killed in MS, LA, AR NM, SC, GA, and AZ. Keep in mind the small and safe states have been removed from the calculation. I don’t like this. If nothing else because I don’t trust the Mississippi numbers.

    2) So let’s just use overall population as the denominator. I’m using 2016 population because that’s what I already have in my file. Some states have grown a lot in the past 20 years. Oh, well. I don’t think it matters that much for these purposes. If it does, we can consider it later. Keep in mind these are ratios, the actual numbers by themselves are meaningless. But as a ratio, yes, a value of 1 means a cop is twice as likely to be killed per capita. It does appear that a cop in Louisiana is about 4 times as likely to be shot and killed as a cop in New Jersey.

    This says that Louisiana, by far, is the most dangerous state to police in. Arizona is next. And given that its population has grown drastically in the past 20 years, it should really be higher. And that would make LA seem like less of an outlier.

    New Jersey, Massachusetts, and New York are all comparably safe. I won’t say the safest because the 17 safest (and smallest) states have all been dropped for the statistical reason of having fewer than 10 cops murdered over the past 20 years.

    I think number 2 (population) is better than number 1 (number of cops). But they’re not drastically different. You get the same states on top and the same states on bottom. But I’m going with state population as the denominator because I don’t trust the count of cops.

    3) Now let’s consider violent crime as an independent variable (which is the variable that affects something else, on which something else is dependent). And back to using all 50 states.

    I just got some crude numbers off wikipedia and then took an average of 5 years of data for each state. (Not the best methods, but probably accurate. Certainly fine for preliminary work.)

    Let’s run some correlations. I like correlations because they’re easy to understand. They also tell you where you should look for deeper answers.

    First question: at the state level, is violent crime rate correlated with cops getting killed? Absolutely (Pearson Correlation = .62, Sig = .000). This is a strong and unsurprising relationship.

    Next, at the state level, is violent crime correlated with being killed by cops? Surprisingly, technically, statistically, no. (correlation = .23 sig = .104) Not at the state level; not with an N of 50. Now I know from other research that violent crime is correlated with being killed by cops, but you’ve have to delve down into the neighborhood level to see that effect. But still, if it that doesn’t come out at the state level, it’s a clue that something else is also at work! This is where things get interesting. Something else is also at play on a state level that is more significant than straight-up levels of violent crime.

    4) What about geographic area? This is where wikipedia is great because you can get state size in seconds. And then if you already have population and you’re handy with cut-and-paste and sorting on spreadsheets, you can get population density in minutes.

    And it turns out the population density is indeed correlated with a lot.

    Lack of density — more space — is correlated with being more likely to be killed by cops. Think of what this means. Common sense tells you it’s not a view of “big sky country” that makes cops shoot someone. Whatever really matters, is correlated to density (or lack thereof). Maybe it’s single person patrol. Or the time for backup to arrive. Or meth labs. Or gun culture. This is why they say “correlation doesn’t equal causation” (which is also the most frustrating phrases in social science, because correlation can very much indicate causality, and the phrase is often used to dismiss meaningful correlations as meaningless.)

    Population density (lack thereof) is also correlated with cops being killed. Density is not at all correlated with crime (like not even leaning in one direction). And yet both crime and density are heavily correlated with a lot of other factors. And both are correlated with cops being killed. More crime = more cops killed; more density = fewer cops being killed.

    So now lets do a brief multivariate analysis, which is about as far as I go. This means that we look at more than variable at the same time. Which is more important (plays a greater role) in cops being shot and cops shooting people? Crime or density? (Or something else.)

    Density seems to be more predictive than crime in terms of cops killing people and less important in terms of cops being killed (though for the latter both are correlated).

    When I move “cops killed” to the independent variable side and keep a focus on people killed by cops, density becomes less important and violent crime becomes more important. This makes intuitive sense. Because the issue with a spread out area is that cop, alone, would face greater threats.

    Keep in mind the above is about cops being killed. Much more talked about (by non cops) is people killed by cops. I wrote about that a few days ago.

    If you’re still with me, kudos. Causes here’s where the whammo happens!

