Friday, July 29, 2016

Chart(s) of the week: Crime -- Just How Bad Is It?

I’m still reeling from Donald Trump’s dystopian vision of America and the world we live in, as laid out in his GOP convention acceptance speech.  The theme of that speech, in a nutshell: America is going to hell, and only Donald Trump can save us.

Now, in case you think I'm concerned about the bleak nature of the speech because I'm a leftie: not so.  For the moment please provisionally entertain my assertion; I'll leave discussion of my own political leanings to another post, where you can learn more about my biases.

One theme of this post is cribbed from Daniel Patrick Moynihan:  "Everybody is entitled to his own opinion, but not his own facts."  In this post, I am inspired partly by The Donald's own call to "present the facts plainly and honestly."  We will see that, in fact, some of Trump's facts are correct, and some of his concerns have some foundation.  We will also see that even when he cites some correct statistics, his discussion is seriously incomplete, and often misleading.

I'm an Economist, not a Political Pundit; so Why This Post?


Why would an economist comment on politics?  I’ve long been interested in politics, including but not limited to their interaction with economics.  My first degree was in Political Science from La Salle University, where I studied U.S. government, foreign policy, international politics and history with Bob Courtney, Fred Foley, Michael Dillon, Ken Hill, Minna Weinstein, and my senior thesis advisor and mentor C. Richard Cleary, among others.  Graduate study in international affairs at GWU followed, with political scientists Burt Sapin and Stephen Shaffer among my teachers.  At Wisconsin I had too many outstanding friends and colleagues in political science and related areas to list here, though I’d be remiss if I didn’t mention Karen Bogenschneder, who involved me in her Family Impact Seminars for the Wisconsin State Legislator; Don Kettl, who recruited me for Governor Tommy Thompson's Blue Ribbon Commission on State and Local Partnerships.

In this post I just want to lay out a few charts that have a bearing on some of the assertions in Trump’s speech.  Specifically, how bad are things?  In future posts we’ll look briefly at incomes, poverty, the state of U.S. manufacturing, trade, and other kinds of violence, including terrorism; all things Trump brought into his speech; these are things that I’ve been concerned about, too.  In those future posts, I’ll also add a few items that Trump did not mention, e.g. life expectancy.

Each chart we present is one simple look at an important and complex topic, certainly worthy of a separate, more detailed blog post in the future.  But for now, let’s just get a few facts about crime, specifically homicides, on the table.

(As discussed briefly at the end of this post, we're going to follow common usage and use "homicide" and "murder" as synonyms, although the homicide data we use includes non-negligent manslaughter, which is technically not a murder.)

Why focus on homicides?  One reason is that depriving someone of their life is such a heinous crime. Another is that they have such large externalities, especially some kinds of murders, like those of children and of police officers.  "Externality" is economist-speak for spillovers, good or bad.  That is, externalities are things that affect not only those directly involved, but also families and friends, neighbors, and even society at large.  Any significant crime is likely to have spillover costs, but murders surely have especially larger spillovers.

Another aspect of homicides is that they are, and have been throughout our history, tied up in our discussion of many other social issues, including those of race and poverty.  Some unlawful deaths are classified as "terrorism," and these have special and very large negative externalities.

Examining statistics on homicide or any other serious crime can seem bloodless and unfeeling.  Reducing the murder rate is a good thing, for sure, but a low murder rate may be cold comfort to remaining individual victims and their families in a "low rate" environment.

One reason to study homicides is that, because they are so heinous, they are measured better, and more consistently over time and across jurisdictions, than other kinds of crimes.  Of course, "measured better and more consistently" doesn't mean measured perfectly.  Murders are also usually found to be correlated with other kinds of violent crime, including those that are less well measured.  So murders are of interest for their own sake, but also give us a start at thinking about other kinds of crimes.

In this post, having noted these shortcomings, we'll focus mostly on the numbers.  Discussion of externalities, of terrorism, of other kinds of crime will wait for another day.

I’m motivated by Donald Trump's comments in this post, but these are important issues whatever your political leanings, in an election year, or not.  In future posts, I certainly won’t let his opponent Hillary Clinton off the hook.  And I’m sure to throw in some charts relevant to pronouncements by Bernie Sanders, Paul Ryan, and others on the way.

Let’s get started!

Early in Trump's speech, he stated:

"Homicides last year increased by 17 percent in America's fifty largest cities. That's the largest increase in 25 years.In our nation's capital, killings have risen by 50 percent. They are up nearly 60% in nearby Baltimore.In the President's hometown of Chicago, more than 2,000 people have been the victims of shootings this year alone.And more than 4,000 have been killed in the Chicago area since he took office."

His grade, if Trump had turned in such a report in my urban economics course?  Incomplete, and misleadingly so.  Revise and resubmit, so I can grade a proper answer.

