Sunday, July 10, 2016

Racism and Bears: How systemic racism leads to minorities viewing police as predators

There is so much I want to talk about right now, but I cannot talk about everything at once.

I've written before about how the social contract between the public and the police in this country needs to be renegotiated.  We've never truly come to a real cultural agreement about police in this country.  And I think that the lack of a real understanding of what police are leads to these cycles of violence.  I wrote that post two years ago during another round of killings by and of police.

I've written before about how gun control, Second Amendment rights, and police militarization are linked and all applied very differently to black people in America.

I've written about how "common sense" type weapons laws are actually used to abuse and persecute minorities.

I've even written about how complicated actually trying to understand violence statistics is, and how many ways there are to mess up data with subtle errors.  And that is assuming that there is no agenda, and when people (including myself) write about violence there is almost always an agenda.

And yet there is ever more to talk about.

I want to talk more plainly about things I've already touched on.

I also want to talk about what police go through.  The dangers they face, the things they have to experience, and how little attention their heroism and sacrifice receives.

I want to talk about the problems of police being used as revenue generators by governments.

I want to talk about so many things, but what I want to start with today is by trying to explain how structural racism works.  I want to explain why complaining about a racist system is not the same as calling police racists.  It's not.  I realize that I will probably fail to change many minds, but I need to at least be able to say that I've tried.

Philando Castile, even though this article doesn't focus as much on Mr. Castile, I still felt that he is emblematic of this current cycle of violence by and against police.  Particularly since he did nothing to become an emblem of violence.

I'm going to start my explanation with an example and position that I am sure will be seen as wrong by many:  The Police Killing of Alton Sterling was justified.

As far as I am concerned, based on the video, the police were totally justified.  Despite having two police officers on top of him trying to subdue him, Sterling was still clearly struggling with the officers.  Even if he had not been armed (which he was), continuing to resist at that point would have been monumentally inadvisable.  Alton Sterling had a gun in his pocket.  You can shoot through a pocket.  Police were responding to a report that he had already threatened someone with his gun.  From the perspective of the police they were going to arrest an armed suspect who had already demonstrated dangerous actions.  When you are armed and police are arresting you because of reports of you threatening people with a deadly weapon, continuing to resist goes from being monumentally inadvisable to being suicidally inadvisable.

But notice that I am not saying it was suicidally stupid.  Because the question of why anyone in that situation would be so unwise as to foolishly invite lethal actions by the police leads us to try to understand the ways that racism affects people and groups.

Why was Alton Sterling struggling with the police?

He knew he was armed.  He knew there were two cops on him.  He sure as heck should have been smart enough to realize that you don't get to shoot your way out of a situation with police (at least almost never).  And yet he continued to fight.  Overwhelmed, outnumbered, and out-gunned, he fought vainly and suicidally.  There was no possibility of anything positive coming from his actions, but he struggled.

There is a comparison that I can offer that I draw from my Alaskan upbringing.  He fought like he was being mauled by a grizzly.  When you are attacked by a grizzly there is really nothing you can realistically do to make yourself survive once they have their jaws and claws on you.  You might survive, but not because you really have a chance of killing the bear.  And yet people fight back.  If you are going to die anyway then maybe you can take the bear with you.  And sometimes miracles happen.

I cannot say what was going through Alton Sterling's mind, or whether or not he thought things through when he decided to struggle, but my suspicion is that he wasn't thinking.  My suspicion was that he went into fight or flight mode, and flight wasn't an option.

But why?

For myself, if police came to arrest me, I would not resist.  Even if they abused me and violated my rights I would not resist, because I know that no good can come from it.  Plus, not resisting increases my chances of surviving, and if they do violate my rights it puts me in a better position to turn around an sue them.  My underlying assumption is that the police are not going to kill me, and if they break the law I have the resources to sue the police.  (This underlying assumption is pretty much the dictionary definition of privilege)

For many white people it is easy to point to the ways that so many people killed by police seem to bring it on themselves.  When you assume that the police aren't going to kill you it changes the way you perceive them.  To go back to the bear analogy, many white people's relationship with police is a lot like urban environmentalists relationship with bears.  People who don't have to worry about getting eaten by bears think of them as beautiful creatures that serve a vital ecological role.  People who do have to worry about getting eaten might still appreciate and respect the ecological role that bears serve, but they don't want the bears near them.  Because they don't want to get eaten.