    Were one to only look at individual variables, the key would seem to be density followed by crime and rate at which cops are killed. But it turns out that much of what is measured in those variables are simply correlated with and less important than the percentage of black population in a state. Crime matters. Police being killed matters (independently of violent crime), population density may matter a little, and of course other variables that I’m not even looking probably matter a lot. The question is always if they can be identified and accurately quantified.

    Last year I observed that cops shoot more often in states that have fewer blacks. So I already had a strong hunch to look in this direction.

    When one puts the state’s percentage of African-American residents into the equation, things start to fall into place. This is also taking into account how often cops get shot, crime, and density (which finally starts to lessen in importance — because as we know is only indicative of other factors — but still probably important in terms of gun laws and culture and police-backup).

    If one considers crime, density, and black percentage — but only when one does so all together — all three are significant (with an R-squared of .55). When one adds the rate at which cops are killed, r-squared goes up to .62.

    [R-squared is technically the distance (squared to take account of negative numbers) that data points are from the trend line of a chart. At some level, r-squared is supposed to indicate how much of what is being looked at is explained by the independent variables in the statistical regression. But that’s more in a statistical sense than a real-world sense. Still, generally, other things being equal, a high r-squared is better than a low r-squared. And an r-squared of 0.63 ain’t shabby for this kind of game.]

    So what does all this mean? Density matters, but not so much for what it is but for things correlated with it (same could be said for race). All these variables have “intervening variables,” the way people act, the choices they make, the factors that make us do what we do. Things that may be harder to measure than crude indicators like “population density” and “race.”

    Still, looking a these variables, density seems mostly to correlate with the lack of African-American in a state. The black percentage of a state seems to be the most significant factor in determining how many people are shot and killed by police (with overall violence and cops being killed also being important). But, contrary to what many people believe — and basically all of the “narrative” of the past few years — the relationship is inverse. The greater the percentage of blacks in a state, the less likely cops are to shoot and kill people.

    This is counter-intuitive to a lot of people, particularly if you think cops only shoot black people. But it makes perfect sense if one thinks about it in two parts:

    1) Whites don’t really care about who police shoot; period; end of story. And without the pressure over bad (or even good) police-involved shootings, cops never learn how to shoot less. Other things being equal, cops simply shoot more people if there isn’t any push-back from (to over-generalize) blacks and liberals and media and anti-police protesters. Call it the Al Sharpton Effect, if you will. Basically, in many places, police organization and culture do need to be pressured into changing for the better.

    2) Police can be recruited, trained, and taught to less often use legally justifiable but not-needed lethal force less. The state variations in police use of lethal force are huge. Some states (and particularly jurisdictions within states) do it better than others. Instead of saying “police are the problem” we could look at the states and cities and department that are doing it better and learn.

    Ultimately what we need are well and better trained police officers who shoot less often, but still shoot when needed.

    I’ll leave you one final bit of data. I don’t know if there’s a there here or not. My guess is this does matter. But maybe it’s just a clue that leads to the above. Or maybe it’s something else. Maybe you can figure it out.

    This is a table that shows a simple ratio: the number of citizens killed for each cop killed. Good people can debate what this ratio should be. I don’t want to go there. The correct ratio is no cops getting killed and few criminals getting killed. But what’s interesting to me is the that there is such a large difference between the states, and by a factor of 10! By and large the states on the high-end (more citizens getting killed) are very white and the states on the low-end (fewer citizens getting killed) are disproportionately black.

    Take Oklahoma. Cops in Oklahoma are not getting killed a lot, per capita or per number (0.6 per year over the past 20 years). There’s not a lot of violent crime, and yet in the past 4 years cops in Oklahoma have killed 118 people. Again, I don’t want to get into what the correct ratio is, but seeing how the national average is 20 civilians-to-cops shot and killed, and seeing how some states are down under 10, why the hell is Oklahoma pushing 50?

    Louisiana cops are getting shot at and killed three-times more often than cops in Oklahoma (and 8 times more often than cops in New Jersey). Both Oklahoma and Louisiana cops shoot a lot of people. But in Louisiana, dare I say, they have good reason to.

  • How to make people care about violence

    How to make people care about violence

    Over at Nola Crime News, Jeff Asher tweeted this graphicjust now.

    Click on it; it moves! So while people are dying, I’m thinking about data presentation. There’s something about a moving line that may make one pay attention to dead people in a way that actual dead people don’t.