Recent Trends in Homicides


I'm giving Trump a big fat "I" even though the data Trump cites are consistent with preliminary FBI crime reports for 2014 and 2015.  As has been widely reported, Trump's speech appears to draw on a Washington Post Wonkblog story by Max Ehrenfreund and Denise Lu that reports preliminary homicide data for 2015 for 50 large cities. The data do, indeed show a disturbingly large increase in murders across these cities.

The basic data we use to analyze murder rates comes from the FBI's Uniform Crime Reporting data. The majority of homicides are reported to state and local law enforcement, not to federal authorities directly, so the more complete data come out with a lag.  As of this writing, 2014 data are the most recent complete data, which is why the preliminary 50 cities data for 2015 have been widely reported.  All in, these 50 cities (n.b. cities, not metropolitan areas!) contain about 50 million people, or about 15 percent of the U.S. population.  The data are not nationally representative; among other issues, murder rates tend to be higher in cities than in suburbs; rural murder rates are usually somewhere in between.  Nevertheless, it's reasonable to start with this preliminary data; certainly they are of intense interest to the people that live there, and they may presage qualitative results that we'll see in the later, more complete data.

In the event, Ehrenfreund and Lu's report is well worth careful reading, including their interactive graphics.  Pull up their article here, in a separate browser window, and look again at their charts along with this commentary.  I'll insert some screenshots of the charts here, but be sure to go to their site; you lose the interactive feature with these screenshots!



Their first chart (screenshot above) is a simple bar chart of changes in homicide rates.  Of their 50 cities, 36 show an increase in homicides between 2014 and 2015; and 13 have increases over 40 percent.  The top five percentage increases were in Cleveland, Nashville, Milwaukee, Denver, and Oklahoma City. Thirteen cities had fewer homicides in 2015 than in 2014; the largest declines were in Boston, Austin, El Paso, Fresno, and Tucson. One city, Mesa, Arizona, had roughly no change.



Rates of change are important, but they only tell part of the story.  Ehrenfreund and Lu's second chart is complex but extremely informative and interesting one.  That chart, screenshot with link above, plots the murder rate (murders per 100,000 people) annually for each of the 50 cities, from 1985 to 2015.  Thus, while Cleveland has the increase in murders from 2014 to 2015 (90%!), and Baltimore ranks "only" seventh, albeit with a still distressing increase of 59%.  Both cities have above average murder rates as well; but here the relative ranking is reversed.  Cleveland's average murder rate came in at 16 per 100,000 in 2014, and 31 per 100,000 in 2015.  That's bad, but Baltimore is even worse; Charm City's murder rate rose from 35 per 100,000 to 53.  For comparison, Ehrenfreund and Lu calculate the average murder rate over the 50 cities 9.3 to 10.8; an increase of about 16 percent.


Their third chart, above, looks at the average yearly percentage change in the murder rate in the 50 large cities.  It puts the large 9.3% increase from 2014 to 2015 in perspective, i.e. it comes after years of mainly declines; and those declines, in turn, came after some large increases in the late 1980s.  We'll have more to say about these trends, too, below.

Ehrenfreund and Lu's UCR Data Transformed


A recurring theme in my teaching is that, when looking at consequential data, it's important to chart it and otherwise analyze it correctly; and in particular to remember that there are almost always several alternative ways to present even basic charts.  Ehrenfreund and Lu present three different looks at the data, and I congratulate them for their data work.  I'm particularly impressed at their second chart, which you should examine carefully, in its full interactive form at their site.

After you examine their charts, come back to this blog and we'll extend their work a little further, starting here.

Our first chart is a simple one, that plots the 2015 homicide rates for the 50 cities against the 2014 rates.  The area of each circle is proportional to each city's population; the circles, and the (admittedly hard to read in some cases) city names are centered on the data points.  The diagonal line shows where 2015 rates equal 2014 rates, so crime rates are getting worse in cities that lie above the line.

Here again Cleveland and Baltimore stand out.  But we also note that the two deadliest cities, New Orleans and Detroit, haven't changed much, in percentage terms. (Not that zero change is good -- we want to see some declines!)

Taken as a whole, this chart shows that there were big changes in a few deadly cities (cities with murder rates over 10 per 100,000), especially Baltimore and Cleveland, but also Milwaukee and Washington DC and Kansas City.  Among the safer cities, Minneapolis, Nashville, Oklahoma City, Omaha saw significant increases.  Boston stands out among the safer cities that became safer still.

Let's look at New Orleans again.  It's growth in murder rates is about 9.3 percent, which means it's average for the 50 cities.  But that growth rate, applied to a higher level of murders, means its murder rate went up by about 3.7 per 100,000.  A similar increase in murder rates in safer cities like Austin or Raleigh would have meant the murder rate roughly doubled!  Put another way, a modest (though still undesirable) increase in murders in safer cities translates into a larger percentage increase than we'd calculate for a more deadly city.  The point is not that growth rates are irrelevant; rather, it's that we should carefully examine both levels and rates of change.  Let's do that next!