If you are one of vast majority of Americans who are statistically less likely to be killed by police it is easy to assume that #blacklivesmatter people are being overly dramatic, and that all people need to do to not get shot by cops is to not do stupid things.  But if you are part of the 14.5% (13% African American, 1.2% Native American, source) of Americans that are more than 3 times as likely to be killed by police as white Americans (source) then it doesn't seem so dramatic (Native Americans are actually more likely to be killed by police than blacks).

It just so happens that this week provided an object lesson in why so many dark skinned people fear the police.  Philando Castile was a law abiding, licensed firearm carrier, and school employee.  He was killed despite following the rules, and complying with the police.  Killed by a police officer while in a car with his girlfriend and her four year old daughter.  After he was shot he was allowed to bleed out rather than receiving prompt medical attention.  The killing of Philando Castile serves to effectively confirm the idea for many people that even if black people follow the rules the police will still kill them.

(Just to be clear here, I do think that the killing of Alton Sterling was justified, and I would be shocked if any inquest found otherwise, but the killing of Philando Castile is unquestionably minimally Manslaughter in my opinion.  If the officer who shot Philado Castile is not minimally charged with manslaughter I feel it would be a gross injustice.  But I want to avoid going too far down the rabbit hole of parsing what killings are acceptable, and which are not.)

This is where we start to see the real impact of systemic racism.  I am NOT talking about the killing of Mr. Castile in this case.  While I have no doubt that race played a key part in Mr. Castile's death, that was an individual act.  The real impact is in the reinforcement of the narrative that police kill black people for no reason.  When black men get killed for no good reason by police officers, and then the police officers get away with it scott free, it supports the idea that cops are looking to kill black people.  And that idea of persecution is self-fulfilling.  If the cops are just going to kill you anyway then why not go down fighting?


We can talk all we want about the importance of respecting police, but you can't really respect someone you think is trying to kill you.  It's like respecting bears.  You can respect bears, even if you are scared of them, but when they get their claws on you then you can't really respect them in that moment.  This is the effect that a racist system has between those who are discriminated against and those who are charged with enforcing the system.  In the aggregate it doesn't matter if most cops are good and non-racist, the environment of fear and distrust that has been created is self-reinforcing.  If a population responds to police doing their jobs like they are rabid bears then those police are going to get used to having to fight, and that population is going to have ever more reasons to believe the cops are after them.


And let us be clear, the police do a dangerous job.  There are roughly 800,000 sworn officers in the US which comes out to roughly %0.26 of the population.  In 2015 130 of those officers died in the line of duty (source).  That means that the death rate, in the line of duty, for police officers in 2015 was 16.25 per 100,000.  For comparison, the homicide rate for the US in 2013 was 5.1 (source, figure 18).  So police are 3 times as likely to die in the line of duty as the average American is to be killed by anyone.  So far this year there have been 59 officers who have died in the line of duty, which puts us on pace for a similar death toll.  One of those officers killed this year was Steven Smith, the brother of a friend, a 27 year veteran, and very much the epitome of a heroic police officer.

Law Enforcement Officers are very much aware that they have dangerous jobs, and they would have to be mentally incompetent not to realize that certain groups are far more likely to get violent with them than others.  On individual levels this is frequently wrong, but people would have to be inhuman not to make associations.  Danger can feel color coded for cops and blacks alike.  African Americans learn to fear blue, and Police learn to fear brown.  That is simply human nature, and there is only so much that can be done to address this problem without addressing the underlying racist structures that create this situation.


Because we really can't understand how this situation where blacks view police as predators arises without understanding the ways that police are saddled with fundamentally racist law enforcement priorities.  Those are not priorities set by the police, those are priorities set by elected leaders.  There is plenty of attention paid to the ways that black people are scapegoated and vilified in the media, but the police are also convenient scapegoats for elected leaders who don't want to admit that their own actions force police to behave in racist ways.

In particular here I am referring to the doctrine of "broken windows policing."  Broken window policing is based around the idea that if you aggressively go after visible symptoms of crime that you will drive down crime rates.  This approach does work, and it is great for property values.  If you eliminate the visible signs of crime then people are more willing to invest in properties, thus property values increase.  New York city is the prime example of this.