    Jeff’s graphic looks at Baltimore City shooting victims over the past 365 days. Each data point tallies the total number of shooting victims over the past 365 day. This nullifies seasonal change, which is worth a lot. But by taking a past-year average, you lose the “BAM” of what happened literally overnight, after six police officers were criminally charged for the death of Freddie Gray. The violence didn’t just “increase.” It stepped up, by two-thirds. Overnight. After April 27, 2015. The visual above indicates a rapid but continuous increase over the course of a year. But it’s still a good visual and can’t think of better one.

    I don’t know how to present a good visual that shows what has happened in Baltimore. In the past I’ve tried with a pre- and post-riot trend line. Not just once, but twice. But that’s hardly convinced the masses that police (or more dead bodies) matter.

    People are already talking about the rise in violence in Baltimore in terms of poverty or drugs or police legitimacy or blah-dee-blah. And sure, all that matters. But stop it! None of that, not any of that, explains the increase in violence. Police because less proactive because A) innocent cops were criminally charged and B) Political pressure (from the mayor, the police commissioner, and the US DOJ) told police to be less proactive as a means to reduce racial disparity in policing. You see it Baltimore. You see it Chicago. You see it in New Orleans. The problem is you’re seeing it basically everywhere.

    Here’s New Orleans, again from Jeff Asher.

    These increases are no joke. This is a “holy shit” type increase in violence. And the chart under-presents the quickness of the increase.

    What happened in New Orleans? I don’t know NOLA as well as Baltimore or New York. But the NOLA PD has seen a 30 percent reduction in manpower and a massive reduction in proactive policing (as measured by drug enforcement. I also suspect the consent decree hasn’t helped police in terms of crime prevention, since, and this is important: crime prevention isn’t one iota of any consent decree. Somehow, crime is supposed to manage itself while police are better managed.

    The only big city of note without an increase in violence is NYC. And even here, people objectto the exact kind of proactive policingthat keeps crime from rising. Luckily, at least in New York, even liberal Mayor de Blasio isn’t listening to the “police are the problem” posse.

  • Spin This: The biggest murder increase in 45 years

    Murder is up. Who knew? (I’ve been saying so since last October.) Eventually, we’re all going to have to accept this (not in a moral sense but in a statistical sense). The accepted liberal reaction to this increase seems to be “it’s not a big deal” and “Don’t freak out.” Let’s not get “hysterical.” Let’s talk about “gun control.” (In the early 1990s, by the way, it was all about “drug treatment.” That didn’t happen either. And crime went down.)

    What I really do not understand is why the Left is willing to concede crime prevention to the Right. (I bet Trump won’t be downplaying this in tonight’s debate.)

    False argument #1: The best violence-reduction strategy is a job-production strategy.

    It sounds nice, but I say bullsh*t. As if unemployed people just can’t help but shoot each other.

    Do not get me wrong: Poverty is bad. But it just so happens that 2015, the year with the big murder increase, also saw the biggest decrease in poverty since 1991. 3.5 million people rose out of poverty last year. That’s great news. It really is. (Full report & summary in the NYT.)

    But we still hear this from people like this St. Louis alderman:

    How do we use that [crime] data to elevate the consciousness of our community? How do we use that data to provide the opportunity for people to get meaningful jobs, with livable wages?

    No. I mean, yes! Please, work on that, too. But the question from these data is how the hell we get police back into policing and crime prevention. Sure, it sucks when dad loses his job, but consider how much worse it is for dad to get killed coming home from work. (I would even say that you can’t have a real job-production strategy until you achieve violence reduction. Who the hell is going to open a business where you will get robbed and workers get mugged walking to their car?)

    The Guardian goes on to summarize the Brennan Center’s position:

    Last year’s national murder increase was not a uniform trend, but a sum of contradictory changes in cities across the country. Early analyses of the 2015 murder increase suggested much of it might be driven by murder spikes in just 10 large cities.

    (Now I see how clever the Brennan Center was to put out their paper last week, so it becomes cited immediately to put things “in context.”)

    False argument #2: It’s just happening a few cities.

    No. It’s not.

