A Critique of Homicide Rates as Commonly Used
This is NOT a political essay. This is a data essay.
I am establishing that firmly before proceeding because I’d like to transcend any lazy, unproductive desires to appeal to coalitions in this newsletter. The world is complex: whatever configuration of moral foundations that one is partial to may be optimal in one situation and not in another. Also, I do not want to alienate any politically-weary readers I might have.
So it is certainly not my intention to offend anyone with what I am about to say. Nevertheless, I am discussing a common misuse of data, as well indirectly addressing some popular-yet-damnable “talking points” that might have won over someone reading this. While I would regret rubbing anyone the wrong way, I have to think that truth—if I’ve stumbled upon that here—is of interest to everyone but the liar. If I am wrong about any of this, please let me know.
Homicide1 rates are not a very useful measure when comparing the severity of deadly crime in different places, and the less alike those places are in terms of population size and area, the worse they get. They are measures better suited for comparing demographically diverse populations, not diverse places. Recent use of homicide rates in public discourse has been outright terrible, involving comparisons of regions of such disparate size and population density that they are almost completely nonsensical. Anyone using homicide rates in this way to make a political point about crime ought to be suspected of ignorance at best, chicanery at worst.
Most people already understand that there is a problem with these comparisons on some level, irrespective of political allegiance. I have lived north of Chicago for decades: conversations about crime in the city are an ordinary occurrence, and among the most commonplace remarks in that regard concern the relative safety of some neighborhoods over others. Surely it is so in every large city that has a problem with violence. If you’ve ever been to (or lived in) one, it is a near certainty that at some point you’ve heard (or said) something like, “Don’t worry: this part of town is safe.” I imagine that that is indeed the case wherever such statements are uttered, which brings us to the first thing to consider about the homicide rate of a particular place, be it city or state or nation: homicides are not evenly distributed. This is so obvious a statement that it almost seems beneath commentary, but I believe that it is too easy to lose sight of this when discussing homicide rates in relation to a large area. Some parts of cities like Chicago are very clearly safer than others, so if a visitor from London wants to go to a Cubs game, no one worries that the city’s overall homicide rate might be about 20 times higher than London’s2. Indeed, if you took the measure of past homicides in the city as an indicator for present risk and used that to estimate the daily distributed risk per individual, the miniscule probability that results would still be a bogus representation of risk in Wrigleyville because there simply isn’t that much deadly crime there.
…for someone living in a dangerous neighborhood that is beset by gun violence, it would be of absolutely no comfort that there might be 999,999 people happily going about their business among the high-rises on the horizon.
Again, almost everyone understands this on some level. A homicide rate isn’t a death lottery that everyone in an area enters. It’s a count of past homicides relative to a given population, bounded by whatever region is examined (something which may be sensible for political purposes and unsuitable for comparative purposes) and which probably includes portions of residents who regularly face greater risk, and others who face far less. Consider that any local homicide rate can be subsumed by expanding the boundaries of the area discussed, and the new, broader figure—while perfectly accurate—would in all likelihood be horribly imprecise if one wanted to understand the situation in relation to any specific part of the whole. Naturally, the boundaries for a homicide rate can even be expanded to encompass entire countries and, again, while that overall rate might be statistically correct, it would be pretty much worthless when discussing any particular place within a nation. It should go without saying, but too often it doesn’t: the level of analysis matters. You lose resolution as the area described increases. When assessing risk, the homicide rate of a place matters less than exactly where most of the violence is happening3.
To look at this from a different perspective than that used in the Wrigleyville example above, for someone living in a dangerous neighborhood that is beset by gun violence, it would be of absolutely no comfort that in the same city there might be 999,999 people happily going about their business among the high-rises on the horizon. Those people are perfectly safe. The person startled by the frequent report of gunshots in the vicinity is not, and the implication there is that with a large, populous city, the homicide rate is actually being lowered by large concentrations of people outside of the danger zones who are relatively at much lower risk from homicide. Therefore, trying to use homicide rate of a city as a measure of risk may in fact understate the problem in dangerous neighborhoods and, inversely, with a sparsely populated yet geographically larger area (like some American states) a lower population density overall would not offset rampant violence which may be occurring somewhere within its borders. This is second thing to consider about homicide rates: population density can inflate or deflate the statistic.
That can be difficult to grasp, so let’s start with the basics. A rate statistic is simply the number of one countable thing in relation to another (like homicides compared to a specific population). If you are comparing two or more rates, each ratio can be expressed relative to the same quantity in the relevant metric, providing a useful reference point which seemingly allows for easier comparison. Without veering off course into an elaborate math lesson, let’s use simple, improbably round numbers to illustrate this: ratios of 60:6,000 and 180:9,000 can be expressed as 1:100 and 2:100, respectively. One percent, two percent.