But visible signs of crime is effectively code for visible signs of poverty.  What broken windows policing does is it criminalizes poverty.  When property values are driven up it drives out lower income people.  In the case of New York, since the 1990's values have gone so high that families with six-figure incomes can barely afford (and often can't afford) apartments.  This is great news if you are a property owner, and as a politician you get to brag about how you've gotten rid of crime.  Rudy Giuliani became a nationally prominent politician based on how effectively he criminalized poverty in New York.

But it is a problem if your skin color is a visible sign of poverty.  Not all brown people are poor, but a higher percentage are than pink skinned people.  And that is really how racism works.  You don't pass a law that makes it illegal for black people to be in a neighborhood, you pass a law that makes it illegal to do things that poor people do.

Here is a scenario:
More lower income people smoke than higher income people, so if you jack up cigarette prices you will disproportionately cause hardship for poor people.
If you make it harder for poor people to buy packs of cigarettes you will increase the purchasing of single cigarettes, so if you pass a law banning single cigarette sales then you will create a demand for an addictive good that can now only be satisfied illegally for many poor people.
Now that you have created a demand for illegally sold single cigarettes the local government can set an enforcement priority on cracking down on people selling single cigarettes.

That is how Eric Garner died.  He was selling cigarettes, because the New York City government created a demand that disproportionately affected poor people, who are disproportionately minorities.  And those laws specifically included criminalizing a behavior common to lower income people (resale of loose cigarettes), and then policing of that crime was prioritized as a part of a broken windows strategy.  Then Eric Garner died while resisting arrest for selling cigarettes.

The national dialog focused on the apparent racism of the police choking a black man to death for selling cigarettes, but it didn't address the fact that those police were doing their jobs enforcing a racist system of policies designed to persecute the poor.  Those laws aren't limited to cigarettes, that is just a specific high profile example.  I have also written in this blog about New York's knife laws that have clearly racist application and orientation.  Police in New York have the power to make almost any modern folding knife illegal if they want (even though those same knives an be purchased legally), and people who work physical jobs (lower income) need knives at work more, and so those laws once again disproportionately affect black people.


This is what people mean when they talk about racist systems.  Saying that there is racial inequality in the US is not about saying that individual cops are all racists, it is about saying that there are structural elements in our society that systematically negatively impact minorities.  Too often people on the side of law and order take cries of racism as personal accusations of bigotry.

Explicit racism is certainly dangerous when we are talking about lynchings and cross burning type stuff, but it is the systemic issues that perpetuate racial inequality.

An individual racist is fairly easy to avoid.  In today's world there aren't many people who tolerate open racism.  But it isn't bigots who are the real power behind racial inequities, it is all the people who fail to acknowledge the racist structures that perpetuate violence and discrimination.  It is politicians who pass laws and policies that discriminate against poor people and then don't acknowledge their complicity in killing poor people.


Eric Garner died because police were enforcing racist laws.  Those kinds of racist laws lead to feelings of persecution among minorities.  Those feelings of persecution lead to inappropriately violent confrontations on both sides.  That is why people like Alton Sterling continue to struggle when they should just lay down with their hands on their head.  And then killings like Alton Sterling's lead to increased agitation, and scapegoating of police leads to retaliatory shootings like in Dallas.  Then the police feel targeted.  That leads to cops getting extra jumpy when doing routine things like pulling someone over for a broken taillight.  That leads to innocent people like Philando Castile getting shot for no good reason.  And those types of unjust killings, precipitated by fear, perpetuate the fear and violence that drives this endless cycle.

That is how racism works.

Thursday, July 7, 2016

Correlations Between Crime, Employment, and Government (Or the lack thereof)