    Homicide (and almost all violent crime) is up in every grouping of towns and cities (such as “under 10,000” and “over 1,000,000”). Period. Now that doesn’t mean it’s up in every city. But what a weird and nonsensical standard. Sure, if we remove all the places where crime is up, crime wouldn’t be up. But that’s we have fancy statistical concepts like “overall,” “in general,” and “trend.”

    Even if we were to remove the 6 cities with the largest increase — and I don’t know why we would — but just to see if the problem is isolated in a few cities, let’s take out Baltimore, Chicago, Milwaukee, Washington, Cleveland, and Houston (collectively those cities saw about 420 more murders in 2015) — even without these cities the rest of America would still have 600 more murders and the biggest homicide increase in 25 years. That’s how bad these just released numbers are.

    Now we can say that violence in concentrated in certain neighborhoods. That’s true. But we’ve long known this. Indeed, as you can tell from looking out your window, there aren’t armed marauders outside your castle gates. What matters, or at least should matter, is that more American are being murdered. I find it distasteful (particularly when it comes from the Left) to say “most people” don’t have to worry about crime because the “average person” is still safe. The fact that violence disproportionately affects a subset of Americans may indeed mean it’s not a “national crime wave,” but it is all the more reason to care.

    False argument #3: It might just be a statistical blip.

    But it’s not. I mean, it could be a statistical blip…. If it were just one or two percent. But it’s up 11 percent. The last time we saw an 11 percent one-year increase in murder was 1971. That’s exactly my entire lifetime. And that was in the middle of eight-year run when homicides doubled from ten to twenty thousand. This “blip” was literally the deaths of 1,600 more Americans. The number of people killed went up from 14,164 in 2014 to 15,696 in 2015. That one-year increase negated 5 years of homicide decline.

    If you think this increase in murder “no cause for alarm” and people who care are “overreacting,” to you, I respectfully say “go to hell.” We worked too hard to get to where we are (or were) with lower crime. And a “don’t-overreact” reaction does not help. And it may lead exactly to the right-wing law-and-order backlash you so fear. (But on the flipside, to those who don’t really care but will use these deaths to make some racist point about “black-on-black” crime and “those people,” I say with all my heart, “no really, to hell with you, too!”)

    Why I care (and why you should, too):

    Among academics, it’s quite uncool to blame criminals for crime or give police credit for crime prevention. But then how many statisticians who use the UCR Homicide Supplement can point to a specific row and say, “Yeah, I handled that one.”

    Too many who say they’re for “justice” never really have to think about the injustice of just even one real murder victim (one not shot by police). But then maybe I care because I was a Baltimore cop. Every single cop can tell you a story about a dead person. Why? Because they care. Granted, some cops do care more than others, but you can’t police and not care.

    I wasn’t a cop for long (less than 2 years in total), and even I lost track of how many victims I dealt with. But a few do stand out. And this isn’t even getting into my cop friends who were shot, killed, nearly killed, had to kill somebody, or carry physical and emotional wounds for life.

    I remember the stare of a young black man at the same track we ran around while in the academy. His backpack made me think he was a good kid, on his way home from school. He was shot, perhaps after being robbed. We made long eye contact, even though he was dead.

    I remember the guy with a gunshot to the head one 321 Post. He was still alive when I got to him. But he clearly a goner, with blood and brain dripping from the hole in his head. His sisters were wailing while he died.

    How many Harvard PhD students have the intimate experience of sorted through a victims’ clothes? Clothes that are literally dripping with blood and yet still reeking of body odor. You’re trying to go through everything, looking for pockets, for any sign of identification of the life that used to be. And then there are the death notifications.

    Think of all those deaths. Last year there were 133 more murders in Baltimore than there were in 2014. [This year the numbers are down slightly compared to 2015, and the chutzpah of some people to herald Baltimore’s “crime drop” is shocking.] Take a moment and picture all those dead bodies, almost all shot young black men and teenagers. Visually stack them up like cordwood if you wish, or lay them all head-to-toe. It’s real human carnage.

