When I moved to New York City in 1977, I invented an explanation of the people I had encountered during my several years in Los Angeles. There are all kinds of people in any city, of course, and I was only half-serious. But the explanation went like this: back in the pilgrim and pioneer days, there were all these immigrants landing on the East Coast. Some of them decided to stay and settle down where they were, in New York and Boston and so forth. But others did not fit in so well, in those heavily populated and civilized lands. To them, it was more appealing to move west, to places like Ohio and Tennessee. And then the same thing happened again, out in Ohio and Tennessee: some of the kids grew up well-suited to stay and settle down, but others did not belong, and found themselves obliged to move along, out to the prairie lands. And the cycle repeated itself across the prairie, through Dodge City and across the Wild West, and again from there until, finally, in L.A., we wound up with the people who did not seem to belong anywhere else.
That story comes to mind at present because I find myself engaged in an effort to develop a sense of crime rates in the various states of the U.S. I started this effort with impressions like those conveyed in these two maps:
In other words, the amount of crime in a city or state can be treated as a straightforward matter. We can look at statistics; find the areas reporting the highest rates of crime; notice that those areas tend to be heavily black, or Hispanic, or poor; and conclude that we will be safer if we stick to the wealthier and non-minority neighborhoods. And there may be some truth to this approach. Researchers commonly observe that, in the words of Ulmer et al. (2012), there are “relatively high levels of violent crime among Blacks and Hispanics.”
One need not be hostile to blacks, Hispanics, or poor people to say such things. As Ulmer et al. observe, to the extent that such crime reports are accurate, they may be due, not to some innate cultural or racial flaw, but rather to “the effects of poverty and unemployment, educational inequality, residential segregation, social disorganization, subcultural adaptations to disadvantage, and the legacy of racism and discrimination on behavior.” Ulmer et al. point out, among other things, the impact of a generation of disproportionate (i.e., racist) imprisonment of black men, leaving young people without their fathers’ influence in their upbringing.
The Chaos of Crime Statistics
However interesting or important such discussions may be, they leave out a separate and equally important area of discussion. The concept of “crime statistics” must be taken with a grain of salt. The classic adage in data processing is, Garbage in, garbage out. It is difficult to reach good conclusions with bad data. And in many ways, crime data in the U.S. are flawed. For example, I have personally observed instances in which law enforcement officers have refused to take crime reports, have declined to prosecute criminal activity, and have prosecuted innocent people. The media recurrently cover cases in which rich and/or white people get reduced sentences or no prosecution at all, where a poor or minority individual would have wound up in prison. Shelden et al. (2015, p. 31) argue that, in fact, self-report and victimization surveys present similar crime rates for blacks and whites, and that higher black arrest rates are primarily due to the fact that the police are more likely to notice (and sometimes invent) reasons to stop and arrest blacks.
Shelden et al. (2015, pp. 23-25) point out other problems with official crime statistics. For one thing, most crime is not reported to police. There are also significant differences in how police departments treat and report crime: they vary in their degree of cooperation with FBI data collection personnel, in their concepts of crime (e.g., whether attempted sexual assault qualifies as rape), and in crime data recording (e.g., treating theft of a bicycle, but not child abuse, as a “serious” crime); they deliberately over- or underreport crime for budgetary or political reasons; in cases where one person has committed several crimes in a single incident, police departments differ in whether they report each crime for which that person is charged, or only the most serious of those crimes; and the alleged crime remains on the record even if charges are dropped or the suspect is acquitted. For instance, Yung (2014) notes that police departments in major cities deliberately undercounted rapes over a period of two decades “because of [police department] cultural hostility to rape complaints and to create the illusion of success in fighting violent crime.”
While this post is focused on street crime, Shelden et al. (2015, pp. 31-36) offer the valid observation that corporations and governmental entities commit some of the most damaging crimes, most of which are excluded from crime statistics:
Occupational diseases, death, and injuries . . . [are far more significant than] the crimes shown on television. . . . Violence committed by corporations . . . exposes people to harmful conditions, products, or substances over time. . . . [Some corporate crimes] were allowed to occur with malice aforethought and recklessly endangered employees or consumers. . . . Direct economic losses from all forms of white collar crime are immense and dwarf those of conventional crime. . . . Crimes of the state . . . are generally committed against powerless people, and they are rarely, if ever, called crimes.