Hello readers, it has been far too long since I last wrote an entry for this blog.  I have been writing a lot, but all of my work has been focused toward school and projects that I hope to have published later.  Since I do not want to scoop myself on my research that means that most of what I am spending my mental energy on these days is stuff that I can't put in the blog.  However, this last term I took a statistics class, and my final project for that was actually stuff I was looking at so that I could use it in the blog.  So for your reading pleasure, here is my unedited paper:
Correlations Between Crime, Employment, and Government (Or the lack thereof)
The objective of this project was exploratory analysis.  In my non-academic writings I frequently write about crime statistics, with a special focus on homicide and gun violence statistics.  There is a wealth of well-done analyses that look various aspects of homicide rates, but there is less attention paid to the relationship that homicide has with other types of crime, or to broader societal trends.  The FBI publishes numerous in depth statistical analyses of crime rates and trends (FBI, Numerous Dates), including ones that look at rates of different types of crime compared to each other, so my goal in this project was not to duplicate those analyses, but rather to look at the relation of crime rates to employment statistics and political leadership to see if there were hidden trends I could uncover and analyze.  
In order to look at the relationships between crime, employment, and government I first had to decide on a scope.  I chose to run the analyses on the national level in order to look at broad trends.  I then needed to source my data and produce a usable dataset for analysis.  I conducted a number of descriptive time-series plots, simple Spearman’s correlations (in recognition of the non-parametric nature of much of the data), Pearson’s pairwise correlations for comparison, and ultimately decided that a Principle Component Analysis (PCA) would be the most useful tool to indicate follow up analysis directions.  
Unfortunately, the data I was using was only appropriate for PCA is I removed all but the crime rate data.  I conducted the analysis, but the results are not anything ground breaking.  While I have not seen a PCA of violent and property crime rates through time in official governmental analyses, the information I was able to glean is ultimately duplicated in a number of easily accessible sources.  My analysis failed to identify hidden trends, and my exploratory analysis failed to explore new areas of inquiry, but I was able to at least conduct sufficient correlational operations to suggest that further analysis was unlikely to result in valuable insight.  This was not the outcome I hoped to achieve, but I had been aware of the high likelihood of negative results heading into the project.


In order to find a dataset that looked at the things that I wanted to look at, on the scale I wanted to look, and at the time resolution that I wanted to use (annual) I needed to create my own dataset.  
For the national crime data, I used the Uniform Crime Reporting (UCR) statistics data tool (UCR, 2010).  This is a tool that allows individuals to access the datasets available to the FBI in order to make customized datasets.  
For the employment data I used Quandl to extract data from 1960 to 2012 (the years that the FBI had national data for the statistics I was interested in).  The Quandl data was published by the Bureau of Labor Statistics (BLS).  Quandl allowed me to take the data and extract is by the dates I needed, and to transform the data from monthly entries to annual entries (thereby obviating the need to seasonally adjust the data in R).  As will be discussed in the Variables section, I unfortunately extracted the job change data as percentage values rather than numerical values, which complicated some analyses.  
For the political data I simply used widely available information on the political affiliation of Presidents and the parties in power in Congress.   The presidential affiliation was straightforward.  I counted years in which a new president comes into office as entirely the year of the incoming president.  For Congress, due to the bicameral nature of the legislature, I used three values, Democratic control, Republican control, or split (for years that the House and Senate are controlled by different parties).  I used text data for these values.  In retrospect, I should have given these variables numerical values to facilitate analysis.  A better approach would also be to incorporate the actual numerical breakdown of party affiliation in Congress to create a finer grained analytical tool.


  • Population – Total US population (all values are by year from 1960 to 2012 unless noted otherwise)
  • Violent crime total – Total for US
  • Murder and nonnegligent Manslaughter – Total for US
  • Forcible rape – Total for US (see forcible rape rate note)
  • Robbery – Total for US
  • Aggravated assault – Total for US
  • Property crime total – Total for US
  • Burglary – Total for US
  • Larceny-theft – Total for US
  • Motor vehicle theft – Total for US
  • Violent Crime rate – Rate per 100K
  • Murder and nonnegligent manslaughter rate – Rate per 100K
  • Forcible rape rate – Rate per 100K. Note: Due to changes in the culture and legal system regarding the willingness to report rape, as well as broadening the definition of rape (for example, during much of the time being looked at in this project a husband could not be considered to have raped his wife, regardless of whether or not she consented) treating this variable as apples to apples through time is very problematic.
  • Robbery rate – Rate per 100K
  • Aggravated assault rate – Rate per 100K
  • Property crime rate – Rate per 100K
  • Burglary rate – Rate per 100K
  • Larceny-theft rate – Rate per 100K
  • Motor vehicle theft rate – Rate per 100K
  • Employment Private Sector – Total, in thousands, of people employed by private businesses
  • Employment Change Private Sector – Total change, in percentage of total, of people employed by private businesses
  • Employment Government Sector – Total, in thousands, of people employed by government
  • Employment Change Government Sector – Total change, in percentage of total, of people employed by government
  • Presidential Party – Which political party the President belonged to by year
  • Legislative Party Majority – Which party was in power in the legislature.  This will be a three value variable: Republican, Democratic, or split.  The variable will only be assigned to a party if that party controlled both houses.