    If you took all the Baltimore murder victims from just last year and laid them head-to-toe where the Ravens play football, that line of dead bloody bodies could score six endzone-to-endzone touchdowns. And the increase in violence last year happened all after April 27th. All it took was one man’s in-custody death coupled with anti-police protests, bad leadership, a riot, and a politician’s horrible choice to press criminal charges against six police officers in the matter of Freddie Gray’s death. (All charges ended up being dropped after multiple trials without a single conviction on any charge.)

    This is actually one time I don’t care about the historical perspective. Less than the 1990’s crack-crazy murder rate is not good enough. We got down to a homicide rate like Canada (about 1/4 of ours), and maybe I’ll be satisfied. We can start caring now. Or we can start caring after a few more thousand people are needless killed. And if you think I’m over-reacting, consider that you might be under-reacting.

    [Posts in this series: 1, 2, 3, 4, 5, 6]

  • 2015 UCR is out

    The 2015 UCRis finally out. This means we have real numbers on last year. And the numbers are not good. Homicide is up 10.8 percent. That’s biggest increase in 45 years. Don’t downplay it.

    I’ll talk about that in my next post, but first the boring roundup:

    Firearms were used in 71.5 percent, which is up from last year’s 67.9 percent. That’s 1,500 more murders by firearm.

    52.3 percent of all victims are black. (Up ever-so-slightly from 51 percent in 2014.) 906 more black men were killed in 2015compared to 2014 (6,115 vs. 5,209). That’s a very big 17.4 percent increase (murder among white men when up 9.2 percent). To put those numbers in perspecdtive, police shot and killed 248 black menlast year (and 10 black women). Most of those were justified.

    21 percentof homicide victims are women, same as last year. And women are 7 percent of known offenders.

    And though I don’t like looking at other crime stats (because I don’t trust their reliability) rape, robbery, and aggravated assult are all up as well. Reported property crimes are down a bit, but I suspect that’s more due to people’s decreasing desire to call the police or waiting for them to show up.

  • Brennan Center: No need for “most Americans” to worry about more murders

    The good people at the Brennan Center for Justice at NYU School of Law have assured us (pdf link to report):

    Reports of a national crime wave were premature and unfounded, and that “the average person in a large urban area is safer walking on the street today than he or she would have been at almost any time in the past 30 years.”

    …

    The authors conclude there is no evidence of a national murder wave…. Most Americans will continue to experience low rates of crime…. There is not a nationwide crime wave, or rising violence across American cities.

    Ah, yes.

    Cause for a moment there, I was kinda worried that more people were getting killed. But it turns out, I guess, that it was illiberal of me to care about people who are particularly at risk of being killed. Also, did you know:

    Homicides are concentrated in the most segregated and poorest areas of the city.

    I hadn’t thought of that. And since that’s not where the “average person” lives, I guess we don’t need to worry.

    Maybe I should just jump on this illogic ideological bandwagon of denial to see where it goes:

    By “historic standards,” racism is pretty low in America. QED: Not a problem.

    #BlackLivesMatter can close up shop because “most Americans” don’t have to worry about being shot by police.

    Enough with all those new letters, the “average American” doesn’t face any LGBTQ discrimination.

    Check, check, and check. Problems solved!

    Oh, but while you’re here. Not that it’s any cause for concern. But there is this little issue…:

    The murder rate is projected to rise 13.1 percent this year…. [and] 31.5 percent from 2014 to 2016.

    Say what?!

    [Update: 2015 stats are out. The rate, based on recorded homicide, increased from 4.26 to 4.75 per 100,000. The rate, based on estimated homicides, increased from 4.44 to 4.90. Recorded 13,594 homicides in 2014 (estimated 14,164). 15,192 in 2015 (estimated 15,696). 2014 estimated population 318,857,056. 2015: 320,090,857.]

    If these numbers are correct — and they may not be (there is some odd math in this report*; and keep in mind 2015’s national UCR stats haven’t yet come out) — but if these Brennen Centers estimates are correct, that would mean 2015 saw a 16.3 percent increase in the homicide rate.

    So all we’ve got is just your average 16.3 percent annual increase in murder. I mean, we had one of those, well, uh, actually, never. This would be the largest increase since the government has been keeping track. (An increase in 1921 might have been greater, but we don’t really know.) The last time the UCR recorded a 31.5 percent increase in two years was, oh, never.

    [In raw numbers the homicide increase is the greatest in 25 years. But it’s standard industry practice to use rates and percentages.]

