Monday, November 12, 2012

Is crime rising or falling? It depends which Justice Department source you ask

Doug Berman recently pointed out an incongruous set of federal reports on crime rates. As background, there are two national reporting measures for crime rates: Uniform Crime Reports (UCR) compiled by the FBI from data submitted by local departments, and the National Crime Victims Survey (NCVS) performed by the Bureau of Justice Statistics, which is a rolling sample surveyed over time (they recently altered their methodology). The UCR measures crimes actually reported by police, while the NCVS uses a national polling sample to perform a longitudinal (over time) analysis of crimes experienced by households, including both crimes reported by police and those which were not. (E.g., when crimes go unreported for personal or family reasons.)

In recent years, both the UCR and the NCVS have shown crime consistently declining, despite myriad naysayers who insisted what they reported could not possibly be true. For the last two decades, the two have more or less gone down in tandem with historic crime declines culminating in 40 year lows by both measures, results which seemed unimaginable not long ago. In recent months, however, the FBI's UCR data said reported crime last year had declined. "In 2011, all four of the violent crime offense categories—murder and non-negligent manslaughter, forcible rape, robbery, and aggravated assault—declined nationwide when compared with data from 2010." See Grits coverage of the UCR report from last June.

Meanwhile, the latest NCVS reported somewhat startlingly countervailing trends, declaring their sample of crime victims reported increased crime rates between 2010 and 2011. The NCVS reported, in part:
Between 2010 and 2011, the rate of violent victimization increased 17 percent, from 19.3 to 22.5 victimizations per 1,000 persons age 12 or older. The increase in total violence was due to a 22 percent increase in the number of aggravated and simple assaults. There was no statistically significant change in the number of rapes or sexual assaults and robberies.

While the percentage change in violent crime from 2010 to 2011 is relatively large, the actual difference between the rates for those years (3.3 victimizations per 1,000) is below the average annual change in violent crime (4.3 victimizations per 1,000) over the past two decades. The low rates make the percentage change large, but crime still remains at historically low levels. Since 1993, the rate of violent victimization declined 72 percent.

The rate of total property crime increased 11 percent, from 125.4 to 138.7 victimizations per 1,000 households between 2010 and 2011. Household burglary increased 14 percent, from 25.8 to 29.4 victimizations per 1,000 households.

In 2011, 49 percent of violent victimizations and 37 percent of property victimization were reported to police. From 2010 to 2011, there was no statistically significant change in the percentage of violent victimizations reported to the police. The percentage of property victimizations reported to the police declined from 39 percent in 2010 to 37 percent in 2011. 
How to reconcile these data? With a few caveats, I'm inclined to believe the UCR, for the moment. That's because it relies on actually reported crime as opposed to statistical estimates from a sample. A tainted sample, even if unintended, taints results, while the UCR just totals up the data reported by local departments and comes up with a number.  It's not based on probabilities or statistical estimates, it's a report of concrete episodes reported and investigated by law enforcement.

I know some commenters will say that police departments underreport crime to make their stats look good, and while Grits grants that happens, let me iterate that I don't believe methods of stat manipulation remotely account for the marked crime declines witnessed at the macro-level over the last two decades. Whether such usually localized tactics contribute to the differentials in these reports no one can say for sure, but I'm inclined to suspect other factors. OTOH, who knows what's behind the conflicting data? As Doug Berman declared upon observing the recent trends, "I do not know whether to worry about crime going up or to worry about whether we can be sure if crime is going up or going down." Certainly it's impossible to productively debate the causes of crime trends if we don't have tools to reliably measure them.

Perhaps the NCVS report indicates declining crime rates have reached a floor, or at least a temporary plateau. As the Bureau of Justice Statistics emphasized, "The low rates make the percentage change large, but crime still remains at historically low levels." The point about crime rates being at modern lows brings us to the unwieldy measuring sticks being used to analyze them. Both the UCR and the NCVS are ridiculously broad  measures that naturally fluctuate - to the point that any responsible criminologist would emphasize that a trend cannot be pronounced until it repeats for 2-3 years - so any such conclusion would be premature. But the conflicting, recent data at least raise the possibility that recent downward crime trends may not hold for the foreseeable future. If the NCVS declines next year along with the UCR data, then this may have been a statistical fluke. If the upward trend continues in victimization reports, however, one would expect reported crime to eventually rise as well.

Your guess is as good as mine so let me know what you think in the comments: Is crime declining or rising? At the moment, you can find a respected federal source to support either view.


Unknown said...

As a person that has been through the UCR training and previously responsible for the monthly reports for 2 police agencies, the coding is left up to interpretation. While the guide book gives specific examples of each category, much remains to interpretation. Another thing about the UCR - it is outdated. It was designed in the 1940's and to date, has not been updated. Cyber crimes, for example, is not included. Rape is still defined as male on female with no exceptions for male on male, or female on male. There are many more examples that can be made, making the point that this is nowhere near the perfect solution, but it's the only one we have. When coding, the crimes Grits mentions are categorized as Part 1 crimes - this is what is reported in the media. Part 2 crimes are not reported to media but that's where several "new" crimes are posted. The UCR needs updating, but for now, it's all we have!

Stephanie said...

With respect, I lean toward NCVS as the more accurate picture. Self-reporting has always shown higher rates of crime because people don't report crime for a variety of reasons. (NCVS also asks people why they don't report - an interesting part of the data as well.) NCVS shows increases in assaults and other "minor" crimes which are exactly the crimes people are less likely to report. I agree that incidents of police-skewed data are few and far between so in a sense, both reports are right. Serious crime continues to decrease - the crimes most likely to come to official attention - and some types of less serious crimes show an increase even as people continue their patterns of not reporting those types of crimes. Why people don't report and what that might mean is a fascinating subject in itself.

Anonymous said...

I don't remember these crime mobs a few years ago. This seems to be a new thing and it is now the accepted thing these days where 20-50 run in at one time and clean out a store and nobody knows who they are. There is too many of them (maybe they have us outnumbered) so we don't count them.