Our second chart uses the same basic data, but another transformation.  This time, we plot the 2014-2015 percentage change against the starting point.  Here we get a clearer look at the crime growth rates of the safer cities; Austin and El Paso and Fresno have noticeable percentage declines as well as the aforementioned Boston.  Some of the safer cities, Nashville, Denver, Oklahoma City, Minneapolis and Long Beach, have some large percentage increases, albeit over a small base.  Again, multiple looks at the data with well-chosen transformations tells us more about what's going on.


Decades of Data Tells a Different Story than Two Years of Data


Preliminary data for 50 cities may be warning us of crime increases in other places, and in future years; or it may not.  Let's look at longer patterns, including but not limited to the 29 years of data presented in Ehrenfreund and Lu that Trump did NOT mention in his address, that we can glean from their second and third figures.  Their second figure, in particular, shows that the recent 50 city increase in murder rates, from a 2014 average of 9.3, to a 2015 average of 10.8, comes after a peak of 27 in 1991; in 1985, their starting point, the 50 city average murder rate was 19.4 per 100,000.  I'm concerned about the recent increase in preliminary data, but we're not back to the high crime 1980s and 90s, at least not yet!


Let's look at this city by city.  I pulled the city murder rates for a few years later, 1993 (average of 25.6), and repeated my first plot from above, but this time used the 1993 homicide rates.  A few cities -- Las Vegas, Milwaukee, Baltimore -- have current murder rates a bit above their 1993 rates; most have fallen, even after the last year's increase. As bad as New Orleans' current murder rate is, it's half the rate they experienced in 1993; DC's improvement is even more striking, though much remains to be done.  Moving away from the extremes of the data, cities including New York, Los Angeles, in fact the great majority of cities are far safer now than they were two decades or so ago.


A Century of Homicide Data -- and Some Related Policy Debates


As noted above, the latest comprehensive murder data we have as of this writing (and the conventions) stops at 2014.  Statistics on murders in the United States are available back to 1900.  Of course, older data, before the modern UCR system was put in place in the 1930s, should be treated with some caution.

Our chart shows some fascinating patterns. I'm not sure what to make of the low murder rates circa 1900; but there is a lot of anecdotal evidence as well as scholarship that suggests murders (and other crimes) increased substantially during Prohibition.  Was the end of Prohibition the sole reason for the post 1933 decline?  Hmmm.

Maybe the end of Prohibition had a lot to do with the 1933 to circa 1960 decline.  But what caused the run back up in the 1960s to a new peak in 1974?  What caused the volatility from that era until the early 1990s?  And, if we can unpack those changes, what might we expect for the future?

Geek Alert: Two Paragraphs on Properties of Time Series!


First, let me state the obvious.  No simple "trend forecasting" is going to tell us much about murder rates.  A time series of data that has a constant mean and variance (among a few other properties) is called a "stationary" series. A stationary series that's "mean reverting" can be forecast by studying how fast we get back to that constant mean if something pushes us away temporarily.  Recognizing that a "trend" is just a mean that is shifting in a very predictable manner, we can also forecast a series that's "trend stationary."

I haven't run any formal tests, but my eyeballs tell me that murder rates are not likely mean reverting, or trend reverting; that the variances aren't constant over time; and that it's not going to be easy to foreast murder rates with some simple trend analysis.  If I had to bet, I'd bet murder rates turn out to be a "random walk," a series with no memory.  Like a drunkard's walk, it's going to be hard to forecast.

End of Geek Alert; Now a Wonk Alert!


How can we explain the variation in murder rates (which, by the way, are correlated with other violent crime rates)?  In particular, why have homicides fallen substantially over the past two decades?  If we can understand these patterns better, can we gain some insight into whether the apparent increase in crime in a number of large cities over the past year is a blip in the data, or a harbinger of other increases?  Even better, can understanding determinants of homicide rates help us reduce it?

The short answer is, much rigorous research on homicide rates, and other crime, remains to be done. My reading of the literature, still in process as of this writing, can be briefly summarized as follows.  Overall, there are no obvious silver bullets.  And a number of contentious points are hotly debated.  Candidate explanations for the post-1993 decline include, but are not limited to:

  • Demographic shifts;
  • Changes in economic opportunity, unemployment rates;
  • Higher rates of incarceration;
  • Improved policing;
  • Changes in the size and organization of drug markets;
  • The availability of firearms;
  • Increased immigration;
  • Declining rates of births of unwanted children;
  • Reduction in childhood exposure to lead;
  • Cultural and institutional changes;
  • Housing conditions and affordability.

