Interpreting the 2018 Statistics on Violent Crime

“The longstanding general trend of declining violent crime in the United States, which began in the 1990s, has reversed direction in recent years.”

This is the first sentence from the Bureau of Justice Statistics’s annual report on crime statistics from the National Crime Victimization Survey (NCVS). The report did not attract much media attention when released in September 2019 — a one-sentence mention in the last paragraph of a story on page A14 of The New York Times, and no mention at all in other major newspapers.

But this should be big news. One of the major stories in crime statistics has been the dramatic drop in violent crime rates since the early 1990s, a trend that has been found in all of the major sources of information about crime: homicide statistics, the FBI statistics of crimes known to police agencies, and the NCVS.* Now Criminal Victimization, 2018 says that violent crime is increasing again. What types of violent crime increased and what features of the survey might affect the statistics?

Let’s look at some graphs. Figure 1** shows the NCVS estimates of the number of victimizations per 1,000 persons age 12 and older from 1993 to 2018. The graph on the left shows the time trends for the “serious violent crimes” of rape and sexual assault, robbery, and aggravated assault (assault involving a weapon or serious injury); the graph on the right shows the time trends for simple assault (any assault that is not aggravated). Simple assaults account for the majority of violent victimizations every year — in a typical year, about two-thirds of the violent victimizations are simple assaults — and simple assaults are less likely to be reported to the police than serious violent crimes. Simple assaults not reported to the police accounted for nearly half (about 45 percent) of the estimated increase in violent victimizations from 2015 to 2018. Serious violent crimes also increased from 2015 to 2018, according to the survey, but not to the same extent.

Figure 1. Victimization rates for serious violent crimes (left) and simple assault (right) from NCVS. The vertical lines represent the 95 percent confidence intervals for each estimate.

Note that 2015 marked the NCVS’s lowest-ever violent crime victimization rate since the survey started, so a comparison of any year with 2015 is taking the smallest ever-observed violent victimization rate as a baseline. It is possible that the 2015 statistics were unusually low, since some of the survey methods changed in 2016. Interviewers for the new sample in 2016 were trained more recently, and research by the Census Bureau showed that interviewers who had received refresher training tended to elicit more victimization reports, particularly for crimes not reported to police, from survey participants.

And, as with other surveys, the response rate for the NCVS has been dropping dramatically. In the early 1990s, almost everyone who was asked to participate in the survey talked to the interviewer. But by 2018 only 73 percent of the households that were randomly selected to participate did so. And even when one member of a household participates, other persons in the household might refuse: only 82 percent of the persons age 12 and older in participating households took the survey. By multiplying these two percentages, we can estimate that in 2018 only about 60 percent of the persons who were randomly selected to be in the survey provided data. The NCVS uses a technique called weighting* to account for the nonresponse, by increasing the weights of persons who respond to the survey in a demographic group to compensate for the nonrespondents in that group. But it’s possible that the nonrespondents in a group might be similar to the respondents in the group with respect to age, race, sex, urban/rural status, household size, and other demographic characteristics — and still have had systematically different victimization experiences. In 2016 earlier years, the persons who did not respond to the survey were more likely to be young, male, or Hispanic, but the Bureau of Justice Statistics has not yet published its analysis of nonresponse patterns for 2017 and 2018.***

No data set is ever perfect, though, and it often helps to look at other data to see if the same trend shows up in multiple places. We don’t have other reliable sources of information about crimes that are not reported to the police (the NCVS, despite its shortcomings, is the best source we have), but we do have another source of information about crimes known to the police.* Figure 2 shows rates of non-homicide violent crime (rape, robbery, and aggravated assault) and homicide from the FBI’s Uniform Crime Reports. These are tabulated from statistics submitted by law enforcement agencies, and thus contain only crimes that have been reported to the police.

Figure 2. Rates of all violent crime (left) and homicide alone (right) from the FBI reports Crime in the United States. These statistics include only crimes that are known to law enforcement agencies.

The FBI uses a different scale, counting crimes per 100,000 persons, including children (as opposed to victimizations per 1,000 persons aged 12 and over from the NCVS). The FBI statistics include different types of violent crime (they exclude simple assault) and use different definitions for the crimes that are included. We would thus not expect exact correspondence with the NCVS statistics of serious violent crimes reported to the police. But we would expect to see roughly the same overall pattern for the FBI violent crime graph on the left in Figure 2 as for the bottom (blue) line in the NCVS serious violent crime graph on the left in Figure 1, and we do. The FBI statistics, like the NCVS statistics, show the general decrease in crime since the early 1990s — since 2010, the FBI rates of homicide, and violent crime as a whole, have been consistently about half of the 1990-1994 rates.

But we don’t see an uptick for 2018 in the FBI statistics. That may be because most of the NCVS increase in 2018 is from violent crimes not reported to the police, which of course would not appear in the FBI statistics. The 2019 statistics should provide information on whether the NCVS increase is a continuing trend, or just a one-year blip in the survey estimates. In the meantime, shouldn’t the 2018 crime statistics be featured in more news reports?

Footnotes

*You can read about the advantages and limitations of different crime data sources, and about how weighting works, in my book Measuring Crime: Behind the Statistics.

**The statistics were downloaded from the NCVS Victimization Analysis Tool on November 4, 2019. The Bureau of Justice Statistics says that the statistics from 2006 are unreliable because the survey design and the methods used to collect data that year changed. Similar problems occurred for 2016, but they did not appear to affect the violent crime rates as much as in 2006; I used the revised estimates for 2016 in the graph but the originally calculated estimates for violent crime were very close to the revised ones.

Note that the statistics in FIgure 1 give victimization rates — that is, the number of violent victimizations that occurred to persons during the year, divided by the population (in thousands). This differs from the percentage of people who experienced at least one violent crime during the year, since people can experience multiple crimes in a year.

***In previous years, the public-use data sets have been released at about the same time as, or shortly after, the publication of the annual report on Criminal Victimization. For example, the data file for the 2016 statistics was released in December 2017, the same month that the Bureau of Justice Statistics published Criminal Victimization, 2016. But the data sets for 2017 and 2018 have not yet been released, so I cannot explore the weighting patterns or other factors that might give insight into potential nonresponse bias.