So if, for instance, a hospital performs a high rate of Cesarean section operations relative to the number of pregnancies they deliver, comparing that rate to the national rate might be a good measure of whether the surgery is over-prescribed at the hospital. (This assumes that the national rate, which functions as an average, is reasonable as opposed to negligent.) Providing the absolute number of C-sections performed at each hospital by itself would not be very useful in making comparisons: perhaps one hospital simply handles more pregnancies than the other, and the absolute number of C-sections alone could give someone the idea that the surgery is performed less or more than the average when that is not the case. So you take the local rate and the national rate, express them in relation to the same practical number of pregnancies (e.g., C-sections per 100 pregnancies—the percent) and the results allow for easier comparison.
An issue too rarely addressed, however, is that the metric which the primary quantity is being measured against in a rate is often not the only relevant consideration (sometimes, as we will see, it’s not even the proper one). There might be several: some which analysis must uncover, and others which must be considered from the outset. If you’ve ever seen “age-adjusted” rates, for instance, you’ve seen another consideration accounted for. (The American Centers for Disease Control typically adjusts mortality rates by age because age is an enormous factor underlying various causes of death. Without that adjustment, rate comparisons of, say, COVID-19 mortality between US states would be misleading when comparing a state with a higher proportion of elderly residents to one with a lower proportion. So the CDC assigns weights to different age groups based on a determined “standard population”, and adjusts their rates so it was as though every state had the same proportion of each group.)
With annual homicide rates, the quantity used for comparison against the number of homicides is the total specified population of a given area, which is usually taken by census at mid-year. Taking the total homicides in that area during the same year and dividing it by the total population produces a per capita statistic (the rate relative to a single person) which is often multiplied by some power of ten to make the data easier to understand (e.g., 4 homicides for every 100,000 people is easier to grasp than 0.00004 homicides per person.) and if the rates of different cities are described in relation to the same quantity of people, this looks like the basis for a reasonable comparison. But it turns out that the size of the locations described, as implied above, is another very important consideration. A comparison of places using two homicide rates might trick people into thinking that they are getting an apples-to-apples comparison when, in fact, such statistics can be very misleading. Consider this: if one place has 4 homicides per 100,000 people and another has 8 homicides per 100,000 people, it intuitively seems that the former place must be safer. After all, 8 is twice as many homicides per 100,000 people as 4. But imagine that the latter homicide rate belonged to a fairly large, sparsely populated state and that 8 constituted the entire count of homicides that occurred there in a year (that is, this imaginary state has a convenient total population of 100,000). Now imagine that former rate represents a geographically smaller, densely populated city that actually has, let’s say, 400,000 people4. The total number of homicides which occurred there was therefore 16 , in a much smaller area. Which place do you suppose is actually safer?
As confusing as this is at first, if you accept that homicide rates are not a death lottery and that proximity to incidences of violence is what actually makes a locale safer or more dangerous, while its population as a whole could be described as less homicidal, the city is clearly more dangerous than the state in the example above. This illustrates how, while we generally associate densely populated urban areas with increased crime and homicides in America, a city’s homicide rate might in fact be lower than a state’s which we might not think of as particularly violent. If you live in that city, there are more homicides in your vicinity; and if you are in the wrong place at the wrong time, it doesn’t mean much that the overall area has more inhabitants. While the additional people do lower the homicide rate, they confer no protective effect on those whose lives actually intersect with violence. On the other hand, if you lived in the sparsely populated state, it doesn’t mean anything if there are fewer inhabitants per square mile if you happen to be in the wrong place at the wrong time either: it just so happens that the chance of being in the wrong place for incidental violence is greatly diminished by the size of the state, and the wrong time could actually occur less frequently (e.g., if we were to use the hypothetical numbers above, being somewhere at the wrong time in the state only occurs at one-half the rate of that in the city over the course of a year). Again, the homicide rate is not a death lottery. Having more people in the geographic area described by a homicide rate statistic doesn’t actually make anyone living there any safer. The density of homicides in an area is more useful than the rate relative to population for assessing danger.
None of this requires any particular statistical sophistication to comprehend or verify. Let’s use one more illustrative example to show how population can distort a homicide rate even when using locations of the same size by imagining two places with roughly the same area, but hugely different populations. Suppose you had a massive residential complex covering 55,000 square feet, and about 2,000 people live there. Now imagine a small rural plot with a single home on it, about 1.3 acres in size. Four people live there. These places are about the same size in area (1 acre = 43,560 square feet) with vastly different population sizes. Imagine also that during the course of one year, a single homicide occurs in both places. Now obviously we wouldn’t report homicide rates for such places, and it doesn’t make much sense to report individual murders as a rate, but for the sake of discussion let’s see what happens when we do. If you took the per capita homicide rate for each area and multiplied it by a multiple of ten to obtain a number comparable to something we’d see with a larger population, the homicide rate of the residential complex is 50 per 100,000; the homicide rate in the rural home is 25,000 per 100,000. Relatively speaking, the 1,999 people who weren’t killed in that residential complex drastically deflate the homicide rate in that case, whereas the few people living on the rural plot drastically inflate the homicide rate there. The larger the population of a place, the less each homicide impacts the homicide rate, so even when using the same area, large differences in population can distort comparisons of homicide rates. And while we have used a highly unconventional example with radically different population sizes here, the effect is very much what happens when places with different population densities are compared, albeit to an extreme degree.
When I stated earlier that everyone understands at some level the problem of apples-to-oranges comparisons of homicide rates, I meant that we all intuitively know and appreciate that the danger represented by the homicide rate of one locale is typically not uniformly distributed within it; just beyond that plain fact lies the understanding that the differences in the portion of the population not proximal to violence affect the statistic. Just as you wouldn’t suppose Chicago’s overall homicide rate meant very much to tourists visiting the Willis Tower, neither you would not either tell someone who resides in a peaceful suburb in Mississippi that it was perfectly safe to venture into the most dangerous part of Chicago because the homicide rate in Illinois is, after all, less than that of their home state. If those two examples make sense to you, trust your intuition. Homicide rates can be slightly useful for establishing an ongoing basis for examining deadly violence in a single given locale over time. If the town or city you live in has a substantial increase in the number of homicides per 10,000 or 100,000 residents in the past few years, that is certainly of concern, and worthy of examination at the level of policy or culture and so forth5. But they are most useful for comparing demographically diverse populations within a locale. (Stratifications of people by age, or income, or sex, or race, etc.) So while it is possible to carefully make inferences about geographically distinct populations when comparing homicide rates, if someone is claiming that ~actually~ a major American city that has a reputation for fatal violence isn’t as dangerous as a much smaller city that does not because, after all, just look at the homicide rates; or, that while certain cities might have problems with violence, comparing states shows where the violence problem really lies, chances are that those are bad comparisons, and the claims, while not incorrect statistically, are sophistry. Comparing the homicide rates of different places is often nothing but a cheap trick, and the person making the comparison might be a fool or is taking you for one. Rate comparisons of crime in different locales are a situation where common sense usually prevails over smart-sounding talking points. If a place has a large number of murders, one is not safer there because many more people happen to live in the area, driving the homicide rate down.
To return to my initial disclaimer, this is not meant to be a persuasive essay in terms of politics. I do not expect anyone to read this and, upon discovering that someone on their team engaged in the manner of glib obfuscation alluded to here, abandon their political allegiances altogether, or even care very much. Political coalitions are broad and no organization operating at a certain scale manages to keep out all the fools and scoundrels.
My goals are here far more modest: that rate comparisons in general (and homicide rates in particular) be considered in a more intelligent manner; that counter-intuitive claims based on such comparisons be treated with greater skepticism; and that all this contributes in some small way to an apolitical movement promoting a more robust standard of honesty.
I might as well start by clarifying that homicide is not exactly the same thing as murder. Murder is the crime of committing an intentional homicide; homicide is a larger category of any killing of another human being, including manslaughter and self-defense. It’s fairly common to see homicide rates referred to interchangeably as murder rates, and this is at minimum imprecise. Given that homicide is the more inclusive term, the discussion here may be presumed to apply to most common discourse regarding murder rate, which is the term that most politicians and commentators seem to prefer (presumably because it sounds more sensational).
While a hypothetical comparison between cities could have served this essay just as well, I did examine Chicago and London’s 2022 homicide rates out of curiosity. Though the data required is fairly straightforward (mid-year population estimates and an accurate count of homicides) I found that even with major metropolises, the quality of reported data is highly inconsistent. Many writers supplying homicide rates for these cities did not provide sources or, when they did, notable details were missing. “Murder” and “homicide” were often used interchangeably, population projections were apparently used in absence of census data without being noted, and certain finer elements (such as whether figures from the separate administrative district of the City of London were excluded from or made no difference to Greater London reporting) were left unexplained. Even “official” figures can be rather sloppy, but it seems that the population of the Windy City was indeed about 20 times more homicidal than that of the rainy one in 2022.
Recent maps of murder or homicide density are a much better tool for someone interested in identifying risk and which violent areas to avoid. I have not heard or seen such maps referred to much in public discourse. (I suspect that is because it’s harder to obfuscate around this kind of data, but I may be cynical about this sort of thing.) If you wanted to compare different places in terms of homicide, you might start with a metric like “homicides per square mile” but as far as I can tell, no such figures are commonly reported. Indeed, an internet search for that term at the time of this writing mainly produced conversations ridiculing the notion, as though assessing risk for crime in a particular place was best determined in relation to how many people live there, rather than how many crimes are actually happening in that space. I have made my problems with this approach as clear as I can in this essay.
The numbers used here are obviously improbable and round, but a number of major US cities do in fact have higher populations than some US states.
This assumes that the population is fairly stable; otherwise, using the absolute count of homicides is preferable, as this prevents the statistic to be skewed by population increases or decreases over time. The homicide rate remains attractive because it permits comparisons between places (which, again, is rarely a good idea).