In the U.S., discussions of crime do not necessarily deny the foregoing thoughts. The consensus seems to be, rather, that we know there are these other kinds and dimensions of crime, but we feel there is little we can do about them, and we have learned to live with them, whereas we continue to feel that traditionally understood categories of violent and property crime are both threatening and relatively avoidable if we avoid states, cities, and neighborhoods perceived as dangerous. It is, indeed, common sense that there are parts of town and/or times of day where you just don’t walk around or carry a lot of cash.
The FBI’s Uniform Crime Reporting (UCR) program appears to be very widely used despite the deep flaws in the FBI’s UCR data (see e.g., Biderman & Lynch, 2012). Sources including Neighborhood Scout and (apparently) City-Data use FBI crime data to identify neighborhoods that appear relatively safe or unsafe, and innumerable media reports purvey those data as gospel.
Often, those reports oversimplify and distort the reality. For instance, MSN claimed that, in 2014, the ten most dangerous cities in America were (in ascending order) Indianapolis, Stockton CA, Cleveland, Baltimore, Milwaukee, Birmingham AL, St. Louis, Oakland, Memphis, and Detroit. Those cities surely do have dangerous neighborhoods. And yet various suburbs of Indianapolis and Detroit have recurrently ranked among the best places to live in America. During my two years of living near downtown and in northwest Indianapolis, I biked and jogged through a variety of neighborhoods with no problems and hardly any evidence of physical danger. Certainly there was evidence of urban blight. But in a place that statistics supposedly confirmed as one of the worst in the nation, I would have expected a war zone environment more like the places where I had lived, worked, and traveled in and around New York, northeastern New Jersey, and Los Angeles. This was definitely not that.
The FBI also offers the National Incident-Based Reporting System (NIBRS). Unfortunately, NIBRS too depends on data provided by law enforcement agencies, and those data are of dubious quality, hampered once again by “low participation rates among the nation’s police agencies” (Bierie, 2014, p. 1; see Nolan et al., 2015). According to the Bureau of Justice Statistics (BJS, 2015?), NIBRS data cover only about 30% of the U.S. population.
There may be good ways to use FBI UCR and other law enforcement crime data. For instance, while comparisons across different cities or states may entail too many major assumptions and inconsistencies, it could be that a source like Neighborhood Scout does provide a relatively reliable sense of which neighborhoods within a single city are relatively safe. There may also be other sources that would be more useful for some purposes. For instance, poor and minority individuals might look into the quality of police forces in various locations. One study measured citizens’ attitudes toward police departments using Twitter comments. That study found negative attitudes toward police in many states, and positive attitudes in few (most notably, North Dakota, Kansas, Mississippi, West Virginia, and New Hampshire). Of course, such comments may reflect preexisting bias (e.g., pro-police in conservative Kansas, anti-police in liberal Massachusetts) as distinct from actual performance. The same is true on the local level (e.g., highly pro-police in San Diego and anti-police in Los Angeles).
The National Crime Victimization Survey (NCVS)
Rather than rely on data assembled and provided by law enforcement agencies, Shelden et al. (2015, pp. 25-26) favor the National Crime Victimization Survey (NCVS) (see also JRank). Unlike the sparse reporting that plagues the FBI data, BJS (2015?) indicates that 87% of eligible persons responded to the most recent NCVS survey. Moreover, these survey data are not filtered by law enforcement: they include information on alleged crimes not reported to police.
According to Shelden et al. (2015, p. 27), problems in NCVS survey data tend to arise from the decisions and situations of the individuals surveyed, as distinct from decisions made by police departments or politicians. NCVS is believed to understate the level of crime somewhat. Among other things, some people forget crimes that have occurred, or whether they occurred within the timeframe being asked about, or they decline to report certain (e.g., domestic) crimes. There are also surveying problems, such as the challenge of identifying and contacting runaways, frequent movers, homeless people, and others likely to be victimized. These sorts of problems appear generally unimportant for present purposes, however: they are likely to occur to some degree everywhere, and thus are not likely to significantly distort comparisons among cities and states.
The Criminal Victimization series is the major annual BJS publication summarizing NCVS data (NCBI, 2014). Unfortunately, the PDF report and the CSV file included in Criminal Victimization 2014 (the most recent year available at this writing) offer no comprehensive geographical breakdown of crimes. This is because NCVS is a survey, not a census of every person within any given area. Due to federal budget cutbacks, in most subnational units (i.e., states, counties, cities), NCVS does not survey enough individuals to provide a statistically reliable estimate by itself.
In a separate BJS project, however, Fay and Diallo (2015, pp. 6-8, 20-23) compared trends in NCVS data within subnational units (states, especially), to estimate where and how much crime appears to be occurring. Fay and Diallo considered this method most reliable for the 36 states whose populations were at least 2 million. For those states, they estimated average violent and property crime rates per 1,000 residents per year during the years 1999 to 2011. As far as I could tell, this was the most reliable available source of data on crime in those states.
For the sake of completeness, of course, I wanted information for all 50 states, even though the least populous states would be relatively unimportant for most purposes. So I tried to develop a quick and dirty way of filling in the blanks for the 14 states on which Fay and Diallo were silent. My approach was to use the averages of crime rates in neighboring states. That approach required me to abandon Alaska and Hawaii, since they had no neighbors. But for the 12 remaining states (i.e., 50 minus 2 minus 36), I used the approach exemplified in the case of Delaware: I averaged the average crime rates from the neighboring states with which Delaware shared relatively long borders (i.e., New Jersey and Maryland, but not Pennsylvania). This approach was problematic in the northern Plains and Rockies, where data were unavailable for a number of neighboring states. In those states, I tried building up averages, moving from south to north, starting with Nebraska (calculated as the averages of the estimates for Iowa and Kansas) and Idaho (using data for Washington, Oregon, Nevada, and Utah). As discussed below, this seemed to work in some states but not others.
Here, then, are the crime data that I found in the NCVS (click to enlarge):
In that table, states are listed in the order shown in the rightmost column (Rank Comparison: Average). That column provides the simple average of each state’s ranks in the areas of violent and property crime. For example, Florida ranked 15th among states in violent crime, with an average of 32.5 violent crimes per 1,000 residents per year, and Florida ranked 13th among states in property crime, with an average of 155 property crimes per 1,000 residents per year. As shown in the Difference column, those two values differed by two places (i.e., 15th minus 13th). For most states, the differences between violent and property crime rankings were not large. For Florida, those values, combined, produce an average rank of 14th among the 48 states in the contiguous U.S.
With those crime rates, Florida had 43% more violent crime (shown in the table as 143%) than in best-ranked Georgia, and 30% more property crime than in best-ranked New York. It may seem preposterous that New York would be best-ranked in anything, where crime is concerned. That reaction leads to an interesting discussion (below). For now, the key point is that NCVS, presenting data taken directly from survey participants, can be expected to vary significantly, in some jurisdictions, from the prevailing wisdom or the official story. Conclusions for some states can be expected to change as data collection continues and as NCVS improves (assuming improved funding). But it is presently not obvious that the state-by-state calculations offered by Fay and Diallo contain any significant errors.
As just indicated, the table’s ranking of states uses a simple average of their violent and property crime ranks (15th and 13th, in the case of Florida). But the table’s P/V column shows that, in most states, there were about five times as many property crimes as violent crimes per capita. Giving equal weight to the violent crime rank and the property crime rank thus implies that a single violent crime is about as important as five property crimes. Such a judgment may not be true. A person who felt that violent crimes were very much more serious than property crimes might give more weight to the violent crime ranking. But that adjustment would probably not cause significant changes in the overall ranking of most states because, again, as shown in the Difference column, most states’ violent crime and property ranks are not extremely different from one another.