Selected surveys of the relevant literature can be found below.  A future post will discuss some of the details of research on these issues.


So: How Bad Is Crime? And What Can We Do About It?


Back to the question we started with.  While there is some data behind Trump's assertion of rising crime rates, the data are highly selective and preliminary.  The recent increase in murder rates for a single year in a number of large cities is troubling.  But our look at a fuller dataset, including more places and longer time periods, tells us the increase comes in the context of twenty five years of surprisingly steady declines in murder and other crime rates.  Overall, based on these data, we're a lot safer than we've been for much of my lifetime.

My students are familiar with "Occam's Razor," the principle by which a simple theory that fits the facts is usually preferred to a complicated theory that also fits.  But what about the facts themselves?  Nobel-prize winning biologist Sydney Brenner is credited with coining the phrase "Occam's Broom," a process by which inconvenient facts are swept under the carpet.  Another term in common usage is "cherry picking the data."  We'll run into this practice a lot, often among pundits and politicians, and (sadly) sometimes among shoddy academics who don't uphold normal research standards.

Whether or not the preliminary 2015 data are one-off, or harbinger of some change in trend, nobody is, or should be, satisfied with a national murder rate of 4 or 5 per 100,000, just because it used to be 10.  Nobody in Detroit or Baltimore or New Orleans or Cleveland or DC should be satisfied with murder rates of 30 to 50 per 100,000 just because we've had some cities hit 80 murders per 100,000 in the past.

So what do we do about it?  As conservative columnist Charlie Sykes ruminated after Trump's acceptance speech, "on day one, he ends crime and violence; you sort of wonder how you accomplish some of it."  That's not much of a plan.

There are ways forward.  For example, a number of serious policy proposals regarding improved policing and criminal justice reform are under discussion (although political progress will likely now be stalled until after the election).  But it's hard to find any mention of these, or other specific proposals that empirical evidence suggests would reduce crime, in Trump's vague pronouncements on the subject..

So, what should we do?  Despite the length of this post, there's a lot more to be said in future posts about what research by economists, criminologists and others have taught us about crime; and about other aspects of safety and security, including terrorism, and natural disasters.  More to come!


A Few Notes on Data and Definitions


In this blog I follow common practice and use "homicide" and "murder" as synonyms.  In fact, the FBI Uniform Crime Report homicide data includes both murders, and non-negligent manslaughter.

The FBI does not classify some unlawful deaths as murders or homicides. Notably, the 2,996 deaths on September 11, 2001 (New York City; the Pentagon; Shanksville, Pa.) were classified as deaths from terrorism, and not included in murder statistics.

On the other hand, the 168 deaths on April 19, 1995, in Oklahoma City from the bombing of the Alfred P. Murrah Federal Building were included in homicide statistics.  See FBI Terrorism Reports; we'll discuss these in a future post.

My charts make use of the data from Max Ehrenfreund and Denise Lu's Wonkblog post.  They discuss the data in some detail at the end of their blog post, I recommend that you read that section carefully.  I didn't find a table or spreadsheet at the site, so I copied down the 1985, 1993, 2014 and 2015 homicide rates from their second chart.  (It's a nicely interactive chart, as you drag the cursor over the picture it pulls up the city, year and value of the variable).  I also copied down the percentage change in murder rates 2014 to 2015 from their first chart.  I obtained Census data for 2014 population as well.  My spreadsheet containing this data, and the three charts posted above can be downloaded here.

For Excel freaks:  I'm using Excel 2010, which does not do true data labels; so I have a VBA macro that labels the points with the city name.  If you are unfamiliar with VBA, see my teaching materials on Excel and VBA here

Reading for Life


Barker, Vanessa. "Explaining the Great American Crime Decline: A Review of Blumstein and Wallman, Goldberger and Rosenfeld, and Zimring." Law & Social Inquiry 35, no. 2 (2010): 489-516.
Blumstein, Alfred, Frederick P Rivara, and Richard Rosenfeld. "The Rise and Decline of Homicide-and Why." Annual review of public health 21, no. 1 (2000): 505-41.
Blumstein, Alfred, and Joel Wallman. The Crime Drop in America. Cambridge University Press, 2006.
Goldberger, Arthur S, and Richard Rosenfeld, eds. Understanding Crime Trends: Workshop Report, Washington, Dc: The National Academies Press, 2008.
Roeder, Oliver K, Lauren-Brooke Eisen, and Julia Bowling. "What Caused the Crime Decline?": New York University, Brennan Center for Justice, 2015.
Yezer, Anthony M. Economics of Crime and Enforcement. ME Sharpe, 2013.
Zimring, Franklin E. The Great American Crime Decline. Oxford University Press, USA, 2007.





1 comment: