Sunday, January 3, 2016

Data contradicting beliefs

From In Denial About Crime by Heather Mac Donald. An instance where ideology occludes seeing what the data suggests. There is an argument going on as to whether there is indeed a Ferguson Effect. The proponents argue something to the effect that as a consequence of the anti-police protests organized during and after the Ferguson riots, and which spread to other big cities such as Baltimore, Milwaukee, etc., the police suspended vigorous community policing in order to reduce the number of potential encounters that might lead to violence. Those arguing this position further argue that reduced policing has resulted in a rise in violent crime in those cities.

This position is rejected by others who argue in effect, that there has been little change in policing, that those changes which have occurred have not affected violent crime and that violent crime is caused by other factors such as poverty.

Right now there is simple disagreement on the data which is slowly emerging and which indicates that indeed there is a link between reduced policing and increased violent crime. This is much like the argument whether there is systemic gender discrimination in terms of compensation. There isn't, as has been known for some 30-40 years. The data analysis gets increasingly sophisticated and the data sets get dramatically larger, consistently showing that differences in gender compensation arise from different choices in terms of degree, field of study, industry sector, years worked, hours worked, duration of employment, etc. No matter how clear it becomes that when you compare apples to apples there is no systemic compensation discrimination, there remains a firm and fixed belief on the part of some that systemic discrimination does indeed exist no matter that the evidence says otherwise.

With the Ferguson Effect, both sides are still a little sloppy in their argumentation, though Mac Donald is much better than most. For those arguing that there is a Ferguson Effect, what would be best is to be able to demonstrate that there is significant step change in violent crime before and after for those cities which have demonstrably made real reductions in policing as a consequence of Ferguson demonstrations.

This would involve identifying a list of those cities and some agreed definitions of what constitutes reduced policing (just some neighborhoods?, for what duration?, what type of reduction?, etc.) Then you would want to take some defined period, say one or two years, before and after the reduction had taken effect and monitor all types of crime and their levels (absolute and rate). This hasn't been done yet. No one has agreed on definitions. There were protests in some cities but no official changes in policing policies. Should they be included? There were cities without protests but that did institute reductions. What about them?

Until such rigorous analysis is done, the question will remain unanswered to everyone's satisfaction. The data that is available, however, is very much consistent with those arguing that there is a Ferguson Effect and against those arguing that there is no such thing.

Mac Donald does a good job of marshalling the evidence to address arguments recently advanced that attempted to use newly available data. What I enjoyed about Mac Donald's article was her delineation of the various strategies used to hide the data about the real rise in violent crime in cities with reduced policing. We are still short of the robust analysis that would be desirable, but we are pretty far past the point where the counterarguments are anything but statements of faith based belief.
The campaign to deny the murder and shooting spike in many American cities continues apace. The latest effort is a report by the Brennan Center for Justice, which the press has hailed ecstatically as a refutation of what I and others have dubbed the “Ferguson effect”—the phenomenon of officers backing off of proactive policing and thereby emboldening criminals. In fact, the report confirms the Ferguson effect, while also showing how clueless the media are about crime and policing.

The Brennan Center researchers gathered homicide data from 25 of the nation’s 30 largest cities for the period January 1, 2015, to October 1, 2015. (Not included were San Francisco, Indianapolis, Columbus, El Paso, and Nashville.) The researchers then tried to estimate what 2015’s full-year homicide numbers for those 25 cities would be, based on the extent to which homicides were up from January to October 2015, compared with the similar period in 2014.

The resulting projected increase for homicides in 2015 in those 25 cities is 11 percent. (By point of comparison, the FiveThirtyEight data blog looked at the 60 largest cities and found a 16 percent increase in homicides by September 2015. On Monday, the Brennan Center revised its own estimate of the 2015 murder increase to 14.6 percent.) An 11 percent one-year increase in any crime category is massive; an equivalent decrease in homicides would be greeted with high-fives by politicians and police chiefs. Yet the media have tried to repackage that 11 percent increase as trivial. They employ several strategies for doing so, the most important of which is simply not disclosing the actual figure. An Atlantic article titled “Debunking the Ferguson Effect” reports: “Based on their data, the Brennan Center projects that homicides will rise slightly overall from 2014 to 2015.” A reader could be forgiven for thinking that that “slight” rise in homicides is of the order of, say, 2 to 3 percent. Nothing in the Atlantic write-up disabuses the reader of that error. Vox, declaring the crime increase “bunk,” is similarly discreet about the actual homicide jump, leaving it to the reader’s imagination. Crime & Justice News, published by the John Jay College of Criminal Justice, coyly admits that “murder is up moderately in some places” without disclosing what that “moderate” increase may be.

A second strategy for brushing off the homicide surge is to contextualize it over a long period of time. Because homicides haven’t returned to their early 1990s or early 2000s levels, the current crime increase is insignificant, the Brennan Center and its media supporters assert, echoing an argument that arose immediately after I first documented the Ferguson effect nationally. “Today’s murder rates are still at all-time historic lows,” write the Brennan Center researchers. “In 1990 there were 29.3 murders per 100,000 residents in these cities. In 2000, there were 13.8 murders per 100,000. Now, there are 9.9 murders per 100,000 residents. Averaged across the cities, we find that while Americans in urban areas have experienced more murders this year than last year, they are safer than they were five years ago and much safer than they were 25 years ago.” The Atlantic is similarly reassuring: “The relative uptick [which he never specifies] is still small compared to the massive two-decade drop [in homicides] that preceded it.”

It’s unlikely, however, that the nation would give back in one year a 50 percent crime drop that unfurled over two decades. The relevant question is: What is the current trend? If 2015’s homicide and shooting outbreak continues, those 1990s violent-crime levels will return sooner than anyone would have imagined. Violent crime was down nearly 5 percent in the first half of 2014; the post-Ferguson violent-crime spike in the second half of 2014 wiped out that earlier crime success, leaving 2014 a wash. Since then, crime has continued rising.

The most desperate tactic for discounting the homicide increase is to disaggregate the average. Yes, some cities have seen a homicide increase, the Ferguson-effect deniers argue, but others have seen a homicide decrease. “Fears of ‘a new nationwide crime wave’ are premature at best and wildly misleading at worst,” asserts The Atlantic, because the “numbers make clear that violent crime is up in some major U.S. cities and down in others.” But such variance is inherent in any average. If there weren’t variation across the members of a set, no average would be needed. Any national crime increase or decrease will have counterexamples of the dominant trend within it, yet policymakers and analysts rightly find the average meaningful. The existence of a Ferguson effect does not require that every city experience de-policing and a resulting crime increase. Enough cities are, however—in particular, those with significant black populations and where antipolice agitation has been most strident—to demand attention.

No comments:

Post a Comment