Reviewing the NCVS Data
I decided to compare the state crime rates estimated by Fay and Diallo (2015) (above) against those produced by the FBI’s UCR data (2014). The following table provides that comparison. First, its NCVS data (as estimated by Fay and Diallo) are restated as crimes per 100,000 (rather than 1,000) residents. Then the NCVS data are compared against FBI UCR data.
The Ratio columns of that table (summarized in the bottom lines) indicate that, on average, the NCVS survey identified about ten times as many violent crimes per capita, and about seven times as many property crimes, as are reported in the FBI UCR data. These ratios suggest that the FBI UCR data fail to capture the large majority of crimes.
(For some of the 12 states not estimated by Fay and Diallo, my attempt to build averages using values from neighboring states (above) resulted in substantial divergence from the FBI UCR data. This table excludes those states — specifically, Maine, Vermont, New Hampshire, Idaho, and Wyoming — for which my method was apparently not very good. Since almost all of those states border Canada, the method might have been improved if I had included Canada’s averages in the calculation.)
The Rank Diff. column of the table shows the amount by which the two rankings differ. For instance, using the NCVS approach, the average of violent and property crime rates in Washington state put it in 46th place, making it one of the most dangerous states. But in the FBI UCR data, Washington ranked 33rd. That difference of 13 places is reflected in the number 13 shown for Washington in the Rank Diff. column.
I was particularly interested in the states with relatively small Rank Diff. values. The table is sorted to list those states at the top. Those small Rank Diff. values indicated that, for those states, the NCVS and FBI UCR approaches arrived at similar conclusions. For instance, both approaches ranked Pennsylvania 15th (resulting in a Rank Diff. value of zero), and they were also in very close agreement on the relative ranking of Virginia (5th and 6th, respectively). In the table, the names and Rank Diff. values are shaded, for the 19 states having a Rank Diff. of 8 or less. There seems to be relatively little disagreement about the levels of crime in these states.
In the table, those 19 shaded states are sorted according to NCVS rank. That sorting shows the extremes in NCVS data. In violent crimes among the shaded states, we find New Jersey, with 2,350 crimes per 100,000 according to the NCVS data, versus Nevada, with 4,660. Similarly, in property crimes, the extremes are New Jersey (12,400) and Arizona (26,100). (Here, again, we have the unexpected ascendance of a place like New Jersey, to be discussed shortly.)
The geographical locations of those 18 states are interesting. Note, first, that the table contains a horizontal line, separating the top nine shaded states (with relatively low crime, according to NCVS) from the bottom nine (with relatively high crime), excluding Maryland as an apparently unusual case. Seven of the top nine are eastern states lying between Virginia and New England. By contrast, of the bottom nine shaded states, all (aside from Maryland and the guesstimated Nebraska) are either (a) the Rust Belt trio of Illinois, Indiana, and Michigan or (b) located at least partly within the south-central and southwestern regions of the country. The latter is true especially at the bottom of the list, where Texas, Nevada, and Arizona offer very poor crime rates, regardless of whether one uses FBI UCR or NCVS data.
In short, the NCVS and FBI UCR data sources appear to agree that there tends to be a substantial difference in crime rates between the East and the Southwest. This observation harks back to the story told at the outset, about relatively flighty people moving across the continent in pioneer times. It really could be that the laws, mentalities, and peoples of the two regions were dichotomized in a grand historical filtering process.
Or one might focus on population density. The foregoing table offers a column presenting each state’s average population density, in terms of persons per square mile. As noted in the bottom lines of the table, the low-crime shaded states have more than three times the population density of the high-crime shaded states.
That is certainly not what one might expect. The common assumption is that there would be less crime where there is more space. But there are reasons to question that assumption. Densely populated and/or older states might in fact be more protected against crime: they might tend to have denser and more established systems of governance, networks of family and neighborhood, and expectations of civil behavior.
In this interpretation, the Wild West was created by those who were inclined toward it, and might not have been created if those who left the East Coast and those who stayed behind had swapped places. Expressed in different terms, and based on personal experience in both places, it is conceivable that the image of a dangerous New York City overlooks the fact that millions of people have been socialized, over a period of many generations, to live together in relative peace, where millions of Texans crammed into that same place might lack the social skills and experience necessary for coexistence. As the following map suggests, it is at least an interesting hypothesis: it does appear that less densely populated areas may contain a disproportionately high percentage of individuals who are not highly adapted to the situations in which they find themselves.
The rows of data at the bottom of the lists of states in the foregoing table present another quandary. In the FBI UCR data, the Deep South is even more dangerous than the Southwest: South Carolina, Florida, Georgia, Alabama, Arkansas, Louisiana, and Tennessee all rank among the most dangerous states in the nation.Those states are listed near the bottom of the table because of their large Rank Diff. values. That is, for these states, the NCVS calculations differ sharply from the FBI UCR data. NCVS indicates that, in fact, South Carolina, Florida, Georgia, and Alabama are among the 15 safest states, and that there are substantial (although not as extreme) discrepancies for the others as well.
So which source is right? There are limits to how much the crime statistics in one state can say about those in another, even when the other is its neighbor. But it does seem reasonable to ask how Tennessee can rank 42nd in crime, in the FBI UCR data, while its geographical twin Kentucky ranks 12th. The NCVS data present the more plausible claim that Tennessee actually ranks 29th and Kentucky 27th. It is likewise believable that, if Virginia ranks 6th in the FBI UCR data and 5th in NCVS data, North Carolina would rank 7th (per NCVS), not 28th (FBI UCR) — in which case one might ask, again, whether South Carolina deserves to be tied for 44th place.
As noted above, the FBI UCR data are vulnerable to exaggeration and distortion for budgetary and political reasons. This becomes interesting in the case of Utah, another state near the bottom of the foregoing table — for which, that is, FBI UCR and NCVS data differ considerably. The discrepancy between FBI UCR and NCVS is particularly noteworthy in the area of violent crime in Utah: the NCVS estimate (4,050 violent crimes per year on average) is 18.8 times the FBI UCR number (216 violent crimes per 100,000 per year). Assuming Utah residents are not extraordinarily inclined to invent nonexistent reports of firsthand encounters with criminal behavior, there is a question as to whether Utah state and/or local authorities would suppress undesirable reports of violent crime. The question seems fair, given that no other state has such an extreme discrepancy between NCVS and FBI UCR. Few others even come close. FBI UCR data report significantly higher rates of violent crime in Utah’s comparably populous neighbors to the east, west, and south, supporting the impression that FBI UCR understates Utah’s violent crime rate. As further support, Merrill (2013, p. 22) cites law enforcement agents who estimated that, due to insufficient staffing, the Utah justice system might be identifying as little as 2% of human trafficking cases in that state. In addition, Goff et al. (2013, p. 256) found that Utah’s passage of a law requiring law enforcement personnel to target undocumented immigrants for deportation had reduced citizen willingness to report crimes to police.
There may be states in which (or among which) the seemingly gross underreporting of crime in the FBI UCR system is nonetheless proportional, capturing at least the differences in crime levels from city to suburb and/or from one city to another. The problem is that it is difficult to know which states those might be. In other words, it is not just that the FBI UCR system is surely erroneous in some states; it is that the unknowns surrounding its errors make it useless for some purposes.
For purposes of figuring out where and why street (as distinct from white-collar, corporate, and/or governmental) crime occurs, there may be some merit in using FBI UCR data as an auxiliary source, as Fay and Diallo (2015) do, while treating the National Crime Victim Survey (NCVS) as a more viable (although regrettably underfunded and thus limited) guide. NCVS data tentatively seem to support the hypothesis that street crime will tend to be higher in states where the behaviors of individuals are less restrained by sociohistorical factors. Such factors may include long histories of settlement, deep and broad ties with family and community, and multigenerational training in the art of coexistence in densely populated areas featuring high levels of interpersonal interaction.
It is not clear that the older places would therefore be better places to live, or that they would enjoy superior levels of crime overall. It could be, for example, that the factors discouraging street crime might simultaneously favor organized crime and governmental corruption. But as long as the discussion is limited to street crime, the information presented here could favor policies that would prioritize the strengthening of communities rather than the spending of money on professional law enforcement services.