One major problem with the variables that I created, which I didn’t discover until the end of analysis, is that I incorporated employment change as a percentage of total employment change from year to year.  This meant that the numbers generated had so many decimal places that they did not function for PCA analysis.  In most years, the total employment change is such a small percentage of the total employment numbers that the percentage does not reach a rounding threshold sufficient for PCA.  Also I extracted the employment change data from 1960 to 2012, instead of 1959-2012 which meant that I had “na” values for 1960.  This meant that I lost a year’s data for some analyses.


While the original order of some of the functions that I scripted during this project were different, as I went on I tried to reorganize my script into a more rational order of analysis types.  This reordering became increasingly important as the script started to grow over 200 lines.  By the time that the script was finished, with over 400 lines, having a logical order of functions became vital to ensuring that everything worked, especially when replacing opaque labels with easier to understand labels.
As described in the introduction, I conducted a number of descriptive time-series plots, simple Spearman’s correlations (in recognition of the non-parametric nature of much of the data), a Pearson’s pairwise correlation for comparison, and ultimately decided that a PCA would be the most useful tool to indicate follow up analysis directions.   
The first step was doing simple plots of some of the variables.  Next was running through some histograms, putting in normal lines, and running Q Plots to check for normalcy.  Unsurprisingly most of the variables were not normal.  After the Q Plots I ran through a series of simple time-series plots, which created visually useful graphs.  I followed this with Spearman’s correlations, since the text suggested Spearman’s for continuous variables that do not meet parametric assumptions.  Just for reference I ran a Pearson’s Pairwise Correlation.  I then ran a series of time-series plots that include a two-sided moving average and a LOWESS.  At first I was thinking of this as simply a prettier way to make the graphs, but it ended up being quite informative.  And the final step was running a PCA.  After switching computers this was able to produce a skree plot, which made the PCA worthwhile after all.


Figure 1: Population Plot
I start with the population plot, simply as an illustration of what turned out to be a thoroughly confounding variable.  The population growth in the US over the time frame of the study was so great that total to total comparisons ended up being essentially useless.  For example, here is the Spearman’s correlation between population and government employment:
Figure 2: Spearmans Population Government Employment
The two variables have a 99 percent correlation.  As it turned out essentially every total measure had extremely high levels of correlation, primarily due to population growth.  The US went from a population of 180 Million people to 320 million over the course of the study.  This near doubling of the population made any non-rate measures effectively simply proxies for population growth.  This does not mean that all of the totals pegged to population growth, murder and crime totals actually diverged pretty sharply from the straight line:
Figure 3:  All Violent Crimes Added Together
This graph is actually all violent crimes, with all of them added again.  Effectively a double picture of violent crime, I use it simply as an illustration of overall trends.  Despite constant population growth, the total numbers of crimes (not just rates) have actually dropped significantly over the past few decades.  This change is actually more pronounced for murder, where even the total numbers are currently around the numbers last seen in the early 1970’s, when the population was 120 million people less.  That said, the murder rate is actually lower now than it was at the beginning of the period being looked at, so it is clear that even in the case of murder, population is a confounding factor.
Simple Plots, Histograms, Q-Plots:
Figure 4: simple Plot of Murder and Violent crime
I include this simple plot of the violent crime rate and murder rate simply to highlight the ways that the two do not match up.  The Murder rate is extremely low both at times when the violent crime rate is low, and when it is at a medium value.
Figure 5: Violent Crime Histogram           Figure 6: Population Histogram