    [Update: I’ve been informed over in the twitter world that when they say “nationwide” they don’t mean “nationwide” but “in the top cities.” I would expect the national increase to be less than what is found in the top cities. But I don’t know. Anyway… the 2015 UCR data will be out this week. And then, at least when it comes to last year, we can all stop speculating and know how big the increase in homicide was.]

    As to their overall point that homicide may be up but “crime” is little changed? I just call bullshit. Not on their analysis, per se. It’s just that crime numbers are not as reliable as homicide numbers. Trust homicide. Crime numbers are heavily influenced A) by proactive police and arrests (which are both down) and B) non-reporting (probably up). I trust the strength of the correlation between homicide and other violent crimes more than I trust the data on other violent crimes. If homicide is up, violent crime is up. Trust me on that one.

    *They’ve got some weird math here I can’t figure out:

    The national murder rate is projected to increase by 13.1 percent. Nearly half of the increase (234 out of 496 new homicides) will occur in Chicago. (page 1)

    But if the national rate goes up 13 percent this year, we’d see something closer to 1,500 more homicides. (Based on 2014 rate of 4.5 and 13,472 homicides.) And Chicago’s numbers will be up by about 200 this year. This is closer to 15 percent. What gives?

    Baltimore, Chicago, and Houston are projected to account for 50 percent (517 of 1041) of new homicides between 2014 and 2016. (page 8)

    But if the murder rate is up 30 percent, we’ll have closer to 2,500 new murders. I do not understand.

    Also, these semi-annual “crime isn’t up” reports from the Brennan Center have this odd habit of saying, “if we remove the cities where the increase is the greatest, the increase really isn’t so great. (An odd statistical proposition, to say the least.) But let’s play along and “pull a Brennan.” Let’s remove Chicago, Houston, and Baltimore because (I think) in terms of raw numbers, those cities have the greatest increase in homicides, 2014 – 2016 (roughly 240, 165 and 115, respectively). After we “pull a Brennan” we lose about 520 murders. That’s a lot, but we’d still have close to 17,000 homicides in 2016, which would be a 2-year increase of 20 percent. And even that should be cause for alarm.

    [Posts in this series: 1, 2, 3, 4, 5, 6]

  • “The Light’s Better Here”

    A nice critique of quantitative data over at City Observatory. It’s about transportation planning, but the lesson can be applied to anything, especially policing:

    Reliance on data to solve complex problems is subject to what’s sometimes called the “drunk under the streetlamp” effect: An obviously intoxicated man is on his hands and knees on the sidewalk, under a streetlamp. A passing cop asks him what he’s doing. “Looking for my keys,” the man replies. “Well, where did you drop them?” the cop inquires. “About a block away, but the light’s better here.

    If anything, we have too much data on arrests, response time, clearance, even (sometimes) use of force. These are easy things to count. That doesn’t make them particularly useful or qualitatively significant. Things you can count won’t lead us to solutions that involve foot patrol, discretion, and positive interactions with the public. In policing, a job well done is just too hard to count.

  • “Communities don’t commit crime; people commit crime”

    I love when other people do my writing for me! Thank you and keep these coming:

    On the racial disparity stuff, at least the DOJ is claiming they controlled for crime rates. (see p.63). But did they really? The report is full of boneheaded statements like “Despite making up only 41 percent of the Northern District’s population, African Americans accounted for 83 percent of stops in the district.” (p.65, n.72). Gee, how much of the violent crime in the Northern District is being committed in Greenmount [ed note: poor black] as opposed to Roland Park [rich white]?

    And people always talk in terms of higher crime rates in certain “communities” — as I just did. But communities don’t commit crime; people commit crime. Saying black people live disproportionately in high-crime neighborhoods is a euphemistic way of saying black people commit a disproportionate amount of crime. (And we’re talking about violent crime here, which is what drives street-level proactive enforcement). [Last year 93 percent of Baltimore homicide were black]

    If they tried to control for crime rates by focusing on the rates of crime in different communities rather than the rates of offending by black and white residents, isn’t that problematic? (Genuine questions here; I’m no statistician) [Ed note: Yes]. Doesn’t this all come down to how the DOJ drew the boundaries for the neighborhoods they studied? What did they use? Police districts? It seems like it. Look at p.63 (referring to disproportionate stops of African Americans despite “significant variation in the districts’ demographic characteristics and crime rates”).