Generally speaking, my variables did not appear to be parametric.  As an example I include the violent crime histogram, which as you can see is pretty platykurtic.  The Population variable strangely actually seemed to be pretty normally distributed.
Some of the variables were more aggressively non-parametric than others.  Obviously, the categorical variables, like Congress below, were non-parametric when Q-plotted.  Surprisingly to me, the population variable still seemed fairly normal, albeit somewhat S-curved.
Figure 7: Congress Q-plot Figure 8: Population Q-plot
Correlations, Spearman’s and Pearson’s:
The following section is, sadly, a rather boring list of Spearman’s correlations.  So to make it a little more bearable, I will simply start with the correlation that I found most interesting, and that ultimately became the correlation I investigated most.
For some reason, there is a significant correlation between change in government employment numbers and the violent crime rate.  In this case, the negative correlation means that as government employment increases violent crime rates drop.  Is it possible that increased government services leads to lower crime rates? Very interesting, so I looked further.
Hmm, government employment changes negatively when Republicans control congress.  This is not surprising, since it is a part of the party plank.
But uh oh, there is no correlation between government (President or Congress) and violent crime rates.  So even though there is a correlation between Republicans and negative government job growth, and there is also a correlation between negative government job growth and violent crime, there is not a correlation between governing party and violent crime rates.  I decided to look closer.
Well clearly there is no significant correlation between negative government job growth and murder rates.  Is there a correlation between murder rates and violent crime?
Yes, a pretty strong one, so how can murder rates not be correlated to negative government job growth?  Is there a correlation between murder rates and congress?
That would be a no.
At this point I decided to let the mystery of the correlations between violent crime and government employment rest for a bit.  I ran a lot of Spearman’s correlations, but I will spare you all of them except for one:

It looks like, despite the Republican reputation as the party of business there is no correlation between republican control of congress and job growth.  Based on the correlations I ran it would
seem that neither party really has any idea how to make an economy work, it appears to be mostly random.

Also, just to show that I ran it, here is the Pearson’s correlation.  It took a long time, but I did not do much with it.

Time Series Plots with Two-Sided Moving Averages and LOWESS:
I spent a lot of time working on the time-series analysis, but it is a very complicated topic.  I satisfied myself with producing lots with a moving average and LOWESS.  At first I
simply assumed that this was a prettier way of plotting, but we will conclude with the graph I found most interesting and illuminating.
Figure 9: Murder Rate Time Series
Figure 10: Murder rate with observations indicated
Figure 11:  private employment change
Figure 12: Property crime rate
The property crime rate graph is worth seeing, just to see how closely the following violent crime rate graph correlates to it.  This correlation is very strong.
Figure 13: Violent Crime Rates
Figure 14: Government employment Change
And here finally, thanks to the LOWESS, we see the why for the negative correlation between violent crime and government change.  The change is due to the way that I extracted the Government employment change data.  Instead of extracting the numbers I extracted the percentage of change, and that means that as employment becomes more stable (percentagewise, partially due to the confounding variable of population growth and the effect on government size) the values drop.  Since violent crime rates overall went up over the course of the study, and the level of government employment instability went down, there is a negative correlation.  But I made the negative correlation.  It was my bad statistics.

Primary Component Analysis:

As a final piece of analysis I ran a PCA.  I wanted to see if there was something I had missed:
Unfortunately, thanks to the way that I extracted the change data, I had to toss it out.  Which meant that my PCA essentially just covered well covered ground analytically.  I was able to run the PCA.  Effective as a proof of concept, but a failure at trying to bring something new to the discussion.  The PCA does seem to indicate that murder and burglary are less responsible for the overall variance in the model, but the eigenvalues are still 0.83 and 0.78 respectively.  Every rate is significant.
Interestingly, looking at the inflection point, the Murder Rate is the second most important Primary Component, and looking at the eigenvalues, Murder rate and Burglary Rate are the strongest positive correlations, and bizarrely Assault Rate is negatively correlated with Murder Rate.  Assault is also the eigenvalue next least associated with violent crime rates after Murder and Burglary.  So, strangely, it looks like Murders, Assaults, and Burglary (while certainly associated with other crime rates) are the least pinned to other general rates.


Ultimately, overall crime rates are strongly correlated.  This is not a very useful or insightful observation.  Also well discussed in the literature is the way that murder does not track perfectly to other rates of crime.  The most important result for me was realizing how easy it is to mess up your own data.  That at least is a useful lesson.


Coghlan, A. (2010). Using R for Time Series Analysis.
Zucchini, W. & Nenadi´c, O. (No Date). Time Series Analysis with R - Part I.
Shumway, R.H. & Stoffer, D.S. (2010). An R time series quick fix. Time Series Analysis and Its Applications: With R Examples - Third Edition.
FBI. (2016, Numerous Dates).  FBI Crime Statistics Home Page. US Department of Justice.  Federal Bureau of Investigation.
Bureau of Labor Statistics. (2016), (Dates extracted: 1960-2012).  BLS Employment statistics.  Quandl.
UCR. (2010). Datatool. US Department of Justice.  Federal Bureau of Investigation.  Uniform Crime Reporting Statistics.