    Let’s say the fictional North Central district is 100% white, and the fictional South Central district is 100% black. The crime rate is twice as high in the South Central, and people in the South Central are twice as likely to get stopped by police. I assume that would pass muster with the DOJ. But, of course, real districts in Baltimore don’t look like that. They’re mixed and largely segregated. They look like the Northern District, with lots of crime in black neighborhoods like Greenmount and hardly any crime in white neighborhoods like Roland Park. They look like the Southeast district, with far less crime in largely white Sector 1 than in Sector 2 (bordering the Eastern), where the population is mostly black. It seems to me this messes up any attempt to control for crime rates if they’re only looking at things at the district level. Better would be to break it down by individual neighborhoods, though that leaves room for error, too. Even better would be to study actual rates of offending by black and white residents. What percentage of robbery suspects were reported as black? What percentage of shooting suspects in cases that had eyewitnesses? What percentage of suspects from “CDS outside” complaints were described by the callers as black?

    Wouldn’t it be a powerful illustration to just look at the gender of people stopped? Men presumably account for about 50% of Baltimore’s population. Surely they make up a much higher percentage of those stopped and arrested. And I’m sure the disparity persists even if one attempts to control for crime rates in different “communities.” (I doubt the male/female ratio differs all that much district-to-district). Is that evidence that BPD cops are profiling men, and have an anti-male bias? Or are cops focusing on men because, well, men commit most of the crime in the city, and therefore it’s mostly going to be men who exhibit suspicious behavior that prompts officers to stop them?

  • Let’s all put our thinking caps, everybody

    Why in the world are people reporting on last year’s data when we have current data at our fingertips? Does reality only happen after numbers are published on DOJ letterhead?

    NBC News reports:

    Number of Police Officer Killings Drops, Reversing 2014 Spike.

    Do reporters not have the wherewithal to see what the current situation is? It is but a mouse-click away. Officer Down Memorial Page keeps an excellent running and historical tab.

    This year’s data show killings are up (albeit with a small n of 17). We’re on pace for 45 cops being shot and killed this year, and that would be much more than last year, and similar to the 47 shot and killed in 2014.

    Now I wouldn’t want the headline to say “Police Killings Spike 55 Percent in 2016,” but at least that would be technically true. But it would be statistically irresponsible, with such a low n.

    Oh, look what I just found. This is too good to be true:

    2016 Sees Year-Over-Year Spike In Cops Killed By Gunfire.

    And it’s also from NBC News! (File under “You can’t make this shit up!”) Well, at least NBC has all their bases covered.

    Anyway, it looks like 2015 is going to be the blip, not 2014.

    [I corrected a wrong number I had regarding 2014.]

  • Null finding

    Null finding

    Wouldn’t it be great if I could show that video games decrease violence because young men, rather than hanging out and starting fights and shootings on the corner, stay inside and play “Call of Duty”?

    I don’t think people over 40 understand just how big video games are. Compare games to movies. “Star Wars: The Force Awakens” has grossed $935,500,000 to date. Grand Theft Auto V grossed more than that in three days.

    Overall I still suspect that video games had something to do with the crime drop in the late 1990s. That was when video games got good enough that even cool kids stayed inside to play them. And if you’re home, you’re not on the street getting into trouble. More recently video game sales have leveled off. And social media has grown, which may contribute to a rise in crime. Who knows? And how would you prove it?

    As preliminary research I thought I’d look some big games and try and link their release to a drop in nationwide homicide.

    But — and not for the first time — I was stymied by the UCR. They don’t break crimes down by days. Only months. And that’s not good enough. Damn them!

    But at least for starters I could go to Opendata Baltimore to see if anything was going on in Baltimore. I looked at all assaults and also homicides in the 7 days both before and after release day (excluding the day of release because of hassles with the early morning hours).

    It looks like nothing is up. The data look random. Anyway, overall there’s certainly no decrease. Homicides are up, but the n is low and it’s probably just random fluctuation (and assaults are down).

    Well, I tried.